.. ...

Allomorphy and ION

Mark Aronoff 20E-221, lVl .I.~. April 6, 197D

0

Principal Conclusions

I. Allomorphy

Any rule which effects a phonological change, and

i) which need not follow any phonological rule

ii) whose environment may be stated morphologically

is a rule of allomorphy.

Rules of allomorphy follow the rules of word formation

and provide the input to the phonology.

Rules of allomor phy are not extrinsically ordered.

II. ben-Moshe's Laws of the h OOt

First Law: Only and most roots which take the marked

form of a suf f ix exhibit allomor phy.

Second Law: If a root is marked, then all occurences

of that root take the marked suffix.

Third Law:If any occurence of a given root exhibits

allomor pny in a given environment, then all

occurences of that root in that environment

will show the same allomorph.

Thanks fo

Morris Halle Hichard Oehrle Alan Prince ~dwin Williams

Crosley Shelvador ben-Moshe

Susan Nartin

1

o. This paper had,originally, a very clever

introductory paragraph, wherein I noted, somewhat

sarcastically, the popularity o_{ the orizing in

linguistics, and warned Lhe reader that they would _rind

little of t1~a t in the toll owing pages. Despite tnis

salutation, about half-way through the aej)aal body of

the work, I realized,to my great chagrin,that ~ was

making, and assuming,some raher strong,and somewhat novel

theoretical claims. these claims have to do with morph­

ology,particularly that aspect of mor~hology which

commonly goe under the rubric of readjustment rules

( cf. Chomsky and 11alle 196B, henceforth SPE). In

particular,I will of ten have recourse to one type of

r~Iorphological rule (1'1-rule), a rule oi allomorphy,

which states generally irregular phonological changes

in the shape of a morpheme in different morphological

environments.

Before going 1'urther into the notion of

allomorphy, let me first elucidate that of morpho­

logical environment, or conditioning, one which, though

it is tacltly assumed in much work on Generative 1honolo-gy,

especially in ~PE, has not, to my knowledge, been explic­

itly isolated. Since I am eneral~y concerned with word

formation, I will give examples o.i' various types of possible

environmen~s for (conditions on) rules of this sor~. Je

2

are familiar with syntactic, semantic,and phonological

conditions on word formation . An example of a syntactic

cundition is the fact that the suffix -ment is generally

attached to Lexical Items which have the syntactic label

Verb • .Nominals in -ment thus have the form X+ment, W11ere

X is a verb. A phomologi cal condition, rirst noted by

J.R. Ross, and discussed in ~iegel ( 1~71), holds on the

.rormation of nouns in -al. Basically, -al may appear - -after a stressed vowel, followed by an optional glide or

sonorant, and no more han one consonant. So denial,

appraisal,and dispersal. 'l'he stem-word is also generally

bi-syllabic 1 .1m example of a semantic c_;ndi tion, auain

a slightly modi ied vers on of one in siegel (1~71; , is

that on the Iormation of words in -ee. the stem of such

a word must be a transitive verb (or the root of a trans­

itive verb), which selects an animate object, or indirect

object. So presentee, evacuee, payee,promisee,patentee. 2

The sole exception to this rule is escapee.

A morphulogical condition, tnen ,is, by eliminatiGn,

one which is neither syntactic, semantic,nor phonological;

one which is grammatically arbitrary, in that it depends

on some >'f eature which usually plays no part in the

derivation of sen~ences. A good example of a morphological

condition is that on the suf.Lix -ity which restricts it

to l.Jatinate stems . Morphologi cal features are sometimes

relevant to phonological processes. In .h.'nglish , all those

3

instances of lf which eventually become s or 3 are J.·ound

in Latinate morphemes, and are usually orthograyhically

c. The rule,or rules, which take k too~ must therefore be

sensitive to the morphological feature +Latinate3 • 1n SPE,

and other works,morphological features are encoded into

the phonology by means of rule-features. Thus all Latinate

morphemes with kin the relevant phonological environmamm

will be marked by a rule-feature as undergoing the k to e

rule. Whether this is the correct device, or whether

morphological features may play a direct role in the

phonology, l do not at present Anow.

A rule ol" all :Jmorphy is thus a special type of

morphologically conditioned rule, which spells out the

various forms of a morpheme, and which, I claim is

distinct from, and, intrinsically, precedes the rules of

the phomology. Allomorphy lS not a novel c .)ncept. It was

widely used in ~tructural linguistics. However, in the

great r~olt against Taxonomic ~honemics, allomorphy,

which,in that framework, covered both regular and irregular

processes indi s criminately, was rejected in all but the

most glaring cases of suppletion (go/went) •. ~hen examples

ofnon-suppletive irregularity were later discovered, thes

were handled by rule-features. But the use of the rule-

eature in cases of morphological alternation is very

4

diffrent from other uses of tnis device. ~n general,

\ cf. Kiparsky 1968 )rule - L· ea tures are used to avoid

having to posit abstract phonological units (phoneues)

which never appear on the surface. However in the cases

of allomorphy which I discuss below, Lhere is no need to

posit such new phonemes . A morpheme can have more than

one allomorph, but each allomor 9h has aphonologically

regular form , i.e. contains no "abstract" ph nemes.

The rules of allomorphy thus do not pretend to replace all

uses of rule-1ea tures, only those which cannot be motivated

in terms of abstarctness, and which invmlve morphological

al tarnation. In general, then, t_le rules of allomorphy

take the place of a very speci1ic class of phonological

rule(P-rule). ~hese rules have the peculiarity that tneir

environment must (or may) be morphologically marked, and

that there is no P-rule which must pre cede them. There are

rules which possess only one of these properties. So the

rule which ta ,< es ,! to y_ in iVn (SPE p.227) must be marked,

for only certain instances of ivn undergo it. However ,

this rule must allow phonological rules, specifically

stress rules, and therefore cannot be a rule of allomorphy .

