TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

24
TURKISH fc-DELETION: SIMPLICITY VS. RETRIEVAL STIG ELIASSON ABSTRACT A phonological model which integratos the study of phonological structure with a competence System of cognitive operations offers a principled account of a crux in linguistic theory. The preference for a concrete rather than an abstract phonological solution for the case of k/0 alternations in Modern Standard Turkish is seen to result from a requirement that underlying repre- sentations be retrievable, i.e. retraceable on the basis of phonological rules in close conjunction with precisely circumscribed and formalizable cognitive operations.* 1. THE SIMPLICITY METRIC OF GENERATIVE PHONOLOGY The notion of a simplicity metric or evaluation measure figures prominently in generative grammar. CHOMSKY & HALLE (1968: 334) formulate such a metric for rules in the following way: (1) The Value' of a sequence of rules is the reciprocal of the number of Symbols in its minimal representation. In the bulk of 8PE (CHOMSKY & HALLE 1968, chs. 1—8), the simplicity metric in conjunction with rule ordering conventions and various abbreviatory devices is assigned a fundamental role in choosing between alternative descriptions within one and the same theoretical framework. However, the authors indicate that in the course of their work they discovered various deficiencies in this formulation, and in tho final chapter of 8PE (eh. 9), they propose to remedy.these shoricomings by integrating the simplicity metric with a theory of märkedness and, furthermore, by intro- ducing a set of auxiliary principles such äs a hierarchy condition and various symmetry conditions for phonological Systems to complement the simplicity metric (ibid. 410). Additional principles 0165—4004/86/0019-289 $ 2,— © Mouton Pubtishers, The Hague; Societas Linguistica Europaea Brought to you by | University of Glasgow Library Authenticated | 130.209.6.50 Download Date | 9/5/13 11:03 PM

Transcript of TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

Page 1: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

TURKISH fc-DELETION: SIMPLICITYVS. RETRIEVAL

STIG ELIASSON

ABSTRACT

A phonological model which integratos the study of phonological structurewith a competence System of cognitive operations offers a principled accountof a crux in linguistic theory. The preference for a concrete rather than anabstract phonological solution for the case of k/0 alternations in ModernStandard Turkish is seen to result from a requirement that underlying repre-sentations be retrievable, i.e. retraceable on the basis of phonological rulesin close conjunction with precisely circumscribed and formalizable cognitiveoperations.*

1. THE SIMPLICITY METRIC OF GENERATIVE PHONOLOGY

The notion of a simplicity metric or evaluation measure figuresprominently in generative grammar. CHOMSKY & HALLE (1968:334) formulate such a metric for rules in the following way:

(1) The Value' of a sequence of rules is the reciprocal of thenumber of Symbols in its minimal representation.

In the bulk of 8PE (CHOMSKY & HALLE 1968, chs. 1—8), thesimplicity metric in conjunction with rule ordering conventionsand various abbreviatory devices is assigned a fundamental rolein choosing between alternative descriptions within one and thesame theoretical framework. However, the authors indicate thatin the course of their work they discovered various deficiencies inthis formulation, and in tho final chapter of 8PE (eh. 9), theypropose to remedy.these shoricomings by integrating the simplicitymetric with a theory of märkedness and, furthermore, by intro-ducing a set of auxiliary principles such äs a hierarchy conditionand various symmetry conditions for phonological Systems tocomplement the simplicity metric (ibid. 410). Additional principles

0165—4004/86/0019-289 $ 2,—© Mouton Pubtishers, The Hague;

Societas Linguistica Europaea

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 2: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

290

for favoring particular Solutions were later proposed by KIPABSKY(1968,1971, 1972, 1973), notably his alternation condition, opacityprinciple, and notions of paradigmatic uniformity. Recently, how-ever,there has been a turn away from such supplementary principlesback to the original formulation in SPE, chs. 1—8, äs evidencede.g. in two crucial studies by HALLE (1979) and KEPARSKY (1978).Under the rubric of form vs. function, HALLE critically examinessome descriptive phonological problems for which functional expla-nations have been invoked in the literature, and he concludes thateach of these problems can be settled by means of the simplicitymetric alone. In a similar vein, KIPABSKY argues that the simplic-ity metric will explain the course of certain historical developmentsand that such supposedly functional considerations äs rule opacityand paradigmatic uniformity belong to a separate, non-linguistic ornon-structural, component of cognitive organization. In what fol-lows, I will reexamine one of HALLE'S examples and suggest thatat least in this instance the explanation lies, not in simplicity, butrather in a notion of retrievability, which may be viewed äs anatural revision and extension of fundamental principles in genera-tive theory.

2. THE &/0-ALTERNATION IN TURKISH

The case I will discuss here is the phonetic alternation betweenJe and 0 in Turkish.1 We may first recall the familiär fact that,while some morpheme-final stops and affricates in Turkish are con-sistently voiceless (2a), others alternate in regard to voicing (2b).2

(2) Absolutesingular

(a) topatsac

(b) mektupöeäitaae

Absoluteplural

top+larat+larsac+lar

mektup+larcesit+leraaö+lar

With 3rd pers.poss. suffix

top-f-uat+isaö+i

mektub+uöeäid+iaaj+i

'baU1

'horse''hair'

'letter''küid, sort*4tree'

All the root morphem^ in (2) involve final non-velar consonants.The Situation in the velar stoßs is süghtly more complex. As in the

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 3: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

291

other consonant series, there are in the velar series instances ofstops that are always voiceless (3a), and stops that alternate withrespect to voicing (3b), but the latter alternation arises only after/n/. For the rest, /k/ alternates with 0 (3c).

(3) (a) ekokytik

(b) renkfrank

(c) köpekkulaksokaktoprakucakkelebek

ek+lerok+laryük+ler

renk+lerfrank +ler

köpek+Ierkulak -f-larsokak +lartoprak+larucak+larkelebek +ler

ek+iok+uyük+ü

reng+ifrang+köpe+ikula+isoka+itopra+iußa+ikelebe+i

'affix''arrow''bürden*

'color''franc'

'dog''ear''street''land, earth*'airplane''butterfly'

Schematically, we may sum up the patterns just mentioned äs inFig. 1.

