Upload
lynne-ramsey
View
214
Download
1
Embed Size (px)
Citation preview
Cognitive Modeling in Linguistics 2011
The 13-th International Conference
Island Corfu in Greece, September 22-29, 2011
THE PERCEPTION - PRODUCTION MODEL IN THE ACQUISITION OF ENGLISH WORD STRESS: THE CASE OF POLISH SPEAKERS
Anna MarczakAdam Mickiewicz University, PoznańPWSZ Płock, Instytut Neofilologii
Aim of the paper
to demonstrate that
universal prosodic preferences guide the acquisition
of English stress perception and production
L2 learners use their prosodic strategies in terms
of learning optimal constraint ranking
consonant final words and syllable weight (WSP
constraint) are the most problematic parameters
in Polish English stress acquisition
Empirical evidence
Polish learners of English (PE learners) of different schools :
Primary, Secondary and students of English Philology
Outline of the talk
1. Hypothesis
2. Description of the experiments
3. WHY Optimality Theory (OT) and WHY Natural Phonology (NP) ?
4. Theoretical part: OT stress systems: English, Polish, Polish English
5. Principles & Parameters (P&P)
6. OT and NP analysis of the selected data
7. Discussion
8. Preliminary conclusions
1. Hypothesis
Assumptions
For Polish English (PE) subjects:
a degree of difficulty in assigning stress patterns would correlate with syllable type
penultimate stress pattern causes incorrect stress placement- due to the positive parameter setting (cf. Altmann 2006)
TROCHAIC and DACTYLIC foot types are expected to be a universal criterion in PE stress perception and production
2. Description of the experiments
Total number of PE subjects:
Primary school:14
Secondary school:14
Students of English Philology: 14
Native speakers of English: 10
English sentences (35), stimuli-only final words,(production task: reading sentences and then underlying stressed syllable in perception task)
An English carrier sentence: I said xxxxx this time, each sentence contains a different nonce word (max 5 syllable), (production & perception tasks). Additional task: metrical beat choice
The recordings were annotated by 2 Phoneticians
Stress shift from a verbal to a nominal stress pattern in verbs
54 %
Stress shift from penult to antepenult
88 %
Stress shift from antepenult to penult
94 %
2. Experiment: real words
• 2008
• 2009
1. In this summer I want to visit Manitoba.
2. When I was in school I learned to throw the javelin.
3. The delegates were still not sure who they should elect.
4. In the opera company, Bob’s the best baritone.
5. We can’t talk about that, it’s not on the agenda.
6. I really didn’t think that the building would collapse.
7. Roberta is not very easy to astonish.
8. The thing I love about coffee is the aroma.
9. On Saturdays I like to go to the cinema.
10. You can’t take part in the class but you’re allowed to observe.
11. Edmonton was devastated by a hurricane.
12. The town asked font a big loan to build an arena.
13. It has strong taste but I really like venison.
14. When we all went to the zoo we saw an antelope.
15. When I came to Canada, it was hard to adapt.
16. They made Tony an offer he’s going to consider.
17. They asked me but I don’t want to be the candidate.
18. This new manuscript is difficult to edit.19. Some of the results were difficult to interpret.20. My brother always wanted to be a matador.21. I have never met anyone from Minnesota.22. I’m thirty years old and I still have my appendix.23. She lives in the United States of America.24. I find that position much too tiring to maintain.25. I don’t think she’s as old as she might appear.26. I can’t come on Friday, I guess I’ll have to cancel.27. You can see the sun a bit above the horizon.28. When it gets hot I like to sit on the veranda.29. You can record over the songs you want to erase.30. He didn’t read the book, he just read a synopsis.31. The committee will support whatever you decide.32. Don’t talk to me, Bob’s the person you have
to convince.33. I was amazed by what you were able to achieve.34. I was trying to fix the doors on the cabinet.35. The exam committee couldn’t reach a consensus.
2. Experiments: nonce words (text for subjects)
Production task: Read the following sentences aloud:
I said fepade this time
I said pomphensus this time
I said kermanic this time
I said lanezalist this time
…
Perception task: Repeat each word after a native speaker of English:
fepade, pomphensus,kermanic, lanezalist
…
3. Why OT ?
Describing PE stress variation within OT framework seems to be an appropriate tool to establish:
Stages of language development as Constraint resettings Learnability processes as learning Constraint rankings OT learning mechanism (adopted from GLA Boersma
1997) Constraint hierarchies of stress grammar in English (E),
Polish(P) and Polish English (PE)
3. Why NP ?
Natural Phonology (NP) (Stampe 1979, Dziubalska-Kołaczyk 2001, 2002) is a preference theory, universal tendency of lg behaviour (compatibility with OT). NP proposes an adequate expalnatory apparatus in L2 prosody
NP is able to explain L2 learner’s learning strategies in the acquisition of prosodic structure
NP has an explanatory power in a language external context (cognitive grounding: pronoucability – speaker’s easiness of pronunciation, that is speaker oriented activity and perceptability- listener oriented activity, Lenition vs Fortition processes)
4. OT Stress Systems: optimal stress in English
The model outlined here is an application of Optimality Theory ( Mc Carthy 2003 Kager 1999, among others) to that off-shoot of Metrical Theory developed by Burzio (1994), itself grounded in the work of Hayes (1995).
