University of Southern California Dissertations and Theses

Correspondence and faithfulness constraints in optimality theory: A study of Korean consonantal phonology

(USC Thesis Other)

Correspondence and faithfulness constraints in optimality theory: A study of Korean consonantal phonology

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content INFORMATION TO USERS This manuscript has been reproduced from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter free, while others may be from any type o f computer printer. The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book. Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6” x 9” black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order. UMI A Bell & Howell Information Company 300 North Zeeb Road, Ann Arbor MI 48106-1346 USA 313/761-4700 800/521-0600 Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CORRESPONDENCE AND FAITHFULNESS CONSTRAINTS IN OPTIMALITY THEORY: A STUDY OF KOREAN CONSONANTAL PHONOLOGY by Sechang Lee A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Linguistics) May 1997 Copyright 1997 Sechang Lee Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 9733085 UMI Microform 9733085 Copyright 1997, by UMI Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. UMI 300 North Zeeb Road Ann Arbor, MI 48103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES. CALIFORNIA 90007 This dissertation, written by Sechang Lee under the direction of k .f.® Dissertation Committee, and approved by all its members, has been presented to and accepted by The Graduate School, in partial fulfillment of re quirements for the degree of DOCTOR OF PHILOSOPHY Dean o f Graduate Studies February 12, 1997 Date DISSERTATION COMMITTEE ' J b o & X x x ' b .... '7 Chairperson ..... v I d # , Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Acknowledgem ents I thank my lucky stars that I met Alicja Gorecka at USC. She has been my academic advisor for the past five years, and led me to the world of phonology with special care. With unfailing patience, she has constantly pointed out the problems with my primitive proposals and suggested right directions for revision and further research. Each time I saw her at discussion, I realized that I knew too little and so must work all the more. It is virtually impossible to imagine what my dissertation would have looked like without her consistent guidance and support. She was also the best teacher I have ever met. It was my great pleasure to have worked for her as a TA for two semesters in the introductory course of linguistics. Working with her was really the invaluable part of my life. Also, I would like to express my appreciation to the committee members, Barry Schein, Bernard Comrie, and Roger Woodard. Their critical comments and suggestions have made this a better dissertation than it would otherwise have been. I have received financial support all during the education at USC. I gratefully acknowledge of our Linguistics Department of USC. During the course of my graduate studies in Korea University, I was lucky enough to have truely exceptional teachers. Very special thanks go to my former advisor, Prof. Yong-Jae Lee to whom I owe an encounter with phonology for the first time in my life. I shall never forget the warmth and affection he has given me ever since I became his student. Thanks to him, I learned something to live for. I am grateful to Prof. Kyung-Ja Park for introducing me to applied linguistics. Prof. Kee-Ho Kim played a central role in my early training as a phonologist. He taught me Underspecification and Feature Theory whose impact on this dissertation cannot be too much emphasized. He gave me tons of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. help in the form of advice, suggestions, and mental support. I feel sincere gratitude for what he has done to me these past several years. Prof. Dong-In Cho has been taking special care of me just as my big-brother ever since I knew him. He babysat me when I first came over to Los Angeles, and showed me how to survive in the competitive society of linguistics. His unusual concern and support for me will always have a special place in my heart. I am indebted to Bruce Hayes (UCLA) for his insightful comments on Korean tensification at the panel discussion of the 4th Japanese/Korean Linguistics Conference. Donca Steriade (UCLA) and Moira Yip (UCI) gave me suggestions and references for place assimilation phenomenon. I am grateful to them, too. There are other people who helped me directly or indirectly. I have greatly benefited from the discussion with Mr. Jongho Jun. I owe him some crucial ideas in my dissertation. I thank Hyeon-Kwan Cho, Young-Kook Kwon, Hye-Young Um, and Hikyoung Lee for sending me some useful papers. I also want to thank my classmate, Hong-Keun Park who was my best friend on every occasion. I could not be more grateful to my parents who made all these possible. I know that they are the very people who would feel happiest with the success of their favorite son. And I especially thank my wife, Lanju Kim, for her assistance and encouragements during the long gestation of this dissertation. Last of all, I thank my three-year-old daughter, Jung-Eun, for growing up without any trouble and making my wife and me always happy. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. iv Table of Contents Chapter 1: Preliminaries §1.1 Introduction 1 § 1.2 Background in Korean phonology 4 § 1.2.1 Underlying inventory of Korean § 1.2.1.1 Consonant inventory §1.2.1.2 Vowel inventory 5 § 1.2.2 Syllable structure of Korean 6 § 1.2.3 Moraic phonology 9 Chapter 2: Irregular Verbs §2.0 Introduction 11 §2.1/p/-irregular verbs 12 §2.2/h/-irregular verbs 24 §2.2.1 Linear approach 26 §2.2.2 Non-linear approach 27 §2.2.3 OT approach 29 §2.3 /t/-irregular verbs 34 §2.3.1 The status of ‘r’ 36 §2.3.2 The analysis 38 §2.4 /ui/-irregular verbs 41 §2.4.1 Compensatory lengthening 42 §2.4.2 The analysis 44 §2.5 /l/-irregular verbs 47 §2.6 /r/-irregular verbs 52 §2.7 /s/-irregular verbs 55 §2.8 Conclusion 58 Chapter 3: Neutralization §3.0 Introduction 60 §3.1 The paradigm §3.2 Earlier treatments 62 §3.2.1 Iverson & Kim (1987) §3.2.2 Clements & Hume (1995) 64 §3.3 Glottalization of unreleased consonants 66 §3.4 Closure and release phase of a stop 67 §3.5 The analysis 69 §3.6 Conclusion 75 Chapter 4: Tensification §4.0 Introduction 76 §4.1 The paradigm §4.2 Earlier treatments 77 §4.3 Tense consonants as geminates 80 §4.4 Onomatopoetic and mimetic words 82 §4.5 Initial geminates in Mokilese 85 §4.6 Tensification in OT 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V §4.6.1 Tensification after stops 88 §4.6.2 Tensification after fricative 90 §4.6.3 No tensification after nasals 92 §4.6.4 Tensification in compounds 94 §4.6.4.1 Sub-compounds 95 §4.6.4.2 Co-compounds 102 §4.8 Conclusion 104 Chapter 5: Place Assimilation §5.0 Introduction 105 §5.1 The paradigm 106 §5.2 The generalizations 110 §5.2.1 Target place 111 §5.2.2 Trigger place 113 §5.3 Theoretical problems in place assimilation 114 §5.3.1 Formal simplicity and naturalness 115 §5.3.2 Cross-linguistic variation 118 §5.3.3 Direction of place assimilation 121 §5.4 The proposal 122 §5.4.1 Licensing place features 123 §5.4.2 Faithfulness constraints and LlCENSE(consonantal place) 129 §5.4.3 Alignment constraints 132 §5.5 Case studies 137 §5.5.1 Korean 138 §5.5.1.1 Heterorganic clusters §5.5.1.2 Homorganic clusters 149 §5.5.1.2.1 Obligatory Contour Principle 151 §5.5.1.2.2 The analysis 154 §5.5.2 English 161 §5.5.3 Other languages 167 §5.5.4 An unsolved problem 168 §5.6 Conclusion 170 Chapter 6: Vowel Change and Umlaut §6.0 Introduction 172 §6.1 Umlaut in Kyungsang dialect §6.2 Earlier treatments 177 §6.3 The proposal 179 §6.4 Assimilation in OT 191 §6.4.1 Markedness in SPE §6.4.2 Palatalization 192 §6.4.3 Umlaut 195 §6.5 Historical vowel change 196 §6.6 Umlaut in Choong-nam dialect 200 §6.7 Conclusion 205 Appendix A: Site-Articulator model 207 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vi Appendix B: Optimality Theory 214 Part 1: Prince & Smolensky (1993) Part 2: McCarthy & Prince (1995) 217 References 221 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I Chpater 1: Preliminaries §1.1 Introduction The aim of this thesis is to show how an optimality-theoretic conception of phonology (McCarthy & Prince, 1993, 1995; Prince & Smolensky, 1993) overcomes some of the limitations of the traditional ways of treating various phonological phenomena in Korean. There can be little doubt that the Optimality Theory (OT, henceforth) has brought a new depth of explanation to a number of areas of phonological analysis (for instance, triggering & blocking effects in phonology, syllable theory and prosodic morphology, among others). Building on the notion of OT, I attempt to shed new light on the properties of some general phonological phenomena of Korean. The facts to be discussed throughout this thesis are primarily from Korean. My approach to them is based on the Site-Articulator model (Gorecka, 1989) and the interaction of universal constraints in terms of OT. In the framework of traditional standard generative phonology (Chomsky & Halle, 1968, henceforth SPE), phonological rules characterize how underlying and surface representations are related; a Structural Description delimits a range of inputs and a Structural Change specifies the operations that altered the input. Since SPE, the central issue of phonological investigation has been to explicate the rule system by defining the operations available for transforming inputs. In the development of the phonological theory, however, the importance of the rule component has been reduced by the discovery that the significant regularities across languages were to be found not in input configurations, but rather in the character of output structures. A significant body of research since the mid 1970’s (Goldsmith, 1976; Clements, 1985; Sagey, 1986; Schein & Steriade, 1986) has focused on the characterization of a well- Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 2 formed representation, attempting to derive a variety of phonological phenomena (harmony, assimilation, dissimilation, etc.) directly from the properties of input representation. In expressing the input-output relations, the shared assumption has been that once the input representation is established, the appropriate output representation is determined automatically by a set of universal well-formed conditions. Therefore, in the unmarked case, the autosegments and the segments are not paired underlyingly but associated by universal well-formedness principles. What is problematic, however, is that the proposed ‘universal and automatic’ conventions have tended to vary from language to language and from case to case, as pointed out in Archangeli & Pulleyblank (1994:13). In other words, input-oriented conceptions of phonology are ill equipped to capture the fact that a number of phonological patterns in natural languages appear in many languages, but the patterns exhibit considerable crosslinguistic variation in the details of their manifestation (Mohanan, 1993:61). OT, in common with much recent work, shifts the explanatory burden from the theory of operations (or rules) to the theory of well-formedness. The central proposal of OT is that Universal Grammar provides a set of highly general well-formedness constraints and they are ranked in a hierarchy of relevance; an individual grammar consists of a ranking of these constraints, and it resolves any conflict in favor of the higher-ranked constraint. In the chapters to come, it will be argued that a variety of phonological effects, previously understood in terms of the triggering or blocking effects of rules by special conditions, will be seen to emerge from the interaction of universal constraints. This thesis is structured as follows. In Chatper 2, I argue that the so-called ‘irregular’ verbs in Korean which have been regarded as phonologically not predictable by many of the traditional grammarians of Korean (Kim-Renaud, 1974, 86; Lee, 1972; Martin, 1992) Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 3 are in fact not irregular (Sohn, 1987). On the contrary, I will show that the irregular verbs