Sound Change, Abstract Representations, and Simplicity Author(s): Wayne P. Lawrence Reviewed work(s): Source: Linguistic Inquiry, Vol.

37, No. 2 (Spring, 2006), pp. 346-350 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4179370 . Accessed: 30/07/2012 13:54
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

346
SOUND CHANGE, ABSTRACT REPRESENTATIONS,AND SIMPLICITY

SQUIBS

AND

DISCUSSION

WayneP. Lawrence The Universityof Auckland

One criticism often leveled against generative/abstract phonology is that there is no psychologicaljustificationfor the view that the brain/ mind operates according to principles of "simplicity" and "economy." For example, Jaeger(1986:74) writes that "[n]early all studies of speech perceptionand productionindicate that something close to surface forms exists in memory ... and that words are stored with much redundancy." In this squib, I will present two historically and geographically unrelated examplesof sporadic,irregular, historicalchangesin individual lexical items which stronglysuggest that,in at least some speakers' mental representationof the prechange form, the words were being not motivatedby any languagegiven a ratherabstractrepresentation specific details of the morphosyntax.Both examples are of words where adjacentsyllables are (near-)identical: kasusuin one case, sizimi in the other. These change to kakasu and hibimi, respectively. These sound changes are amenableto a naturalexplanationonly if they are that explicitly consideredto have taken place on some representation What is noteworthy,hownotatedthem as instances of reduplication. ever, is that these words are not historicallydue to any reduplication process, nor is thereany synchronicreduplication process thatappears to be relevant to them. In other words, the language speaker/learner has made the abstraction thatthese words containa reduplicated syllaof the ble, with no more evidence for it than the actualpronunciation words. 1 Two Sound Changes The NorthernOkinawan dialect of the community of Yaka has the word kakasu '(edible) sea urchin',the Proto-Okinawan form of which is reconstructed as *gacucu.1 In Proto-Northern-Okinawan, both affricates (c = [ts]) in this form irregularlybecame s. The immediate predecessor of Yaka kakasu is thought to have been *kasusu, and indeed this is the form attestedin the dialect of the Kin community, located about seven kilometerseast of Yaka. Assuming some kind of here as a postposedR), reduplicativefunction (informallyrepresented
the change *kasusu -- kakasu can be interpreted as *kasuR -> kaRsu

(i.e., as a movement of the reduplicativefunction within the word), on the reflectinga probableconfusion aboutthe locus of reduplication part of the speakerwho triggeredthis historicalchange.2
I would like to thanktwo anonymousreviewers for LinguisticInquiryfor very helpful comments. I am also grateful to the members of the University of Auckland informal formal-grammar discussion group for discussion at an early stage of this squib. ' The Okinawandialects are a subgroupof the Ryukyuanlanguagegroup. is a sister language of the ancestorlanguage of the mainland Proto-Ryukyuan Japanesedialects. 2 A Google searchreveals examples like Hononolu (11 hits) and Holululu (8 hits) for Honolulu, Kadudu (for Kakadu) National Park (2 hits), Jobaba (for Jojoba) oil (2 hits), and Tititica for Titicaca (9 hits). As the anonymous

SQUIBS

AND

DISCUSSION

347

In this squib, I use R as a segmentallyempty reduplicativefunction that is subject to specification of unit (mora, syllable, foot, etc.) and direction. I have arbitrarily selected reference to the left (i.e., to the preceding unit), but reference to the right would work as well: *kaRsu-- Rkasu.R differs from the function RED used in Optimality Theory analyses insofar as RED iS restricted to being a morpheme (specifically, an affix) (McCarthy andPrince 1999:232), whereasthere is no evidence to supportmorphemicstatus of the reduplicantin the examples given in this squib. Standard Japanesesizimi 'corbicula(a small freshwater bivalve)'3 is recordedin written form as hibimi on at least two separateoccasions.4 Confusion of si [Si] and hi [ci] is common in many dialects of Japanese,with hi tending to be pronouncedas si in Tokyo Japanese (Vance 1987:22); so, in the Tokyo Japanesesituation,si -> hi would be an instance of hypercorrection. However, the change zi [(d)3i] -> bi in the second syllable is a very unnaturalchange that can be explained only if the second syllable is viewed as related to the wordinitial syllable throughreduplication. In Japanese,b is functionallythe voiced equivalent of h, so zi stands in the same relationshipto si as bi does to hi. Both the prechange and the postchange forms can be viewed as involving reduplication of the word-initialsyllable, coupled with voicing (1).5
(1) s i R m i -->h i R m i [Si(d)3imi-ibimi]

