UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Korean vowel harmony: an optimality account Choi, Sohee 1994

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_1994-0254.pdf [ 2.71MB ]
JSON: 831-1.0087341.json
JSON-LD: 831-1.0087341-ld.json
RDF/XML (Pretty): 831-1.0087341-rdf.xml
RDF/JSON: 831-1.0087341-rdf.json
Turtle: 831-1.0087341-turtle.txt
N-Triples: 831-1.0087341-rdf-ntriples.txt
Original Record: 831-1.0087341-source.json
Full Text

Full Text

KOREAN VOWEL HARMONY - AN OPTIMALITY ACCOUNTbySOHEE CHOIB.A., Yonsei University, 1991A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF ARTSinTHE FACULTY OF GRADUATE STUDIESDepartment of LinguisticsWe accepts this thesis as conformingto the required standard...THE UNIVERSITY OF BRITISH COLUMBIAApril 1994© Sohee Choi, 1994in presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.Department of E&A ISftCSThe University of British ColumbiaVancouver, CanadaDate &pr.DE-6 (2/88)11ABS TR1.CTThis thesis examines two aspects of Korean vowelharmony: the identification of the correct harmonizingfeature and neutrality in the harmony system.In this study, I investigate the hypothesis that Koreanvowel harmony can be described as tongue root harmony andthat the neutrality in the harmony system can be accountedfor through the interaction of various constraints withinthe framework of Optimality Theory; the Theory of ConstraintInteraction (Prince and Smolensky 1993, McCarthy and Prince1993)In Chapter One, the basic phenomena and the problems ofKorean vowel harmony are presented. Chapter Two discussesthe theoretical background of Optimality Theory. The basicclaims and motivation for the theory are dealt with in thefirst section. In the following section, I consider majorstudies applying this theory, including: syllable theory(Prince and Smolensky 1993); prosodic morphology (McCarthyand Prince 1993, Prince and Smolensky 1993); and vowelharmony (Pulleyblank 1993) . In Chapter Three, the Optimalityaccount of Present—Day Korean and Later Middle Korean ispresented. I further discuss Optimality Theory’s treatmentof diachronic change in harmony systems. Concluding remarksand areas for further research are given in Chapter Four.111TABLE OF CONTENTSAbstract iiTable of contents iiiAcknowledgement vCHAPTER ONE. INTRODUCTION 11.1. A Summary Introduction of KoreanVowel Harmony Phenomenon 21.2. Optimality Theory and Applicationthereof in Vowel Harmony 41.3. Organization of Content 7CHAPTER TWO. THEORETICAL BACKGROUND 92.1. Optimality Theory 102.2. Major Application 182.2.1. Syllable Theory 182.2.2. Prosodic Morphology 242.2.3. Vowel Harmony 31CHAPTER THREE. KOREAN VOWEL HARMONY 443.1. Present-Day Korean 453.1.1. Sound Symbolic Words 493.1.1.1. Previous Accounts 523.1.1.2. Optimality Account 58iv3.1.2. A—initial Suffixes 773.2. Later Middle Korean 833.2.1. Optimality Account 87CHAPTER FOUR. CONCLUSION 95BIBLIOGRAPHY 97APPENDIX: List of Constraints 102VACKNOWLEDGEMENTI would like to thank the members of my committee, Dr.Douglas Pulleyblank, Dr. Patricia A. Shaw, and Dr. DavidIngram. I am especially indebted to my supervisor Dr.Douglas Pulleyblank whose guidance, and encouragement havemade this process especially rewarding.I would also like to acknowledge that work on thisthesis was supported by grant no. 410-91—0204 from theSocial and Humanities Research Council of Canada.Special thanks are extended to my parents who havesupported and encouraged me throughout this work.1CHAPTER 1INTRODUCTIONVowel harmony is a well known phonological phenomenonfound in numerous languages. Until now, it has beengenerally assumed that palatal (front-back) harmony is foundtypically in Uralic and Altaic languages, while tongue rootharmony occurs mainly in language families in Africa (Vago1980) . But recent studies (Svantesson 1985, Song 1990) showthat tongue root harmony is also found in some of the Altaiclanguages such as Korean and Mongolian.The vowel harmony of Korean has been a controversialtopic not only due to the difficulty in finding a correctharmonizing feature but also due to problems in explainingthe identification and properties of neutral vowels inharmony. The neutrality of high unrounded vowels in theKorean harmony system has hitherto been treated merely as anexception to the harmony in most studies.In this study, the hypothesis that Korean vowel harmonycan be described as tongue root harmony and that theneutrality in the harmony system can be accounted for asresulting from the interaction of various constraints willbe investigated within the framework of Optirnaiity Theory;the Theory of Constraint Interaction (Prince & Smolensky1993, McCarthy & Prince 1993a,b)21.1. A Summary Introduction of Korean Vowel HarmonyPhenomenaIn Present-Day Korean (henceforth; PDK) the vowelharmony phenomenon can be found in sound symbolic words andso called ‘A—initial suffixes’ (C.W.Kim 1978, Kim-Renaud1986, etc.). In PDK, there are 10 monophthongs which resulthistorically from the loss of [A] and themonophthongization of diphthongs (y->e, ay->c, oy->ö, uy->U)after Later Middle Korean (henceforth; LMK) The vowelsystems of LMK and PDK are shown in (1) and (2)respectively.(l)Later Middle Korean (2)Present-Day Koreani i U i u i U0 0 C O 0 0a A e aIn the sound symbolic words of PDK, vowels are dividedinto two sets. The vowels [i,e,U,i,o,u] form a harmonic grouptraditionally called ‘dark’ vowels, and the vowels [e,O,a,o]form the other set called ‘light’ vowels. Within a soundsymbolic word, vowels must be either all dark or all light.The only exceptions for this are the vowels [ii and [i] whichare neutral in non-initial syllables and therefore maycooccur with either dark or light vowels. Some examples ofsound symbolic words are given in (3)3(3) Dark Lightcik’l cek’al ‘chattering’spk sapak ‘crunching’tils’ok tals’ak ‘lifting’p’yulutuij p’yolotoij ‘pouting’pisil pesil ‘staggering’sinkil senkil ‘smiling’The last two examples ‘staggering’ and ‘smiling’ showthe neutrality of the high unrounded vowels in the secondsyllables. Vowel harmony in sound symbolic words has beendifficult to describe because the vowels of each harmonicclass do not clearly constitute a natural class.Traditionally, it has been treated as palatal harmony of thetype common in Altaic languages including Turkish andClassical Mongolian. However, since each harmonic groupincludes both back and nonback vowels, it cannot simply betreated as strictly palatal harmony. McCarthy (1983)described vowel harmony in sound symbolic words as a heightharmony arguing that [low] is the feature that distinguisheseach harmonic set. This analysis is not very plausible sincehe had to include an ad hoc context-free rule ([+round]->[-low]) for the nonlow rounded vowels (o,ö). Kim-Renaud(1986) described vowel harmony in sound symbolic words bysemantic features like [dark] and [light], which does notseem to have strong phonological motivation. In other4studies, the harmony was considered as exceptional like NezPerce, since the harmony was not considered to becharacterized by any single distinctive feature(Tohsaku, 1983) . I will adhere to the arguments presented inSong (1990) in stating that {ATR] is the feature whichdistinguishes the dark vowels from the light ones in PDK.1.2. Optimality Theory and Application thereof inVowel HarmonyOptimality Theory, which is a theory of constraintinteraction, offers a principled account for the problem ofneutral vowels in Korean vowel harmony, using the mechanismof ranking a set of conflicting constraints that play acrucial role in the language.Since Chomsky and Halle (1968), it has been standardprocedure to derive output forms from input structure byapplying a set of rules. However, the Optimality conceptsuggests a rather different method. Instead of taking theunderlying form (an input) and transforming it into itsassociated output in a step-by-step manner, OptimalityTheory allows the generation of a large set of candidateoutputs and evaluates them by the mechanism of wellformedness constraints. The constraints are part ofUniversal Grammar, and each language has a different rankingof the constraints constituting the individual grammar. Bygiving an appropriate ranking for Korean, therefore, we can5solve the problem of neutral vowels within the framework ofOptimality Theory. The neutral vowels in the harmony systemof Korean result simply from the interaction between thegrounded condition RTR/LO (Archangeli & Pulleyblank, inpress) and right and left edge constraints on alignment:(4) ConstraintsALIGN-ct-L/R: The left/right edge defined byfeature a. aligns with the left/right edge ofDomain D.RTR/LO: If [-ATR] then [+10w]; if [—ATR] then not[—low]The ALIGN constraint links the feature to the left/rightedge of each domain. For example, if [-ATR] fails to link tothe first mora of the word then it violates ALIGN-L.Likewise, a violation of ALIGN-R will occur if the featureis not aligned to the right edge of a morpheme. The groundedcondition RTR/LO prevents the [—ATR] feature from linking tononlow vowels.To illustrate this, let us discuss the differencebetween ‘crunching’ (sapak) and ‘staggering’ (pesil) from theexamples in (3) . I assume that [-ATR] is the harmonizingfeature in harmony and that the light sound symbolic wordshave an underlying [-ATR] specification while dark soundsymbolic words do not. The properties of the neutral vowelsare explained by the relative order of the constraints6RTR/LQ and ALIGN-R (RTR/LO>> ALIGN-R); the basic hypothesisis that harmony is interrupted by the effect of the groundedcondition RTR/LO whereas the constraints on alignmentencourage it. ALTGN—L ensures that the feature [—ATR] alignsto the left-edge of the words. ALIGN—R causes this [-ATRIfeature to spread to the following vowels. Thus, ALIGN-L andALIGN-R together derive the harmony effect. However, theRTR/LO condition which prevents [-ATR] from linking tononlow vowels blocks this harmony when the non-initialvowels are not low. As a result, the non—initial moras ofeach word surface as [+ATR] (i,i) in such cases. In otherwords, the feature [-ATRI aligns to the right vowel of sapakwithout incurring any violation. If [-ATR] is not aligned tothe right, then the form violates ALIGN-R; as it is, harmonyproceeds to the right in normal cases. However, in the caseof high unrounded vowels, harmony cannot proceed as suchbecause it will incur an RTR/LO violation. The candidateform pesil will be the optimal form even though it violatesALIGN-R because it is more important to obey the higherranked constraint RTR/LO. [i] and [ii are neutral only innon-initial syllables because ALIGN-L is ranked higher thanRTR/LO. To sum up, the ranking with respect to these threeconstraints is: ALIGN-L>> RTR/LO>> ALIGN-R.The basic analysis of the most crucial type of case inour investigation has been outlined here and will bediscussed in detail in Chapter Three. A more profound7analysis, however, will require a complete ranking of allthe constraints.We will also examine the diachronic change from LaterMiddle Korean to Present—Day Korean; this change will beaccounted for by reranking some of the constraints for eachstage.1.3. Organization of ContentChapter Two will discuss the theoretical background ofOptimality Theory. The first subsection will deal with thebasic claims and the motivation/development of OptimalityTheory. In the following section, we will look at the majorstudies or applications concerning this theory includingsyllable theory (Prince & Smolensky 1993), prosodicmorphology (McCarthy & Prince 1993a, Prince & Smolensky1993) and vowel harmony (Pulleyblank 1993) . In ChapterThree, the main body of the thesis, we will look at theKorean vowel harmony system. The first section will followthe vowel system of Korean through its diachronic changesfrom Later Middle Korean to Present-Day Korean. Previousaccounts of Present-Day Korean vowel harmony will bepresented briefly, followed by the Optimality account of theharmony system. In the subsequent section, the Optimalityaccount of Later Middle Korean will be discussed. OptimalityTheory’s treatment of diachronic change in harmony systems8will also be discussed. The concluding remarks and areas forfurther research will be given in Chapter Four.9CHAPTER 2THEORET I CAL BACKGROUNDIn this chapter, we will discuss the basiccharacterization of Optimality Theory (henceforth; OT)presented in Prince and Smolensky (1993) and McCarthy andPrince (1993a) . In order to derive the hypothesis regardingthe harmony system of Korean discussed in the introduction,it is important to lay out some of the properties ofOptimality Theory.Some of the major applications of this theory will alsobe presented in this chapter. The discussion will help oneto understand how OT treats some major areas of phonology.Furthermore, the discussion of syllable theory and prosodicmorphology will show that crucial properties of OT areindependently motivated --not introduced specifically toaccount for the harmony phenomenon of Korean, but to explainsimilar problems in other areas of phonology; the use