Non-alternating AlternatingVoiced Voiceless Zero

bd

'

(only after/n/)

k 0 l (elsewhere)

FIG. 1. Basic alternation patterns of morpheme-final stops and affricatesin Turkwh

3. PHONOLOGICAL SOLUTIONS

As a background to our later discussion let us now briefly reviewthe essentials of some of the phonological Solutions that have beenadvanced for the &/0-alternations in Turkish. As is well known, LEES(1961: 7—9) put forth an analysis of the above data where he lined

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 4: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

292

np the &/0-alternations in (3c) with the voiced/voiceless alternationsin (2b) and (3b). He accounted for the voiced/voiceless alternationsby means of a rule of devoicing which operates before a wordboundary or before a consonant. Furthermore, for morphemesexhibiting phonetic alternations between k and 0, he postulatedan abstract underlying //g//.3 This //g//, which never appears pho-netically in the environments under consideration, becomes voice-less before a word boundary or a consonant by the devoicing rulejust referred to and drops before a vowel by the following pair ofrules (ibid. 52, 56, 68, 70) :4

(4) (a) Conversion of //g// to |g| (Voiced velar softening')

(b) Intervocalic ^-deletionVg(+)V-V(+)V

LEES' reason for formulating (4) in two steps is that he is therebyable to tie in his analysis of final &/0-alternations with his assump-tion of another abstract underlying segment, the velar glide //g//,which always disappears phonetically.

Disregarding LEES' idea of an underlying //g//, ZIMMER (1975),in a paper that has attracted considerable attention, reformulatesthe rules in (4) äs a rule of ^-deletion (ibid. 557)

(5) g-0/V_ + V

and goes on to challenge LEES' analysis of Turkish A/0-alternations,whether in the form of (4) or (5). ZIMMER, in contrast to LEES,suggests that there exists a phonetically overt basis for the alterna-tions. Leaning on an observation cited by JANSKY (1954: 22f.)(also cf. KISSUNG 1960: 22 f., KONONOV 1956: 47, LEWIS 1967: 10,and DENY 1921 : 145, 147), he shows that the appearance of alterna-tion is correlated with the number of syllables in the forms. Forwhereas monosyllables generally remain unchanged, äs exemplifiedin (3a) above, polysyllableß commonly alternate> äs (3c) illustrates.This structural regularity is, in addition, extended to many loan-words in Turkish; ZIMMER cites words such äs bek, bek+i *back(in soccer)' without alternation and frikik, friki+i 'free kick (insoccer)' with the k/0 change. He further notes that deviations fromthis basic pattern for polysyllables tand to fall into certain definite

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 5: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

293

classes, a point later refined by SEZER (1981) (cf. also section 4below). Hence, for ZIMMER the crucial difference between non-alter-nating and alternating morphemes in final k is not that the formerend in /k/ underlyingly and the latter in underlying //g//·, but ratherthat all end in /k/ phonologically and that monosyllables retaintheir k's, while polysyllables typically drop them. For LEES' rules(4a) and (4b) or the alternative in (5), ZIMMER (ibid. 561) accord-ingly substitutes the rule

(6) k-0/VC0V_ + V

which says that final k's of polysyllabic forms drop before a fol-lowing morpheme beginning with a vowel. Moreover, in a separatestudy, ZIMMER & ABBOTT (1978) tested native Speakers for theirtreatment of Turkish-sounding non-sense words ending in k andfound that their subjects tended to preserve the k in monosyllablesand delete it in polysyllables, thus providing important psycho-Imguistie support for ZIMMER'S structurally motivated solution.

In view of the obvious distributional basis for phonetic i/0-alter-nations, ZIMMER does suggest, however, that it is possible to amendLEES' <7~analysis by adding a phonotactic condition which saysthat "post-vocalic morpheme-final /k/ occurs only in monosyllabicmorphemes, and post-vocalic morpheme-final /g/ only in polysyllabicones" (ibid.). Let us refer to this solution äs the 'amended ^-anal-ysis'. An actual solution essentially along these lines is implied by

(1976: 42):

"Any word of more than one syllable that is found in the dictionarywith final k almost always ends basically with /g/. For this reason,we will normally not need to indicate final /g/ in transcription: if yousee, for example, the word küpük 'sinalP, you can assume that its basicform is /kügüg/. On the other hand, words of one syllable with finalJe normally really end in /k/. There are three one-syllable words withfinal /g/: pok (/gog/) 'much', gök (/gög/) 'sky', and yok (/yog/) 'non-existent'." . . . , ' ' i

;4. EV^TJATIQN

Proceeding now to compare and evalüate the amended ̂ -analysisand the A-analysis, ZIMMER submits that it is not possible to choosebetween these by means of the 'customary simplicity ineasure ofgenerative phonology. In terms of the evaluation metric, rule (5) ofthe amended <7-analysis is, he says, somewhat simpler than rule (6) of

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 6: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

294

the &-analysis, but, on the other band, the phonotactic conditionof the amended ^-analysis will be more complex than that of thei-analysis which states merely that "/g/ does not occur morpheme-finally after vowels" (ibid. 561). So instead of resorting to formalsimplicity to account for the choice of solution, ZIMMER seeks theexplanation in the notion of function, suggesting that ease oflearn-ingmaybethedecisivefactor and that "the k-analysis does every-thhig that needs to be done in functional terms" (ibid. 562).

HALLE (1979) takes issue with ZIMMER'S (1975: 556) statementthat the Turkish data illustrate a type of Situation not handledeither by the generative simplicity metric itself or the metric inconjunction with supplementary principles. HALLE (1979: 328)"attempt[s] to show . . . that ZIMMER ismistaken in this assertion>that the old simplicity measure does not only address itself to thisproblem, but in fact forces the solution that ZIMMER regards ästhe correct one". First, HALLE (ibid. 329) presents the phonologicalrules required by the amended ^-analysis and the &-analysis. Sincehis exact formulation of these rules will be relevant shortly, wereproduce them here äs (7) and (8), these being the feature formula-tions of (5) and (6), respectively.

