Upload
pavel-iosad
View
1.049
Download
2
Embed Size (px)
DESCRIPTION
Presented at the Toronto-Tromsø Phonology Worksrshop, University of Toronto, Canada, October 2009
Citation preview
The problemAnalysis
Implications
Laryngeal phonology in Plougrescant Bretonsandhi mutation and contrast
Pavel IosadUniversitetet i TromsoslashCASTL
paveliosaduitno
TorontoslashndashTromsoslash Phonoloslashgy WorkshopOctober 9ndash11 2009
University of Toronto
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect
2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature
3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)
4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]
5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do
6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect
2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature
3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)
4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]
5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do
6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature
3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)
4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]
5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do
6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)
4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]
5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do
6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]
5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do
6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do
6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Talk outline
1 Laryngeal phonology in a Breton dialect2 Final devoicing is loss of contrast not loss of feature3 Sandhi voicing is phonetic implementation (mostly)4 Devoicing sandhi do not need [minusvoice]5 Privative laryngeal features will do6 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Background
Breton a Celtic language closely related to Cornish and Welsh
Mostly described by Celtologists dialectologists and historicallinguists
Breton phonology remains seriously understudied (as opposedto syntax)
Few proper phonetic studies mostly aural transcriptions
What can we do
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Previous work
Kraumlmer (2000)
Icircle de Groix Breton (Ternes 1970)
Argued to exhibit a ternary contrast between [+voice][minusvoice] and [0voice] segments
Evidence for binary features
Final devoicing is loss of features
Hall (2008)
Same dialect same source
Privative features with feature geometry
Feature disalignment
Final devoicing is loss of features and loss of contrast
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
The present approach
Work in progress (almost) nothing is Vnal
Features are privative with feature geometry
ldquoFinal devoicingrdquo is loss of contrastDevoicing sandhi is
Either lexical phonologyOr failed mutation due to geminate inalterability
Argument for substance-free phonology
Tested on Plougrescant Breton (Jackson 1960)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Breton dialects
Traditionally divided into four groupsCornouaillais Treacutegorrois Leacuteonais (KLT) relativelyhomogeneous basis for standard languageVannetais (south-east) very divergent sometime even served byown literary tradition (Guillevic amp Le GoU 1902)
Icircle de Groix is a Vannetais dialect
Source rather messy (ldquophonemicrdquo approach not verysystematic)
Here attempt to look at a less messy data point
Plougrescant is a Treacutegorrois dialect description by Jackson(1960) more systematic
Further outlook extend approach to Icircle de Groix if possible
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Consonant inventory
Place
Manner Labial Alveolar Postalveolar Palatal Velar Uvular Laryngeal
Stop p b t d c eacute k gFricative f v s z S Z X hNasal m n ntildeLateral l LRhotic rGlide w j
Length contrast for all consonants except voiced obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Vowel inventory
E()bull
ebullebull
ubull
YbullIbull
ybullibull
œ()bull
O()bull
o()bull
amacrbullabull
Length is only licensed by (main) stress
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Voiced and voiceless obstruents contrast word-initially shortallophones
(1) a [pesk] lsquoVshrsquob [bœrE] lsquomorningrsquoc [logOt] lsquomicersquo
Voiced and voiceless obstruents contrast immediately followingunstressed vowels short allophones