The metrical parameters (to be visible to surface representation (SR): feet are either binary or maximal ternary ( trochees or dactyls), extrametricality is right-edge (limitation: syll. consonant, schwa, (ø) empty vowel), they constitute a third syllable weight, ultralight (U)
4. Constraint hierarchy in English
IDMETRIFYU
METRIFYUVV Ø]
METRIFYUVERB
RTMOSTPRIMARY
IDSTRESS
*LONGFT
*UNARYFT
FTWEIGHTMAX26
*STRESSWEAK
FTHEADLT
*METRIFYU *STRAYSTRESS
WDINTFINSTRAY
PRESERVESTRESS
*UNARYFT, *LONGFT, FTHEADLT, FTWEIGHTMAX26, RTMOSTPRIMARY, *STRESSWEAK, IDSTRESS >>
*STRAYSTRING >> *WDINTFINSTRAY >> PRESERVESTRESS (/ FTWEIGHTMIN13)
RTMOSTPRIMARY
IDSTRESS
*LONGFT
*UNARYFT
*WDINTFINSTRAY
FTHEADLT
FTINIINT
*METRIFYU
BINARYFT
LTMOSTDACTYL
Constraint hierarchy in Polish
*UNARYFT, *LONGFT, RTMOSTPRIMARY, FTHEADLT, IDSTRESS, *WDINTFINSTRAY, *FTINIINTO>> *METRIFYU >> BINARYFT >> LTMOSTDACTYL4. Constraint hierarchy in
Polish
IDSTRESS
4. Constraint hierarchy in Polish-English
RTMOSTPRIMARY
*LONGFT
*UNARYFT
FTHEADLT
*METRIFYU
BINARYFTFTWEIGHTMAX
Intermediate and ustable constraint hierarchy in PE
METRIFYUVERB
Results – Metrical Parameters 5
In my error analysis on stress placement and production I employed a parametric tool designed by Dresher and Kaye (1990) (Appendix 3), to assume a metrical comparison of languages.
Let us look at a juxtaposition of metrical parameter settings for Polish and English proposed by Archibald (1998: 180) (Appendix 4). We will use this tool for analytical purposes of this study.
5) Parameters Polish English
P1 (word tree) right right
P2 (foot type) binary binary
P3 (built from) left left
P4 (strong on) right right
P5 (QI/QS) QI QS
P6 (sensitive to) NA rhyme
P8 (extrametrical) no yes
P8A (extrametrical on)
NA right
P1: The word tree is strong on the [Left/Right].P2: Feet are [Binary/Unbounded].P3: Feet are built from the [Left/Right].P4: Feet are strong on the [Left/Right].P5: Feet are quantity-sensitive (QS) [No].P6: Feet are QS to the [Rime/Nucleus].P8: The extrametrical syllable occurs on the
[Left/Right].P8A: There is an extrametrical syllable
[No/Yes].
6. Polish English variation: OT procedure
kəm'pju:tə TROCHAIC FOOT DACTYLIC FOOT
kəm'pju:tə
*
'kɒmpju:tə
*
kəm'pju:tə
*
6. Polish English variation : Natural Phonology interpretation
The above data can be interpreted in NP approach in terms of 3 developmental stages of English word stress acquisition :
• Transfer stage: correct stress placement only in the case of penultimate stress pattern
• Transition stage: positive parameter setting causes incorrect stress location (cf. Altmann 2006)
• Final stage: parameter is learnt to be placed correct
6. Problem
Stress variation is within the English grammarQUESTION:
How to learn word stress variability?