ofconstraints other than those directly involved in harmonyprovides independent motivation for the types of theoreticalproperties crucial to the account of Korean vowel harmony.102.1. Optimality TheoryThe goal of OT is to develop and explore a theorywhereby representational well-formedness determines theassignment of grammatical structure. In the standard theoryof generative phonology (since Chomsky and Halle, 1968), thephonological rule aims to encode grammatical generalizationsas follows:(5) A->B/C DSuch a rule examines its input for the pattern CAD andchanges element A into B, producing an output that istypically subject to further rules of the same type (Chomsky& Halle 1968, McCarthy & Prince l993a) . Challenges to thisderivational approach have recently been observed, as it hasbeen found that linguistic patterning in many areas isactually governed by structural constraints on the outputlevel -—constraints which furthermore hold generally acrossforms that would be processed by many distinct rewriterules. A number of phonologists (Goldsmith 1990,1993,Paradis 1988a, Singh 1987, Coleman 1991, Scobbie 1991,1992,Bird 1990, Myers 1991 and many others) have proposedconstraint-based models such as “The Theory of Constraintsand Repair Strategies” and “The Persistent Rule Theory”which shift the explanatory burden from the input-drivenrewrite rules to output constraints.11Optimality Theory is the outgrowth of a research programfirst proposed in Smolensky (1988) which seeks toinvestigate the insights of connectionist and symboliccomputation. Smolensky demonstrates that certainconnectionist networks can be analyzed as algorithms formaximizing “Harmony”, a numerical measure of wellformedness. OT takes this connectionist insight of harmonymaximization as a basic cognitive principle and imports itinto grammatical theory as optimization. Unlike otherconnectionist approaches to language, however, OT does notseek to replace symbolic representations and linguisticanalyses with low-level numerical computation (Kirchner1993b, Prince & Smolensky 1993)The basic idea of OT is that Universal Grammar consistsof a set of constraints on representational well—formedness,out of which individual grammars are constructed. Individualgrammars are constructed by imposing a ranking on auniversal constraint set. According to this theory,language—particular variation is due to differences in theranking of the constraints. OT shifts the explanatory burdenof linguistic theory from input-based rules to output-basedconstraints. OT allows for the specification of a large setof candidate outputs instead of taking an underlying formand transforming it in a step-by-step manner into itsassociated output. The candidate set is evaluated by thesystem of well-formedness constraints, which selects the12actual output (optimal) forms from the candidate set.Therefore, the grammar is configured like this:(6) Gen (in) -> {cand1, cand2, .Eval ({cand1, cand2, ... }) =(McCarthy & Prince, 1993a)The grammar defines a pairing of underlying and surfaceforms (input,output) by the function Gen (‘generator’)which is a fixed part of Universal Grammar. The functionEval determines the relative harmony of the candidates. Anoptimal output is at the top of the harmonic order on thecandidate set, which best satisfies the constraint system.The set of constraints, denoted by is provided byUniversal Grammar and an individual grammar G1 is obtainedby imposing a strict dominance order, denoted by ‘>>‘, onthe elements of C.(7) UG = {C1,2.., } = CG1 ( C ; >>(Ito, Mester & Padgett 1993)OT has the attractive trait of allowing the output formto violate some of the relevant constraints. Prince andSmolensky (1992, 1993) have argued that the goal ofdeveloping a restrictive theory of Universal Grammar can bebest served by allowing constraints to be violated. The13output, therefore, will typically fail to meet everyconstraint. For a given input, the candidate that best-satisfies the constraint system will be optimal and is bydefinition the output that the grammar associates with theinput (McCarthy & Prince 1993a)OT relies on the notion of constraint interactionwhereby the satisfaction of one constraint can be designatedto take absolute priority over the satisfaction of another.The means that a grammar uses to resolve conflicts is torank constraints in a “strict dominance hierarchy”. Eachconstraint has absolute priority over all the constraintslower in the hierarchy. That is, lower-ranked constraintscan be violated in an optimal output form when suchviolation secures success in higher-ranked constraints.The following four principles are hallmark properties ofOT presented in McCarthy and Prince (1993a) among others.(8) Principles of Optimality Theorya.Violability; Constraints are violable, butviolation is minimal.b.Ranking; Constraints are ranked on a languageparticular basis; the notion of minimal violation isdefined in terms of this ranking.c.Inclusiveness; The constraint hierarchy evaluates aset of candidate analyses that are admitted by verygeneral considerations of structural well-14formedness. There are no specific rules or repairstrategies.d.Parallelism; Best-satisfaction of the constrainthierarchy is computed over the whole hierarchy andthe whole candidate set. There is no serialderivation.As mentioned above, ranking and violability are the keycharacteristics of OT. Let us, then, consider the notions ofconstraint ranking and constraint violability. Suppose wehave a granuuar consisting of two constraints, A and B. Thegrammar functions to pair underlying forms with surfaceforms: Eval(Gen(in1))->out1, Eval(Gen(in2))—>out2, and soon. Suppose that we have an underlying form /ink/ whichgives rise to a candidate set {k-cand1,k-2}through thefunction Gen (Mccarthy & Prince,1993a,b).The following tableau illustrates how the satisfactionof a constraint hierarchy proceeds.(9) Constraint Tableau, A>>BCandidates A Bk-cand1k-cand2(Mccarthy & Prince, 1993a,b)U)Di CtH-U) Di C) CtH 0 H-U) H H C) Di CtCD Di H DiC) (DC)CtCto CDCDHH••r1DiU)F-3H--CtCtDiCDDiCDoQ..HH-U)?oCtCD0HI—’HDiCDCtHC)phoP’H--Nt--CDDiCtU)itU)CDCtCDCDCt-_CtP’-H-CDHt-0CDH‘U)H0CDCD-DiCtIi0C)C)CDCtCDCD H0CDH)Ct0CtCDD’CtIIIICDH-CDCDC)DiU)0H?IDiU)C)Q-CtDiCDI-QDiQoH-HDlCtphoCtCtMCDCtCDCD‘-100DiCtU]H-C)DiECtCtC)0C)DiI-hDiH-U)0C)DiDiCDCt0CtCtH-CDU)CDDiCtHC)U)HHCDitCDU)H••XCDI-ICDDiitDiCDCD•U)IICtU)<CDLC)DiCtQIIHDiI-<HCDCDDiH)Di‘dH0C)I-hDiH-0CDDiH-0H0Cti—i-HDiCtCtbH-0DiCtCtCtDi‘<CtCtCDHCtU)CtH-Ct0<CDDi<C))U)CDCtU)CtCDC)H0-I0Di‘‘CDDiCDCtiH-H-H-H)DiHD-DiU)Q0V0Ct-HU)iV-CDCDDiH-DiM0rtHCtDiCtH-0CT)WCDCDDiC)C)HQ•H-Q0CDCtDiHCtCDCDH-U)DiU)HCtCtU)H,CDCCDçoo‘-DiCtCDCtDiZH-CDHH-HCDCDCtU)0H00DiHDiaDiCt8HDiHCD•H-QIICtCDH,Ct‘-<W-DiHDiCDH-Di0DiDi0H0QCtHCDCDH-H-<H-itQ.U)IICDU)CDCt..CDDi1i-DiHCDHH-0H-DiCDCDH-Di1<CDCDCt0CtDiDiC)H-CDC)C)CD-0CD0:CDH-CDH-0C’)Ct•—CtU)H-U)H-H-H-Ct:CDC)0H-DlCt0)Di•CD0CtCtC)HCtCtIICD-CL<CDDiDiCDi<0CDHDiH-itH-PhCtI0C)CDDi0H-CD*Di0t-1CLCDCD(UCtDiH-CDH-U)IIQ..0-U)Di0C)Di0HDi0CtH-DiCtCtH-‘-<CDH0HIH•Di-CDCDCDCLCDI-DiU)U)itCtCtcøCtphCLCtCtCDCt0CDL.iHCDHCtH-DiCtCDII—CDI-U)CD-U)CtCDCn16An interesting configuration arises when a constraintin the grammar admits multiple (or gradient) violations. Wewill see in the following chapter that the analysis ofKorean vowel harmony involves both single and multipleviolations.Suppose we have a pair of candidates that ties on allconstraints higher-ranked than C, and with C itself we havethe following.(1O)Multiple ViolationCandidates ... Ccand1 ... *cand2 ...(McCarthy & Prince, 1993b)The principle of Harmonic Ordering (Prince &Smolensky,1993)’ entails the desirable result that anysingle constraint will only be violated minimally in anoptimal form. In this case, we say that cand1 is the optimaloutput because its accumulated violations of C are fewerthan those of cand2: C is violated minimally.1The notion of Harmonic Ordering defines best-satisfactionin a way that encompasses hierarchical ranking of violation(‘violate the lowest-ranked constraint’) and nonranking(‘violate any single constraint to the least degreepossible’) (McCarthy & Prince 1993).17Let us conclude the discussion of the theoreticalbackground by looking at three principles that underlie thetheory of Gen presented in Mccarthy and Prince (1993a).(i)Freedom of Analysis; Any amount of structure maybe posited.(ii)Contairiment; No element may be literally removedfrom the input form. The input is thus containedin every candidate form.(iii)Consistency of Exponence; No changes in theexponence of phonologically-specified morphemesare permitted.Freedom of Analysis allows Gen to supply candidateoutputs that include any degree of structure, both in termsof autosegmental features and associations and in terms ofprosody. It requires no specific rules or repair strategiessince the basic principles of representational form supply arange of candidates that are inclusive. The Containmentproperty limits this freedom by requiring the input to bepresent in any licit candidates. Therefore, if an input formincludes a feature specification (i.e. [-ATRJ), the correctoutput form must have this feature. Consistency of Exponencemeans that the lexical specifications of a morpheme cannotbe affected by Gen. Thus, any given morpheme’s phonologicalexponents must be identical in underlying and surface form.182.2. Major Application2.2.1. Syllable TheoryOne of the major applications of Optimality Theory is toestablish a typology of syllable structure systems. The ideais that Universal Grammar provides a set of violableconstraints on syllable structure, and that individualgrammars fix the relative ranking of these constraints(Prince & Smolensky 1993) . Therefore, the typology ofpossible languages is given by the set of all possiblerankings. Prince and Smolensky (1993) derives CV syllablestructure typology by using the following constraints andthe various rankings of these constraints.(11) The Basic Syllable Constraintsi)ONS; Syllables must have onsets.ii)-COD; Syllables must not have a coda.iii)PARSE; Underlying segments must be parsed intosyllable structure.iv)FILL; Syllable positions must be filled withunderlying segments.PARSE and FILL comprise what Prince and Smolensky (1993)calls the “Faithfulness” family of constraints in the sensethat well-formed syllable structures are those in whichinput segments are in one-to-one correspondence with19syllable positions. This faithfulness constraint set isdenoted as !I (and a number of it as 1Ft) in the followingtable.(12) CV Syllable Structure Typology(Prince & Smolensky, 1993)OnsetsONS>>F F>>ONSCodas -COD>>F1F>>—COD ZCV(C) Z(C)V(C)The above typology is consistent with that of Jakobsonabout syllable structure:“There are languages lacking syllables with initialvowels and/or syllables with final consonants, butthere are no languages devoid of syllables withinitial consonants or of syllables with finalvowels.”(Jakobson 1962, Quoted in Prince & Smolensky 1993)When the Faithfulness constraints dominate bothstructural constraints (ONS,-COD), respecting the input ismore important than obeying ONS and -COD. For example, thestring /CVC/ will be parsed as a closed syllable, violating-COD, and the string /V/ will be parsed as an onsetless20syllable, violating ONS. This gives the language (C)V(C)2 asthe table in (12) shows. On the contrary, a more aggressiveparsing of the input will result when a member of theFaithfulness constraints is dominated by either, or both of,the structural constraints. In the case where ONS dominatesthe Faithfulness constraints, every syllable must have anonset. The string /V/ can either remain unsyllabified,violating PARSE, or be parsed as .LI3V., violating FILL. Thisgives the language CV or cv(c)• The case where -CODdominates a Faithfulness constraint corresponds to languagesin which codas are forbidden (XCV orWe can extend our explication by showing the interactionof the structural constraints ONS and -COD with theFaithfulness (PARSE, FILL) constraints. Let us discuss theinteraction of ONS with Faithfulness first. The followingtableaus and explanations are largely from Prince andSmolensky (1993) chapter 6.Suppose we have an input /V/. If ONS is dominated byboth of the Faithfulness constraints, the optimal parse willbe .V. as the tableau (13) shows.2The notation Yz denotes the language whose syllables fitthe pattern XYZ (Prince & Smolensky 1993)O represents an empty structural position.21(13)Onset Not Requiredtr.v.FILL PARSE ONS**1Notice here that the relative ranking of FILL and PARSEhas no effect on the outcome. The violation of PARSE andFILL is fatal because there is a candidate which satisfiesboth constraints: namely .V., since the syllable does nothave to have an onset in this case. The candidates <V> and.LIV. can be the optimal parse when we have the followingranking of the constraints.(14)Enforcement by Overparsing (Epenthesis)/V/ ONS PARSE FILLV I....... ........................................................................................