(7)

—sonorant—coronal—labial-f-voice

(8) [-son, -cor, -lab] -> 0 / [+syl] [-Syl]0[+syl]_ [+syl]

His main argument, however, zeroes in on the lexical redundancyconditions discussed by ZIMMER, which HALLE (ibid. 330) formulatesexplicitly äs in (9) and (10).· —- sonl

~cor-labj

f [-voice] / * ([-syl]) [+syl]_ +

[t+voice]/ [+syl] [-syl]0 [+syl]_

(10) [—son, —cor, —lab] -* [—voice]/ [+syl] +

(9), then, is the redundancy condition associated with the amended0r-analysis, while (10) goes with ZIMMER'S &-solution. HALLE pointsout that these redundancy conditions must be included in the

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 7: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

295

evaluation of the two alternative phonological Solutions. Comparingthe phonological rulesalone, the amended gr-analysis will be simplerthan the Ä-solution. But if we consider redundancy äs well, we willhave the Situation summed up here in Table 1.

TABLE lThe complexity of two rule formulations for k/0-alternations in Turkish

according the simplicity metric in SPE

Complexity of phonological ruleComplexity of special redundancy

condition

Total complexity:

The amended(7-analysis

[including rule (7)and condition (9)]

6

10

16

The Ä-analysis[including rule (8)and condition (10)]

7

5

12

Table l shows that the £-analysis is simpler than the amended<7-analysis by four features, and, consequently, the SPE simplicitymeasure selects the solution favored on other grounds. HALLE(ibid. 331) therefore "concludefs] that the Turkish example providesno evidence whatever for the proposition that purely formal con-siderations may be overridden by considerations of a functionalsort"*

While not in the first place concerned with the matter of formversus function, SEZEB (1981) investigates the Turkish problemfurther and provides essential new data. First, he shows that theredundancy condition of the amended ^-analysis will in fact haveto be even more complex than had earlier been assumed, since post-vocalic final k is retained not only in monosyllables, but also inpolysyllables after long vowels. Thus, although k drops after theshort vowel in (lla), it stays after the long vowel in the variantform in (llb) (ibid. 359).

(11) Absolute(a) tastik(b) tastik

Possessivetasti+itastik+i

'certification'(same)

Accordingly, SEZEB suggests that the redundancy condition of theamended ^-analysis, which we reproduced in (9) above, will have

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 8: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

296

to be elaborated in the following way (ibid. 362; emendationsitalicized here):

(12) a. Morpheme-final post-vocalic /k/ occurs:1. in monosyllabic morphemes2. in polysyllabic morphemes after long vowels;

b. Morpheme-final post-vocalic /g/ occurs in polysyllabic mor-phemes after short vowels

In the &-solution, on the other hand, the feature [—long] willhave to be added to the 4-deletion rule so that the rule takes theform (13) (ibid. 360).5

(13)

—son—cor-lab—voiced

Furthermore, SEZER adduces certain data that, he submits, createinsurmountable difficulties for any kiiid of ^-analysis. All thesedata involve morpheme-final post-vocalic k9 s which drop under theusual cönditions except befpre vowel-initial borrowed suffixes andbefore the auxiliaries olmak 'to be, to become' and etmek 4to do'which elsewhere behave like native vowel-initial suffixes. Cf. (14).

(14) Citation form k dropped k retained

(a) meslek mesle+im meslek+'profession' 'my profession' 'professionaF

nazik näzi-|-im näzik+en'kind' am kindV 'kindly'

(b) destek deste+i destek+olmak'support' (n.) 'his support" 'suppOrt*

yasak yasa+i yasak+etmek'prohibition' 'his prohibitionj 'prohibit^

SEZEB mentions that one might attempt to handle the unexpectedi-retöntiori in the columii to t&e right in (14) by representing theforeign suffixes and thetiative atixiliaries wfth apreceding ==bound*

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 9: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

297

ary which would be independently justified by the fact that theseitems do not undergo vowel harmony. We would then have

(15) meslek=inäzik=endestek=olmakyasak=etmek

and the =boundary could be made to trigger final devoicing ofunderlying //g// under the gr-analysis or prevent &-dropping uijderthe i-analysis. However, SEZER suggests that the representationwith a =boundary is not really open to the gr-analysis, since simplyadding the =boundary to the environment of the final devoicingrule would erroneously induce devoicing of morpheme-final /b, d, j/before the non-native and auxiliary Suffixes (cf. kalbten 'heartily',not *kalp=en,a,ndharäb=olmak 'to be ruined', not *haräp=olmak).The (/-analysis and the &-analysis therefore 'do not display equallevels of empirical adequacy' (ibid. 357). Hence, according to SEZERthe correctness of ZIMMER'S analysis "can be argued exclusively onformal grounds without recourse to functional considerations"(ibid. 375).

5. SIMPLICITY AND THE GRAMMAR

The above conclusions about the importance of form would pre-sumably seem quite appealing in almost any brand of formal pho-nology. Nevertheless, some problems do remain. For one thing,the new data in (14) do not actually eliminate the ̂ -analysis. Whilethe differing behavior of non-velar and velar non-continuantsbefore foreign Suffixes and the auxiliaries is certainly a considera-tion that weighs heavily against the ^-analysis, this analysis canstill be maintained. The formalism of generative phonology caneasily capture such an asymmetry by incorporating its angledbrackets notation into the final devoicing rule, äs shown in (16).