One might vie .. Lhe rules of allomorphy as very

arly P-rules. However , there is some indication that this

5

is not correct. ~her seems to be no extrinsic ordering

among the rules ol" allomorphy( with one type of exception

which I will discuss below). If this is so, vhen one can

say that these rules precede,but are not included in,the

P-rules. ~hey are in fact, as I noted above, more or less

equivalent to a class of readjustment rules. So, for example,

the rule .luat spells out the irre0 ular, and single regular

var·ant of the English plural morpheme can be considered

a rule of allomorphy.

By separating rules of allomorphy from the

phonology proper, I am, inciden'tally,making a claim that

cer'tain alternations are not phonological. One interes~ing

variant of this is t..lle claim 1:;hat theses alternations are

not necessarily those which cannot be stated phonologically,

i.e. without recourse to morphological conditions, but rather

those which need not be so stated. This claim is corroborated

by some fascinating work done on the Maori passive morpheme

( cf. Hale 196?), and gains some support from a few o1' the

alternations discussed below.

RE~KK : ~his paper, uespite the last five pages ,should not be taken as a theoretical polemic in favor of allomorphy.

I did not set out with ~ hat purpose i~ mind, but rather to describe a set of data. Allomorphy proved to be a useful tool in the description of that data;it seemed to unite diverse phenomena in an interesting way,and it is with this in mind that I would nope the paper will be read and considered.

6

1.1 Armed ~ith this i orwarning, let me tell the

still curious Gnat t iis is a,hopefully,detai led s ~udy

of the English nominal marker -At+iVn (IO~). This suff ix

is very widespread. ~ alker lists about 2000 words ending

in it (4% of the items in that dictionary). Normally,it

is a dverbal nominalizing sufi ix (fascination/fascinate),

however there are nominals in lON whose s t em l S not a free

word (compunction/compunct,salvation,salvate). There are

also some very lew which have a corresponding adjective

or noun, instead of t he expec~ed verb ,contr i tion/contrite,

ideation/idea). The suffix is usually restricted to Latinate

stems, agin with a few excep~ions (botheration,flirtation).

Not all instances of orthographic Xion are to

be taken a s instance;:; of -ro.~.~. All forms in which the i

is syllabic are not (accordion,ganglion). Also disregarded

are words such as onion,companion,million, which can pro­

bably be excluded on semantic grounds.

I will not deal in this paper with the actual

formation, or analysis (morphological) of vJords in -lON,

and will have correspondingly little to say about the

productivity o1· the suf .r ix. This work is rather a pre­

liminary,in many ways , to a morpholog~ cal statement.

1.2. The phonology of -~ ON is dea lt WLth very

thoroughly and c onv ~ncingly i n ~PE. There an underlying

phonological t orm (+At)iVn is posited 1"or it. iVn must

be bi-syllabic because of stress facts, namely the

7

placement of primary stress on the syllaole preceding it

(prohibition,SPE p.87),and the operation of the tri-syllabic

laxing rule(decide/decision,S~E p.182). A later rule takes

ito :t.. (SPE p.2 <' 5-2 .c 7) • .v· urther rules of spiran oization

and palatalization yield the correct output.

On then, to the allomorphs of ION. As many

peoplehave,I am sure,noticed, this sui .t' ix has several

underlying phonological .{orms , as shown by t 1e following

alternations:

realize realization *realizion

" educate education *educatation •.

repeat repetition *repetation *repetion

commune communion *communation

resume resumption *resumion *resumation

resolve resolution *resolvation

I believe that from these, anu other examples, we can

dmonstrate that ION exhibits the following allomorphs:

+iVn, +At+iVn, +t+iVn, and possibly +it+iVn, +U~+iVn

The distribution of these forms is somewhat complex,

but 1 will try to give as complete a description as

possible.

2.1. At+i Vn this is the unmarked variant. I~ is

freeest in terms of conditions on its attachment, and

alsotne most productive . As evidence of its productivity

let us look at verbs in the suffix +ize. These verbs may,

8

almost without exception, form nominals in +At+iVn. so,

if I create the verb communalize, I may automa~cally

create the corresponding nominal communalization. ln the

cases where a nominal in +At+iVn may not be 1.'ormed , it is

because there already exists a parallel nominal of another

form,as i'or example criticize,*criticizaLion, criticism.

The strong productivity of +At+iVn suffixation with verbs

in +ize can be strikingly compared with suf ixation in

+ment, a Latinate nominal suffix with almost parallel

semantics. /Jalker lists only the following 1'i ve nominals

in .X.izemen"t:

afiranchisement, advertisement,chastisement, divertisement agg:randizemen t.

So we see an almost exceptionless tendency for verbs in

+ize to have nominals in +At+iVn, as opposed to +ment.

How this fact is to be captured I."ormally , I do not at

present know.

~-2 As far as I can see,there is no restriction on

the iinal consonant of verbs to which: +At+iVn may attach.

The following list demonstrates this point.

9

List of +At+iVn nominals

LABIAL

perturb(ation) l:: orm(ation) exhum(ation) usurp(ation)

CORO.NAL

cess(ation; de grad (a ti on) elicit(ation) accus(ation) revel(ation) declar(ation) examin(ation) represent(ation) deport(ation) manifest(ation) COnSU.lt(ation) affect (a ti on) commend(ation) sens(ation) indors(ation)

VBL.AR

evoc(ation) purg , ation) prolong(ation)

~here do not appear to be any principled gaps.

?v • >· There are only a few instances o1· +At+i Vn

at"ter a vowel-final stem. 1'his can simply be traced to

the relative paucity of ~ vowel-final Latinate stems.