(2) a [bOto] lsquoshoesrsquob [SadEnt] lsquochained (participle)rsquoc [kYryno] lsquopeals of thunderrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Following long stressed vowels consonants can only be shortvoiceless obstruents do not occur
(3) a [ober] lsquoto do to make to workrsquob [lizr] lsquoletterrsquoc [meln] lsquoyellowrsquo
Following short stressed vowels consonants are long voicedobstruents cannot be long so they are excluded
(4) a [taput] lsquoto takersquob [jaXOX] lsquomore healthyrsquoc [skYdElo] lsquobasinsrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Restrictions on laryngeal features
Word-Vnally following a stressed vowel voiced obstruents arenot permitted Consonants are short following long stressedvowels and long following short stressed vowels
(5) a [tok] lsquohatrsquob [mel] lsquohoneyrsquo
(6) a [grwEk] lsquowoman wifersquob [mEl] lsquoballrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Summary
Leaving Vnal devoicing aside for a moment laryngeal featuresare mostly predictable
Laryngeal contrasts are allowed in the onset of the Vrst syllableand of the stressed syllableOtherwise they are predictable
Voiced following unstressed (always short) vowelsVoiced when single and following long stressed vowelsVoiceless (and long) when single and following short stressedvowels
What is contrastive What is marked
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing
At Vrst blush Vnal devoicing looks normal
(7) a [bYgalEeacutejo] lsquochildrenrsquob [bYga
macrlIc] lsquochildrsquo
But what about vowel length
This is a good question
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
The really interesting part is when a stressed vowel precedes
Stress is normally penultimate in KLT (but not in Vannetais)so this is mostly monosyllables and a few words with Vnalstress
If it is vowel length that is distinctive we expect VC
(8) a [togo] lsquohatsrsquob [tok] lsquohatrsquo
And cf minimal pairs like
(9) a [kas] lsquosendrsquo ([s] never voiced Frenchborrowing)
b [kamacrs] lsquocatrsquo (cf orthographic kaz)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
This isnrsquot really devoicing in view of what we know aboutquantity and voicing
This is incomplete neutralization
Confer real devoicing
(10) a [lOgodn] lsquomousersquob [lOgOta] lsquoto hunt micersquo
Side note it isnrsquot always about voicing per se
(11) a [rOhis] lsquopeople of ar Rocrsquohrsquob [rOX] lsquoar Rocrsquoh (placename)rsquo
Not really surprising if you know (some) [h] is historically Gbut must be accounted for
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Final devoicing in monosyllables
Does real Vnal devoicing happen Well yes
There is variation described by Jackson (1960) as ldquofreerdquo andespecially with coronals
Context probably unknowable the ambition here is at best toVnd which representations are involved
(12) [tyt]sim[tYt] lsquopeoplersquo (orthographic tud)
More examples to come immediately below as they involvesandhi to which we now turn
What about lexically voiceless Vnals These are relatively fewFrench borrowings of various antiquity and behave asexpected cf (9-a)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
The traditional view (Stephens 1993 Favereau 2001) isessentially that all consonants are voiced in sandhi before[+voice] segments
(13) a [pwelz aO] lsquoif you saw mersquob [shyma
macrb newe] lsquonew sonrsquo
c [shypOb bin] lsquolittle youthrsquo
And voiceless before voiceless consonants
(14) a [shymamacrp hir] lsquotall sonrsquo
b [n shydyt kapap] lsquothe able peoplersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
Plus there is the devoicing sandhi that is the focus of Kraumlmer(2000) and Hall (2008)
For Icircle de Groix Ternes (1970) describes it as a lexicaldistribution some words and only these words devoice initialobstruents following an obstruent
For Plougrescant Jackson (1960) is less concerned ldquosometimesrdquo
(15) a [lamacrt tı] lsquosaid to mersquo cf [dı] lsquoto mersquo
b [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
MotivationBasic dataFinal devoicing and sandhi
Sandhi
In the narrative texts given by Jackson (1960) the sandhi rulesare often violatedEspecially with regard to sandhi voicing
(16) a [shymamacrp dy] lsquoblack sonrsquo
b [shymErX vamacrt] lsquogood girlrsquo
c [dwan tœs diwı]lsquothe fear that you have of mersquo