[ + voi] 2 Discussion

[+ voi]

The two examples given above are monomorphemic, both synchronically and historicallyas far back as their history can be traced.Word formationprocesses in these languagesprovideno evidence thatreduplication is involved in the make-up of these words. In spite of this, sound changes indicate that the representation of these forms at the

reviewer who alertedme to such examples puts it, "These seem to supportthe idea that the speaker knows there's reduplicationsomewhere but isn't sure where." Whereasthese exampleshave possibly arisenin the processof committing words to writing, the Yaka dialect example differs in that the change has taken place without any influence of a writing system, Yaka (like most Ryukyuan dialects) being a purely spoken language with no orthography or written tradition. 3 This is a common lexical item in oral Japanese, since sizimis are frequently used in making miso soup. 4 One is reported in GengoseikatuEditorialBoard 1979:43, and a separate instance is reported on a number of Web sites (e.g., www.geocities.cojp/ HeartLand-Himawari/9254/top/k/k3._html). 5 The voicing in sizimi, hibimi is not a case of the sequential voicing (rendaku) process that is observed in certain other reduplicativeforms such as hito-bito 'people', sju-zju'variety'.This is becausesequentialvoicing applies only if the compound-finalconstituentis a free form, and sizimi is not a compound of /si-si/ or /si-simi/ where either si or simi is a free form.

348

SQUIBS

AND

DISCUSSION

time of the soundchange involved the mappingof phonologicalinforTo the pre-soundmationof one syllable onto another(reduplication). change speaker,the only externalevidence for such a representation lay in the phonetic form itself. In the frameworkof OptimalityTheory, Prince and Smolensky (2004:225-230) propose a principleof Lexicon Optimizationaccordlearnersconstructlexical ing to which, in the absence of alternations, representationsthat are identical to the surface representations they hear.6Inkelas(1994:1) arguesthatthe use of underspecification arises "only when there are alternantsurfaceforms all of which are predictable from context or grammaticaldefaults." Under this conception, the Yakaand Japaneseunderlyingforms priorto soundchanges would be /kakasu/and/Ii3imi/, respectively,providingno base for the attested historicalchanges to occur. A proposedexpansionof Lexicon OptimiLexicon Optimization(Harrisonand Kaun zation, Pattern-Responsive 2000), would allow pervasive phonological patterns,be they due to alternation,analogy, or predictablepatternsof surfacedistribution,to yield underspecifiedinput representations, but is there such evidence for a reduplicative structure(underspecifiedsyllable) in Yaka and Japanese? The Yaka dialect has yet to be described in any detail, but a glossary of the closely relatedKin dialect (Okamura1994) containing approximately 1,120 different vocabulary items (excluding compounds, conjugationalvariants, child language forms, and exclamations) contains only 3 words of the shapes CIV2CIV2CV and (C)VCIV2CIV2 (of a total of 79 trimoraictrisyllabicwords, or 3.8%), and only two words of the shape CIV2CIV2 (of a total of 86 bimoraic bisyllabic words, or 2.3%), where the voicing of the consonantshas been ignored.7In StandardJapanese,a count of trimoraictrisyllabic noncompoundwords listed in Terakawaand Kusaka1944 reveals that 6.9% (51 out of 741 such words) are of the shape C1V2C1V2(C)V or in differences consonant The (again ignoring voicing). (C)VCIV2CIV2 same dictionarylists 627 non-Sino-Japanese noncompoundnouns of the shape CVCV, and only 17 (2.7%) of these exhibit a reduplicative structure.8 There is thus no pervasive phonologicalpatterning in these languages that would motivate underspecificationunder Pattern-Responsive Lexicon Optimization.

6 Prince and Smolensky (2004:30, 230) do permit the use of constraints of the *STRUC family to producemaximally simple lexical representations, but it is unclearwhat would motivatethe rankingof the relevant*STRUC constraints above the relevantfaithfulnessconstraintsin the two cases discussed here. For argumentsagainst *STRUC constraints,see Gouskova 2003. 7 The forms found with potentiallyreduplicatedsyllables are asasa 'kind of cicada', kasusu 'sea urchin', suzumi 'drum', nunu 'cloth', susu 'hem'. 8 There are no Sino-Japanesemorphemesof the form CIV2C1V2, so including monomorphemicSino-JapaneseCVCV words would furtherreduce the percentage.There are also no verbs of the form CIV2CIV2.