Examples are renounce/renunciation, vary/variation,

continue/continuation. 8he 1nte~~sting cla§s of vowel-,'

final stems does exist. As noted in SP~ (p.201),verbs

in +fz and +£1l generally nave nominals in ication,as

ampli1·yj ampli1" i ca tipn, imply/ implication. 'l'his al terna t­

ion is covered by the.following "ad-hoc" P-rule:

(SP~ p.201,rule (b2))

ampliiy is thus derived from ampli i ik, and the i in

ampli.Lication ls shortened by -,;he rule tnat shortens

10

the r"irst a of explanation. :rhe rule is "ad-hoc 11 in a

non-pejorative sense. lt is so rormulated as to have its

structural desc ·iption met only by verbs .. vi th the roots

+1y, ±E!l· ~ven the + ooundary does its job, serving

to prevent dis=lik rrom being converted into dis=li.

The rule is peculiar, t1ough, in that though u~E lists

it among the P-rules, there is no other P-rule ~hat

mutJ~; precede it ~,.vowel shi . .L t must follow it). lt could

be the ~irst rule of the phonology. This leads us quite

naturally to ask whether this is indeed a Y-rule at all •

..t'erhaps it is a rule of allomorphy. Is ~here any evidence

for this proposalY

~irst of all, stating \62; as a r ule of

allomorphy ~·wuld immedia Lely account f Jr its ordering

properties, for such rules must vrecede all P-rules.

Secondly, (62; could just as easily have a mor phological

as a phonological environment, since its domain or

application is limited to two roots. utating the

environment phonologically does, however, make a certain

prediction, namely ~hat no wurd of the form HX+C 1Ik#

will ever reach ~he s urface in ~nglish. aS noted

above the + is crucial. ~here are words such as dis=like

and outpsyche. these ~ = boundaries have a nasty habit of

fading away to +4 . If,perchance, outpsyche, a very common

11

verb,did lose its =, would it then become *awtsi.

Doubtful. This intuition is a strong indication

that the environment ;or the k-dropping is indeed

not c1Ik, but rather vhe two roots ln question. lf

we persist in stating the rule as phonological, we

must mark them ~ith a rule feature as +(62). If

we simply regard (62) as a rule of allomorphy, we

lose a rule-feature and minor rule combination. The

rule-feature in question is, in addition, not

motivated on grounds of abstraction.

ords in +fy provide us V;i th another example

of a rule-feature minor rule combination which may be

included in "the rules o:t: allomoryhy. As mentioned in

SPE, certain verbs in +fy have nominals in +faction

(putrefy/putrefaction). SPE accounts J.'or this by having

the verb marked oy a rQle-feature for undergoing

::>hart Vowel Shil"t, the minor rule tnat accounts .for

the sing/sang alternation. aore 1.)recisely, the rule that

assigns the rule-feature (+0VSJ must assign it only

to the morpheme +fy, in order to avoid having the

other vowels of the word in question undergo SVS.

The deivation of satisfaction thus proceeds as J.ollows

(cf. ::>PE pp.201-202):

12

(+8VSJ satis+(~Ik)+t+iVn.

fik fak

satisi'ak~-i:n

laxing ;_cc SV0 etc.

There is some doubt as to tne status of SVS. The sing/sang

alternation has been claimed (Hoard & Sloat 1j 73J to

be the result of other distinct rules. There is also a

quetion as to .• hether rule :eeatures may be assigned to

constituent morphemes, ra t. her tnan entire Lexical

Items (SPE 373 ff.). The rule-feature is also, i r, cidentally,

of the non-abstract variety. This suggests that the

fik to fak process may be morphological. In tnis case,

there is very strong evidence, from the .{ orm of t.heiO.N

suft"ix, that this is so.

All nominals in +fak are oi· the 1orm i+factian

(liguefaction/*liguefacation). All nominals in +fik, with

the sole exce~tion of crucifixion, are of the I orm

X+fication. +fak is thus a root With the property that

the ~ of At+iVn drops after it. •rhis property is very

common; +t+iVn is the marked form of the sufLix, after

non-coronals, as will be seen below. There are , i urther-

more, no cases where the rule that assigns the marked

suf f ix must make reference to a derived phonologl cal

property(other than perhaps the root final consonantJ

~ e must, therefore,presume that it is the feature (+SVS)

which determines the form of the suffix here:

13

R 1. At+iVn t+iVn I fik(+~V~J+ __

But there is no other case of a rule feature determining

the form of the suffix either. (R1), which must be ~osited

if the fik/fak alternation is phonological, is thus a

very s i ngular rule. If, on the other hand, we state the

alternation at a mor phological level, we obviate the need

for both such a bizarre rule as (R1), and the suspec~

rule feature and minor rule.~he morphological solution

is thus to be preferred on several gr Junds.

2.4. I can feel the critical reader, squirm~ng

in their seat, hand outs tretched in objection; it

appears that sut·f icient evidence has not been presented

with regard to the last two issues. ~he alternate

solutions have not been discredited;both generate the

required output. 1 will remind this reader,though, that

the question here is not one of generation, but of

evaluation; when presented with a form, which may be

dealt with in either of two ways, which one does the

5Tammar choose? I have tried to show that when faced

with the dilemma, rule-feature plus minor rule, or

mor pholog1cal rule, the grammar chooses the latter.

Of course this decision calls into question the status

of rule-feature minor rule pairs. If they can be

eliminatd from the grammar entirely, then tnere is

no longer a dilemma.