Jackson (1960) explains the the texts were dictated at a slowpaceHowever some (in fact most) of the examples such as (16-a)and (16-b) are transcribed with a secondaryndashmain stressrhythm these are possibly genuine connected phrasesThus failure of sandhi is not necessarily an artefact of dictationNote that vowels outside main-stressed syllables are shortenedso the preservation of length contrasts under devoicing doesnot work in the same way when stress is secondary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Outline of analysis
Outline feature analysis
Argue that Vnal devoicing without length permutations is aphonetic process
Argue that sandhi voicing is the Wip side of Vnal devoicing
Unify some devoicing sandhi with ldquofailure of mutationrdquo
Tentatively propose that other devoicing sandhi are an artifactof univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
Before we even discuss Vnal devoicing we should solve the[voice][spread glottis] problemPhonetics rather poorly understoodVoiceless stops are described as aspirated (at least initially) atLe Bourg Blanc (Falcrsquohun 1951) and Saint-Pol-de-Leacuteon(Sommerfelt 1978) but these are both LeacuteonaisNo mention of aspiration is made for Plougrescant by Jackson(1960 1967)In all cases the voiced stops are described or assumed to bevoicedOne possible point at Plougrescant fricatives underwent acontext-free voicing (ldquonew lenitionrdquo) cf Southern EnglishFricative Voicing which Honeybone (2005a) takes as evidencefor [spread glottis]emptyBut Honeybone (2005a) himself admits the analysis of fricativesshould not be spread to stops uncritically
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Feature analysis
In substance-free phonology with emergent privative featuresthis point is rather mootWe are interested in the patterning whether the ldquovoicelessrdquoobstruents are labelled [spread glottis] or [voiceless] (cf Blaho2008) is irrelevantOr voiced stops are [voice] or [stiU] of courseI propose that in Plougrescant Breton ldquovoiceless stopsrdquo are[voiceless] and ldquovoiced stopsrdquo do not bear a laryngeal featurebut do have a laryngeal nodeI return below to why nodes are better than featuresMain reason is restricted distribution only initial and stressedsyllables both reasonable contexts for positional faithfulness(Beckman 1999 Smith 2002)We need to make reference to this feature to derive therestrictions (but not to describe Vnal devoicing as I argue below)In that sense it is ldquomarkedrdquo (Trubetzkoyan markedness)Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing
I propose that Vnal devoicing is in fact loss of the laryngealnode i e it is the exclusion of the very possibility ofcontrasting for laryngeal features
Devoiced stops are a third phonological category they behavediUerently from true voiceless stops in that they do not obeylength-related restrictions
True voiceless stops cannot follow long vowels devoiced stopscan
In particular what is the diUerence between Vnal devoicing asin [tyt] and Vnal devoicing with gemination as in [tYt]
No tableaux in analysis (but hopefully it is prettytheory-independent)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Assumptions of analysis
Vowel length distinctive in main-stressed syllablesfaithfulness markedness in this context
[voiceless] above Max[vcl]
Except for positional faithfulness Max[vcl]Initial andMax[vcl]σ above [vcl]
Bimoraic template for main-stress syllable (Main-to-Weight)McGarrity (2003) Bye amp de Lacy (2008)
Final devoicing driven by a constraint Lar_]Wd militatingagainst any segments with a laryngeal node at the end of a(morphological) Word
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
Obstruents are long and voiceless following short stressedvowels
σ
a p u tt
σ
Lar
[vcl]
The voiceless obstruent piggybacks on Main-to-Weight to beparsed into the stressed syllable and thus keep [vcl]This is assuming (as I do) that faithfulness to vowel length isundominated
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b u tt
σ
Lar
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
Assuming richness of the base what happens with voicedobstruents after short vowels
σ
a b˚
u tt
σ
Lar=
Assume a constraint Larmicro geminates without laryngealspeciVcations exist in the language (geminate sonorants)
This is of course outranked by positional faithfulness to [vcl] toderive the previous case
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vd