(16) [—voiced]/

soncontant\cor/b-Condition: If a, then b

#—\ l—/a |c

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 10: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

298

Thus we still have to wrestle with the theoretical possibility of a0r-analysis. In addition, it appears that there is reason to doubtcertain assumptions underlying the attempt to settle the choice ofsolution on the basis of simplicity. In particular, I want to questionthe morpheme structure condition (MSC) discussed above on whichthese conclusions are crucially based.

Although ZIMMER (1975), for ease of reference, calls the statementexpressed by (9) above a "morpheme structure condition", he indi-cates quite clearly that it is not in fact a statement about mor-pheme structure. Rather, he says that "strictly speaking this isnot an MSC, but something like a Vord structure condition', sincefinal velars are deleted in monosyllabic äs well äs in polysyllabicSuffixes" (ibid. 565). Some examples of this, partly extracted fromSWIFT (1963, eh. 3), are reproduced in (17).

(17) Suffix Base Derived form Possessive

(a) (a)k

(b) lik

(c) jak

top'ball'

erkek'male'

oyungame

(d) mak ye'eat'

top+ak'dough ball'

erkek+lik'masculinity'

oyun+jak'toy'

ye+mek'food'

(e) dik tani tani+dik'recognize' 'acquaintance'

(f) (y)ajak yaz'write'

yaz+ajakVriting-in-the-future'

top+a+i'his dough ball*

erkek+H+i'bis masculinity'

oyun+ja+i'his toy'

ye+me+i'bis food'

tani+di+i'his acquaintance'

yaz+a}a+i'his writing-in-the-future'

But to say that the redundancy condition in question is basicallya word structure condition means in effect that it behaves like aphonological rule. The inevitable conclusion is therefore that NOMSC of the special form (9) should be added to LEES' solution.Thus we either have a solution with a contrast betweeh morpheme-final /k/ and //g// and no MSC (the simple gr-analysis) or a solutionwith an MSC and no contrast between /k/ and //g// (the &-analysis).

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 11: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

299

However, if the MSC vanishes and the simple ^-analysis re-emerges äs the basic internally-based alternative to the &-solution,we face a difficulty with respect to the evaluation measure, äs Table2 indicates.

TABLB 2

The complexity of LEES' and ZIMMEB'S rule formulations for Turkishk/0 alternations according to the simplicity metric in SPE

The simpleo" änälysis

[including rule (7)]

The Ä-analysis[including rule (8)

and MSC (10)]

Complexity of phonological ruleComplexity of morpheme structure

condition

Total complexity: 12

For in terms of the number of distinctive feature specificationsneeded in phonological rules and morpheme structure conditions,the simple ̂ -analysis will turn out to be less costly than the jfc-anal-ysis. As things stand, then, the evaluation metric will select the<7-analysis, which we do not want, and discard the i-analysis,which, äs ZIMMEB demonstrates, is the solution that must befavored on other grounds.6 Hence, simplicity will not to our satisfac-tion settle this particular portion of the analysis.

6. SIMPLICITY AND THE LEXICON

In view of this startling result we may ponder the possibilitythat we have not taken into account all things that should beconsidered in applying the evaluation measure. Indeed, even thoughHALLE does not mention this in his 1979 paper, it has been repeat-edly emphasized elsewhere that the evaluation measure must beapplied to the total grammar, including the lexicon. For instance,CHOMSKY & HALLE (1968: 297) state that "we postulate a set oflexical matrices and a System of phonological rules which JOINTLYmaximize value" (emphasis added).7

In the Turkish case, when we include the lexicon in the evalua-tion, the relative complexity of the two competing Solutions willbe quite different from what we saw before. LEES' solution involvesa distinction between morpheme-final /k/ and //g// which has tö be

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 12: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

300

marked for every lexical entry in question by means of the specifi-cations [—voiced] and [+voiced], respectively. ZIMMER'S solution,on the other band, just leaves out the specification of the value ofthe feature [voiced] for morpheme-final velar stops in the lexiconand later supplies the appropriate value by means of MSC (10).Since the feature savings in the lexicon will be quite substantial(cf. VIETZE et al. 1975: 69—123 passim), ZIMMER'S solution willconsequently be much more economical than the one proposed byLEE s.

But a problem still remains. In his 1979 article, HALLE alsocomments on the widely discussed case of the passive and gerundiveallomorphy in Maori (HALLE 1979: 326—328). As we recall, theissue there is whether certain consonants that appear in the passiveand gerundive forms should be assigned to the verb stem or to theending. For instance, do active-passive pairs like niki — nikitia'lift', tomo — tomokia 'begin', kimi — kimihia 'look for', andapiti — apitiria 'add' reflect word-final consonant deletion in theactive (C -> 0 / *) and the addition of a suffix -ia in the passive(a so-called phonological solution), or do these pairs involve verbstems invariably ending in vowels and diacritically marked for theparticular passive allomorph -tia, -kia, -hia, -ria, etc., thatthey take (a lexical solution) ? Contrasting the lexical solution,which he prefers, with the phönological solution, HALLE mentionsthat the lexical solution gets rid of the rule of consonant deletionand regularizes the morpheme structure of the language, and headds (ibid. 327 f.):

"Arrayed against this is, of course, the additional cos t of a complicatedand ad hoc rule governing the distribution of the two suffixes. Al-though there does not exist at present a proposal of how to convertthese eonsiderations into a mechanical evaluation prpcedure, such anevaluation is surely not unimaginable and has impficitly been utilizedquite widely. Consider, for example, the treatment of the verb to bein phönological descriptions of English. It is invariably treated ästotally exceptional and attempts are not made to integrate it/into thephönological and morphological rules of the language. Whatever sim-plicity considerations underlie this decision — which to most linguistsis so uncontroversial äs not even to merit serious discussion — willalso correetly handle the case of the Maori passive."