14

2.5. The proposed derivation of nominals from

stems in +fy and ±ElY provides us with one case in .... , ~

which there is an ordering between morpnological~

The r' ik/fak rule must apply to the stem beiore the

condition on the attachment of At+iVn can be stated,

r'or the marked variant t+iVn is attached to fak, and

not to fik. It is probable, however, that rules of

stem allomorphy always precede rules of suffix

allomorphy. lf this indeed turns out to be the general

case, then we can say that though tnere may be ordering

among morphological rules, the ordering may never be

extrinsic. ln any case, the issue a v, ai ts further

evidence.

3.1. We have now encoun1.ereu. a set of ION nom..Lnal-

izations, the X+faktion type whose formation, or analysis

is not purely agglutinative. I shall turn then to other

forms in which there appears to be a change in the

suffix from the unmarked f orm At+iVn, and sometimes,

concomitant change in the root of the stem verb. In this

regard, we have already seen that fak is the marked

nominal forming variant of f.Y: it only appears with cer­

tain verbs, such as satisfy, and with others which are

marked by the purely non-phonological pre ence of an

~before the +fy (liquefy, putrefy). It is this marked

variant which takes the marked form of the suffix. This

15

phenomenon is quite general. T'lany of the roots which

take marked forms of the suffix exhibit mor 1)hological

variation. This is called ben-Moshe's first law of

the marked root.

3.2. Verbs in +At (equivocate, prevaricate) ,form

IO~ nominals about as productively as those in +ize,

with a corresponding lack of nominals in +ment(only

abatement and (rein,mis,under)statement. As has been 11

observed before (Siegel1971), the derived nominals do

not reduplicate the At (equivocation/*eguivocatation,

prevarication/*prevaricatation). It appears that the

restriction does not merely hold on*+At+At sequences

but on any XAt+At; so relate must be analyzed as

re+lat, yet we have relation and not*relatation.

Contrary to this tendency, however, we have dilatation,

where the unanalyzability of +lat seems to indlcate that

the narrowform (+At+At) of the restriction holds.

~imilarly with natation • The existence of dilation,

though, does point to a preference for the extended form.

Perhaps a look at other unanalyzable Latinate stems in

Xate will help here. There are three, to my knowledge:

abate, sate, and state. sate has an alternate satiate

from which t he nominal satiation is ~L"ormed. Tricky.

·r~ i th abate and s ~,ate something quite fantastic happens.

16

They are, as no~ed above, the only verbs of ~ne form

Xate which 1orm nominals in +ment. 1 can only concluue

!rom this that faced with Lhe crucial case, which will

decide be~ween two possible verslons of s rule, the

grammars equivoca~es, ana avoids having ~o maKe a uecisionJ.

Other cases o~ ~he extension of an environment

~rom morphological to phonological nave been uiscussed

by Pr1nce ~1972). He points out tirstly tnat nouns in

osis(arteriosclerosis) nave adjectives in otlc and never

o~ical. However ~he environment covers all cases of Xot,

and not just X+ot; nence iulotic(*alJ,uespotic(*al).

~uio~ical and uespo~ical were posslble words not too

long ago. ~nis sugges~s a historical change in ~ne rule,

from restricted ~o extended, morphological to phonolog­

ical. ~imilarly, verbs in i+fy have adjectives in i+fic

(specify/specific), and not i+fic+al (*specifical). This

has been extended in some cases (scientific(*al)), but

not all (Pontiff,*pontific,pontifical).

It appears ,the~, that the~ is a constraint

prohibiting +At+At orXAt+At sequences. How one might ask,

does the constraint operate? Does some rule delete the

first At, or the second? ~·rom looking at other suffixes

we can gain some insight. It appears that stem final +At

deletes generally before suffixes, so celebrate/celebrant,

nominate/nominee. If we wish to make a general statement,

17

we ean say that +At deletes before a suf1.ix, i.e. it

is the firs:t .A.t that deletes. One might claim as a

rule that .tit+iVn, being the unmarked varian"t ol" the

suffix ION, is also the only productive form of the suffix,

and the fac"t that it is the first At which deletes in n

the formation of this very productive class of derived

nominals (Xate/Xation) is evidence for this claim. How­

ever, in order to have any strength, the claim snould

have relevance to other suffixation processes. If one

could show that, in general , the unmarked .rorm of a

suffix is the only productive form, then t:nis particular

caes would be decided by the theory. I am , at present,

not in any pos 1tion to discuss the matter further .

4. The Harked Roots

4. 1 • We come now to those cases which I consider

the most interesting, the marked roots. I have already

mentioned that fak is such a root. It is marked, and -.

takes t+iVn, rather than At+iVn. Note that fak is not

a lexical item, it is a morpheme, distributionally

determined, with very little sema: tic content(compare

the semantics of putrel"y and satisfy, in particular,

what does sa~is+ mean?). One might think that this is

an exceptional case, and that when a given lexical

i tern has a nominal in A t+i vn, the form of the s u.f i ix

is determined by that individual and entire lexical

18

item. This is not so. It is rather the last morpheme

of the stem, which can usually be called the root,

which determines the form of the suffix. lf a root is

marked, then all occurences of that root take the marked

suffix. 1'his ben-Moshe' s second law of the root • .J.!'urther ­

more, if any occurence of a given root exhibits

allomorphy (/~At+iVn), then all occurences of that root

will show the~sam£ allomorph. fy/fik/fak is the only

exception to this last rule.