The obstruent loses its laryngeal speciVcation in order tobecome moraic for the beneVt of Main-to-Weight
Laryngeally unspeciVed obstruent geminates are realized asvoiceless for obvious phonetic reasons
Maybe these are excluded by Lexicon Optimization since thelearner never really has to posit b
˚
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o k O tl
σ
Lar
[vcl]
micro micro
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Medial obstruents Vt
This is a simple case
σ
o g O tl
σ
Lar
[vcl]
micro micro
=
No superheavy syllables so [vcl] cannot be saved
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o gt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
No Lar node word-VnallyFinal consonant is extrametrical (so maybe no Lar node notlicensed by prosodic structure)
Stress ultimate if V in Vnal syllable else penultimate Moraictrochee but then Vnal VC must be L
σ
o g˚
t
Lar
micro micro
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing voiced stops
Laryngeally unspeciVed obstruents in pausa are realized asvoiceless phonetic reasons are well-known
What if our [vcl] is really [spread glottis] in this dialect
It is apparently unproblematic to have aspiration as thephonetically natural realization of phonologicalunderspeciVcation (Vaux amp Samuels 2005)
What about cases such as [tyt]sim[tYt]
I propose this is real Vnal devoicing i e the imposition of the[vcl] feature at word (phrase) edges (Iverson amp Salmons 2007)
First letrsquos look at underlying voiceless obstruents
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final voiceless stops
The [vcl] obstruent becomes moraic to satisfyMain-to-Weight so the restrictions on vocalic quantity hold
σ
a sk
Lar
[vcl]
Wd
micro micro
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y dt
Lar
micro micro
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
True Vnal devoicing
In this scenario forms such as [tYt] for tyd imply that theconstraint driving Vnal devoicing is ranked over faithfulness forvowel length
σ
y tt
Lar
[vcl]
micro micro
=
=
Wd
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Final devoicing summary
I have argued that what looks like normal Vnal devoicing is infact the deletion of a Lar node or absence of contrast
Further evidence Vnal v does not always neutralize with fphonetically Jackson (1960) writes [v
˚]
We know [v] is aerodynamically complicated (Padgett toappear)
So this would be consistent with a phonologicallyunderspeciVed v
˚
Final devoicing as Vnal fortition (Iverson amp Salmons 2007) isdistinct from this process and also attested
Grazing other dialects Vnal devoicing is optional atSaint-Pol-de-Leacuteon (Sommerfelt 1978) ()
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing sandhi
In this system voicing sandhi arise from two sources
Before sonorants laryngeally unmarked stops are voiced in thephonetics
Sonorants do not contrast for laryngeal features so they do nothave a [Lar] to spread
Explains variability (pause-sensitivity)
No need to have (contrastive) laryngeal features for sonorants(Kraumlmer 2000 Blaho 2008 Hall 2008)
[shymamacrb newe] = mab
˚newe
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Voicing assimilation sandhi
Before obstruents we are faced with two optionsSame as above
Explains possible devoicing even before voiced obstruentsPossibly predicts that under certain phonetic circumstances Vnalconsonants may be voiced before voiceless consonants
Spread of Lar with [vcl] if need beVariation must have a phonological explanation (stochasticranking)Devoicing sandhi crucial piece of evidence in favour
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Some examples of devoicing sandhi
(17) a [lamacrt tı] lsquosaid to mersquo
b [me gaf tı] lsquoI Vnd I considerrsquo (lit lsquoI get tomersquo)
c [shydO wenk tit] lsquoyour two sousrsquo (lit lsquotwo sous toyoursquo)
Prepositions are overrepresented
Actually this is also true of Icircle de Groix
(18) [tra nvaNk temp]lsquowe donrsquot miss anythingrsquo (litlsquonothing is missing to usrsquo)
Whatrsquos with the prepositions
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 1 mutation
Breton is (widely) known for its initial consonant mutation
Here we are only interested in lenition
Underlying p t k b d g m
Mutated b d g v z h v
The interesting bit is the voicing of voiceless stops