6. Solution
by modeling
word stress perception and production on three levels of representation:
Underlying Representation (UR)
Surface Representation (SR)
Overt representation (OR)
6. Solution
by operating on 3 levels of reperesentation, where
OT is responsible for (UR=input) and (SR=output) eventually communicating with NP (OR=phonetic representation=auditory, articulatory)
OT representations are capable to describe metrical (mental) processes
NP representations are capable to explain stress beats choices as natural preferences, Stress Beat Decoding-Encoding Detector (SBD-ED), (Marczak A. 2011)
6. The PE learner’s task is
to rank L2 stress grammar according to English-like interpretation of overt forms
to undergo at least 3 stages of L2 stress acquisition
6. Natural Phonology stages in L2 word stress acquisition
Transfer stage/Polish-like:mostly penultimate stress patterns are active (more predictable, incorrect patterns), ex. fre’qwency
Transition stage /parameter learning stage: Polish English (PE), (not predictable),variability, ex.’computer/ com’puter
Final stage/English-like: (predictable), ex. com’puter
6. Optimality Theory stages in L2 word stress acquisition:
1st Stage of Markedness dominance: Markedness >>Faithfulness, unranked constraints: (L1=L2), constraint interplay
2nd Stage of Demotion-Promotion mechanism/Gradual Learning Algorithm (Boersma 1997), a stage of variable PE stress system,
unstable constraint hierarchy:L1>>L2), (L2>>L1),
3rd Stage of fixed constraint hierarchy
(English-like ), completed learning: (L2>>L1)
7. OT-PE diagnosis: Gradual Learning Algorithm (Marczak 2011, adopted from Boersma ,1998)
The two s are so close to each other that they may swap places as they
wobble. and are too far away to ever swap. may cross the BOTTOM Line as it wobbles and become inactive.
is extent
C
C C
CC
C
C
C
INACTIVE
ACTIVEC
C
C
7. PE Stress Placement Mechanism
L2 STRESS RANKING TARGET STRESS PLACEMENT STRATEGIES
processing
PE STRESS VARIATION STRATEGIES
*LO
NG
FT
BIN
AR
YFT
1)
RT
MO
STPR
IMA
RY
FTH
EA
DLT
IDST
RE
SS
*UN
AR
YFT
L1 AND L2 STRESS CONSTRAINT CONFLICT (Stress Typology Interaction L1 ↔ L2)
Stage of Markedness Dominance
Stage of Gradual Learning Algorithm
Stage of Fixed Constraint Hierarchy
PE STRESS LEARNING PROCESS IN NP& OT: 1) the most frequent constraints are red (preferences)
7. Discussion: local variability
can be determined due to observations:
L2 sounds that are dissimilar to the L1(Flege 1987) -acquired more easily than similar ones. This postulation may hold in the perception and production of stressed syllables. Similar L2 prosodic structures to the L1 ones may cause problems in L2 acquisition.
7. Discussion: global uniformity
Polish English system eventually stabilizes with ranking values due to:
language exposure and received feedback (Transfer stage, learning)
local variability stops occuring ( Final stage)
global uniformity is achieved (Final stage)
L'H
U
"LH
U
H'L
U
"HL
U
L'H
LU
"LH
LU
LH
'LU
H'H
LU
HH
'LU
"HH
LU
('LL
L)<
U>
(L'L
L)<
U>
('HH
L)<
U>
(HH
'L)<
U>
(H'H
L)<
U>
L('L
LH
)<U
>
L(L
'LH
)<U
>
"L(L
LH
)<U
>
L(L
L'H
) <
U>
H'L
L<
U>
"HL
L<
U>
HL
'L<
U>
Fepade Untrist Kermanic Donphortable Bomizon Pomphensus Lanezalist Anescelate
0
2
4
6
8
10
12
14
16
18
20
10
4
2
12
5
8
1
3
7
4
8
6
32
9
6
1
6
1
7 7
13
1
6
8
5
7
23
11
13
1
56
3
65
3
1
10
3
9
5
8
6
3
1110
3
1
14
5
9
13
12
12
PRODUCTION8 Nonce words, 3 groups of lg proficiency, 112 tokens
Total number of subjects: 42
Native speaker Primary School Secondary School English Philology
7. 2011 Experiments:Production - Language Proficiency factor
7. 2011 Experiments:Perception - Language Proficiency factor
L'HU "LHU H'LU "HLU L'HLU LH'LU H'HLU HH'LU ('LLL)<U>
(LL'L)<U>
('HHL)<U>
(H'HL)<U>
(H'HL)<U>
H'LL<U>
"HLL<U>
HL'L<U>
L('LLH)<U>
L(L'LH)<U>
L(LL'H)<U>
Fepade Untrist Kermanic Donphortable Bomizon Pomphensus Lanezalist Anescelate
0
2
4
6
8
10
12
14
16
18
20
14
12
2
14 14
12
2
11
12
12
2
13
1
13
1
13
1
12
2
9
5
13
1
13
1
10
3
1
12
1 1
14
13
1
13
1
12
2
14 14
12
2
14
PERCEPTION8 Nonce words, 3 groups of lg proficiency, 112 tokens
Total number of subjects: 42
Native speaker Primary School Secondary School English Philology
7. 2011 Experiments: Production vs Perception (real words) - Language Proficiency factor
94%
6%
Primary SchoolPerception correctness
Correct Incorrect
42%
58%
Primary SchoolProduction correctness
Correct Incorrect
7. 2011 Experiments: Production vs Perception (real words) - Language Proficiency factor
66%34%
Polish Students of English Philology1st year 2011
Production correctness
Correct Incorrect
87%13%
Polish MA students of English Philology3rd year 2011
Production correctness
Correct Incorrect
66%34%
Polish Students of English Philology1st year 2011
Perception correctness
Correct Incorrect
98% 2%
Polish MA Students of English Philology3rd year 2011
Perception correctness
Correct Incorrect
8. Eksperiments: conclusions
Basically, the subjects have mastered the English stress patterns•One thing, they do not apply correctly: word-internal feet are not quantity sensitive•They make a big distinction between words ending in a consonant, and words ending in a vowel•The clear tendency is that in words ending in a vowel, the subjects come quite close to what is done in English•The results of the experiments demonstrate higher accuracy in Nonce words ( both in perception and production) which may be interpreted that some of the real words were incorrectly lexicalized• Language proficiency factor is a crucial variable both in L2 stress perception and production accuracy.•A number and a type of syllables in a word decide about L2 stress location strategies
8. Preliminary Conclusions
Data obtained in experiments (Marczak A. 2007, 2009, 2010, 2011) show constant alternations in stress production between two foot types which are essential for all three stress systems (PP, PE, EE), namely trochaic and dactylic, which appear to constitute two free variants in the PE stress grammar: TROCHAIC and DACTYLIC.