I will give an example. Take tne list of all

verbs of the form xsume:

subsume,resume,presume,consume,assume

~he first elements of ~hese verbs are common Latinate

prefixes, which may be isolated distributionally . The

verbs have very little in common semantically, and there

is no possible way inwhich, by giving the morpheme

+sume some meaning, h~wever broad, we may derive the

meanings of the separate verbs compositionally. ~

is therefore a distributionally determined morpheme ,

with little, if any, semantic content. now all of

these verbs have nominals in I v~:

subsumption,resumption,presumption,consump$ion,assumption

The .form of the suf1.·ix must be t+iVn (as with fak),

and is never At+iVn (consummation, as in ~hakespeare

is from consummate). This fact is not determined

19

phonologically , for we have exhume/exhumation,

de·plWile/deplumation. Yet every verb in +sume takes

the same marked Jorm of the suf...._" ix. 11 e can conclude

either that this fact is the product oi chance,or

that it is determined by +sume. ~·urther examples will

demonstrate that it is no matter of chance. v~e must

therefore conclude that tne root +sume is marke u to

form nominals in+t+ivn.

'l'he verbs ln+sume provide a good example o1'

another phenomenon too . If a verb containing a marked

root permits nominalization by rrn , it is usually the

case that all verbs in that root will have a parallel

nominal f orm. This is a tendency, and not s u strict

as the second law.

4 .2. ~·, hat are the marked .forms of vhe af .L'.L X ION?

For roots endlng in non-coronals, i.e. labials and

velars, the marked .L·J rm is +t+iVn. +fak and +sume

are two examples of t his. others are listed below

verb root /_+ION

d4=k+e d4=k deduce deduction

skrib skrib prescribe prescription - conceive conception sev sep

- - redeem redemption em em

sorb sorb absorb absorption

stroy struk destroy destruction

20

These are the marked non-coronal roots . By ben-lVioshe 's

laws, no verb containing any one of these roots will

have a nominal in At+iVn . This is the case . The point

about +sume having little or no intrinsic meaning

holds for these roots too, as 1well as the tendency

.{ or all verbs in a given root to have ION nominals .

A list of the verbs and IO~ nominals in +ceive and

+duce demonstrates these two independent points

quite well:

receive reception deceive deception conce l ve conception perceive perception apperceive apperception

deduce deduction reduce reduction seduce seduction

? induce induction ? conduce conduction

produce production introduce introduction reproduce reproduction

Marked coronal-final rou"ts c·,J mprise the most

irregular and complex class, especially with regard to

changes of the root. E~en the form of the marked ION

sufix after coronals is unclear. hany investigators

(cf Schnitzer(1 ~ 71 1 , Householder(1 972)) list it as

21

+t+iVn, i.e. the same suffix as with marked non-cor onals.

This cannot be so. nather the suf f ix must be +ivn,as

in SP~, f or the following rasons.

¥irstly, alternations such as rebel/rebellion,

commune/communion, demand the positing of +iVn, at

leas~ after some occurences of nasals and liquids.

Secondly, as alternat ions like decide/decision

revise/revision, argue, the vowel preceding Iu1~ must

be laxed by tri-syllabic laxing. In cases such as

abrade/abrasion, ro~ate/ro~ation, t his vowel has

further undergone a rule that tenses non-high vowels

in the following environment(S~E p.1 d1):

-low I -hi -cons V

-back -stress

This rule also operates in alternations s uch as

Canada/Canadian,Abel/~belian. Crucially, there must oe

one and only one consonant af ter t he af f ected vowel.

If the suffix in abrasion is +t+iVn, the environment

will not be met. ~here must, tnerefore, be a rule which

deletes t he ; t, bef ore the above rule applies. Since

this t-deleti on rule is not crucially ordered after

any ~-rule, the suf f ix can just as easily be +iVn,

exactly as it is in commlli4ion and rebellion. We can

thus nave a uni1orm suffix for all marked coronals.

22

Note, by the way, that the environment for the rejected

rule of t-deletion cannot be stated phonologically,but

rather in terms of marked coronal roots,or stems.

~he root ven (convene/convention) shows an

interesting interaction oi the two phenomena just

dealt with under the two argumen~s. One might be

tempted to regard the alternation of this root as

evidence of the suffix being +t+iVn after at least

some occurences of g. However, if the suffix is

+t+iVn, the t must be deleted, in most cases,before

the tensing rule applies, as I have just demonstrated.

One would, therefore ,have to mark ven as an exception

to this !-deletion rule. If, however,the root ven

has the form vent as an allomorph /_10~, there is no

need to mark this root as an exception to any rule.

The root change is also equivalent in effect to

marking ven as taKing +t+iVn, instead mf +iVn, like

communion. The t-dropping thus requires an ex~ra

marking. What,in the t+iVn SJlution was two separated

unrelated marks, thus becomes one und..1orm morphological

change, by a rule of allomorphy.I give the two

derivations,r·cr comparison.

23

Rival derivations of vention

+iVn

Morphological. markings 1.ven=vent/At+iVn

underlying forms vent+At+iVn

(mark1)

Rules of suffix At+iVn=iVn

/ marke ~ coronals (mark1)

Phonolog1cal vent+iVn rules 1.~-dropping n.a.

2. '.tri-syllabic laxing

(or laxing /vu) n.a.

3. tensing n.a.

+t+iVn

1.At+iVn=t+iVn/ven __ ( As opposed to iVn/mune __ )

2.ven is an exception to t-deletion.

ven+At+iVn (mark1, 2)

A-c+iVn=t+iVn \mark1, +mark)

ven+t+iVn

ven+t+iVn \mark2 J

ven+t.+iVn

n.a.

UUT.PU'r ven:s~n

et cetera ven~~n

24

ls~id above that ~he rule o~ !-dropping is

suspect, in that there is no P-rule which it must

!allow, and at least one ,~ensing; which it must

precede6 • There is one example in the li teratUL·e, however,

in which the t is put to some phonological use.Schnitzer '-(1971) attempts to derive succession !rom sub=Ked+t+iVn.