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Historically prepositions in Brythonic have tended to undergothe eUects of soft mutationlenition in a context-free way
Old Welsh and Old Breton gurth lsquothroughrsquo Modern Welsh wrthModern Breton ouzh
Old Welsh di lsquotorsquo Modern Welsh i (via [Di])
Modern Welsh variation trwysim drwy lsquothroughrsquo
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Detour 2 prepositions
Why is this important
At least in Welsh there is evidence that the new initialconsonant is not fully lexicalized
In particular gan lsquowithrsquo is historically kant
The conjunction a lsquoandrsquo causes a mutation whereby voicelessstops are spirantized to [f T X] but voiced ones are unaUected
We expect a gan for lsquoand withrsquo but it is actually a chan(Morgan 1952 Ball amp Muumlller 1992)
The same is true of dros and drwy though there the variantswith the voiceless stop survive in the modern language
So maybe gan is really [L]can underlyingly
Where [L] is the autosegment (Wolf 2007)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t˚
[L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Back to Breton devoicing sandhi
I propose that (some) Breton devoicing sandhi reWect the sameincomplete lexicalization of the voiced stops
Consider lavare[t t]intilde
t [L] t
Lar
[vcl]
Normally [L] docks to the following t e g due to MaxFlt(Wolf 2007)
But not when the Lar node spreads to a preceding root node
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This can be for any number of reasonsSome version of geminate inalterabilityStructure sharing inhibits weakening processes (Honeybone2005b)Under certain assumptions the structure shown is not convex(Scobbie 1997)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
Further evidence for this approach comes from so-calledldquofailure of mutationrdquo (Jackson 1967 sect481)Lenition of voiceless stops is said to ldquofailrdquo when an adjective(given the necessary morphosyntactic conditions) follows anobstruent-Vnal nounBut with sonorant-Vnal nouns or voiced stops mutationhappensCf kaer lsquobeautifulrsquo
(19) a una
dimezellmaiden
gaerbeautiful
b ura
vaouezwoman
kaerbeautiful
Morphosyntax actually irrelevant since other triggers of thismutation are sonorant-Vnal
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Devoicing sandhi
This is the same phenomenon an autosegment normallyleading to voicing is inhibited by spreading of the Lar node
Following sonorants the Lar node canrsquot spread since sonorantswith a Lar node are never well-formed
But this time we have much better evidence for theautosegment being there
The same data are described by Ternes (1970) in an extremelyconvoluted way
The generalization if an obstruent is voiced by an autosegmentit can resist voicing by spreading Lar to a preceding obstruent
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
What autosegments
In previous work I have doubted that the autosegmentalapproach is suited to Brythonic Celtic mutations (cf also Green2006)
I think these data are actually pretty solid evidence forautosegments or at least for a phonological analysis
Breton is less problematic than Welsh morphosyntactically
Breton mutation seems to be genuinely sensitive to prosody(Pyatt 2003)
There is still the problem of doing mutation phonologicallyWolf (2007) covers only a small subset
In particular the autosegment should cause deletion of [vcl] inthe current approach
Problem But see Bye amp Svenonius (2009) for an approach
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Other types of devoicing sandhi do not seem to fall under thisrubric
(20) a [san kOneri] lsquoSaint Goneryrsquob [kankuS] lsquo100 timesrsquo cf [tErguS] lsquothricersquo
I propose that here devoicing is due to univerbation i e therelevant words are now compounds
Word-internally voiceless obstruent clusters are (nearly)universal (also noted by Hall 2008 for Icircle de Groix)
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
More devoicing sandhi
Jackson (1967 sect487) ldquoprovection in common phrasesrdquoAre these actually phrases or words
Saint Goneacutery is the patron saint of the local chapel
lsquoThricersquo might well be a single word cf Welsh dwywaith lsquotwicersquoand in fact [guS] is the reduced form cf stressed gwej lsquotimeoccasionrsquoEtc
Photo credit SteUen Heilfort SourcePavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
What featuresAnalysis of length and Vnal devoicingAnalysis of sandhi