<V> *1LV *(15)Enforcement by Underparsing (Deletion)/V/ FILL ONS PARSEV *1crr<v> *II liii!! III!!!!!! I 111111 I I II I I II*1/V/.LW.22The above cases show that we must have an onset. If FILLis the lowest ranked constraint, a syllable is enforced byoverparsing (.LIV.). If PARSE is the lowest, then a syllableis enforced by underparsing (<V>) . Therefore, we can drawthe following conclusion: onsets are optional if PARSEand/or FILL dominates ONS, and mandatory if ONS dominateseither PARSE or FILL.The same line of argument can be presented for Codacases. The following three rankings determines appropriateoutputs for each case.(16)Coda Not Forbidden/CVC/ FILL PARSE -CODcvc. ‘SI<C)’ . !.CV.CO. *!(17)Enforcement by Overparsing (Epenthesis)/CvC/ -COD PARSE FILLCvC *1111111 I I!. .* !crCVCLI *23(18)Enforcement by Underparsing (Deletion)/CVC/ FILL -COD1PARSECVC *1c.Cv<C>._____________ ____________________________.CV.CEI.Codas are optional in a language if -COD is dominated byboth PARSE and FILL as in (16). Codas are forbidden if -CODdominates either PARSE or FILL. If PARSE has the lowestranking, then the optimal form will be realized byunderparsing (phonetic deletion) as in (18) . If FILL is thelowest then it is realized by overparsing (epenthesis) as in(17)The heart of constraint interaction in CT is to buildindividual phonologies directly from universal principles ofstructural well—formedness. Universal Grammar provides a setof highly general well-formedness constraints. These oftenconflicting constraints are all operative in individuallanguages. Languages differ primarily in the way they rankthese universal constraints in a strict dominance hierarchythat determines the circumstances under which constraintsare violated. Therefore, a language-particular grammar is ameans of resolving the conflicts among universal constraints(Prince & Smolensky 1993)In this section, we have seen that these changes in theranking of universal constraints give different types of24syllable typologies. Later, we will see that the changes inranking can also characterize the different historicalstages in a single language as we discuss Korean vowelharmony.2.2.2. Prosodic MorphologyAs we have briefly seen in Chapter One, OptimalityTheory’s treatment of the harmony system is quite differentfrom the standard approach. In this section, we will observehow Optimality Theory treats prosodic morphology, and theway in which OT’s prosodic morphology differs from thestandard prosodic morphology based on the argumentspresented in Mccarthy and Prince (1993a)The standard theory of prosodic morphology is embodiedin the following three core principles.(19) Principles of Standard Prosodic Morphology(McCarthy & Prince, 1986)a.Prosodic Morphology HypothesisTemplates are defined in terms of the authenticunits of prosody: mora(i), syllab1e(),foot(Ft), prosodic word (PrWd).b.Templatic Satisfaction ConditionSatisfaction of templatic constraints isobligatory and is determined by the principles25of prosody, both universal and language-specific.c.Prosodic CircumscriptionThe domain to which morphological operationsapply may be circumscribed by prosodiccriteria as well as the more familiarmorphological ones.The essence of the above principles is that templaticand circumscriptional morphology are governed by universaland language-particular constraints on prosodic well—formedness. Optimality Theory (McCarthy & Prince 1993a),however, gives a rather different perspective on theseprinciples. The Template Satisfaction Condition and ProsodicCircumscription can be seen as fixing a dominance relationbetween the well-formedness constraints of two differentdomains: namely, prosody(P) and morphology(M): P>>M. Simplyput, the constraints on prosodic structure take precedenceover the constraints on morphological structure in templaticand circumscriptional morphology (McCarthy & Prince1993a:102) . This is illustrated by showing the AxinincaCampa Reduplication case from McCarthy and Prince (1993a)chapter 5•4 The reduplication process is suffixal inI this study, we will only discuss the cases which arerelevant to show the ranking schema P>>M.26Axininca Campa. The crucial data that we will discuss isgiven in (20)(20) Base Reduplicationa. /osampi/ osampi-sampi ‘ask’b. /osaijkina/ osaijkina-sagkina ‘writec. /n-osampi/ n-osampi-sampi I ask’d. /n-osaijkina/ n-osaijkina-sakina ‘write’e. /n-apii/ n-apii-napii ‘repeat’f./naa/ naa-naa ‘chew’Axininca Campa reduplication involves C-initial roots aswell as V—initial roots. We will, however, omit anydiscussion of the C-initial roots since they are not thecase of interest here. The relevant constraints arepresented in (21)(21) Constraintsi)ONS; Syllables must have an onset.ii)FILL; Syllable positions must be filled withan underlying segment.iii)M1X; The reduplication is phonologicallyidentical to the Base.iv)R<ROOT; The Reduplicant (R) contains only theroot.v)DISYLL; The Reduplicant is minimally disyllabic.27The ONS constraint, already familiar from the previoussection, requires an onset for each syllable. FILL forcesthe node in the prosodic tree to be filled with a segment.MAX is a familiar feature of reduplicative theory thatnecessitates that the Reduplicant be an exact copy of itsbase (Steriade 1988a) . R<ROOT characterizes themorphological composition of the source of the Reduplicant,demanding a kind of morphological integrity (Mutaka & Hyman1990). Finally, DISYLL requires a reduplicant of a certainminimal size (two syllables), a kind of generalizedtemplatic restriction. The last three constraints pertain tomorphological structure while ONS and FILL apply to prosodicstructure.Let us, then, look at reduplication with unprefixed V—initial roots ((20) a,b) Since osampi-sampi is the optimalform, more harmonic than the total reduplication case,osampi-osampi, we can see that ONS dominates MAX.(22)Candidates ONS MAXo.sarn.pi-o.sam.pi *o.sam.pi-sarn.pi *The Base o.sam.pi must violate ONS, because other optionssuch as <o>.sam.pi or Lio.sam.pi are excluded by higher rankingconstraints PARSE and FILL. But the Reduplicant should not28violate ONS even at the price of MAX violation. This showsthat MAX is crucially dominated by ONS. The prefixed formsof V-initial roots copy neither the prefix nor the root-initial syllable as in (20)c,d. Just as in the above—mentioned case (22), the constraint ONS excludes the totalroot reduplication candidate, n-osampi-osampi. Total Basereduplication (which includes prefix n- in this case) isruled out by R<ROOT because the Reduplicant must containonly the root according to this constraint. Therefore, theactual output form violates only low-ranking MAX as tableau(23) shows.(23)Candidates ONS R<ROOT MAXn—osampl—nosampin—osam1i osampi * !n—osamp1—samp1Here, the ranking between ONS and R<ROOT is not clearwhile MAX is the lowest ranking constraint for it isviolated in the actual output form. The next case ((20)e),however, demonstrates that the morphological constraintR<ROOT is indeed ranked lower than prosodic constraint, ONS.Consider tableau (24)29(24)____________ _______________________________________Candidates ONS Disyll R<ROOT Max*4,4,The form that violates R<ROQT can still be optimal sinceall the other candidates violate highly ranked constraints,namely, ONS and DISYLL.The next example, (20)f, shows that another prosodicconstraint FILL dominates the morphological constraintDISYLL.(25)________________Candidates FILL DISYLLnaa-naa *naa.E1L1—iia.a.E1L1 * ! * * *The form naa-naa is optimal even though it violatesDISYLL since the other candidate is a violation of thehigher constraint, FILL.The analysis of Axininca Campa shows that the prosodicconstraints ONS and FILL crucially dominate reduplicativeconstraints such as DISYLL, R<ROOT and M7X. That is, thereduplicative constraints can be violated even in an optimalform. Frequently, the Reduplicant is an inexact copy of then-apii-pjjn-apii.pjj *!n-apii-napii30base, violating MAX. Sometimes the Reduplicant containsaffixal material, in violation of R<ROOT (n-apii-napii)Sometimes, the Reduplicant is monosyllabic, in violation ofDISYLL (naa-naa).Certain aspects of CT lead to a very differentconception of templates and the Template SatisfactionCondition. Templates are constraints on theprosody/morphology interface (McCarthy & Prince 1993a) andtemplatic constraints are violable in principle like allother constraints in CT. We will conclude this section byviewing the principles of Prosodic Morphology within CT fromMcCarthy and Prince (1993a).(26) Prosodic Morphology within Optimality Theory(Mccarthy & Prince 1993a:103)a.Prosodic Morphology HypothesisTemplates are constraints on the prosody/morphology interface, asserting the coincidenceof morphological and prosodic constituents.b.Template Satisfaction ConditionTemplatic constraints may be undominated, inwhich case they are satisfied fully, and theymay be dominated, in which case they areviolated minimally, in accordance with generalprinciples of Optimality Theory.31c.Ranking SchemaP>>MThe ranking schema (P>>M) is a broad assertion about thenature of prosodic morphology: if some morphological domainis to be prosodically conditioned, then in that domain Pmust dominate M (Mccarthy & Prince 1993a). This developmentreduces the theory of Prosodic Morphology to the P>>M schemaand a set of constraints, all violable in principle.As mentioned earlier, CT gives a very differentperspective and conception regarding Prosodic Morphology. Inthe following section, we will show that certain aspects ofOT lead to a rather different view of harmony systems aswell. Furthermore, P0K vowel harmony, which will bepresented in the next chapter, contrasts CT’s treatment ofneutrality in harmony systems with that of conventionalderivation-based theory.2.2.3. Vowel HarmonyIn this section, we will examine in greater detail theimplications of CT for the treatment of vowel harmonysystems. We will begin our analysis of CT’s treatment ofvowel harmony with a summary of the argument presented inPulleyblank (1993) regarding tongue root harmony patterns inStandard Yoruba, a Kwa language spoken in Nigeria. The basichypothesis of his analysis is that formal patterns of32harmony are derived by the alignment constraints of OT(Prince & Smolensky 1993, McCarthy & Prince 1993a,b,Kirchner 1993a) while substantive properties of harmony arederived by Grounding Theory (Archangeli & Pulleyblank, inpress) . That is, the central properties of harmony systemsare derived from the interaction of three types ofphonological constraints: faithfulness constraints,alignment constraints and grounded conditions.Pulleyblank (1993) defines these families of constraintsas follows:(27)Faithfulness Conditions (Pulleyblank 1993)PARSE: An F-element Ct. must be dominated by anappropriate node in the prosodic tree.ReoF (Recoverability of F-element) : An F—element(feature or node) a. that is present in an output formis also present in the input.RecP (Recoverability of Path): For any path betweenan F-element a and some anchor f3, if ais associatedto 13 in the output then a is associated to f3 in theinput.The faithfulness conditions, as we have alreadydiscussed in the previous section, are required to parsefeatures into prosodically well-formed structures with aslittle modification as possible. For example, the parse33constraint5 requires a feature (if it is present in theinput) to link to an appropriate mora. Recoverabilityconstraints require that material in the output form also bepresent in the input. The insertion of any feature orassociation line will be proscribed by these constraints.Rec-F prohibits changes in featural content, and Rec-Pprohibits changes in associative paths.Alignment constraints cause edges of featural domains tocoincide with the edges of phonologically, ormorphologically, defined domains. This is stated in (28).(28) Alignment Constraints (Pulleyblank 1993)ALIGN-L(o,L;D,L) : The left—edge of feature c alignswith the left-edge of domain D.ALIGN-R(c,R;D,L): The right-edge of feature c alignswith the right-edge of domain D.For example, ALIGN-L (-ATR,L;Word,L) will require theleft edge of any [-ATR] specification to be aligned with theleft edge of the word.Finally, the grounded conditions, which are proposed inArchangeli and Pulleyblank (in press), function to increasethe phonetic well-formedness of feature combinations. Thebasic claim is that implicational relations between features5This constraint is, in essence, the same as (7) whichdeals with prosodic structure.34must be phonetically motivated or grounded. “Such groundedconditions govern properties of segmental inventories,harmonic neutrality (opacity and transparency), rules ofredundancy, conditions on the targets and triggers ofphonological rules, and so on.” (Pulleyblank 1993:8)The first substantive type of implication advocated bygrounded conditions in Archangeli and Pulleyblank (in press)involves the movement of the tongue root implied by a changein tongue height. Consider, for instance, the tongue rootwith an upward movement of the tongue body (formally[+highj) . Raising of the tongue body ([+high]) tends tocorrelate with tongue root advancement ([+ATR]) because ofthe physiological fact that tongue is a single volume.6 Onthe other hand, the downward movement of the tongue body([+ low]) tends to correlate with tongue root retraction([-ATR]) . These properties can be phonologically expressedas follows:(29) Grounded Path Condition:c/ATR(Archangeli & Pulleyblank, in press:148)a.HI/ATR Condition:If [+high] then [+ATR],not{-ATR].b.LO/ATR Condition:If [+low] then [-ATR],not[+ATR].6See the diagrams in Archangeli & Pulleyblank (inpress:147)35The second type7 involves changes in tongue heightimplied by a tongue root movement. Movement of the tongueroot tends to be accompanied by a sympathetic movement ofthe tongue body. These dependencies are expressed formallyin the following four path conditions.