He uses the ! t.,O devoice the _9;. Hov,ever there is no way

.f:or him to shorten the e, and his 1inal !·arm is

suk=ses+iVn , the incorrect outpu~.,. uEE (p.1b2) ~ists

~ as exemp~ !rom the ~ensing rule; e is thUS snortened

by tri-syllabic laxing, and markedly not lengthened.

Using the device of root allomorphy, we can list ced

as ~/_+At+iVn ~_or shorten the ~ !_co. The double

consonan~ prevents the e from tensing, and is in

accora with the distributional fact that there are

no Latinate roots of the rorm *XVs+.

~hus the only caes l know 01 where the t

has been claimed to perform some phonological

runction can be aealt with in a princ~pled fashion

if the marked coronal su1fix is iVn, and if we make

use o.i hte device of allomorphy in marked roo~s

\ ben-I•wshe' s t irst law) • lu 1 act tne princip: al

01 allomorphy is very strongly supported oy both of the

last examples \the derivation of Xvention and acession).

whe.L·e t ormerly i l.ems had to be marked with rule

reatures, in one case several unrelated rule features ,

the same effect can be gai~ed by a simple rule of

2)

allomorphy, whicn serves in addition ~o mark ~he root

in question.

4.4. Hartin's observations

Beiore aelving urther into coronals, we

must maKe a short siae-trip, and look into t he c ;nclus­

ions or some wor k uone by :::>usan 1·1art1n ~ 1972), which • ...

will be 01. great nelp 1n the analysis ot the data.

Martin observes tnat derived words in t he s ,!: t ixes

lUN,-ive,-ory,=£!, are built on what seems to be the

same form of a given root, a .r orm which, as we have

seen in t he case of lON, of ten differs fr om that of

the stem verb. The following alternations de monstrate

this point:

divide

compel

subvert assert

retain

accuse

excrete apprehend

division compuls i on

subvers ion

assertion

divisive compulsive

s ubversive

assertive

re t ention retentive

accusation ~accysative)

excretion apprehension a ppr ehensive

percuss percus s ion percussive

divisor compulsory

accusatory

excretory

Let us accept,then, in the light o!: theabove alternations,

that the various suffixes are attached to t he same allo-

morph of a given stem,i.e. to t ne same underlying phono­

logical form. Since t here is so much neutralization of

26

contrast in the lU~ forms, we may now discover the

underlying rinal consonant(s) of a : r root / ION -~ -

by inspection of the corresponding +ive or+ory

form.

Martin also maKes the i'urther point t .t1at

in general the IO~ suffixea word is primary. One

very sel dom finds a +ive or +ory word which does

not have a corresponding +IO~ form, and, even

historically, of the ION +ive pairs, the . ION form

can generally be shown to have entered the language

before the +ive form. This second point of Nartin's

proves to be crucial when we try to ~ ormalize the first

one,i.e. at what s t age in a derivation do the IO~ and

+ive f orms have the same final consonant(s)?

There is a ;· great deal of allomorphy

exhibited by coronal f inal stems. I will discuss

as best I can all the al~ernations, and have,as a

prelude to this,listed them inthe table below. The

table may be used as a reference in t.t1e discussion

that fol l ows.

27

Table of marked coronal alternations

Samnle verbs verb-f inal -· c I +iVn I +ive

excrete ,X+sert t t t

X+mit (permit) t t s

X+vert (convert) t t s

digest st s~ st

connect kt k~ kt

decide,explode d t s

X+ cede (concede) d ~ s

apprehend nd n~ ns

commune n n

scan n nt

convene ,retain n nt nt

prevent nt n~ nt

recense ns nt

coerce rs rt rs

disperse rs rt./rt rs

submerge,asperge rdt. rt. rs

adhere r t s

recur r rt. rs

rebel 1 1

.X.+pel ~expel) 1 lt ls

convulse ls 1~ ls

revise z t percuss s t s

admonition t ~ t

28

ti e can extract from this list some very general properties of the /_ION and /_ive forms.

Firstly, of the full consonants, only ~ and t occur

before +ive. The absence of any VJiced full conso-

nants before this suffix can be easily captured by the

following ad-hoc rule:

C = -voice/ __ +ive

The fact that there are no voiceless counterparts to

1 and g might account for two interesting gaps in our

paradigm. There are no cases of *Xlive or *Xnive.

communion has no corresponding adjec~ive, and rebellion

has rebellious. Just as with abatement and statement,

the morphology avoids a phonological dilemma.

~he second general fac~ to be noted is that

except after 1 and g, +iVn is preceded only by the pala~als

~,~,~. This is the result of palatalization, an apparently

simple process (but see below, and SP~ 229-231).

Looking at the alternations, we note only

eight cases where the .final consonant (cluster) of the

verbal stem is in a one-to-one correspondence (one way)

with the c .nsonant(s) preceding ION and+ive. These are

~,Vz,st,kt,nt,nd,~,ls. Except for nd, all these have

exactly the same consonant before +ive as they do word­

finally. This suggests that whatever alternation is

manifested I iVn, can be accounted for phonologically •

~he general correspondence is

z/<t, t/~ s/~

29

After!! however, we have t/~ (question). these l"acts can

be handled by the following simple rules:

palatalization 1

palatalization 2

t=~/s_yV

+C(tllS

-voc = -ant I +strid _yv

Palatalization before lY . seems . to be a very general

phenomenon;however, the a~tempt to state it as one rule

runs into many phonological problems, which 1 will not

d1scuss here. r1lost fo -che relevant !"acts can be found

in ~PE (229-235).

Turning now to nd, we find the following:

-n~4:n# -nsiv#

~her are several roots in -nd

+hend +tend +pand +cend

apprehend, comprehend pre~end, contend, ex~end expand ascend

n54:n# ~ells us that the nominal/adjec-cival stem must be

either Xnt or Xns. nsiv# tells us that it must be Xns.

lfl.J.!j may therefore posit the following rule:

d= s/n ~on -lve

~his may not be the proper way to state the rule, and

1 will re~urn to it later.