Summary and outlook sandhi
Voicing sandhi are mostly due to phonetic implementation oflaryngeally unspeciVed obstruents in a phrasal context
Some devoicing sandhi are due to inhibition of autosegmentallyinduced voicing
Others might possibly be not phrasal sandhi at allBoth of these phenomena seem to be cross-dialectal so theaccount possibly extends to Icircle de Groix
PrepositionsMore examples the ldquodevoicingrdquo word [bnak] lsquoanyrsquo is MiddleBreton pennac (Lewis amp Piette 1962 sect45)The ldquoprovection in common phrasesrdquo (univerbation) is describedas pan-Breton Examples of devoicing sandhi in Icircle de Groixinclude lsquogrey peasrsquo and lsquolittle Vngerrsquomdashintuitively goodcandidates for univerbation
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Outline
1 The problem
2 Analysis
3 Implications
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Loss of feature or loss of contrast
Here I have argued that Breton presents examples two types ofVnal devoicing
Final devoicing as loss of contrast cf the arguments of Harris(2009) for FD as weakeningFinal devoicing as edge alignment Vnal fortition (Iverson ampSalmons 2007)
Take-home message here there is no process of ldquoVnaldevoicingrdquo ldquoVnal weakeningrdquo or ldquoVnal fortitionrdquo that we canspeak of in universal terms
Argument for substance-free phonology
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Final devoicing as phonetics
Growing body of work on Vnal devoicing (and generallylaryngeal assimilation) as a ldquolow-level phonetic processrdquo
The Paradestuumlck here is of course Dutch (Ernestus amp Baayen2006 2007 Jansen 2007)
Possibly others (e g the disputed claim for Polish)
Breton seems to show quite good evidence for incompleteneutralization
Laryngeally unspeciVed segments interpreted by the phoneticsas devoiced or aspirated rather than [minusvoice] or [spread glottis]speciVed
Needs careful cross-linguistic study
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
Kraumlmer (2000) argues that the presence of both voicing anddevoicing necessitates binary features i e a ternary contrast
Related issue UUmann (2009) asks how to distinguish betweencategorically voiceless and laryngeally unspeciVed stops in aprivative system
The answer is of course feature geometry
Objection of UUmann (2009) but this is an overgeneratingnotational variant of binary features
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Ternary contrasts
One answer who says we never need bigger feature geometrytrees It is correct that arboreal representations can have manylevels but maybe this is empirically better
Related answer binary features are no more God-givenlessstipulative [0voice] [1voice] and [2voice] are also a notationalvariant but these are as overgenerating as trees
Reason three independent values of [F] cannot captureimplication relations in the same way that feature geometry can
Here I argue that the feature geometryunderspeciVcationapproach is empirically more adequate than one based on[plusmnvoice] spreading
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
Here I use class nodes (as in e g Avery 1996)
Blaho (2008) no need for nodes if features can do the job e gsubstitute Lar with [obst] since only obstruents are laryngeallyspeciVed
Gives strange results for Breton since Vnal devoicing is drivenby [obst] works formally but how insightful is it Are thedevoiced obstruents sonorants (Well why not)
Here nodes are necessary
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Tiers or features
If features can only attach to nodes the presence of a node(even with no features) is the formal correspondent ofcontrastive speciVcation
Sort of answers the concern of UUmann (2009) on thediUerence between two types of feature absence
Without nodes how do we deVne tiers and all theautosegmental phenomena that come with them
Null hypothesis all and only features dependent on a speciVcnode are on the same autosegmental tier
Field of empirical inquiry
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton
The problemAnalysis
Implications
Interpretaton of Vnal devoicingTernary contrasts and feature geometry
Summary
New interpretation of Breton data
Possible cross-dialectal extension
Privative features can do the job
Featurenode geometry is preferable to binary features and(possibly) to node-less geometry
Trugarez
Pavel Iosad Laryngeal phonology in Breton