It is interesting to note here that, exactly äs in the Turkish case,HALLE considers only the complexity of rules, not the complexityof lexical matrices. Moreover, taking lexical complexity into accountin the Maori case turns out to yield a rather different result than

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 13: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

301

in the Turkish one. Even if we were to grant MCÖARTHY'S (1981: 241)contention that the number of feature specifications of the totalset of rules required under the phonological solution to the Maoriproblem (a little over "approximately 49 feature specifications")exceeds the cost of the rules required under the lexical solution("a total of 30 feature specifications", "counting a stipulation ofa morpheme äs the equivalent of a single feature"), we must stillkeep in mind that the diacritic markings in the Maori lexicon underthe lexical solution will run into the hundreds.8 Furthermore, it ispertinent to recall that it has been a wide-spread practice ingenerative phonology to favor phonological Solutions over diacriticones. For instance, KENSTOWICZ & KISSEBEKTH (1979: 142) for-mulate this principle äs follows (italics subtracted):

(18) All other things being equal, a phonological solution is preferred overa solution that divides the lexicon into arbitrary classes (a lexica-solution) or over a solution that lists the morphological/syntactic conltexts in which a rule applies (a grammatical solution).

This principle incorporates the awareness that it would lead toabsurd results if the cost of a diacritic feature were equated to theeost of an ordinary phonological feature in the evaluation of gram-mars.9 Thus, for example, we cannot solve the Maori problemessentially by saying that lexical entries of the type /nikit/ wouldbe costlier than lexical entries of the type /niki/ plus a diacriticfeature [+£ia-conjugation] because more features are needed tocharacterize the final consonants of the former than to specify theconjugational classes of the latter. For if we did this, we could ingeneral reduce the cost of lexical entries by replacing specific con-figurations of two or more phonological features by single diacriticfeatures. Hence, it is evident that, because it requires every Maoriverb that is not inflected with -tia to be marked diacritically,the lexical solution for Maori will entail a much higher total costthan the phonological solution.

7. RETBIEVABILITY OF UNDERLYING FORMS

If the foregoing remarks are correct, we are confronted with adilemma. On the one hand, rule simplicity alone does not force ustö choose the solution for Turkish that is preferred on other, struc-tnral and psycholinguistic, grounds. On the other hand, if rtde

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 14: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

302

simplicity is combined with lexical simplicity, this does give theright result in the Turkish case, but yields an undesirable result forthe Maori example, äs this is generally construed in theoreticalwritings. In other words, the current evaluation measure fails us,and there exist, äs far äs I can see, no immediately obvious modifi-cations of it that will not falter on some other score. It thereforeseems legitimate to seareh for some alternative explanatory prin-ciple. ZIMMER (1975: 562) and KENSTOWICZ (1981: 435) have sug-gested that there may be a functional motivation for the choice ofsolution for Turkish. Is this true, and how could we give substanceto such an idea in the form and Operation of our model of phonol-ogy ? In what follows, I will apply certain principles, developedon the basis of totally independent data (see ELIASSON 1975, 1977,1979, 1980, 1981, forthc., etc.), to the case in Turkish to see howit fares in this light.

In short, the conception outlined in the papers just mentionedassümes that phonological structure has to be approached fromtwo largely interwoven, but partly diverging perspectives — fromthe point of view of generation from phonological representationsto phonetic forms, and from the point of view of retrieval fromphonetic forms t o phonological representations. The interlevel trans-ductions which can occur in a phonological System so understoodare of two basic types: bidirectional and unidirectional. To theextent that interlevel relations are bidirectional, we can freelygenerate phonetic forms and retrieve phonological representationswithin the confines of the structural System itself. Insofar äs inter-level relations are unidirectional, however, we need to have accessto strategies outside the linguistic System proper in order to resolveresulting structural ambiguities. Moreover, these strategies maynot be ad hoc or specific to individual languages, but must becompletely general and available to all users of human language.Such a conception imposes relatively severe constraints on whatare possible phonologies for a given language, characterizes fairlyprecisely certain kinds of interactions between a structural lin-guistic System and general cognitive strategies, and would seem tobe at the root of some noteworthy phenomena in linguistie change(ibid.).

From this broadened perspective, let us compare, in äs simpleterms äs possible, the gr-analysis and the i-analysis of the Turkish4/0-alternations. Consider first the gr-solution. This solution crucially

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 15: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

303

employs the two rules of devoicing and ^-deletion which we repro-duce here äs follows:

(19) Devoicing[+voiced]

[— voiced]

(20) Intervocalic ^-deletion

gl / VC0V_ + V0

We have. indicated in the formalism by means of directed arrowsthat both of these rules are unidirectional in the direction from thephonological to the phonetic level. The unidirectionality of thedevoicing rule is easy to establish on the basis of data such ästhose in (2) in section 2 above. The directionality of ̂ -deletion, onthe other hand, is slightly harder to determine. As UNDEBHILL(1976: 29) notes:

"In general, two vowels may not eome together in Turkish, exceptin some loan words and except in the case where g drops out betweenvowels. When a suffix basically beginning in a vowel is to be attachedto a preceding vowel, either one of the vowels drops out, or a consonant('buffer consonant') is inserted between the two vowels. This consonantis nearly always y."

Hence, there is extensive predictability from the phonic environ-ment 7+ F to underlying //g//. Nevertheless, this is not entirely so.ZIMMEB'S revision of LEE s' solution encompasses no underlying//g// for items like [otä] 'tabernacle' — [otai] 'its tabernacle' (cf .ZIMMER 1975: 56f. fn. 5) and presumably therefore includes therule, cited by SEZEB (1981 : 380), to the effect that "final long vowelsare shortened before vowel-initial Suffixes", or in other terms:

(21) Vowel shortening before vowel-initial Suffixes

[+long]

[—long]

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 16: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