~-final stems are curious:

abolish abolition admonish admoni ti.on punish

admonitive punitive

30

~he fact that we have ~/ __ +ive shows that these stems

have a combina~ry variant Xt. Here is a case where

roots show the same sur1ace segment ~ in two environ­

ments, but where . there is good evidence that these

two segments must be deriveu from two distinct

underlying segments, in distinct allomorphs. One might

try to state the ~ to t rule phonologically. Note

however that the change only takes place before iVn

and ive, i.e. in a morphological environment. ~his is

important.

t-final stems show interesting alternations.

~he must common one is t/~/t (excrete,assert,transitJ

This involves no morphological change of any sort.

with the root +mit, however, we .f i nd t1~/s (permit,

permissive,permission). There are no cases of -~~.x.mitive

or Xmitative, facts which, we recall, are predi~ted by

ben-Moshe's laws. This can be accoun~ed for by pos i ting

for +mit the allomorph +miss / __ Io~.

+Vert is another

root which exhibits allomorphy. ~nbtead of the expec ~ ed

subver~~n,subver~iv, we ge~ subver~~n,subversiv. The

occurence of : .1: tells us that the relevant allomorpn

ml.lst be ~·

r~/r1J:

31

.!!'rom vert we may turn to other cases of

coerce disperse immerse emerge asperge/se submerge/se deterge

coer~.fn

di sper 1J.fn (? ~.fn)

immer ~.fn (? 1J.fn) emer1J.fn(*~.fn)

asper1J.fn(*~.fn)

submer1J.fn(*~.fn)

deter1J.fn(*~~n)

coersiv

detersiv

With stems in xrge we always nave r1J. The s in detersive suggests a rule similsr to that involving Xnd:

g =z/_IO.N

Note that this orthographic g is most likely velar

underlyingly (purge/purgaii.on). If we wish to preserve

the uniformity of the statement that +iVn attaches

to coronals and +t+iVn to non-coronals, in marked

stems, then the g to z rule must apply before ~his state­

men~ can be made. rtowever, the g to z rule seems to be

phonologically mo~i vated by the glide 01 +1:..:!.£, just~ · ..

as wi~n the nd to~ rule, and thus to follow the suffix

allomorpny rule or ION to +ivn. ~noosing this ~atter order

would rorce us to state the distribution rule as

IO~ =iVn 1af~er marked coronals, ana illarkedg (not lf)

The distribution rule canno~ be stated ar~er g to !'

sine it feeds i~ ,ir the~ is indeed the p~onological

trigger.We thus have the follow.hg alternative solutions:

32

1

1. g=z (allomorphy)

~. IO~ =ivu/coronals __

2

lO~=iVn (with skewea en vir onmen t.)

g=z (pnonological)

The existence OI back forms asperse,submerse, is

relevant here, tor 1t shows tha~ the g to~ rule is

opaque. disperse,immerse are also back-formations.

uispersion is M~, disperse 1450, immersion is 1450,

immerse 1650. This is aaaea eviaence for the opaci~i or

g to ~' which leads me to support the hypothesis

expressea by the oraering in \1) 1•

The two roots in r are good examples or roots

marked by allomorphy. Lhey are the onli r-Iinal roots

~hat take IO.L~, and are both marked by unpredic~able

allomorphy (.kur/kurz,her/hez). +ne:re is also a good

example ot· the ex~ension or a smi ix to all verbs in a

given root:

inhere adhere cohere

inhesion adnesion conesion

innesive adhesive conesive

+tain/+ten~ (retain/retetion) is curious. lt.

shows a vowel cnange;we would expect *ret.antio11. uP~\p.202;

accounlis tor th1s change by making +~ain undergo the

short vowel shift rule, as was done witn ra.k. ~his seems

53

to be another unwarran1ied. use of a rule i'eature

minor rue pair. Au inspection ot all the items that

undergo ~v~ shows that this putatively general process

applies only to items which are marked. bya non-abs"tract

rule Ieature vO undergo it.

4.6. case,

1 have lei't i'or last tne most regular and. puzzling

the £-i'inal roo"ts. ~'t i'irst glance, they appear

to be no puzzle at all. Except for the root cede, which

was discussed above, we always have the same alternation.

£/~/~ (decid.e/uec~sion/decisive; • Thre is a rule "taking

! "to ~ in some environmen"t. ~he question is, as in many

cases above, how to state the environment, and what sort

of' rule we are dealing with, phonological or morpological.

Keeping in mind Martin s observation ~hat

+ive and +ory, are secondary, and Iu~ is primary,we may

disregard. +ive ana +ory dS environments i'or the rule's

application. ~ne rule appl~s bei'ore 10~ orivn, and the

secondary suliixes take their stems i'rom the output of

'tnat rule. This way of dealing wi1..h the change allows

us both to restrict ~ts environment, and Iormalize the

notion of seconaary sufiix.

~o d=z /_101'-l or~ .n. ls ~ t the sUIIix or the

i tnat conditions this change,i.e. is the rule morphol­

ogically or phonologically conditioned? Generally,£

aoes not appear bei'ore Y:.- (i=y in lVl'l, as noted above).

34

ln fact, the rule that takes i "tOY:. after coronals (..::>_t)~ p.225)

is idiosyncratically blG>cked by ,9; (pavilion/enchiridion).

1n the only case v11ere a.+y does arise, mther than bet ore

ION , we get~~ (cordia}) Thus the rule that takes d to

~ applies only before IO~/ivn, am its derivatives.~ote

in ada.i "tion, that ~ a.oes not spiran tize before IO .....