(30) Grounded Path Conditions: ATR(RTR)/cL(Archangeli & Pulleyblank, in press:150)a.ATR/HI Condition:If[+ATR] then[+high],not[-high].b.ATR/LO Condition:If[+ATR] then[-lowJ, not[+low].c.RTR/HI Condition: If[-ATRJ then[-highj ,not [+highj.d.RTR/LO Condition:If[-ATR] then[+low], not[-low].Archangeli & Pulleyblank (in press) claims that theabove conditions (29) and (30) alone reflect thephysiologically preferred configurations of tongue bodyheight and tongue root retraction/advancement.The three types of phonological constraints that wehave discussed ——namely, faithfulness constraints,alignment constraints and grounding conditions-- will alsobe the central constraints that derive the Korean harmonysystem even though each stage of the language has adifferent ranking of these constraints.7Even though we only need the above two conditions for thediscussion of Yoruba case here, I will include othergrounded conditions to give a general picture of GroundingTheory. Moreover, we make use of the RTR/LO condition, (30)d,in the discussion of Korean vowel harmony in Chapter Three.36With the above mentioned constraints in mind, let usconsider the case of Yoruba tongue root harmony8. Tongueroot harmony in Yoruba involves the spreading of the feature[-ATR] (Archangeli & Pulleyblank 1989, 1993, Pulleyblank1993) as in Korean. Pulleyblank (1993), as in Archangeli &Pulleyblank (1989), assumes that there are two types ofmorphemes: morphemes with and without the lexical [-ATRJspecification.(31) Mid vowels in Yorubaa.ègè ‘dirge’ ‘cassava’èké ‘lie’ eke ‘forked stick’èse ‘cat’ ese ‘row’The morphemes in (31)a do not have lexical [—ATRIspecification and therefore surface as advanced vowels.Vowels in (31)b appear retracted since they have a lexical[-ATRJ specification. Harmony of this type is explained bythe interaction of the ALIGN constraint with theFaithfulness constraints (PARSE,RecF,RecP)8All the data, tableaus and relevant discussions are fromPulleyblank (1993)37(32) Mid vowels, with and without a lexical [—ATRIspeci ficat ionRecF prevents the insertion of [-ATR] where it is notpresent lexically as in (32)a. Therefore, the optimal formwill be the one with advanced vowels (ege). However, whenthere is a lexical [—ATRJ specification, as in (32)b, theform in which [—ATR] is linked to both vowels will be theoptimal one, even with the two violations of RecP. OtherREC—F TT(ThL..12 T -—c-pYoruba ][PRSEa. egeV-atr_______/EgE/ ege\-atr‘dirge’ ege/-atregeb. eseV-atr/E5E/ ese\[—atr) -atr_‘row’ ese/-atrese-atr38candidates are worse because they violate higherconstraints, ALIGN and PARSE. In this case, the ALIGNconstraints play a crucial role; these constraints, however,are not as important when we discuss cases like high vowels(34) and the Later Middle Korean data in Chapter Three.The high and low vowel cases involve groundedconditions as well as ALIGN and Faithfulness constraints.Let us consider high vowels first. As shown in (33), highvowels can cooccur with both advanced and retracted vowels.(33) High vowels in Yorubaa.bi ‘guilt’ b. ebi ‘hunger’etiri ‘difficult’ èñgI ‘molar tooth’ide ‘brass’ ilé ‘house’The cases in (33)a are problematic if we expectcomplete harmony without considering any feature cooccurenceconditions like HI/ATR. Since the cases in (33)b arestraightforward (these morphemes do not have [-ATR]specification, and therefore, surface as all advancedvowels), the following tableau will include the data from(33)a only.39The second candidate of (34)a is optimal even though itviolates ALIGN—R, because the grounded condition HI/ATR isranked higher. Therefore, any form with a retracted highvowel(I) will be ruled out in the first place. The lastcandidate of (34)a is a violation of PARSE since the feature[-ATRI does not associate to the appropriate mora. (34)b canbe explained in the same manner. The incompatibility of highYoruba(34)High vowels with a lexical {-ATR] specificationHI/ATR PARSE REc-F ALIGN—R ALIGN-L REC-Pa ebi I **V—atr/EbI/ ccbi\[—atrj —atr‘gui_it’ eI,i *!I-atr11111 IllIllIllIll liii I 111111 II Iebi *1\\-atrb ide *1V-atr1111111 111111111 I 11111111111 111111 I IlIlIllIllIll 11111111 I II II 111111/IdE/ * *[-atr)111111111 I I II II!! II 1111111 IllIllIllIll liii‘brass’ ide *I-atride-atride\-atr40vowels with retraction, represented by the highly rankedconstraint HI/ATR, causes the general pattern of harmony tobe interrupted in the appropriate configuration(Pulleyblank, 1993)The low vowel cases involve another grounded condition,namely, LO/ATR.(35) Low vowels in Yorubaa.àj ‘witch’ b.âfè ‘Spotted Grass—mouse’abbé ‘needle’ àwo ‘plate’a13 ‘cloth’ àdI ‘palm-nut oil’c.êbà ‘food made from gari’ *ebaegba ‘whip’ *egbacrákpó ‘type of plant’ *erapoThe low vowels are systematically retracted. Thisproperty is accounted for by ranking the LO/ATR conditionhigher than recoverability. Consider the followingconstraint tableau for a form like eba, ‘food made fromgari’.41[REC_F ALIGN-R [IGN_L REC-P* **I:+ :“I’llThe form eba (as opposed to ebe) is optimal even thoughit requires the insertion of a feature (a RecF violation)since the presence of a low advanced vowel () violates thehigher ranked constraint LO/ATR.9The following tableau shows morphemes with initial lowvowels ((35) a,b).9Pulleyblank(1993) mentions that the input with lexical[-ATRJ specification will also derive an identical surfaceform (see Pulleyblank 1993:15)(36) Morphemes with final low vowels, no [-ATR]specificationYoruba LO/ATR HI/ATR PJ\RSEebaV-atr/EbA/‘food’ cbc\-atrebaI-atreb *!Pulleyblank (1993) relies, crucially, on the assumptionthat the ALIGN-R condition is restricted to the domain ofthe ROOT in this case. In other words, the inserted feature([—ATR]) is not subject to the ALIGN-R condition as theoptimal form in (37)b (afe) shows, since the [-ATRI value insuch a case is not a ROOT specification. However, thelexical [—ATRJ in (37)a is a ROOT specification and,therefore, is subject to the ALIGN—R condition (i.e. *aje)Yoruba(37)Morphemes with initial low vowelsLO/ATR HI/ATR42PARSEajeV-atrALIGN-R ALIGN-L REC-Paje\-atra./AjE/[-atr]‘witch’jeI-atreje-atr*!afeV-atrb./AfEl‘mouse’* **!afe-atrefe/-atrefe* ** ‘H *43The important point emerging from this work is thatgrounding condition theory (Archangeli & Pulleyblank, inpress) can be directly incorporated into the fabric ofphonological theory within the Optimality Theory framework.In the next chapter, we will see that the harmonyphenomenon of Korean can also be explained in terms of theinteraction of alignment, faithfulness, and groundedconditions, as in Yoruba. Harmony in Korean is intimatelytied in to the discussion of Yoruba vowel harmony: not onlyare the patterns of vowel harmony formally similar, butKorean, like Yoruba, observes tongue root harmony. Thegrounded conditions have the effect of blocking the effectsof harmony in both systems. Differences are also observed inthe two harmony systems: grounded conditions override REC-Fin Yoruba low vowel cases in (36); we will see, however,that REC-F is still inviolate in the optimal form ——REC—F isranked higher than grounded conditions-- in Korean. Anotherdifference exists in that grounded conditions --as they areranked higher than ALIGN-R-- define the harmonic edges inKorean, but not in Yoruba. Furthermore, it is interesting tonote that the language-particular variation (i.e. Yoruba vs.Korean) is due to the difference in ranking of members ofthe universal constraint set.44CHAPTER 3KOREAN VOWEL HARMONYThis chapter will examine the vowel harmony systems inPresent—Day Korean and in Later Middle Korean. The focus ofthis study is to determine how Optimality Theory accountsfor the difficulties in the harmony system of this language.As we have seen in Chapter One, the vowel harmony ofKorean has long been a problem for two reasons: it is notclear how to characterize the features that distinguish theharmonizing sets; and the neutrality of front unroundedvowels ([i] and [ii) in the harmony system is yet to beexplained, having been treated just as an exception in mostanalyses proposed for Korean vowel harmony.In this study, we will investigate the hypothesis thatKorean vowel harmony can be described by the feature [ATRi(Song,1990) and that OT can give a principled account of theneutral vowels in the harmony system through an interactionof constraints on alignment and grounded conditions. We willalso discuss how CT treats diachronic change in a harmonysystem by showing the development of vowel harmony throughLater Middle Korean to Present-Day Korean. We know that allconstraints are present in principle in the CT hypothesis.The diachronic change in the harmony system can be accounted45for by reranking some of the constraints. Certainconstraints still play a crucial role after the rerankingwhile some of the constraints are no longer crucial aftermoving down in the hierarchy. We will see that both types ofchanges are found in Korean vowel harmony.3.1. Present-Day KoreanThe vowel system is the preliminary basis for the studyof vowel harmony in any language. It is, therefore,worthwhile to review the vowel system of Korean in order todiscuss vowel harmony. We will look at the development ofthe vowel system from Later Middle Korean (henceforth; LMK)to Present-Day Korean (henceforth; PDK) •1 LMK (from thefifteenth to the sixteenth century) had seven underlyingvowels:(38) i i u0a AOf these seven vowels [o,A,aJ have been called ‘light’(yang) vowels and [u,i,j ‘dark’ (yin) vowels; [ii isconsidered neutral. These two vowel sets, namely ‘dark’ and1The vowel shift in the previous stage (Old Korean to MiddleKorean) is disregarded in this study.46‘light’, each form a harmonic group as in (39). (The arrow<--> denotes harmonic pairs.)(39) neutral dark lighti U <---> 0i A<--> aIn the eighteenth century, the vowel [A] was lost. As aconsequence, [a] came into opposition with [ii while itretained the old opposition with [a]. Then, in thenineteenth century, we encounter two more front vowels [e]and [eJ caused by monophthongization (y --> e, ay --> c) . Sincethe beginning of the twentieth century, Korean has had aflourishing vowel system with two additional front roundvowels, [ö] and [U], brought about throughmonophthongization (oy --> o, uy --> U) . This derives the PDK 10vowel system shown in (40).(40) i U i uC O 0C aPDK obtained four additional vowels [e,e,U,O] throughmonophthongization and lost one vowel [A]; and [ci isderived from both [Ay] and [ay]. In this study, we willdevise a phonological 12 vowel system for PDK based on47historical grounds. The vowels [e] and [a] have twodifferent sources due to the loss of [A]. That is, PDK [ciis derived from LMK [Ay] and [ay], and PDK [a] from LMK [A]and [a]. Therefore, the phonological 12 vowel system will beas follows:(41) i U (uy) i ue(y) o(oy) o(Ay) a1 (A)c2 (ay) a2The vowels in parentheses indicates the LMK vowels fromwhich the P0K vowels are derived. [e1] is distinguished from[e2] in that it is developed from [Ay] while [c2] is from[ayj . [a1] indicates the reflex of [A], and [a2] is the LMK[a]. The following cognates show that the above claims areborn out.(42) LNK P0Kuy U ‘up’ny ne ‘you’èamoy ëamö (ye11ow) cucumber’OPAY pope ‘treasure’hay he ‘to do’pAlAm palam ‘wrnd’na na ‘I’48Although V1 and V2 do not have any phoneticdifferences, it will be shown in the following section thatthey play a phonologically distinctive role in the harmonysystem. In other words, the neutralization between [a] and[A] (as well as [ay] and {Ay]) in PDK is realized only atthe phonetic level and the distinctions still remain activephonologically. We may assume that the vowels derived fromLMK [+low] vowels2 ——namely [j and [a], remain as [+low]vowels ([e], [a], [e]’ [a2]) in PDK phonology. Hence, the followingfeature specifications for PDK vowels can be drawn.