304

But this rule and the ^-deletion rule will yield identieal derivationalOutput strings. In other words, intervocalic gr-deletion äs well äsfinal devoicing are unidirectional in the direction from the phono-logical to the phonetic level, since they both involve phoneticneutralizations of distinct phonological representations. So, if wenow want to retrieve the derivational inputs to these rules on thebasis of phonetic data, we cannot merely run these rules back-wards. More specifically, faced with a phonetic form such äs[mektup], we cannot teil from rule (19) alone whether the cor-responding underlying form is /mektub/ or /mektup/. Similarly,given a phonic form |misra+i|, one does not know from rule (20)or (21) whether the deeper representation of the form is |misrag+i|or |misra+i|· Hence, to attempt to retrieve the underlying repre-sentations we have to go outside the formal grammar in a st ictsense, taking our recourse to a general cognitive procedure orstrategy that involves a comparison of phonetic forms ftogetherwith portions of their associated structural frames. For such anitem-and-frame (IF) comparison let us, for the present purposes,imagine the following very rough format, where = and ^ standfor 'not distinct from' and 'distinct from', respectively:10

(22) PHONIC FORMa = / ^ PHONIC FORMb

RELEVANT RELEVANTGRAMMATICAL == / ^ GRAMMATICALFEATURESa FEATURESb

RELEVANT RELEVANTSEMANTIC = / ^ SEMANTICFEATURESa FEATURESb

In terms of this format, the phonic forms [mektup] and [mektub-]exhibit non-distinct grammatical-semantic properties, but a pho-netic difference [—voiced] vs. [+voiced]. However, this phonicdifference is precisely parallel to the effect of the unidirectionalphonological rule (19) and hence the underlying indeterminacy(/p/ or /b/ ?) of the phonic segment [p] in [mektup] 'letter' isresolved. In an analogous fashion, one can resolve the underlyingindeterminacy of the phonic form [misra-] on the basis of therelated phonic form [misrä] 'line (in a pöem)' and the V^owel short-ening rule (21). Thus far in our discussion of retrieval we have

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 17: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

305

considered only ihe rules of devoicing and prevocalic vowel short-ening. But if we apply this procedure to the &/0-alternations äsdescribed under the gr-solution, we run into a problem. Supposewe have the Situation in (23), where sokak means 'street'.

(23) sokak ^ soka-

RELEVANT RELEVANTGRAMMATICAL = GRAMMATICALFEATURESSokak FEATURESgoka-

RELEVANT RELEVANTSEMANTIC = SEMANTICFEATURESSokak

Then we immediately notice that the devoicing rule (19) cannot beapplied to resolve the item-and-frame format (23) because a crucialpart of the environment for the voicing alternation is not met,namely, the specification [— son, — cont] is missing in [soka-].On the other band, if we try to apply the ^-deletion rule (20) tothe IF format (23), this will not work either because the differenceindicated in the format and the difference specified by the rule donot coincide, äs noted in (24).

(24) IF difference: k ^ 0Rule difference: g ^ 0Result: No match

Furthermore, retracing both devoicing (19) and ^-deletion (20)simultaneously is prohibited by the generative assumption ofsequential ordering. Thus, from the standpoint of the notions andprinciples which we have adhered to here, the ^-analysis is not apossible solution for the Turkish dialect in question.

Consider next the &-analysis. This analysis presupposes the pho-nological rule which we restate here äs (25).n

(25) ^-deletion

k

/ VC0V_ + V

0 ·. . . : · . · · · .· . - "

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 18: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

306

As is clear from what was said above about the putative rule (20),also (25) will be unidirectional in that it merges its derivationalOutputs with the derivational Outputs of vowel shortening (21).Now, if we apply rule (25) to the IF format (23), we see that thephonic discrepancy there will immediately be resolved, äs in (26).

(26) IF difference: Je ^ 0Rule difference: k ^= 0Result: Match

There is a complete match between the IF difference and thedifference specified by the phonological rule and consequently theunderlying form of [soka-] must end in a /k/.12

8. CONCLUDING REMARK

In conclusion, we may draw a distinction in regard to linguisticSolutions between the notions 'simple' (i.e, expressed by fewerdistinctive features) and 'possible' (i.e. admitted by a certain con-figuration of properties in the primary linguistic data). In theTurkish case, the simplicity and retrieval approaches make thepredictions about possible Solutions shown in Table 3.

TABLB 3

Predictions about possible (äs opposed to favored) Solutions for Turkishk/0-alternations under simplicity and retrieval approaches to phonology

Simplicity theoryRetrievability

Possible Solutions?

The simplegr-analysis

YesNo

TheÄ-analysis

YesYes

The basic assertion of simplicity theory is that, all other thingsbeing essentially equal, Speakers select a particular solution Xbecause it is at least one feature less costly than a competingsolution X+l. In other words, the simplicity metric constitutes alinear scale along which lower values are preferable to higher ones,

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 19: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

307

but which places no upper limits whatsoever on complexity. On theother band the basic assertion of the alternative approach discussedhere is that underlying linguistic representations must be retriev-able. Therefore, whenever a specific solution will not allow usto retrieve its phonological representations, we must search for analternative which contains representations and rules that do admitretrieval. It is essential to note that the notion of retrieval addressesitself directly to the point at issue, namely, the degree of abstract-ness in representations and rules. Retrievability requires interleveltransductions to be minimized and phonological representationsto be relatively close to the phonetic surface. The simple ̂ -analysisand the i-analysis make claims which seem totally unrelated tosimplicity considerations of the sort countenanced above. Basically,the question is not at which point along a simplicity scale one ofthese Solutions is located relative to the other one. Rather, the g-analysis says that certain abstractions are possible which the &-anal-ysis does not encompass. Furthermore, the notions of retrieval,which were not developed for this specific case, happen to select,indeed force, the particular solution for Turkish which is nowgeuerally viewed äs the appropriate one.

These notions of retrieval, which are largely formal or structuralin nature and only in part functional or extra-structural, deviatefrom certain traditional ideas in generative phonology such äs uni-directionality, but are in agreement with even more basic generativeconcepts and principles like the notion of rule. It is only on thebasis of process theories of phonology that the problem of retrievalcan be raised in a meaningful way, and considerations of retrievalseem to constitute a natural extension of these fundamental phono-logical principles.