1nere is no rule taking t to ~' fer the .! appears before

+ive (secretive/decisive).

Thus, the change, if phonological, is highly

marke~. lt can just as easily be stated with a

morphological environment, and, since it follows no

~-rule, as a rule of allomorphy. Lhis rule can also,

incidentally, be applied to the instances of lli!

discussea. earlier. Such an extension would necessitate

a rule devoicing ~/n_. ~his latter rule does seem to be

well motivated. Lhere is only one case of word fmnal

~' bronze, and no cases oi lz. X+pulsion,propuls~n)

and a+vulsion(revul~~n).

The critical hand is rising again. 1 have

argued now, in several dEtinet instances, against a

phonological rule, and have posited instead a rule

of allomorphy, which applies to more than one root,

nd to ~'~ to t, g to ~, £ to ~· But these cases are

really the e<n>e of the paJW_er. It appears, from Nartin' s

remarks, that a speaker, when coining a word in +ive

35

or +ory, takes the requisite word in ION, and

performs a kind of unravelling of phonological rules,

until they can go no further. It is tm this last

form, the product of the last unravelling,that they

add the suffix +ive,~+ory. Everything beyond is

opaque, unavailable. By positing the above rules as

morphological rather than phonological rules, I am

claiming that this last 1'\l'm is significant, it is

the ultimate and first form, the basic phonological

form. ~he rules that prod~e this form, are, I claim,

in every case of an entirely different sort from the

P-rules. ~hey are morphological, and their environments

are stated morphologically. I have also shown that if

we always sta~ these rules as rules of allomorphy, we

gain something. We can preserve the generality and

position of the statement of suffix alternation • .Now

any one of these points may be challenged, particularly

these last rules o1' allomorphy; but they cohere.

'.rake just one away, and the others 1·a11.

5.1. ~here remains only a little mopping up to be

done. Two more possible variants o1' ION are in need of

comment. ~hey are +it+iVn and +~t+iVn. Evidence i or the

first is the following set of alternations:

add ?rend vend define .X.+pose compete repeat imbibe

addition rendition vendition definition .X.+posi tion competition repetition imbibition

36.

additive

definitive .X.ot-positive/ory competitive repetitive

~hese can be handled in one of twm ways. ~ither we mark these rules for a special suffix

IO~o it+iVnl root+M

orwe change the root by adding !!· So, for example

add=addit/ __ IO~

The first solution entails having more than one marked

suffix. A given markeo root will either ~ake (t)+iVn

arit+iVn. and will have to be markedfor the given variant.

~his is more complicated than the second . Other than that

1 can see no !actor that determines a choice here.

5.2. :J.:he possibility of there being a suffix

+~t+iVn is provided by the following:

revolve resolve dissolve solve

revolution resolution dissolution solution

The two roots are odd, both eding in lv. One could

simply mark them . ~he suffix would then be t+iVn,

(after a non~cornal), and we would nave, for example,

revolvtion. ~he v could then go to u /_t.

37.

Alternatively, we could have a suffix

~t+iVn, and drop the v. The first s olution seems slightly

better. we can use a sUf f ix we have already. Both require

an ad-hoc rule.

.~ , > t I '-i..,.. I

l tf~) /

38

~-- OOT.N OTES

1. The sole exception to the stress condition is burial.

Exce~ions to the second are trial,rental,cital,quittal,

with mono-syllabic stems, and presupposal,disavowal,

disapproval witn tri-syllabic stems. ~ote that in the

latter cases all are prefixed, and there exist supeosal

avowal,approval.

2. The indirect object case is no longer productive, and

words formed in it are felt to be somewha t obsolete.

3. The reader should note that such a morphological

property as +Latinate is the pr operty not of a phono­

logical form, but of a morpheme. l or example there are

in English two homophomous morphemes ride. The one is

-Latinate, and appears in the verbs ride,override.

the other is +Latinate, and appears in the verb deride •

The suff ix ION attaches to +Latinate rCots, am theref~

may attach to the latter to form derision, but not to

the f ormer, *overrision. I owe this poin"t to l"larchand

(1 960).

4. See Shelvador (fo rthcoming)

5. The "teleological nature of this phenomenon has be en

taken by E. S .Williams as a sign of the hand of God at

work. The gap can be explained, he no"tes,only as the

result of an ac"tive in"telligence's wish to confuse the

investigator,

39.

6. ~he comparison of digest/digestion with fwt/factual

will reveal another such rule (cf. SP~ 230-233).

7. Some readers may s corn the use of philology, and

historbal information aoout the chicken and the egg

in arguments about synchronie description. no .vever, we

are dealing here with frozen, learneu, Latinate vocab­

ulary, not with any general or producrtive phenomenon ,

and in such cases, whatthe linguist often sees as a

general1zation, may in fact be a mere artifact.

40

BIBLIOGRAPlH

1. Chomsky,A.N. and M. Halle (1968) The sound Pattern

of English,New york 2. nale,Kenneth(196?) Paper on the Maori passive 3. Householder,F .w.(1 972) A Problem in rule ordering???'

in Linguistic Inguiry(III,3) 4. Kiparsky, Paul ( 1968) How Abstract is Phonology ("-'4s) 5. Marchand,Hans (1960) The 0ategories and ~ypes of

Present-lJay English \•lord iormation , Wiesbaden 6. Martin, Susan -ive arl other -ion based suf1ixes(Ms) 1. Prince, Alan s. IC (Ms) 8. Schni~zer, Marc. L. A problem in rule ordering

in Linguistic Inquiry (11,3) 9. Shelvador,0rosley(fotthcoming) The phonetics of = and+ 10. :::>iegel, Dorothy Some Lexical 'l'ransderivational

Constraints in ~nglish (ms)