(43) PDK vowel feature specification3U O i a1 e c2 U 0 0 a2atr (_)4 () (-) (-) (-) (-)md + + + +bk + + + + + +low + + + +In the above specifications, [e] and [e2] are treatedas [+low], whereas {c1] is not [+low] even though it isphonetically lower than [e]. Archangeli and Pulleyblank (inpress) argue that a retracted high vowel [I] —--which is2We will discuss LMI< vowel feature specifications in a latersection.I assume that values not specified in this table will beassigned by context—free redundancy rules:]—>[+atr], [ ]—>[—rnd], [ ]—>[—bk], [ ]—>[—low].4Parentheses indicate that [-ATR] is not specifiedsegmentally but it is a property of a morpheme.49phonetically “lower” than an advanced mid vowel ([e])-— mayoccur. That is, even though the vowel [e] is phoneticallyhigher than [i], [ii is phonologically considered [+high],while Eel is not. Similarly, [e] is considered [+10w] while[e1] is not, even though the vowel [e1] is phonetically lowerthan [e] in Korean.In PDK, the vowel harmony phenomena can only be foundin sound symbolic words and in the alternation of the socalled, A—initial suffixes5 following verb or adjectivestems (Ahn 1985, C.W.Kim 1978, J.M.Kim 1986, Kim—Renaud 1986,etc.)3.1.1. Sound Symbolic WordsSound symbolic words (or ideophones) are words denotingsound, smell, taste, color, mood, size, type of movement, orother perceptual experiences (Kim-Renaud, 1986). As in LMKvowels, PDK vowels are divided into two sets in soundsymbolic words. The vowels [i,e,U,i,,u] form a harmonic grouptraditionally called ‘dark’ vowels which denote darkness,depth, bigness, etc., and the vowels [c,ö,a,o] form the otherset called ‘light’ vowels which gives an impression oflightness, shallowness, smallness, etc. (Kim-Renaud,1986).Within a sound symbolic word, vowels must be either all darkor all light. The only exceptions to this rule are the high5We will use ‘A’ to indicate [] or [a] as Song (1990) did.50unrounded vowels [i] and [ii which are neutral in non-initialsyllables and therefore may cooccur with either dark orlight vowels. Some examples of sound symbolic words aregiven in (44)(44) Dark Lighta. dilc’l cek’al ‘chattering’kilc’uk kelc’ok ‘tall’b. kelkel kelkel ‘exhausted’c. tils’k tals’ak ‘lifting’kilc’k kalc’ak ‘scratching’silc’k sale’ ak ‘stealthy’k’it’k k’at’ak ‘nodding’d. p’11cn p’alkan ‘red’spk sapak ‘crunching’kntuij kantoij ‘jumping up anddown’e. p’yulutuij p’yolotoij ‘pouting’sukun sokon ‘whispering’kulkul kolkol ‘sleeping’k’ult’k k’olt’ak ‘gulping’f. hUhU höhO ‘round about’kUcUcU kökökö ‘shabby’g. pisil pesil ‘staggering’k’ ecikk k’ ecilak ‘half-heartedly’pnsil pansil ‘smiling’pusisi posisi ‘shining’51k’umcilok k’omcilak ‘budging’k’ukick k’okicak ‘crumpling’h. sinkil senkil ‘smiling’p’intil p’entil ‘staggering’t’erjkibrj t’erjkilarj ‘clanging’silkim salaam ‘stealthy’hintil hantil ‘rocking’kntiknj kantilaij ‘wobbling’pusil posil ‘drizzling’k’umtil k’omtil ‘wriggling’(Kim—Renaud 1986, McCarthy 1983, Sohn 1986, Song 1990)The data in (44)g and (44)h (the bold faced characters)show the neutrality of high unrounded vowels ([i],[i]) in noninitial syllables. Vowel harmony in sound symbolic words hasbeen difficult to describe because the dark and light vowelsets are not natural classes in the SPE feature system.Hitherto, the neutrality of the high unrounded vowels hasbeen treated merely as an exception to the harmony system.In the following sections, the previous accounts proposedfor PDK sound symbolic words and the problems thereof willbe discussed. We will then view the Optimality account whichgives a better understanding of these problems.523.1.1.1. Previous AccountsSemantic Feature AnalysisKim-Renaud (1986) describes vowel harmony in soundsymbolic words by semantic features based on the traditionaldichotomy, [dark] and [bright]. According to her, vowelharmony results from velar (nonback/back) harmony in EarlyMiddle Korean. Since all and only the back vowels werebright vowels in that stage in time, the feature [bright]was presumably a redundant feature of the feature [back].She claims that phonetic symbolism found in variouslanguages proves that certain semantic features arepredictable from some phonetic features (Sapir 1911, Poppe1950, Haas 1970) . However, a historic vowel shift in Koreanmade the phonetic distinction between the two harmonizinggroups of vowels opaque, since some nonback vowels becameback and vice versa. Therefore, the phonetic feature([back]) distinguishing the two groups seems to have beenreplaced by the semantic feature ([bright]). Kim-Renaud(1986) further claims that the case of Korean suggests thatsemantic features have a role to play in phonologicalanalysis in general. Although her explanation using semanticdiacritic features for PDK vowel harmony gives differentnuances in sound symbolic words, it lacks an exact mechanismfor including the feature [bright] in phonology. Moreover,the direct phonological use of a nonphonetic diacriticfeature requires considerable clarification. We may53consequently conclude, therefore, that her analysis islacking in strong phonological grounding.Diagonal AnalysisVowel harmony systems (Aoki,1968) can be dividedtypologically into two major groups: horizontal (height)harmony and vertical (front/back) harmony. C.W.Kim (1978)suggests a third type, “diagonal”, which is found in suchlanguages as Korean or Nez Perce, etc. This particular typeof vowel harmony is called “diagonal” because the linedividing the harmonic groups runs diagonally in the vowelchart as in (45);(45) a. u/ b. i u/e /o /oNez Perce Korean(C.W.Kim, 1978:221)C.W.Kim (1978) writes that this vowel harmony system isregarded as an unusual phonological phenomenon since nosingle parameter or distinctive feature of vowels candistinguish the two harmonic classes. He concludes that thisdiagonal harmony is a transitional phenomenon from a fullvowel harmony system in Early Middle and Old Korean , andsuggests a sort of adjustment rule to explain the vowelharmony in synchronic grammar. C.W.Kim (1978), however, does54not include front vowels ([i,e,e,ü,ö]) in the inventory; andeven a full inventory including the front vowels cannotsupport the diagonal harmony system. This analysis doeslittle more than recapitulate a historic process; it merelycites diagonal harmony as a result of vowel shift, withoutoffering an in-depth analysis of the phenomenon.Underlying Vowel AnalysisMccarthy (1983) proposed an abstract version of theKorean vowel system in an attempt to represent thedark/light distinction under the universal feature system,assigning [+10w] and [-low] to dark and light vowels,respectively. Vowels are divided into two groups accordingto the values of the feature [low] as in (46)(46) i u i ue dark =[-low]c a light=[+low]The values of the feature [low] constitute morphemes:[-low] is the ‘dark’ morpheme and [+10w] is the ‘light’morpheme. The harmony phenomena are the result of spreadingthe [low] feature. A context free rule, applied subsequentlyto harmony, then takes all [+round] vowels to [-low] (Re]changes to [ö] and [] changes to [o]) . Although this systemworks very well in grouping the harmonizing vowels for thesound symbolic vowel harmony into natural classes, it55arbitrarily sets up low vowels [o] and [cc] as underlyingforms of [01 and [O] respectively, without providing anyplausible grounds for doing so (Park,1990). Moreover,McCarthy (1986) is obligated to include an ad hoc contextfree rule in order to explain the harmony system.Acoustic Feature AnalysisY.S.Kim (1984), in an attempt to find an acoustic basisfor the harmonic distinction of PDK, suggests the feature[Deep Voice Resonance] to group each harmonic class morenaturally, as in (47)(47) i u i u [—DVR]e O c [+DVRJc aHis acoustic feature [DVRI6 is a good device forgrouping the light vs. dark vowels according to a singlefeature. However, he neither describes this feature clearlynor gives its source, lacking any acoustic clue to justifyits use. Furthermore, the necessity of establishing a newdistinctive feature like [±DVR] remains yet to be justifiedfrom a universal viewpoint.6Further study is needed to find a relationship between thefeature [ATRI and [DVR].56Tongue Root Feature AnalysisSong (1990) proposes that Korean vowel harmony can bedescribed as tongue root harmony based on the formantanalysis of Hwang (1983). She claims that vowels in soundsymbolic words can be divided into two sets by the feature[ATRI as below.(48) i U i U dark =[+ATR]eJ ö j o light=[—ATR]e aSong’s (1990) analysis maintains that the harmonysystem is characterized by spreading of the feature [-ATRIShe claims that a light sound symbolic word has theharmonizing feature [-ATR] and harmony takes place byspreading this feature. A dark sound symbolic word, however,has no harmonizing feature. Therefore, each vowel of a darksound symbolic word gets [+ATR] by a redundancy rule. Theoperation of vowel harmony is illustrated in (49) below.7The formant chart and the relevant explanation will begiven in the following section.57(49) /sapak/ — /spk/ ‘crunching’i. sApAk ii. sApAkx x x x x x x x x x SkeletonI I Root nodeI I { SupralaryngealA A A A Place n.- Dorsal,Radical n.-. ——ATR +ATR Association&Spreading (i)Redundancy Rule (ii)/sapak/ /spk/The neutrality of [ii and [ii, in her analysis, isaccounted for by the Locality Condition8. She assumes thatthe feature structures of [ii and [ii are different,depending on whether these are neutral vowels or harmonicvowels. That is, if the vowels are neutral, they do not havea Radical node (spreading takes place from one radical nodeto another in her analysis); therefore, these vowels areskipped by the harmonic process as a result of the LocalityCondition. The redundant feature [+ATR} is subsequentlyassigned to these vowels. She also assumes that theharmonizing feature value [—ATR] and the redundant featurevalue [+ATR] are present on different planes to avoid aviolation of the Line Crossing Prohibition (Goldsmith 1976)8See Song (1990:22)58Song’s analysis is attractive in that she takes [ATR]as the harmonizing feature. However, her treatment ofneutral vowels relies only upon the assumption that vowelshave different feature structures depending on whether theyare neutral or harmonic vowels; Song (1990) fails to giveany independent motivation in defense of her approach. Optimality AccountAs we have already seen, researchers have encounteredconsiderable difficulty in finding a single feature thatcharacterizes PDK vowel harmony in sound symbolic words.Song (1990) argues against McCarthy’s analysis and proposesthe tongue root feature [ATR] as the harmonizing feature toaccount for vowel harmony systems such as Korean and NezPerce. In this study, we will basically follow Song (1990)as she takes the feature [ATRI to distinguish dark vowelsfrom light vowels in PDK. She provides acoustic evidencebased on Hwang’s (1986) formant analysis to argue for thefeature [ATRI as the harmonizing feature. The formant chart9is given in (50)9Because Hwang (1986) considered [U] a diphthong while heconsidered [ö] a monophthong, he deals with only [ö] but not[U] . Song (1990), therefore, omits the value of [O] in theformant chart and deduces the contrast of [ATRI valuebetween [U] and [O] from that between [U] and [0].59:r::::::::::::::::::::::::::——i———- -——- -—. .— — -— — I — — — - - — — - - — . — — - - — —— —j2I —- — — - - — — - - — - - — — — - - — — - - - — — — —::::s::::::::::::::::::::::‘::::::::::::::::::::::;::::‘-—- - -—— -———--3•“--r------i_I--—- -——- - —— - -s—- -—— -- -—- -———- - —-- -———- . . -: : ::: :-- ---‘;- - :3- --- -I——————• c— — ———— p——— —-- --- - ---.,—---——---—--—-.———--——---————)— - -—— -tL__7__EEEEEIEEEEII 10 • I 7unIt 100 qsFormant Chart“The major acoustic characteristic of [+ATR] is alowering of Fl and concomitantly a downward shift of P2 forback vowels and an upward shift of P2 for front vowels. Inthe formant analysis of present—day Korean the frequenciesof Fl for [e,a,o] are higher than those of [i,e,i,,u] . Thefrequencies of P2 for [i] and [e] are higher than that of[e], and the frequencies of P2 for [ii, [],and [U) are lowerthan those of [a] and [o], respectively.” (Song,1990:40)Despite the fact that there has been a vowel shift fromLater Middle Korean to Present-Day Korean, both systems canbe characterized by the same feature. Hunminjongum (the(50)F242$ 21 27 2$ 21 14 12 12 21 20 10 II Ti II II 14 12 II II 10 I I 7 I60Korean Script of circa 1400 A.D.) describes [o,A.aj as“retracted”, [u,i,] “somewhat retracted” and [i]“unretracted”. Based on this description, C.H.Park (1983)suggests an interesting hypothesis on the six vowel systemof Later Middle Korean. He uses [±ATR] instead of“retracted”. In his system, [+ATR] includes [i,u,i,], the darkvowels, and [-ATRI includes [A,a,o], the light vowels. Wewill see in the following section that this is consistentwith the featural specification of LMK.In light of these analyses, we may conclude that thevowel harmony in PDK and LMK is characterized by the feature[ATRI . Vowels can be divided into two sets by the feature[ATRI as in (51)(51) i U i u dark [+ATR]o light=[—ATRJTo capture the morphological relation between dark andlight sound symbolic words, Song (1990) argues that there isa dependence of light sound symbolic words on dark soundsymbolic words. She claims that there are ‘dark’ soundsymbolic words without ‘light’ counterparts, but there areno ‘light’ sound symbolic words without ‘dark’ counterparts.She presents the following examples to back up the claim.61(52) dark lightmullcij *mo&aU ‘soft’isisi *asisi ‘chilly’Based on this dependence of light sound symbolic wordson dark sound symbolic words, Song (1990) concludes thatonly [-ATR] can be considered a morpheme since dark soundsymbolic words are considered underlying forms. Contrary toher assertions, however, there are light sound symbolicwords without dark counterparts.