SQUIBS

AND

DISCUSSION

349

Zuraw (2002a) proposes that "there is a general drive, repreREDUP, for all wordsto be construedas reduplisentedby the constraint cated" (Zuraw2002b). This "aggressive reduplication"strategyreceives strong support from the sound changes discussed above.9 However, one may ask why this general drive exists. It is plausible that this "aggressive reduplication"is a reflex of a more generalprinciplethat values less "complex" underlyingforms is over more complex candidates.This would mean that C1V2C1V2 more complex than CVR. deemed by the grammarto be structurally One way in which C1V2CIV2 is conceivably more complex thanCVR The two is in the numberof featuresin the underlyingrepresentation. sound changes discussed in this squib provide some supportfor such an evaluationmetric,but furtherevidence will be requiredto establish and refine it. 3 Conclusion The two sporadicsound changes discussed in this squib indicatethat, in the speech of at least the speakers who instigated the changes, both the pre- and postchange forms were given an underspecified, As this canin theirunderlyingrepresentations. reduplicativestructure not be attributed to any language-specificdetail of the languagesconwe shouldseek a moregeneralexplanationfor why the shortest cermed, is identical to the possible derivation(i.e., underlyingrepresentation is not selected. These sound changes suggest surface representation) "complexity" thatthe languagelearnerstrives to reducethe structural of the underlying form of words, adopting underspecifiedstructure where this is recoverable. References
Gengoseikatu Editorial Board, ed. 1979. Me: Kakikotoba memo-tyou (Eyes: Memo-book of written language). Gengoseikatu (Language Life) 336:43. Gouskova, Maria. 2003. Deriving economy: Syncope in Optimality Theory. Doctoral dissertation, University of Massachusetts, Amherst. Harrison, K. David, and Abigail Kaun. 2000. Pattern-Responsive Lexicon Optimization. In NELS 30, ed. by Ji-Yung Kim and Masako Hirotani, 327-340. Amherst: University of Massachusetts, GLSA. 9 An anonymousreviewer points out that the Yaka example, *kasusu > kakasu,bears on the question of whetherconstrualof a word as reduplicated occurs during learning,because the resulting lexical representation would be simpler, or duringgeneration,because the grammarenforces it (Zuraw2002a: 424-430). The reviewernotes that [kakasu]appearsto arguefor the lexicalization interpretation, since under the generationaccount /kasusu/ would simply map to [kasusu], which does not violate REDUP constraints;there is no way for REDUP to predict the change to [kakasu].

350

SQUIBS

AND

DISCUSSION

Inkelas, Sharon. 1994. The consequences of optimizationfor underspecification. Rutgers Optimality Archive ROA-40. http:// roa.rutgers.edu. for Jaeger,Jeri J. 1986. On the acquisitionof abstractrepresentations English vowels. Phonology Yearbook3:71-97. McCarthy,JohnJ., and Alan S. Prince. 1999. Faithfulnessand identity in Prosodic Morphology. In The prosody-morphologyinterface, ed. by Rene Kager, Harryvan der Hulst, and Wim Zonneveld, 218-309. Cambridge:CambridgeUniversity Press. Okamura,Toyo. 1994. Cin-kutuba(Kin dialect). Privateedition. Prince, Alan, and Paul Smolensky. 2004. OptimalityTheory: Constraint interaction in generative grammar. Malden, Mass.: Blackwell. nihongohatuon Terakawa,Kisio, andKusakaMiyosi. 1944. Hyouzyun dictionary).Kyoto: Japanesepronunciation daiziten (Standard Taigadou. Vance, Timothy J. 1987. An introductionto Japanese phonology. Albany, N.Y.: State University of New York Press. Phonology 19:395-439. Zuraw,Kie. 2002a. Aggressive reduplication. UCLA LinguisZuraw, Kie. 2002b. Near- and pseudo-reduplication. tics Talk abstract. http://www.linguistics.ucla.edu/colloquia/ abstracts/2002/zuraw-2002-main.html.