Address of the author: Stig EliassonDepartment of LinguisticsUppsala UniversityBox 5138-751 20 Uppsala (Sweden)

NOTES

* This paper was written during a term äs a visiting scholar in the Depart-ment of Linguistics and Philosophy at M. I. T. I am most grateful tp PaulKiparsky and Morris Halle for numerous interesting discussions which in·spired the paper. In addition, Karl Zimmer, University of California, Berkeley,

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 20: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

308

has conomented extensively on various portions of it. I also thank Engin SezerHalettepe Üniversitesi, Ankara, for valuable comments on the Turkish data.Earlier versions of the paper were presented at the University of Alberta,Edmonton, Canada, the University of Minnesota, Twin Cities, and at the 15thAnnual Conference of the Societas Linguistica Europaea, Athens, Greece,September 8—11, 1982.1 HALLE also discusses the well-known case of the Maori passive allomor-phy in similar terms. For some comments on this problem, see section 6belöw and ELIASSON (forthc.).2 Following common practice, Turkish examples are rendered here in es-sentially their Orthographie form except that phonetically mute, so-calledsoft g9s (<g)) have been omitted and Orthographie <c, c, s, i) replaced by theSymbols c, ji, s, i, respectively. Vowel length is shown by a macron (~). Mor-pheme boundaries ( + ) have been added to indicate morpheme divisions.3 I will use double slashes //// to indicate abstract Segments which are nevermanifested äs such in the corresponding phonetic or shallow phonic represen*tations. (On the term 'phonic', see footnote 10 below.) Vertical lines | | willmark intermediate representations.4 LEES* symbol G subsumes the front velar /g/ and the mid or back velar/ / in his System. V Stands for any vowel and + is m his usage a fyarmonicsuffix boundary.5 Actually, SEZER'S final version of the &-deletion rule (ibid. 377) is alsorestricted in regard to the set of Suffixes before which it applies. See also DENY(1921: 146).6 Cf. also ZIMMER & ABBOTT (1978).7 Similarly, e.g. HALLE (1964: 340): "Ingeneral, we must omit features inall dictionary representations, whenever these can be introduced by a rulethat is less costly than the saving it effects."8 Just to give a very rough idea of the proportions under the 'lexicaF ap-proach between unmarked and diacritically marked verbs in Maori, we maynote the following. Throughout the theoretical linguistic literature, the ending-tia is taken to be the regulär passive allomorph for Maori and the other end-ings to be irregulär. Now, if we consider the verb bases whose passives arelisted in BIGGS (1966) (excluding, though, for practical reasons verb formsconstructed with the causative prefix whaka- äs well äs partially or fully re-duplicated forms of attested simple verbs), we find that out of a total of 449verb bases, 304 form their passives by means of other endings than -tia (7more bases optionally with other endings). All these non-tia verbs will haveto be marked diacritically under the lexical approach.9 Cf. also ZWICKY (1975: 156) who extracts from the practice in the genera-tive literature the following two methodological principles, which he regardsäs parts of the naturalness condition described by POSTAL (1968: 55 — 57):

(a) Insofar äs possible, the content of Segments in remote represent-ations is phonological rather than abstract.Thus, morphophonemes are not distinguished by diacritics but by ap-propriate phonetically-based distinctive features, wherever possible.(b) Insofar äs possible, phonological rules are conditioned phonolog-ically rather than arbitrarily. That is, phonetically-based featuresare preferred to lexical markings.

Moreover, POSTAL (1968: 134f.) observes:In view of the differential arbitrariness of exception features and mor-phological features, and of morphological features and phonologicalfeatures, it seems incorrect to consider them on a par from the pointof view of simplicity considerations. In other words, in consideringthe representation of individual morphemes in the dictionary part ofthe grammar, it is likely that differential cost should be assigned tofeatures of differenttypes.

POSTAL goeson to suggest a "cost hierarchy of features" where, under icertainconditions yet to be specified, phonological, morphological, and exceptionfeatures would be graded along a scale of increasing complexity, with phono-

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 21: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

309

logical features being the simples t and the exception features the most com-plex.10 The expression 'relevant features' includes, among other things, con-textually supplied Information about the categqry membership (noun, adjec-tive, etc.) of the compared forme. The word 'grammaticaP subsumes morpho-logical and syntactic Information. Phonic forms are either customary phoneticforme or eise phonetic forms from which the effects of low-level bidirectionalrules have been removed.11 HALT/R (1979: 329) formulates this rule more generally äs a rule whichdeletes "velar obstruents", hence not just //g//, but also /k/. Cf. (8) above.Such a general formulation of classes of derivational input Segments isStandard and principled practice in the unidirectional framework of genera-tive phonology. It is not possible for our purposes, however. Since our ap-proach must allow us not only to derive, but also to retrieve, derivationalinputs to rules must be uniquely specified and no empty members of setsallowed.12 The question which of [k] or [0] should be considered to be underlying isdecided by the fact that alternating roots behave in certain other respectsexactly like consonant-final items. See, e.g., ZIMMER (1975: 556).

REFERENCES

BIGGS, Bruce. 1966.English-Maori dictionary. Wellington, A. H. & A. W. Reed.

CHOMSKY, Noam & Morris HALLE. 1968.The sound pattern of English. New York, Harper & Row.

DENY, J. 1921.Orammaire de la langue turque (dialecte osmanli). Leroux.

ELIASSON, Stig. 1975."On the issue of directionality", in: The Nordic Languages and ModernLinguistics 2. Proceedings of the Second International Conference of Nordicand General Linguistics, ed. by K.-H. DAHLSTEDT, 421 —444. Stockholm,Almqvist & Wiksell.

ELIASSON, Stig. 1977."Inferential aspects of phonological rules", in: Phonologica 1976, ed. byU. DBESSLEB & O. E. PFEIFFEB, 103 — 110. Innsbruck, Institut für Sprach-wissenschaft der Universität.