(53) dark light*c’1c’l c’alc’al ‘busy’*smsm samsam ‘not too salty’Therefore, the argument that dark sound symbolic words areunderlying forms has no support. McCarthy (1983) argues thatthe values of the feature [low] constitute morphemes for PDKsound symbolic words: [+10w] is the ‘light’ morpheme, [-low]is the ‘dark’ morpheme. Our account on the morphology ofdark vs. light sound symbolic words is similar to McCarthy’sidea in that there is no dependence of light sound symbolicwords on dark sound symbolic words. But unlike McCarthy, Ijust assume that the feature [—ATR] is the morpheme thatmarks the light class of sound symbolic words, while darksound symbolic words do not have this morpheme. That is,words that denote light sound symbolism have lexical [—ATRJ62specification as a result of morpheme concatenation, whiledark sound symbolic words do not have such a specification.Before proceeding with the actual analysis of sound symbolicwords, we will discuss some of the constraints that areneeded for the task.As we have already seen in Chapter Two, the basic ideaof optimization is to formulate a set of constraints andrank them in a “strict dominance hierarchy” (Prince &Smolensky 1993) instead of having ordered rules. Therefore,there is no notion of step-by-step input-output derivation.What we consider is a set of candidates produced by Gen;then we can evaluate them in relation to the appropriateconstraint ranking for the particular language. (54) showsthe constraints for the grammar of PDK sound symbolic words.Some of the constraints have already been discussed.(54) ConstraintsPARSE: An F-element must be dominated by anappropriate node in the prosodic tree (Pulleyblank1993)REC—F (Recoverability of F-element) : An F—elementthat is present in an output form is also presentin the input (Pulleyblank 1993)ALIGN-x—L/R:The left/right edge defined by feature a.aligns with the left/right edge of Domain D(Pulleyblank 1993)RTR/LO: If [-ATR] then [+LO]; if [-ATR] then not63[-LOJ (Archangeli & Pulleyblank 1993)RD/RTR: If [+RD] then [-ATR]; if [+RDJ then not[+ATRJPARSE forces a feature to be dominated by a prosodicnode: if [—ATRI is present in the input, it must be linkedto the appropriate mora. Otherwise, it will incur aviolation of the constraint as we will see in light soundsymbolic words which have the lexical feature [-ATR]. REC—Fensures that no feature can be added freely. When there isno [—ATRI feature in the input, as in dark sound symbolicwords, the insertion of [-ATR] will incur the violation ofthis constraint. The ALIGN constraint has the role oflinking a feature to the left/right edge of each domain; if[—ATR] fails to link with the first mora of the word, itwill be a violation of ALIGN-L. ALIGN-R will be violatedwhen [-ATR] is not linked to the right edge of a morpheme.The grounded condition RTR/LO prevents the [-ATRI featurefrom linking with nonlow vowels. Round vowels must beassociated to [-ATR] according to RD/RTR condition; thestatus of the RD/RTR condition, however, is yet to beestablished. According to Archangeli & Pulleyblank (inpress), some languages involve less strongly groundedconditions such as BK/ATR or L-tone/ATR. Manifestations ofthese conditions appear to be much less frequent. The RD/RTRcondition is an example of a weakly grounded condition in64that it does not have a strong phonetic basis and is notfound in many languages.Words with lexical [-ATRI specification; Light Sound Symbolic WordsLet us begin our analysis with the case of light soundsymbolic words which have a lexical [-ATR] specification.The symbols in the input forms of tableau (56) and thefeature specification of these symbols devoid of [—ATRJ areas follows.(55)__ ____O A E Ai C1 U O i a1 e u o a2md + + + +bk + + + + + +low + + + +Let us examine the candidate forms for the word ‘tall’.65(56)_____ ____Korean PARSEkElc’Ok kilc’uk *!-atr___[—atr] kilc’ok/-atr‘tall’ ke1lc’uk\-atrke lc’ok\/____________-atrHarmony of this basic type is accounted for by theinteraction of Faithfulness constraints (REC-F, PARSE) withconstraints on alignment (ALIGN-L, ALIGN-R), as in YorubaATR harmony.’° Since the constraint PARSE is ranked highestin the hierarchy, the optimal form --when there is a lexical[-ATRI specification in the input—— will be the one in whichthe feature [-ATR] is parsed to vowels. REC-F is irrelevantin this case because the light sound symbolic morphemes havethe lexical [-ATRI specification as a result of thespecification for “lightness”; REC—F will, however, becomerelevant when we deal with the cases with no lexical [—ATRIspecifications. The first candidate for the word ‘tall’ isruled out since it violates PARSE. The next form kilc’ok‘°The grounded conditions will play a crucial role when wedeal with forms including neutral vowels.-F_ALIGN-L RD/RTR i:66violates ALIGN-L ,a fatal flaw since all the rest of thecandidates violate only lower ranked constraints; kllc’ok alsoviolates the grounded condition RTR/LO because the feature[—ATRJ is linked to vowel [o] which is not [+low]. ke1lc’uk isruled out for violating the RD/RTR condition as the roundvowel [u] is advanced. Therefore, the form ke1lc’ok, since itsatisfies all of the higher constraints with the exceptionof RTR/LO, is the optimal one.The following is a crucial example showcasirig thenecessity for distinguishing [c1] and [e2].(57)The optimal form kc2lk which does not violate anyconstraint at all would have violated RTR/LO twice if [c2]were not considered [+low]. In such cases, the third formwould have been the optimal form giving a surface patternPARSE REC-F ALIGN-L RD/RTR RTR/LKoreankElkEl kelkel-atr[—atr] kelke2l/-atrexhausted’ ke2lk l\-atrke21kcV-atr67[...c...i...j. While a possible Korean form (see (60)), thiswould be incorrect for [kelkel). Therefore, it is importantto note the phonological distinction regarding the surfacevowel [e]: the vowel [e2j is [+low] while [e1j is not.The following cases as well as (57) demonstrate the needfor the ALIGN-R condition.(58)tis’k-atrtis’a2k/-atrta1s’k\-atrta1s’a2kV-atrspok-atrspa2k-atrsa2pk-atrsa2pkV-atrKorean PARSE rr rnT_LtAls ‘Ak[-atr]lifting’* *!**1sApAk[-atr]‘crunching’68For each example, the third candidate is ruled outbecause it violates ALIGN-R. The vowel [a1] violates RTR/LOcondition since it is not [+low]. However, [a2] which is[+low] satisfies this condition.(59)_______ ______________________________________The RD/RTR condition is motivated by the above case (aswell as (56)). Without the RD/RTR condition, koilcul wouldhave been the optimal form since ALIGN-R is ranked lowerthan RTR/LO as motivated in (60) below. kolkol is the optimalform ——even though it violates RTR/LO twice—— since all ofthe other forms, which have a round advanced vowel [U],violate the higher ranked RD/RTR condition.”“The fatal violations of the first and second forms,however, are PARSE and ALIGN-L, respectively.Korean PARSE REC-? ALIGN-L RD/RTR RTR/LO ALIGN-RkOlkOl kuilcul-atr[-atr] kulkol/-atr___ ___‘sleeping’ kolicul-atrko11co1V-atr69We can extend the scope of our analysis to the caseswith neutral vowels. This neutrality of the front unroundedvowels ([iJ,[iJ) in the harmony system has been treated merelyas an exception in most analyses proposed for Korean vowelharmony. In this study, we will see that the problem ofneutral vowels can be solved by the interaction between thegrounded condition RTR/LO and the alignment constraints.Consider the following tableau.(60)Korean PA AT1*!pnsi’-atrI I IpAnsl[—atri‘smiling’-Rpnsy1-atrpa2nsi1\-atrpa2nse1lV-atrpisil-atrpsl[—atristaggeringpisc1l/-atrpc1si1\-atrpe1sclV-atr***!70As we have already seen, high unrounded vowels in non-initial syllables do not participate in harmony. Theneutrality of these vowels in non-initial syllables is theresult of ranking grounded conditions (RTR/LO, RD/RTR) lowerthan ALIGN-L, and higher than ALIGN-R. Simply put, the firstvowels must always be parsed given the fact that PARSE andALIGN-L are higher ranked constraints; however, as ALIGN-Ris ranked lower than any other constraints, [—ATRI tends notto be parsed to the second vowel if it incurs a groundedcondition violation (namely, an RTR/LO violation)To illustrate this, let us look at each candidate forthe word ‘smiling’ in tableau (60) . The first form, pnsi1, isa violation of PARSE, the most highly ranked constraint. Thesecond candidate violates ALIGN—L as well as RTR/LO. Thethird form is optimal even though it violates ALIGN-R; forthe form that does satisfy ALIGN—R (pa2nse1l) violates RTR/LOwhich is ranked higher. Similarly, both the third and fourthcandidates for ‘staggering’ violate RTR/LO; however, theform in which [-ATR] is parsed to both moras (pc1sel) incursan additional RTR/LO violation. The optimal form, therefore,is the one with the ALTGN—R violation (pe1sil)A similar explanation can hold for the neutral vowel [iias tableau (61) will show.71pOsAl[—atr]drizzling’silkim-atrsilka1m/-atrsa1kim\-atrsa lka1mV-atrsinkil-atrsinka1l/-atrsc,a1-atrpusi’-atrpusa1l/-atrp,1’-atr(61)Korean PARSE REC-? -sAlkAm[-atr]‘stealthy’ZT. T (TT—R*!* *!sEnkAl[-atr]‘smiling’ se1nkil\-atr________*posi1 *-atr* *!72The cases in the above tableau demonstrate the necessityfor a distinction between [a1) and [a2]. If [a1) were [+10w),the candidates in which [-ATRI is linked to the second vowelwould not have violated RTR/LO. If this were the case, wecould not explain the neutrality of the vowel [ii.The following tableau shows polysyllabic cases withneutral vowels. Here, the graphic representations are givenonly for the optimal forms.(62)tenka1te2nkibn *1Koreantenki1ntEnkAlAn[-atrjtenkila7nPARSE REC-F ALIGN-L RD/RTR RTR/LO ALIGN-R*1 j I*1**---tenka, 1a,n‘clanging’te2nki1a-atrtenka, iontenka,1an *!73As in bisyllabic cases, [—ATR] spreadsbecause of ALIGN-R, unless the RTR/LO conditionas the examples tc2nkilaij ‘clanging’ and k’okica2show. The representation of the optimal forms‘clanging’ and k’okica2 ‘crumpling’ haverightwardsis violated‘crumpling’of tc2nkilarja “gappedKoreank’ukickk’OkEcAk[-atricrumpling’k’ukica2 *1k’uke1c *!k’uke, cakPARSE REC-F ALIGN-L RD/RTR RTR/LO AIIGN-R_________________________________* 1____________________ ___________________________________________________________________________________________4 4 411111111 p1111111111 1111 IplIpplIlIp IIIllII•,IlI I III* **k’ okick*1k’okica2V-atr*k’oke, cakk’oke1c * *!pusisipOsEsE[—atrishining’pusisel*puselsi *!pusel sd[-ar]-- posisdl*!*!*POSC1Sipose1sc1* *174configuration”. According to Archangeli and Pulleyblank (inpress), the gapped configuration has received virtually nomotivation, although it has frequently appeared in theautosegmental literature (cf. McCarthy 1984a, Cole 1987,Archangeli and Pulleyblank 1987, Vago 1988, etc.)Archangeli and Pulleyblank (in press), among others, claimsthat the gapped configuration is ill-formed since itviolates the locality (adjacency) condition, while the twinpeak’2 configuration is well—formed. The case of Korean,however, shows that the twin peak configuration is notsuitable in this granunar because this representation’3,having an additional [—ATR] feature, violates the highlyranked constraint REC-F.To summarize, the basic hypothesis here is that harmonyis interrupted by the effect of the grounded conditionRTR/LO. ALIGN-L ensures that the feature [-ATRJ aligns tothe left edge of the words. ALIGN—R causes this [-ATRJfeature to spread to the following vowels. Therefore, ALIGNL and ALIGN-R together derive the harmonizing effect(spreading). This harmony is blocked, however, by the RTR/LOcondition when the non-initial vowels are not low. As aresult, the non—initial moras of each word surface as [+ATR]([ij, [i]) in such cases. In other words, the feature [—ATR]aligns to the right vowel of words like sa2pk ‘crunching’‘2See Archangeli and Pulleyblank (in press:18)‘3For example, te2nkila-atr -atr75without incurring any violation. If [-ATRI is not aligned tothe right, then the form will violate ALIGN-R. Therefore,harmony proceeds to the right in normal cases. In cases withhigh unrounded vowels, however, harmony cannot proceed forit will incur an (additional) RTR/LO violation. For example,the form pa2nsil for the word ‘smiling’ will be the optimalone even though it violates ALIGN-R because it is moreimportant to obey the higher ranked constraint RTR/LO. Thefact that [ii and [i] are neutral only in non—initialsyllables is because ALIGN-L is ranked higher than RTR/LO.However, the high round vowels ([U], [ü]) are not neutral;they participate in harmony even though they incur an RTR/LOviolation because of the more highly ranked RD/RTRcondition. Hence, we can conclude that high vowels do notparticipate in harmony (RTR/LO) only in non-initialsyllables (ALIGN-L>> RTR/LO>> ALIGN-R), but