• TROCHAIC is identified as a similar prosodic structure existing in both systems: PP and EE, therefore a positive paramater setting from PP (Altmann 2006) may cause a major difficulty in a PE stress learning and resulting in PE stress variation.
• In general the accuracy in stress perception, both in nonce words and real words, display higher correctness than in the production data.
8. Preliminary Conclusions
NP explanation for the three stages (PP, PE, EE):• NP can explain the transfer stage on the basis of natural (PP)
preferences, NP provides a greater explanatory force than OT for this stage.
• For the learning mechanisms that take place within the transition stage NP cannot offer a satisfactory account of the data as natural preferences have to be “unlearned” at that stage. Correct stress placement in the final stage could be explained by the speaker’s exposure to the (natural) language environment.
• Problem: There is No account for optionality within NP (contrary to OT)
8. Preliminary Conclusions
• L2 stress Perception >> L2 stress Production• OT is responsible for describing L2 stress behaviour
whereas NP is responsible for explaining natural preferences of L2 stress grammar
• My future plans for further research: using PE stress p&p data to feed Gradual Learning Algorithm (Boersma 1999, 2008) so as to test predictability in L2 stress variation.
• Impossible stress assignment can be ruled out by Constraint-Based Model (Marczak A. 2011), the examples:
compu’ ter, frequen’cy
Selected references
Altmann, H. 2006. The Perseption and Production of Second Language Stress: A Cross-Linguistic Experimental Study. PhD dissertation. University of Delaware. Archibald, J. 1993. Language Learnability and L2 Phonology: The Acquisition of Metrical Parameters. Kluwer Academic Publishers, Dordrecht-Boston-London Boersma, P. 1997. Functioning Phonology. Formalising the interactions between articulatory and perceptual drives. Universitet van Amsterdam, Amsterdam Boersma, P. & Pater, J. 2008. Convergence Properties of a Gradual Learning Algorithm for Harmonic Grammar. [ROA 970] Coetzee, A. & Pater, J. 2009. The place of variation in phonological theory. [ROA 946]. Hayes, B. 2004. Phonological acquisition in Optimality Theory: the early stages. In: Kager, R. et al. (eds.) Constraints in Phonological Acquisition. CUP. Cambridge. Dziubalska-Kołaczyk, K. Beats and Binding Phonology.Peter Lang, Frankfurt am Main 2002 Dziubalska-Kolaczyk, K. Natural Phonology: Universal Principles for the Study of Llanguag Insiders meet Outsiders 71-75, 2007 , www.ichps.2007.de Kager, R. 1999. Optimality Theory. Cambridge University Press, Cambridge. Marczak, A 2008 Production of English Stress Beats: the case of Polish speakers. Poznan Studies in Contemporary Linguistics Publisher Versita, Marczak, A 2011 A Constraint-Based Model of Word Stress in Polish English Word Stress Acquisition. Proceedings in Sixth Cambridge Postgraduate Conference in Language Research 7-8 December 2010 (108-118), Computer Laboratory, University of Cambridge , McCarthy, J. 2003. Optimality Theory in Phonology. A reader. Malden, MA& Oxford: Blackwell. Müller, G. 2011. Optimality-theoretic Syntax. http://www.uni-leipzig.de/~muellerg/mu245.pdf. Pater, J., R. Bhatt and C. Potts. 2007. Linguistic Optimization [ROA 924]
θæŋks °