ELIASSON, Stig. 1979."Retrieval of phonological representations", in: Papers from the Fifth Scan-dinavian Conference of Linguistics, part I: section papers, ed. T. PETTEBS-SON, 79-92. Stockhohn, Almqvist & Wiksell.

ELIASSON, Stig. 1980."Gase, word order and coding in a historical linguistic perspective", in:Historical morpJiology, ed. by J. FISIAK, 127 — 139. The Hague, Mouton.

ELIASSON, Stig. 1981."Analytic vs. synthetic aspects of phonological structure", in: GOYVAEBTS(1981), 483-524.

ELIASSON, Stig. Forthc."Historical change and phonological directionality".

FUJIMUBA, Osamu (ed.). 1973.Three dimensions of linguistic theory. Tokyo, TEC Company.

GOYVAEBTS, Didier L. (ed.). 1981.Phonology in the 1980's. Ghent, Story-Scientia.

HALLE, Morris. 1964."Phonology in generative grammar", in: The structure of language. Readingsin the philosophy of language, ed. J. A. FODOB & J. J. KATZ, 334—352.Englewood Cliffs, N. J., Prentice-Hall.

HALLE, Morris. 1979."Formal vs. functional considerations in phonology", in:Studies in diachron-

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 22: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

310

ic, synchronic, and typological linguistics. Festschrift for Oswald Szemerenyied. B. BBOGYANYI, 325 — 341. Amsterdam, Benjamins. [Also in KACHBU(1978, 123-134).]

JANSKY, Herbert. 1954.Lehrbuch der türkischen Sprache. 2nd ed. Wiesbaden. [6th ed. 1966.]

KACHBU, Braj (ed.). 1978.Linguistics in the seventies: Directions andprospects. (Special issue of Studiesin the Linguistic Sciences, vol. 8, no. 2.) Urbana, 111., Department of Lin-guistics, University of Illinois.

KENSTOWICZ, Michael. 1981."Functional explanations in generative phonology", in: GOYVAEBTS (1981,431-444).

KENSTOWICZ, Michael & Charles KISSEBEBTH. 1977.Topics in phonological theory. New York, Academic Press.

KENSTOWICZ, Michael & Charles KISSEBEBTH. 1979. iGenerative phonology. Description and theory. New York, Academic Press.

KIPABSKY, Paul. 1968.·"How abstract is phonology?" Bloomington, Indiana University Linguis-tics Club. [Published in FUJIMUBA (1973: 5 — 56). Reprinted in KIPABSKY(1982: 119-163).]

KIPABSKY, Paul. 1971."Historical linguistics", in: A survey of linguistic science, ed. by W. O.DINGWALL, 576 — 642. College Park, Linguistics Program, University ofMaryland. [Reprinted in: KIPABSKY (1982: 57 — 80).]

KIPABSKY, Paul. 1972."Explanation in phonology", in: Goals of linguistic theory, ed. S. PETEBS,189 — 227. Englewood Cliffs, N. J., Prentice-Hall, Inc. [Reprinted in:KIPABSKY (1982: 81-118).]

KIPABSKY, Paul. 1973."Abstractness, opacity, and global rules", in: FUJIMUBA (1973: 57^-86).[Also in: The application and ordering of grammatical rules, ed. A. -SOUDAS, 160—184. The Hague, Mouton, 1976.]

KIPABSKY, Paul. 1978."Analogical change äs a problem for linguistic theory", in: KACHBU (1978:77-96). [Reprinted in KIPABSKY (1982: 217-236).]

KIPABSKY, Paul. 1982.Explanation in phonology. Dordrecht, Foris.

KISSLING, Hans Joachim. 1960.Osmanisch-türkische Grammatik. Wiesbaden, Harrassowitz.

KONONOV, A. N. 1956. ·! "'Grammatika sovremennogo tureckogo literaturnogo jazyka. Moscow, Izda-tel'stvo Akademii Nauk SSSR.

LEES, Robert B. 1961.The phonology of Modern Standard Turkish. Bloomington, Indiana Univer-sity Press.

LEWIS, G. L. 1967.Turkish grammar. Oxford, Oxford University Press.

MCCABTHY, John J. 1981."The role of the evaluation metric in the acquisition of phonology", in:The logical problem of language acquisition, ed. C. L. BAKEB & J. J. McCÄB-THY, 218 —248. Cambridge, MIT Press.

POSTAL, Paul M. 1968.Aspects of phonological theory. New York, Harper & Row.

SEZEB, Engin. 1981."The k/0-alternation in Turkish", in: Harvard Studies in Phonology\2j ed.G. N. CLEMENTS, 354-^382. Bloomington, Indiana University LinguisticsClub.

SWIFT, Lloyd B. 1963. . ·A reference grammar of Modern Turkish. Bloomington, Indiana Uniyersity.

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 23: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

311

UNDEBHILL, Robert. 1976.Turkish grammar. Cambridge, The MIT Press.

VIETZE, Hans-Peter, Ludwig ZENKEB & Ingrid WABNKE. 1975.Bückläufiges Wörterbuch der türkischen Sprache. Leipzig, VEB Verlag En-zyklopädie.

ZIMMER, Karl E. 1975."Some thoughts on likely phonologies for non-ideal Speakers", in: Papersfrom ihe Parasession on Functionalism, ed. R. E. GBOSSMAN, L. J. SAN &T. J. VANGE, 556 — 567. Chicago, Chicago Linguistic Society.

ZIMMER, Karl E. & Barbara ABBOTT. 1978."The ß/0-alternation in Turkish: Some experimental evidence for its pro-ductivity", Journal of Psycholinguistic Research, 7, 35 — 46.

ZWICKY, Arnold M. 1975."The strategy of generative phonology", in: Phonologica 1972, ed. W. U.DBESSLEB & F. V. MABE§, 151-165. München, Fink.

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM

Page 24: TURKISH k-DELETION: SIMPLICITY VS. RETRIEVAL

Brought to you by | University of Glasgow LibraryAuthenticated | 130.209.6.50

Download Date | 9/5/13 11:03 PM