they do so whenthey are rounded (RD/RTR>> RTR/LO).It is interesting to note that Optimality Theory, inaddition to explaining the harmony effect through theinteraction of faithfulness and alignment constraints, givesa principled account of the neutrality of high unroundedvowels in terms of grounded conditions.76Words without lexical [-ATR] Specifications; Dark Sound SymbolicWordsDark sound symbolic words do not have lexical [-ATR)specifications; and as the input of each word lacks thisfeature, REC—F becomes relevant for this class.In the examples shown in (63), REC—F eliminates allpotential candidates that involve a [—ATRI specification.Such a feature is not present in the input forms; any output(63)KoreankElc’Ok‘tall’PARSEki1c’ukREC-F AT, T (Thi- Lkilc’okI-atrkc1lc’uk\-atrke1lc’okV-atr*!pnsi1pAnsEl‘smiling’pnse1-atrpa2nsil-atrpa2nse1l-atr*!77faithful to the input will be ranked above a form thatviolates REC—F. These cases are trivial since all the non—optimal forms violate the highest ranked REC—F (PARSE isirrelevant as there is no [—ATR] feature to be parsed in theinput).The vowel harmony in sound symbolic words is describedby tongue root harmony in this study. Light sound symbolicwords are characterized by the addition of a [-ATRJ morphemeand the harmony effect is derived by the interaction ofPARSE, ALIGN, and grounded conditions. The dark soundsymbolic words do not have this [-ATR] morpheme underlyingly,and, therefore, any forms that have retracted vowels areruled out by the constraint REC-F. As a consequence, allvowels surface as advanced in dark sound symbolic words.3.1.2. A—initial SuffixesIn the initial syllable position of A-initial suffixes,[o] alternates with [a] as demonstrated in (64)(64)a.po-a ‘Look!’ka-a>ka ‘Go!’ko-a ‘beautiful’malk-a ‘clean’b.cip-o ‘Pick (it) up’78pe- ‘Cut (it)cü- ‘Hold (it)ki- ‘Mark (it)mu1- ‘Bite!’s->s ‘Stop!’me-a ‘Tie (it)k’O- ‘Tempt (him)[a] appears after an adjective or a verb stem whose lastvowel is either [a] or [0] as in (64)a. [] appears afterall other vowels as with the examples in (64)b. When thesuffix is followed by other A—initial suffixes, vowelharmony is not observed in these subsequent suffixes(i . e . mak-as’-s’- , *makas’as’a ‘ (He) had blocked (it) ‘) (Kim—Renaud 1986) . The domain of harmony, therefore, isrestricted to the area between the stem final vowel of verbor adjective and the initial vowel of A-initial suffixes.The vowel harmony in A—initial suffixes is related tothe harmony in sound symbolic words in that both of themobserve ATR harmony. Unlike sound symbolic words, however,the vowel harmony in A-initial suffixes is conditioned bythe feature of the stem vowel. If the stem vowel of theadjective or the verb is [+BK], harmony may proceed (i.e.p0-a ‘Look!’, malk-a ‘clean’). However, if the triggeringvowel is [-BK], harmony is blocked (i.e. me-a ‘Tie(it), k’ö-79‘Tempt (him) )14• To account for the differences, we need toinvoke the additional constraint ALIGN—RP which isformulated as follows:(65) .ALIGN-ct-RJ: The right edge defined by feature a.aligns with the right edge of Domain D, when the stemvowel is [+BKJ(Domain D = area between the stem final vowel ofverb or adjective and the initial vowel ofA—initial suffixes)The above condition is restricted to the case in whichthe feature aligns to the suffixal vowels. Therefore, byranking ALIGN-RI3 higher than ALIGN-R, we can get the correctresult for the harmony in both sound symbolic words and A-initial suffixes. That is, ALIGN-RP is irrelevant for thesound symbolic word case since the application is restrictedto the suffix vowels, and is thus vacuously satisfied.ALIGN-R does not have a crucial effect on A-initial suffixcase because ALIGN-RI3 is ranked higher. The followingtableau is submitted for the explanation of PDI< A-initialsuffix vowel harmony.‘4The harmony in sound symbolic words proceeds regardless ofthe feature of the stem vowel (i.e.kelcok ‘tall’)80In the above tableau, only the relevant constraints forthe analysis are highlighted although all the constraintsare present in the grammar. That is, the grounded conditionsRD/RTR and RTR/LO are not included because they do not haveany crucial effect on the analysis. Of the four candidates81for the word ‘Look’, the RD/RTR condition is violated in thefirst two forms. But these forms are already ruled out bythe higher constraints PARSE and ALIGN-L, respectively. TheRTR/LO condition is violated in the third and fourthcandidates. The fatal violation, however, is that of theALIGN-RI3 constraint which rules out the third form. In thesecond example, ‘Tie it’, the RD/RTR condition is irrelevantsince there are no round vowels and the RTR/LO condition issatisfied by all the candidates. Hence, the effects of theconditions RD/RTR and RTR/LO are not crucial in this case.The ranking which includes the entire set of constraints forPDK is: P.4RSE>> REC-F>> ALIGN-L>> RD/RTR>> RTR/LO>>ALIGN-Pfl>> ALIGN-R.ALIGN-RI3 is slightly different from ALIGN-R in that theapplication of this constraint is restricted to cases wherea stem vowel is [-‘-BK]. To illustrate the effect of ALIGN—R,let us compare po- and mc-, the third candidates of eachexample. po- violates ALIGN—RI3 since the stem vowel is[+bk]. Therefore, the optimal candidate for the word ‘Look!’is p0-a2. me-a is the optimal form of ‘Tie it’ for the stemvowel is [—BK], therefore, it does not violate ALIGN—RI3 eventhough it violates ALIGN-R. me-a2 is ruled out by ALIGN-RI3because the feature [-ATR] aligns rightward from a [-BK]vowel.Stems without lexical [—ATR] specification take thesuffix [] as shown in the tableau (67)82(67)Korean PARSE REC-F ALIGN-L ALIGN-R IGN-Rmu1-‘Bite!’mul-a2 *!/-atrmol-o\-atrmol-aV2-atrThis case, too, is as simple as that of dark soundsymbolic words, since the candidates with retracted vowelsviolate the highly ranked constraint REC-F as well asothers.The harmony in A-initial suffixes differs from theharmony in sound symbolic words in that it is conditioned bythe feature of the stem vowel, and shall be explained byinvoking an additional constraint ALIGN-RP. Here, neutralityis not observed since we are only looking at low back vowels([a], {a]) in non—initial syllables. Nevertheless, theharmony in both sound symbolic words and A-initial suffixesis explained by the same ranking in PDI< grammar.833.2. Later Middle KoreanBefore we begin, let us review the vowel system of LaterMiddle Korean. LMK has an underlying seven-vowel system;this was shown in (38) and is here repeated as (68)(68) i i u0a AThe feature specifications of these vowels are asfollows:(69) LMK vowel feature specificationi I A U 0 aatr (—) (—) (—)md + +bk + + + + + +low + +These vowels are divided into two sets as we havediscussed in the beginning of section 3.1.: the ‘light’(yang) vowels [o,A,a] and the ‘dark’ (yin) vowels [u,i,o] . Thistraditional yin/yang division is based on the harmonysystem. Within a word, vowels must be all dark or lightexcept that the vowel [i] may cooccur with either dark or84light vowels. LMK vowel harmony is said to have beenobserved morpheme internally as well as in the combinationof a stem with suffixes (Park,1990). Unlike PDK, LMK vowelharmony involves almost every word. Some examples are givenin (70)(70)a.stem internallyDark Lightkulum ‘cloud’ namo ‘tree’pnky ‘lightning’ sarm ‘man’simil ‘twenty’ kaci ‘branch’miciky ‘rainbow’ kamakoy ‘crow’nilkup ‘seven’ kiliima ‘saddle’kiypyol ‘news’mili ‘in advance’tili ‘field’ tozak ‘thief’tuk ‘very’ maissAm ‘saying’chzom ‘beginning’ api ‘father’mi ‘mother’nmu ‘too much’b.between a stem and a suffixDark Lightpl- ‘to eam-INF’ sal-a ‘to live-INF’pul-o ‘to blow-INF’ tol-a ‘to tum-INF’85ml- ‘far’ mi’l-a ‘ignorantly’n-nin ‘you-TOP’ na-nAn ‘I-TOP’nun-in ‘eye-TOP’ son-An ‘hand-TOP’ki-lil ‘he-ACC’ halmi-hi ‘grandmother-ACC’k-lil ‘vehicle-ACCkypy-lil ‘news-ACC’ tozAk-Al ‘thief-ACC’sli-lil ‘frost-ACC’ api-lil ‘father-ACC’kulum-il ‘cloud-ACC’ mzAm-Al ‘mind-ACC’mm-il ‘people-ACC’ li-Ll ‘village-ACC’cip-iy ‘house-GEN’ kot-Ay ‘flower-GEN’mk-um ‘eating’ cap-om ‘holding’tuyh-y ‘behind-LOC’ pAhm-ay ‘wind-LOC’kulum-oy ‘cloud-LOC’ namo-ay ‘tree-LOC’kith-oy ‘street-LOC’ kaci-ay ‘branch-LOC’tuy-y-nin ‘behind-LOC-TOP’ alp-ay-nAn ‘front-LOC-TOP’tl-um-il ‘dcduct-Nom-Acc’ al-om-Al ‘know-Nom-Acc’(S.O.Lee 1984, S.I-I.Park 1990, Song 1990)In the harmony system of LMK, [U] alternates with [o],[i] alternates with [A], and [o] alternates with [a]. As wecan see from the above data, [ii alone is a neutral vowel inLMK, whereas both [ij and [ii are neutral in PDK.In previous studies of LMK (Ledyard 1966; B.G.Lee1985’s; Park 1982), the harmonizing feature of vowel harmony‘5Although Lee (1985) advocates a horizontal harmonyanalysis for Korean vowel harmony, he also suggests adifferent type of harmonic distinction, [ATRI, based on his86has largely been assumed to be the tongue body feature[back] or [low]. None of these analyses, however, correctlydistinguishes between the two harmony groups according tothe feature specifications in (69) . The analysis that adoptsthe feature [back] as the harmonizing feature (Ledyard 1966,etc.) takes [a,A,o] as [+back] and [u,i,,i] as [—back]. On theother hand, the analysis in which the harmonizing feature is[low] groups [a,o,A] as H-low] while all others are regardedas [-low]. These hypotheses, however, are suspicious becauseit is generally accepted that the features of the LMK vowelsystem are similar to those of the PDK vowel system’6 (Huh1985, Hwang 1986, K.M.Lee 1987) . Therefore, the harmonizingfeature of LMK vowel harmony can be specified as neither[back] nor [low]. In this study, as in PDK, we will take[ATRI as the harmonizing feature in accordance with theargument presented in Song (1990:62);“The contrast between [ii vs. [A], [j vs. [a], and [U]vs. [o] can be described by referring to tongue rootmovement from an articulatory point of view: as thetongue root is moved forward, the tongue body iscompressed and therefore raised, conversely, as thetongue root is retracted, the tongue body is pulleddown and therefore lowered (Hall & Hall,1980:207).interpretation of Hunmin-chongum-haerye (Explanation of theKorean Script of fifteenth century).‘6Compare (43) with (69)87Therefore, we can assume that [i,A,uJ are {+ATRJ and[a,o,Aj are [—ATRI .“Hence, vowel harmony in LMK ,as in PDK, can becharacterized by the feature [ATRI. The LMK vowels aredivided into two sets by this feature:(71) i i U dark=[+ATRJ__r o light=[—ATR]a A3.2.1. Optimality AccountThis section will give the optiniality account for LaterMiddle Korean. We will see the changes in the grammarthrough LMK to PDK and discuss how CT can explain thisvariation. The hypothesis is that the differences in thegrammar of each stage (i.e.LMK and PDK) are the result ofchanges in the ranking of the universal constraint set.Let us consider some of the crucial data from (70) whichare repeated here as (72)(72) Dark Lightnmu ‘too much’ namo ‘tree’mi ‘mother’ kaci ‘branch’mm-il ‘people-ACC’ li-hi ‘village-ACC’mi- ‘far’ mAi-a ‘ignorantly’88LNK vowel harmony involves nearly the entire vocabularywhile P0K vowel harmony is found only in sound symbolicwords and A—initial suffixes. The other difference is foundin regard to the neutral vowels: P0K has two neutral vowels,[i] and [i], in non-initial syllables; whereas LMK has onlyone neutral vowel, [i], regardless of position. For example,the vowel [ii is neutral even in syllable initial positionsas the word li-1AI shows. Also, we can see that the vowel [ijis not neutral when we compare suffixes like -in VS. —An‘TOP’ from the examples in (70)b. These differences areexplained by the ranking of constraints shown in tableau(73). Again, we are assuming that the morphemes which havelight vowels have lexical [-ATRI specification while thosewith dark vowels do not.89(73) Words with lexical [—ATRI Specification*!*!PARSE REC-F BK/RTR RTR/LQ ALIGN-L ALIGN-R***Lz* *1 I I 11111111!!**1*1In PDK, the constraint ALIGN-L is ranked higher than thegrounded conditions since the high unrounded vowels [i,i] areneutral only in non-initial syllables. However, the vowel[iJ in LMK is neutral regardless of its position in thesyllable. Therefore, both of the alignment conditions arenmu-atrnmoI-atrnamu\-atrnamoV-atr‘tree’koci *!-atrkociI-atrkaci\-atrbranch’kaciV-atr90ranked lower than the grounded conditions. In fact, thealignment constraints do not play a crucial role in LMK. TheRD/RTR condition of PDK is replaced and therefore pushedmuch lower in the hierarchy by the BK/RTR17 condition inLMK. In LNK, there is no need to group round vowels togetherbecause all round vowels are back. We must, however,distinguish round vowels in PDK since front rounded vowels([ö,ü]) do exist.Let us consider the word for ‘tree’. The first candidateviolates PARSE, the highest ranked constraint. The secondand third forms are ruled out by BK/RTR because the [+BKJvowels [] and [U] are [+ATR]. They also violate lowerranked constraints such as RTR/LO and ALIGN. The form namois optimal even with the violation of RTR/LO ([-ATR] isparsed to nonlow vowel [o] .) because it beats all othercandidates. However, the violation of this constraint(RTR/LO) is fatal when we have the high front vowel [ii. Thefirst and second forms for ‘branch’ are ruled out by PARSEand BK/RTR, respectively, in the same manner as in ‘tree’.The fourth candidate kaci is ruled out by RTR/LO because theoptimal form kaci passes this constraint and violates onlyALIGN-R which is ranked lower in the hierarchy. That is,when there is the vowel [i], it is better not to align[-ATR] to this vowel because it will incur an RTR/LO17BK/RTR Condition: If [+BK] then [-ATR], not [+ATR]91violation. [—ATR], however, can be parsed to vowels like [ijand [U] due to the BK/RTR condition.The following tableau shows that it is more important tosatisfy the BK/RTR condition when there is competitionbetween BK/RTR and RTR/LO conditions.(74)92The form mAl-a is optimal even with an RTR/LO violationbecause the form without this violation mu-a violates BK/RTRwhich is ranked higher. The forms that have the vowel [iiviolate BK/RTR. The BK/RTR condition prevents [i] as well asall other advanced back vowels ({u,]) from surfacing in sucha case. Hence, we may conclude that the fact that [ii is notneutral in LMK is due to the relatively high ranking of theBK/RTR condition in this grammar.Morphemes with dark vowels can be explained easily as inthe dark sound symbolic words case in PDK. The forms inwhich [-ATRJ is parsed to a vowel will violate REC—F. Theoptimal form, therefore, is the one without retracted vowelsas the following tableau shows.93(75) Words without lexical [—ATR] specificationnamo *1/-atrnamu *1\-atrnamo *7V-atrmin-i1‘people-ACC’mm-Al *7/-atrmm-il *7\-atrmm-Al *7V-atrIn this section, the diachronic changes in the vowelharmony phenomena in LMK and PDK have been explained by thedifferences in the two grammars’ ranking systems within theOptimality Theory framework. The diachronic changes areillustrated in the following scales.namu‘too much’PARSE REc-F BK/RTR RTR/LO ALIGN-L ALIGN—R**4*1111111 IlIllIllIllIll II I IllIllIllIlIlIll II 111111111111 IIIN**N‘liii IlIllIllIl 111111111111 II IIIIIIIIIIIIIIIIIII I 1111 II ‘1111*NIII 111111111 I 1111111 11111111 1 1111111 I I 1111 I 1111111 1111* * *-94(76) Diachronic ChangeLMK: PARSE>>REC-F>>BK/RTR>>RTR/LO>>ALIGN-L>>ALIGN-RPDK: PARSE>>REC-F>>ALIGN-L>>RD/RTR>>RTR/LO>>ALIGN-RWe claimed in the beginning of this chapter thatcertain constraints are still crucial after the rerankingwhile certain constraints cease to play a crucial role aftermoving down in the hierarchy. As we can see from the abovescales, the ALIGN-L constraint was moved up to a positionhigher than the relevant grounded conditions in the changefrom LMK to PDK, and both ALIGN—L and the groundedconditions were crucial in PDK after the reranking from LMK.However, the BK/RTR condition, which was moved down in PDK,ceased to play any crucial role in the grammar after thechange in ranking.95CHAPTER 4CONCLUSIONThis study presents the hypothesis that the vowelharmony patterns of Korean can be described as ATR harmonyand that a better grasp of this harmony system can beachieved with the application of Optimality Theory. Theconcept of constraint interaction allows the characteristicsof neutral vowels in the harmony process to be accounted forby the interaction of grounded conditions with constraintson alignment and the appropriate ranking of theseconstraints, namely RTR/LO>> ALIGN-R.The standard OT claim that an individual grammar is theresult of variation in the ranking of the universalconstraint set is further developed here to conclude thatdifferent historical stages in a single language can also betreated as the result of changes in the ranking. We haveshown how LNK grammar develops into PDK grammar by rerankingsome of the constraints: while the grounded conditions areranked higher than ALIGN-L in LMK, the ALIGN-L constraint ismoved up in PDK because it is important to align the feature[—ATR] to the left edge of a word. The fact that neutralvowels are found only in non-initial syllables in PDKwhereas they are found in any position in LMK can be96accounted for by the changes in ranking from LMI< to PDK. TheBK/RTR condition is moved down since this constraint cannotgroup PDK round vowels including front round vowels. Thedifferences in the inventories of LMK and PDK areresponsible for the different role of the BK/RTR conditionin each stage.Further work, however, is needed on other languages, inorder to substantiate the claim that diachronic change ischaracterized by variation in constraint rankings.97BIBLIOGRAPHYAhn, S.C. (1985) The Interplay of Phonology and Morphologyin Korean, Ph.D. dissertation, University of Illinois.Archangeli, D. & D. Pulleyblank (1987) ‘Maximal andMinimal Rules: Effects of Tier Scansion’,Proceedings of NELS 17, 16-35.Archangeli, D. & D. Pulleyblank (1989) ‘Yoruba VowelHarmony’, Linguistic Inquiry 20, 173—217.Archangeli, D. & D. Pulleyblank (in Press) GroundedPhonology, MIT Press.Archangeli, D. & D. Pulleyblank (1993) ‘Two rules orone. • .or none? [ATRI in Yoruba’, Proceedings of BLS19, 13—26.Ard, J. (1984) ‘Vowel Harmony in Manchu: a criticaloverview’, Journal of Linguistics 20, 57—80.Bickmore, Lee (1993) ‘Tone Distribution in OptimalityTheory’, ms. University at Albany.Binnick, R. (1980) ‘The underlying representation ofharmonizing vowels: Evidence from Modern Mongolian’in R. Vago (ed.), Issues in Vowel Harmony, 113-134.Bird, Steven (1990) Constraint-based Phonology, Ph.Ddissertation, University of Edinburgh.Chang, N.G. (1982) ‘Phonological Variations in 15thCentury Korean’, Journal of Chinese LinguisticsMonograph Series No.1.Chomsky, N. & M. Halle (1968) The Sound Pattern ofEnglish. New York: Harper and Row.Cole, J. (1987) Planar Phonology and Morphology, Ph.D.dissertation, MIT.Coleman, John (1991) Phonological Representation - theirNames, Forms, and Powers. Ph.D. dissertation.University of York.98Goldsmith, John (1976) Autosegmental Phonology, IndianaUniversity Linguistics Club. [Published by GarlandPress, 1979.]Goldsmith, John (1990) Autosegmental and MetricalPhonology. Oxford: Basil Blackwell.Goldsmith, John (1993) ‘Phonology as an IntelligentSystem’, Bridges between Psychology and Linguistics:A Swarthmore Festschrift for Lila Gleitman, ed.DonnaJo Napoli and Judy Kegl, 247-267., Hillsdale, NY:Lawrence Eribaum Associates.Hall, B.L. and R.M.R. Hall (1980) ‘Nez Perce VowelHarmony: An Africanist explanation and sometheoretical questions’, in R.Vago (ed), Issues inVowel Harmony, John Benjamins, Jinsterdam, 201-236.Hamp, P. (1958) ‘Vowel Harmony in Classical Mongolian’,Word 14, 291—294.Hamp, P. (1980) ‘Mongolian vocalic features and theproblems of harmony’, in R.Vago (ed), Issues in Vowelharmony, 101-111., John Benjamins, .1msterdam.Hwang, ELY. (1986) Introduction to Korean Phonology,Leewoo Express: Seoul (in Korean).Huh, U. (1984) Korean Phonology, Jung-Eum sa: Seoul (inKorean).Ito, Junko, Armin Mester & Jaye Padgett (1993) ‘Licensingand Redundancy: Underspecification in OptimalityTheory’, LRC—93-07, University of California, SantaCruz.Kim. C.W. (1968) ‘The Vowel System of Korean’,Language 44, 516—527Kim, C.W. (1978) ‘“Diagonal” Vowel Harmony?: SomeImplications for Historical Phonology’, RecentDevelopments in Historical Phonology, 221-236Mouton, The Hague.Kim, J.M. (1986) Phonology and Syntax of KoreanMorphology, Ph.D dissertation, University ofSouthern California: Los Angeles.99Kim, K.O. (1977) ‘Vowel System of Korean Revisited’,Papers in Korean Linguistics.Kim-Renaud, Y.K. (1986) ‘Semantic Features in Phonology:Evidence from Vowel Harmony in Korean’, Studies inKorean Linguistics, 63—76.Kim, Y.S. (1984) ‘Ideophones in Korean’, paper presentedat the meeting of Linguistic Society of Korea.Kim, Y.S. (1985) Aspects of Korean Morphology. Ph.D.dissertation, University of Texas, [Reproduced byPan Korean Book Corporation, Seoul).Kiparsky, P. (1973) ‘How abstract is phonology?’, in 0.Fujimura (ed), Three Dimentions of LinguisticTheory, TEC Co.: Tokyo, 57-86.Kirchner, Robert (1993a) ‘Turkish Vowel Harmony andDisharmony in Optimality Theory’, ms. UCLA.Kirchner, Robert (1993b) ‘Optimizing YidinY Phonology’,ms. UCLA.Ledyard, G.K. (1966) The Korean Language Reform of 1446:The Origin, Background and Early History of theKorean Alphabet, Ph.D dissertation, University ofCalifornia: Berkeley.Lee, B.G. (1985) The Generative Phonology of ModernKorean, Ilgisa: Seoul (in Korean)Lee, K.M. (1985) Studies in the Phonological History ofKorean, Korea Research Foundation: Seoul (inKorean).Lee, S.O. (1984) ‘An overview of issues in the VowelSystem and Vowel Harmony in Korean’, LanguageResearch, 417-451.Lighter, T.M. (1965) ‘On the description of vowel andconsonant harmony’, Word 21, 244-250.McCarthy, J.J. (1983) ‘Phonological Features andMorphological Structure’, in CLS (ed.), The Interplayof Phonology, Morphology, and Syntax, 135-161.100McCarthy, J.J. (1984) ‘Theoretical Consequences ofMontañes Vowel Harmony’, Linguistic Inquiry 15, 291-318.McCarthy, John & Alan Prince (1993a) Prosodic MorphologyI: Constraint Interaction and Satisfaction, ms.,University of Massachusetts, mherst, and RutgersUniversity.McCarthy, John & Alan Prince (1993b) ‘GeneralizedAlignment’, ms. University of Massachusetts, 1mherst& Rutgers University.Mutaka, N. & L. Hyman (1990) ‘Syllable and MorphemeIntegrity in Kinande Reduplication’, Phonology 7,73—120.Myers, Scott (1991) ‘Persistent Rules’, LinguisticInquiry 22, 315—344.Myers, Scott (1993) ‘OCP Effects in Optimality Theory’,ms. University of Texas, Austin.Paradis, Carole (1988) ‘on constraints and RepairStrategies’, The Linguistic Review 6, 71—97.Park, B.C. (1982) The Study of Old Korean, Korea UniversityPress: Seoul (in Korean)Park, C.H. (1983) A Study on Korean Phonology, WonkwangUniversity Press, In (in Korean)Park, S.H. (1990) ‘Vowel Harmony in Korean’, LanguageResearch, 469—499.Poppe, N. (1965) Introduction to Altaic Linguistics,Wiesbaden: Harrassowitz.Prince, Alan and Paul Smolensky (1993) Optimality Theory:Constraint Interaction in Generative Grammar, ms.,Rutgers University, New Brunswick, and University ofColorado, Boulder.Pulleyblank, D. (1993) ‘Vowel Harmony and OptimalityTheory’, to appear, Proceedings of the Workshop onPhonology, University of Coimbra, Portugal.101Pulleyblank, E. (in press) ‘Phonetics, East Asian:History of’, to appear in Encyclopedia of Languageand Linguistics, Edinburgh: Pergamon Press.Ramstedt, G.J. (1968) A Korean Grammar, AnthropologicalPublications.Scobbie, James (1991) Attribute Value Phonology. Ph.D.dissertation, University of Edinburgh.Scobbie, James (1992) ‘Towards Declarative Phonology’,Edinburgh Working Papers in Cognitive Science 7,1-26.Singh, Raj (1987) ‘Well—formedness Conditions andPhonological Theory’, in W.Dressler et al., ed.,Phonologica 1984, Cambridge University Press,Cambridge.Smolensky, P. (1988) ‘On the proper treatment ofconnectionism’, The Behavioral and Brain Sciences11, 1—74.Sohn, H.S. (1986) Toward an Integrated Theory ofMorphology: Vowel Harmony in Korean, Studies in theLinguistic Sciences 16, No.2, 157-184.Song, J.S. (1990) Vowel Harmony in Nez Perce and Korean,M.A.thesis, University of Ottawa.Steriade, Donca (1988) ‘Reduplication and SyllableTransfer in Sanskrit and Elsewhere’, Phonology 5,73—155.Stewart, J.M. (1967) ‘Tongue Root Position in Akan VowelHarmony’, Phonetica 16, 111-139.Svantesson, J.0. (1985) ‘Vowel Harmony Shift inMongolian’, Lingua 67, 283—327.Tohsaku,Y.H. (1983) A Formal Theory of Vowel Harmony, Ph.DDissertation, University of California, San Diego.Vago, R.M. (1973) ‘Abstract Vowel Harmony in Uralic andAltaic Languages’, Language 44, 579—605.Vago, R.M. (1988) ‘Underspecification in the HeightHarmony System of Pasiego’, Phonology 5, 343-362.102APPENDIX : List of Constraints1.ALIGN-a-L/R: The left/right edge defined by feature xaligns with the left/right edge of Domain D.2.ALIGN-x-RP: The right edge defined by feature a. alignswith the right edge of Domain D, when the stem vowel is[+BK].3.ATR/HI: If [+ATR] then [+high], not [-high].4.ATR/LO: If [+ATR] then [-low], not [+10w].5.—COD: Syllables must not have a coda.6.DISYLL: The Reduplicant is minimally disyllabic.7.FILL: Syllable positions must be filled with underlyingsegments.8.HI/ATR: If [+high] then [+ATR], not [—ATR].9.LO/ATR: If [+10w] then [-ATR], not [+ATR].10.MAX: The reduplication is phonologically identical to thebase.11.ONS: Syllables must have an onset.12.P.ARSE: An F-element (feature or node) a. must be dominatedby an appropriate node in the prosodic tree.13.REC—F (Recoverability of F-element) : An F—element a. thatis present in an output form is also present in theinput.14.REC-P (Recoverability of Path): For any path between anF-element a. and some anchor f3, if a. is associated to13 in the output then a. is associated to 13 in theinput.15.RD/RTR: If [+RD] then [-ATR], not [+ATRJ.16.R<ROOT: The Reduplicant (R) contains only the root.17.RTR/HI: If [-ATR] then [-high], not [+high].18.RTR/LO: If [—ATR] then [+low], not [—low].103


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items