turn out to be phonologically predictable, using the language-specific ranking of universal constraints that is central to OT. Chapter 3 discusses the syllable-final neutralization. I will show that the triggering of neutralization can be expressed by ranking a constraint forbiding a release feature over a faithfulness constraint. In Chapter 4 ,1 argue that the syllable-initial tensification of obstruents is an automatic consequence of the immediately preceding syllable-final neutralization of obstruents. I aim to correlate the syllable-initial strengthening process to the syllable-final weakening process in terms of constraint interaction. OT will be able to provide an account for this correlation. Chapter 5 studies the place assimilation phenomenon with a primary focus in Korean. I propose that the licensing constraint of the consonantal place nodes trigger place assimilation. It will be shown that the variation of place assimilation pattern across languages is due to the re ranking of a single universal constraint in the same constraint hierarchy. Taken as a whole, the analyses throughout this thesis provide strong support for the basic tenet of OT, as developed in McCarthy & Prince (1993, 1995). Unfortunately, there are still examples that OT account will not cover. In Chapter 6 ,1 will discuss the difficulty of treating assimilation in OT. I argue that a historical vowel shift in 19C played an important role in Kyungsang dialect and is the source of the unpredictable pattern of umlaut in this dialect. It will be shown that the vowel change in 17C has affected the outlook of umlaut that took place around 19C. Finally, I outline the major points of the Site- Articulator model and the Optimality Theory in Appendix A and Appendix B, respectively. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 §1.2 Background in Korean phonology In this section, I intend to provide readers with background information on the Korean phonology, particularly the information that are relevant to the discussion in the following chapters. Two basic topics will be dealt with in this section: segment inventory (§1.2.1) and syllable structure (§1.2.2). §1.2.1 Underlying inventory of Korean I will briefly introduce the consonant and vowel system of Korean which I adopt throughout this thesis. §1.2.1.1 Consonant inventory The Korean segment system consists of 19 consonants and 10 vowels. The consonant inventory is shown in (1). There is a three way contrast between lenis (C), tense (C’) and aspirated (Cty obstruents except /s/ that has only two way of contrast with a tense /s’/: (1) 19 consonants Bilabial Alveolar Palatal Velar Stop P. P \ Ph t, t \ th k, k \ kh Affricate 9. 9'. < fh Fricative s, s’ h Nasal m n n Liquid 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 Underlyingly, all the consonants can occur syllable-initially as well as syllable-finally, except for IhJ and /r/ which are not allowed in syllabe coda and onset position, respectively. In syllable-final position, all stops and fricatives are neutralized to unreleased homorganic counterpart. The lateral f\J is pronounced as [r] intervocalically, and does not occur word-initially in (South) Korean.1 §1.2.1.2 Vowel inventory The vowel system consists of 10 simple vowels, 2 glides (j, w) and 12 diphthongs (je, jo, ju, je, ja, jo, wi, we, ws, we, wa, rcj) that are compounds of a simple vowel and a glide: (2) Simple vowels (10 vowels) Front Central Back High l y i (ui) u Higher-mid e, 0 3 (o) o Lower-mid e Low a The status of central vowels /7t/and /a/ is somewhat controversial: they are sometimes regarded as back vowels /ui/ and /o/, respectively.2 In Korean, vowel length is a historical residue and arguably distinctive (cf. Huh, 1965). Martin (1992:32) reports that many Korean speakers do not use long vowels in all 1 This will be discussed in Chapter 2 in detail. 1 cf. Maddieson (1984) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 the words for which some speakers retain them, so that most words with a long vowel within a morpheme have short variants.3 Compare the following pronunciations: (3) Vowel length variants (expanded on Martin 1992:35) [mal] ‘horse’ vs. [ma:l] ‘words’ [pe] ‘pear’ vs. [pe:] ‘stomach’ [son] ‘hand’ vs. [so:n] ‘loss’ [kup=.t’a] ‘be bent’ vs. [ku:p=.t’a] ‘to roast, bake’ [kot=.t’a] ‘to roll up’ vs. [ko:t=.t’a] ‘to walk’ [it=.t’a] ‘to exist’ vs. [i:t=.t’a] ‘to connect’ In Modem Korean, there are only a handful of examples with a vowel length constrast. This constrast is very easily obscured depending on the age of speakers or the rate of speech. § 1.2.2 Syllable structure of Korean Korean has a set of core syllable types consisting of four sequences (CV, V, CVC, VC). I assume the following as a canonical syllable structure of Korean: 1 “... Younger Seoul speakers have largely lost the old vowel-length distinctions... Decisions on noting vowel length for certain common words can be troublesome... The student need not worry about vowel length except when he hears it, for younger Koreans pronounce most of the older long vowels as short, maintaining only those long vowels that are the result of constraction, such as m a:m < m aum ‘soul’, ...” (after Matin 1992: 33-34). Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 7 (4) Syllable structure with glides as onset a On Nu Co C V C A C G (abbreviations: c = syllable, On = onset, Nu = nucleus, Co = coda, C = consonant, V = vowel, G = glide) I adopt the view that the glides ‘ j ’ and ‘w’ should be regarded as a part of the syllable onset, rather than a part of the nucleus (Lee, 1982; Ahn, 1985; Sohn, 1987). The glides /j, w/ are actually the features of /i, u/ with different syllabification. Whenever possible, the syllable structure shuns an onsetless syllable. Korean syllables like to begin with a consonant. When an onsetless syllable is appended to a syllable with a final consonant, that consonant shifts over to become the onset of the second syllable: for example, /pap/ ‘rice’ + III ‘Nominative’ — » [pa.bi], *[pab.i]. Korean in general bars clustering of consonants in syllable codas. Let us consider the following: (5) Consonant cluster simplification (Ahn, 1985) a. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8 /kp/ — > [p] kap<s> ‘price’ /ks/ — > [k] mok<s> ‘share’ /ntf/ — > [n] an<^> ‘to sit’ /nh/ — > [n] an<h> ‘not’ /lh/ -> [1 ] il<h> ‘to lose’ /Is/ — > [1 ] tol<s> ‘anniversary’ Atty [1 ] hal<t^> ‘to lick’ b. Am/ [m] ku<l>m ‘to starve’ Ap/ — > [P] pa<l>p ‘to tread on’ /rclph/ — > [p] 7t<I>p ‘to recite’ /lk/ — > [k] h7C<l>k ‘soil’ The cluster in ‘CVCC’ stems surfaces as a coda plus onset sequence: for example, /kaps/ ‘price’ + /U ‘Nominative’ -> [kap.sji]. But before a consonant or pause, one of the consonants truncates to satisfy the syllable structure of Korean in (4): for example, /kaps/ ‘price’ — > [kap=]; /kaps/ ‘price’ +/-kwa/ ‘and’ — » [kap=.k’wa].4 4 In Standard dialect, sometimes the second consonant is excluded (5a) while at other times the first one is (5b). The choice between the two is not relevant to the present discussion of syllable structure. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 9 § 1.2.3 Moraic phonology Languages that have a vowel length distinction typically also exhibit a syllable weight distinction, and vice versa. This is expected in moraic theory in the sense of Hayes (1989). The existence of language-specific moraic structure will be relevant throughout the phonology of a single language. In Korean, for instance, CVC and C W count as heavy for the assignment of stress and the vowel lengthening. Following Hayes (1989), I assume that contrastive vowel length is reflected directly in underlying forms: (6) a. p p b. p = /a:/ | = /a/ a a Lenis consonants do not bear any underlying mora, but are supplied with one when they close a syllable: (7) Weight by Position (Hayes, 1989:258) a a I l \ p => p p where a dominates only p I I I a ( 3 a p Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 I assume that tense and aspirated consonants have a geminate structure, always bearing a mora. For example, a sequence like [op.p’a] ‘brother’ has three moras, versus two for [o.ba] ‘overcoat’: (8) a. [op.p’a] ‘brother’ b. [o.ba] ‘overcoat’ C T C T c o K /I 1 /I n n / M - / ^ 1 \ / 1 1 / 1 o p’ a 0 b a As shown in (8a), an intervocalic aspirated or tense consonant will become ambisyllabic by linking its underlying mora to the immediately preceding syllable rendering it heavy. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11 Chapter 2: Irregular Verbs §2.0 Introduction Korean verbs5 are typically divided into two groups: regular and irregular (Choy, 1959:320-337; Huh, 1965:271-273). Regular verbs are invariant throughout the paradigm. Irregular verbs, on the other hand, show alternations before certain suffixes. In the literature on Korean (e.g., Choy, 1959; Huh, 1965; Martin, 1992), the behavior of irregular verbs is usually left unanalyzed as the alternations are considered to be phonologically unmotivated. I show in this chapter that irregular alternations are not really irregular but phonologically predictable. Below, I discuss in detail seven types of irregular verbs, which is actually in accord with Martin’s (1992:230-243) classification. This chapter is organized as follows. After a presentation of surface [p]/[u](or [w]) alternations of /p/-irregular verbs, I argue that the alternation in question should be explained in terms of simplification in coda position, which will be captured through the interaction of structural and faithfulness constraints (§2.1). The apparently dual behavior of /h/ will receive a unified account through the interactions among the universal constraints I propose (§2.2). I discuss the /t/-irregular verbs with the introduction of an underlying segment /r/ in Korean (§2.3). I take compensatory lengthening into consideration in dealing with the /ui/-irregular verbs (§2.4). The gemination of /II instead of vowel lengthening is explained by making certain faithfulness constraint protect the derivation of long vowels (§2.5). The decision to recognize /r/ as a phoneme in Korean has the advantage of dealing with an unexpected appearance of [r] in the alternation of certain verbs 5 The term “verbs”, in this chapter, includes what has traditionally been called “verbs” and “adjectives". Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 (§2.6). I point out in §2.7 as well as §2.6 that it is difficult to account for the behavior of /s/-irregular verbs in the manner consistent with the analyses of other irregulr verbs without introducing a standard Lexical Phonology of the grammar. I conclude the chapter with a summary of the main results (§2.8). §2.1 /p/-irregular verbs The /p/-regular verbs are those verbs that end in /p/. They do not show any alternations before suffixes, as shown in (9): an underlying phoneme /p/ is phonetically realized in its unchanged form when either a vowel or a consonant follows: (9) /p/-regular verbs Indicative Connective a. /kop + ta/ Ji [kop-.t’a]6 b. /kup + ta/ /kop + ko/ li [kop=.k’o] /kup + ko/ Stative /kop + a/ Jl [ko.ba] /kup + o/ u Nominal /kop + m/ U [ko.buim] /kup + m/ u Gloss 'to be numb' 'to bend' [kup-.t’a] [kup=.k’o] [ku.bo] [ku.buim] s In Korean, obstruents are unreleased in coda (p*), and the subsequent onset consonants undergo tensification (t'). This obstruent unreleasing and subsequent tensification will be discussed in detail in Chapter 4. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 13 c. /s’ip + ta/ /s’ip + ko/ /s’ip + o/ /s’ip + m/ 'to chew' U U J i U [s’ip— .t'a] [s’ip=.k’ o] [s’i.bo] [s’i.buim] However, the /p/-irregular verbs show alternations, as illustrated in (10): the stems apparently end in /p/, but show an alternation of [p] - [w] before a vowel (i.e., Indicative and Connective forms) while they have no such alternation before a consonant (i.e., Stative and Nominal forms). The question then is how to distinguish between those two types of verbs in a natural way: (10) /p/-irregular verbs Indicative Connective a. /kop + ta/ u [kop=.t’ a] b. /kup + ta/ [kup=.t'a] c. /mip + ta/ li /kop + ko/ [kop=.k'o] /kup + ko/ [kup=.k'o] /mip + ko/ li Stative /kop + a/ U [ko.wa] /kup + o/ li [ku.wo] Nominal /kop + m/ U [ko.um] /kup + m/ [ku.um] /mip + o/ /mip + m/ Gloss 'to be beautiful' 'to roast' 'to be hateful' [mip-.t'a] [mip~.k'o] [mi.wo] [mi.um] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 From the point of view of the Site-Articulator model (Gorecka, 1989) and its conception of constriction location, the alternation between [p] and [w] can be straightforwardly accounted for. I suggest two different underlying representations for the two types of verbs in question, as given in (11) and (12): (11) /p/-regular verbs /kop/ ‘to be numb’ I Place I Constriction I Lower Lip I Labial (12) /p/-irregular verbs /kow/ ‘to be beautiful’ Place C C Lower Lip Dorsal Labial Velar Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15 The /p/-regular verb stem in (11) has a Labial constriction under the Place node while the /p/-irregular counterpart in (12) is assumed to have both Labial and Velar constrictions. A fact of special interest here is that positing two different underlying representations for the same stem-final /p/ does not complicate the underlying segment inventory of Korean. That is because the particular double constriction in (12) is in fact exactly the configuration of /u/ which already exists in the phonemic inventory of Korean: (11) is simply a representation of /p/ while (12) is that of /u/ or /w/ in Gorecka (1989). I assume that the glide /w/ is actually identical with /u/ in constriction locations, but the one differs from the other in syllable positions: /w/ is a nonsyllabic counterpart of /u/, and therefore it occupies an onset position.7 In fact, Martin (1992) makes the same observation: he posits a stem-final ‘w’ and observes that it alternates with ‘p’ before a consonant-initial suffix: (13) (after Martin, 1992:223) /tow-/ ‘to help’ + /-ko/ => [top.ko] /tow-/ + /-a/ => [to.wa] /tow-/ + /-una/ => [to.wu.na] Given this, we are in a position to translate the irregularity of the /w/-ending stems into OT framework. In the analysis of the /p/-irregular verbs, I do not argue for the OT approach, but will show that they can also be analyzed in the manner consistent with other aspects of Korean phonology throughout this thesis. I propose that the alternation of the 7 cf. Levin (1985) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 irregular stem-final /w/ in question result from the interaction between the constraint which forbids double constriction in coda position (14a) and a faithfulness constraint (14b): (14) a. CODACOND: Complex segments (i.e., segments that have more than one constriction node) are not allowed in coda.8 b. MAX-IO(C): A constriction node in the input must have a correspondent in the output. From the OT perspective in the sense of McCarthy & Prince (1995), deletion of the Velar constriction means that CODACOND dominates MAX-IO(velar constriction). With this ranking, obedience to the structural constraint takes precedence over preservation of the input form, as illustrated below: * I argue that C o d aC ond is different from *COMPLEX: the former penalizes complex segments in coda position only while the latter is concerned with consonant clusters either in onset or coda. In Korean, such complex segments as [u] or [w] are allowed other than in coda position. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17 (15) Candidates CodaCond MAX-IO(velar constriction) a- w]0 Place 1 1 Lower Lip Dorsal 1 1 Labial Velar *! « ■ b. w]B 1 Place 1 C 1 Lower Lip Labial * It is necessary to ask why there should be a deletion of Velar constriction instead of Labial constriction to satisfy the coda condition in (15). Assuming that Labial constriction is more prominent than Velar constriction,9 MAX-IO(labiaI constriction) dominates Max- IO(velar constriction), which means that the actual output violates the latter rather than the former: (16) CODACOND » MAX-IO(labial constriction) » MAX-IO(veIar constriction)1 0 9 A similar phenomenon is reported in Shimizu (1971) and also cited in Gorecka (1989:92) that in Kpan (a Yukunoid language), a back glide surfaces as a labial after a velar, which supports the argument that the velar constriction is the more likely target of deletion than the labial constriction. (Unfortunately, the relevant data are not available.) 1 1 1 The ranking ‘MAX-IO(labial constriction) » MAX-IO(veIar constriction)’ will conflict with the one for place assimilation which will be discussed in Chapter 5: MAX-IO(velar place) » MAX-IO(labial place). Since glide [w] does not participate in place assimilation, I assume that the ranking ‘MAX-IO(labial Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 We can see the interaction of these constraints in the following tableau, in which unfettered CodaCond always yields a simplification effect of complex segments in coda: (17) Input: /kow + ko/ ‘to be beautiful’ (Connective) Place C C Lower Lip Dorsal Labial Velar constriction) » MAX-IO(veIar constriction)’ is a sub-hierarchy which makes reference to the points of articulation of glides, specifically. I am grateful to Alicja Gorecka (p.c.) for pointing this out to me. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 19 Candidates I CodaCond MAX-IO(labial constr.) MAX-IO(velar constr.) a. [kow.ko] 1 Place C ^ ^ C 1 1 Lower Lip Dorsal Labial Velar *! b. [kok=.k’o] 1 Place 1 C 1 Dorsal 1 Velar *! ra- c. [k o ^ .k ’o] Place 1 C 1 Lower Lip Labial * (17a) is faithful to the input and so has two constrictions in coda, which is forbidden by CODACOND. Since the violation of CODACOND by (17a) is fatal, the decision has to be made between (17b) and (17c). (17b) lacks the Labial constriction, crucially violating MAX-IO(labial constriction). The optimal output is therefore (17c), which violates neither of these high-ranked constraints. (17c) is in violation of MAX-IO(velar constriction), which is irrelevant because the constraint is low-ranked. Therefore, this analysis captures the fact that the glide [w] is not found in coda position in Korean. Another possible Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 candidate would be [ko.u.o]1 1 in which the suffix-initial consonant DsJ is deleted instead of the Velar constriction of/w/. I claim that [ko.u.o] will involve a severe violation of ONS, which militates against onsetless syllables: (18) The Onset Constraint (Ons) (after Prince & Smolensky, 1993:16) : Syllables must have onsets (except phrase-initially). ONS must be ranked higher than M\X-IO(veIar constriction), but does not have any dominance relation with CODACOND or MAX-IO(labial constriction). (19) is a case in which the stem-final /w/ is followed by a vowel-initial suffix. The following tableau compares three plausible candidates: (19) Input: /kow + a/ ‘to be beautiful’ (Stative) I Place C C I I Lower Lip Dorsal I I Labial Velar 1 1 [ko.wo] is not a possible candidate because [wo] is not allowed in Korean. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 21 Candidates CodaCond MAX-IO(labial constr.) Max-IO( velar constr.) « ■ a. [ko.wa] 1 Place 1 1 Lower Lip Dorsal 1 1 Labial Velar b. [ko.ka] 1 Place 1 C 1 Dorsal 1 Velar *! c. [ko.jja] Place 1 C 1 Lower Lip Labial *! Three candidates fare equally well in terms of CODACOND: none of them has a syllable coda. (19b) is out because it violates MAX-IO(labiaI constriction), and there are candidates available that do not violate the constraint. Of the two candidates that pass this second test, (19c) is excluded because it violates MAX-IO(velar constriction). We can see that Max- IO(velar constriction), while low-ranking, is still active here. ( 19a) is faithful to the input, obeying both MAX-IO(labial constriction) and MAX-IO(veIar constriction). Thus, it is syllabified as [ko.wa], with the stem-final /w/ keeping its double constriction in surface. In such a case, CODACOND is inapplicable, and so it is obeyed. The optimal form (19a) uniquely satisfies all the constraints. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22 The irregular nominal form in (10) can receive a similar treatment. Let us consider the following tableau: (20) Input: /kow + m/ ‘to be beautiful’ (Nominal) I Place C C i I Lower Lip Dorsal Labial Velar Candidates CodaCond MAX-IO(lab. C) MAX-IO(vel. C) DEP-IO(seg.) a. [ko.um] 1 Place 1 1 Lower Lip Dorsal 1 1 Labial Velar b. [ko.kmrn] 1 Place 1 C 1 Dorsal 1 Velar *! * c. [ko.puim] 1 Place ! c 1 Lower Lip Labial *! * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 If the stem-final /w/ is syllabified as the onset of the following syllable as in (20b) and (20c), there must be an epenthesis of a nucleus to support the onset (e.g., /kow + m/ — » *[ko.wuim], *[ko.guim], *[ko.buim]), which is not allowed in Korean. It means that the constraint against epenthesis as shown in (21) is low-ranked in the relevant constraint hierarchy: (21) DEP-IO(segment) : Every segment in the output has a correspondent in the input. In (20a), the stem-final /w/ is syllabified as the nucleus of the following syllable. Thus, /kow + m/becomes [ko.um]. Again, the faithful output in (20a) obeys all the constraints and turns out to be optimal. Therefore, the irregular nominal forms in (10) are again an automatic consequence of faithful syllabification, given the input representation in (12). An important consequence of this approach is that the behavior of /p/-irregular verbs is not irregular any more; the stem-final complex segment /w/ suffers simplification when it happens to be in coda as a result of syllabification. Therefore, it does not come as a particular surprise that the irregular stem-final /p/ is in fact /w/, and realized as [w] or [u] other than in coda position.1 2 1 2 It is reported by Kim-Renaud (1986:49) that Kyungsang dialect has the following form. (i) ki:p-t’a ‘to mend’ ki:p-k’o kib-ass kib-ara From the aspects o f alternation in (i), the stem /kip-/ must belong to the regular class o f verbs. However, I observed in Standard dialect that exactly the same stem behaves as if it were an irregular verb, as shown in (ii). Thus, (i) challenges the basic generalization. Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24 §2.2/h/-irregular verbs The stems that belong to this class end in ‘h \ This group of verbs is called irregular because the stem-final ‘h’ is optionally1 3 deleted between a voiced segment and a vowel or glide, but coalesces with a following consonant. This is shown in (22): (22) flfoh-/ ‘good’ a. Indicative /tfoh + ta1 if [tfoA] b. Causative /tfoh +aso/ if [fo.a.so] (ii) ki:p= -t’a ‘to mend’ ki:p= -k’o ki:-was3 ki:-wara I assume that in Kyungsang dialect, the stem-final segment /p/ in /kip-/ is underlyingly represented as dominating just a Labial constriction. Therefore, it behaves just like a regular verb. 1 3 According to Kim-Renaud (1986:77), the optionality seems to be governed by degree of formality and/or rate of speech. In very careful pronunciation, one hears a voiceless [h] in the voiced surroundings, but in faster, more casual speech the ‘h’ is absent. Connective /tfoh + ko/ if [tfo.kh o] Conditional /If oh + mjan/ if [tfo.ui.mjon] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25 c. Additional /h/-•irregular stems nah- ‘to bear’ noh- ‘to put’ ph arah- ‘be blue’ norah- ‘be yellow’ k’amah- ‘be black’ p’alkah- ‘be red’ hajah- ‘be white’ At issue is the following. At first glance, the behavior of /h i does not seem to be completely predictable. The ‘h’ alternates when these verbs are followed by vowel- or consonant-initial suffixes. As shown in (22a), when /h/ is followed by a lenis stop, they together produce a homorganic aspirated counterpart of the latter. (22b) shows that the verb loses the stem-final /h/ obligatorily when followed by a vowel-inital suffix. For example, the Causative form /tfoh + aso/ avoids intervocalic ‘h’ at the expense of unfaithfulness: the segment fh / in the input has no correspondent in the output. However, /h i is never deleted in stem-initial position, as illustrated in (23): (23)1 4 /halmoni/ => [hal.mo.ni] ‘grandmother’ -noun- /ha + ta! => [ha.ta] ‘to do’ -verb- 1 4 Korean does not have any prefixes to complete the paradigm to show what happens in cases like ‘Prefix + h V ...\ Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 26 In the remainder of this section, I review the traditional ways of treating Pa/ in non linear (§2.2.2) as well as linear (§2.2.1) phonology and show that all the relevant cases can be dealt with more adequately in a framework that has no recourse to rules and parameters (§2.2.3). §2.2.1 Linear approach The facts in (22) and (23) do not receive a unified explanation under the traditional rule- based analysis. The deletion of PaJ between a voiced segment and a vowel or glide is stated within the standard linear phonology as follows (Kim-Renaud, 1974): (24) h-Deletion h — » 0 I [+voice] _____ [-cons] When Pal is adjacent to a lenis stop, they together merge to produce a corresponding aspirated stop. This aspiration has been analyzed and expressed by employing a mirror- image convention in the following manner (Kim-Renaud, 1986): (25) Aspiration (mirror image) [-cont] [h] [I +asp] [2 0] 1 2 This approach expresses the alternations by imposing individual stipulations on the various relevant rules. With these stipulations, it is difficult to account for the repeated appearance Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 27 of ‘h’ in the rules motivated independently. Besides, it does not provide any motivation as to why the ‘h’ should be deleted in the particular environment shown in (24). In the same vein, the aspiration rule in (25) is not motivated, either. §2.2.2 Non-linear approach There is an alternative approach to these peculiar behaviors of /h/ within non-linear framework. Kim (1989) attempts to explain the behavior of /h/ as due to some inherent structural properties. According to her, when a stem-final /h/ is followd by a vowel as in (26), the root node of Ihl is delinked by the syllable structure of Korean: spreading does not occur in this case because long vowels are not a part of Korean phonemic inventory and cannot be derived under Structure P reservation,1 5 Consider the following derivation (Kim, 1989:124): (26) /tfoh-/ ‘good’ + /a/ => [tfo.a] a a X I I On Nu Nu I I I => X X X I I I f o a 1 5 Structure Preservation (Kiparsky, 1985) : No lexical rule application will generate structures prohibited underlyingly. c o XT\ I On Nu Co Nu I I I I X X X + X I I % o h a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28 Within non-linear phonology, the aspiration rule in (25) can be interpreted as a merger of Ihl with the adjacent stop. According to Kim,1 6 the derivation proceeds as in (27): (27) /tfoh-/ ‘good’ + /ta/ => [tfo.tha] / K / I y \ ^ On Nu Co On Nu On Nu On Nu III II II A I X X X + X X =» x x x x x I I I I I I I \ / I t f o h t a tf o t*1 a In non-linear approach, since Ihl has only the laryngeal node, its realization is determined by the adjacent segment. The behavior of Ihl can be accounted for in terms of a universal principle, the Structure Preservation, and a mechanism within non-linear phonology, the spreading. But the aspiration of a lenis stop still remains as a rule that has to be posited without motivation. Therefore, this approach cannot explain why ‘h’ is disallowed syllable-finally in Korean. 1 6 She assumes that aspirated consnants have a geminate structure. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29 §2.2.3 OT approach The basic problem with the derivational approaches is that they fail to provide a unified treatment of the data in (22) and (23). The OT approach, on the other hand, can solve this problem. I suggest that these two separate patterns of Ihl in (22) and (23) be related in terms of constraint domination specific to Korean. The apparent dual behavior of Ihl will become clearer when it is considered within a theory of constraint interaction in the sense of McCarthy & Prince (1995). OT makes no special stipulations about process-specificity. Rather, those speculations derive from the interaction between universal constraints. Applying the alignment type of analysis to the /h/-irregular verbs, I adopt a positive constraint for laryngeal neutralization by Lombardi (1994): (28) Laryngeal neutralization (after Lombardi, 1994:8) ALIGN-LEFT (laryngeal node, a ) : Every laryngeal node stands at the left edge of a syllable. The alignment constraint in (28) requires that ‘h’ stand at the onset of a syllable. The very effect of this constraint is that /h/ is licensed only in onset position. Since this constraint forces an unfaithful analysis of the input, it must dominate MAX-IO(segment), which bans deletion: (29) ALIGN-L(laryngeal node, a) » MAX-IO(segment) Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 30 In addition, there must be an independently motivated constraint requiring that each syllable onset have a place node: (30) ONSET(PL) : Every onset must have a place node. The idea of this constraint is based on a requirement found in many languages, restricting codas to the first segment of a geminate or a consonant homorganic to the onset of the following syllable (Steriade, 1982; Ito, 1986).1 7 For example, Ito (1989) expresses such a condition as a filter referring to the syllable-final position and the melody: (31) Coda Filter (Ito, 1989:224) * a I [PLACE], The top priority is given to ALIGN-L(laryngeal node, a) in (28). The force of ONSET(PL) is felt only when the alignment issue is out of the way, at the expense of violating MAX-IO(segment). The constraint ONSET(PL) in (30) looks at a candidate to check whether a syllable onset dominates a place node. When ONSET(PL) dominates MAX-IO(segment), it has the effect of blocking a placeless ‘h’ in the syllable onset at the 1 7 The O n set(PL) also captures the generalization that the onset is typically a trigger o f place assim ilation while the coda is a target o f the assimilation (cf. C hapter 5). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31 expense of deleting it. The high-ranking two constraints in the hierarchy conspire to compel the violation of MAX-IO(segment). The following tableau illustrates the situation: (32) /tfoh-/ ‘good’ + /a/ =* [tfo.a] Candidates ALIGN-L(laryngeal node, a ) onset(PL) MAX-IO(segment) a. tfoh.a *! b. tfo.ha *! « * * c. fo.a * This is one remarkable advantage of OT over the previous approaches. It can show that two separate patterns ((26) and (27) or (24) and (25)) are triggered by a single alignment constraint, ALIGN-L(laryngeal node, c). The notion of disjointness of morphemic content is relevant to the discussion of merger in /h/-irregular verbs. The constraint MORPHDIS introduced by McCarthy & Prince (1995) discriminates against the case in which the contents of two morphemes overlap: (33) Morphemic Disjointness (McCarthy & Prince, 1995:62) x c M j -» x < Z Mj, for instances of morphemes M^Mj and for x a specific segmental (autosegmental) token. “Distinct instances of morphemes have distinct contents, tokenwise.” When straightforward concatenation of morphemes leads to a merger, the process also violates UNIFORMITY (McCarthy & Prince, 1995:66), which is a string-based constraint against mapping multiple elements to a single correspondent in the output. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32 (34) shows that the very alignment constraint in (28) is also responsible for the merger of Ihl with the following lenis stop as well. (34d) clearly violates MORPHDlS and UNIFORMITY, since both the stem and the suffix share the merged ‘h’ in the output. Therefore, both MORPHDlS and UNIFORMITY must be subordinated to the alignment constraint. By virtue of the ranking given in (32), coalescence is the favored outcome, (34d): (34) /tfoh- (stem)/ ‘good’ + /ta (suffix)/ ‘Indicative’ => [tfo.tha] Candidates ALIGN-L(lary.) ONSET(PL) MAX-IO(seg.) MORPHDlS i UNIFORM a. tfoh.ta *! b. tfo.ha *! * c. tfo.ta *! d. tfo.tha * : * When ‘h’ appears in syllable-final position as in (34a), ALIGN-L(laryngeal node, < J) always decides the matter, because ‘h’ is simply not licensed in coda. (34b) crucially violates ONSET(PL). The lowly-ranked MAX-IO(segment) turns ot to be decisive in eliminating (34c) from the competition. (34d) satisfies the alignment constraint through coalescence: it trades obedience to ALIGN-L(laryngeal node, a), Onset(PL), and MAX-IO(segment) for MORPHDlS and UNIFORMITY-10, a desirable exchange given their subordinate position in the hierarchy.1 8 Therefore, in OT framework, the apparently dual behavior of ‘h’ shown in (22a, b) receives a theoretically consistent explanation. 1 8 The dotted line in the tableau indicates that MORPHDlS and UNIFORMITY-10 are not crucially ranked with respect to each other. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33 Finally, let us consider what protects a word-initial ‘h’ against deletion. The relevant data are repeated in (35): (35) stem-initial /h/ /ha (stem) + ta (suffix)/ => [ha.ta] ‘to do’ -verb- /halmoni/ => [hal.mo.ni] ‘grandmother’ -noun- In fact, word-initial position enjoys a special status across languages. McCarthy & Prince (1993:48) claim that no epenthesis of any kind ever occurs at the beginning of words because of a constraint ALIGN-L: (36) ALIGN-L (McCarthy & Prince, 1993:48) • [stem = [prWd The constraint demands that stem and prosodic word begin together. If epenthesis occurs in word-initial position, prosodic word and stem are mis-ALIGNed (e.g., [S le m □ [P rW d , where □ is an epenthetic element). Therefore, it correctly rules out all initial epenthesis. I propose that the lack of deletion in word-initial onset position results from a similar alignment constraint, as shown below: (37) ALIGN-LEFT (stem, a) : Every stem begins with a syllable (i.e., [s te m = [„). Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 34 In the case of the ‘h’ in word-inital position in (35), it is clear that ALIGN-LEFT (stem, o) dominates Onset(PL) to ensure the appearance of ‘h’ stem-initially: (38) Align-Left (stem, o), ALIGN-L(laryngeal node, a) » ONSET(PL) Let us now consider how the constraints are supposed to work. The important candidates generated from/hajah + ta/are contrasted below: (39) Input: /hajah + ta/ ‘(It’s) white.’ Candidates Align-L (stem, a) ALIGN-L(laryngeal, a) Onset(PL) "■a. [ste m [0ha.ja.th a * * > • [ste m <h> [« a.ja.tha *! In the above tableau, (39a) succeeds on ALIGN-LEFT (stem, a) while (39b) violates it: <h> in (39) is a syllabically unparsed segment. (39a) wins out because ALIGN-L (stem, a ) takes priority over ONSET(PL) in determining the actual output. §2.3 /t/-irregular verbs In (40), we have some examples of /t/-regular verbs in which the stem-final /t/ does not show any alternations throughout the verbal paradigm except for voicing, unreleasing and tensification: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 35 (40) /t/-regular verb Indicative Connective Stative Nominal Gloss a. /kot + ta/ /kot + ko f /lot + 0/ /kot + m/ ‘to pull up’ II II II II [kot=.t’a] [kot=.k’o] [ko.do] [lo.duim] b. /mut + ta/ /mut + ko/ /mut + 0/ /mut + m/ ‘to bury’ II II It II [mut=.t’a] [mut=.k’o] [mu.do] [mu.duim] c. /tit + ta/ /tit + ko/ /tit + 0/ /tit + uim/ ‘to tread’ II II II II [tit=.t’a] [tit=.k’o] [ti.do] [ti.duim] The following class of verbs is called ‘/t/-irregular’. The verbs in this class show a [t]~[r] alternation in the stem-final position (i.e., [r] before vowels, [t] elsewhere): (41) /t/-irregular verb Indicative Connective Stative Nominal Gloss a. /lot + ta/ /kot + ko/ /kot + 0/ /kot + m/ ‘to walk’ II II II II [kot=.t’a] [kot=.k’o] [ko.ro] [ko.ruim] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 36 b. /mut + ta/ /mut + ko/ /mut + 3/ /mut + m/ ‘to ask’ U U U li [mut=.t’a] [mut=.k’o] [mu.ro] [mu.num] c. /tuit + ta/ /tuit + ko/ /tuit + o/ /tuit + m/ ‘to hear’ H li H H [tuit=.t’a] [tuit=.ko] [ttu.ro] [tui.ruim] A question arises as to why ‘t’ and ‘r’ should alternate in just this way. In §2.3.2, I make a brief review of two traditional approaches: the one treating [r] as an allophone of the phoneme /I/, the other recognizing both N and /r/ as underlying segments of Korean. Then, I argue in §2.3.2 that the behavior of /t/-irregular verbs in question can be handled within the latter view. §2.3.1 The status of ‘r’ The behavior of the liquid is one of the most complicated aspects of Korean phonology. According to Kim-Renaud (1974), N and /r/ are not distinctive in Korean. She assumes an underlying N but excludes /r/ on the basis of the observation that the underlying /V is weakened to [r] in intervocalic position, as illustrated below. In (42), the first morpheme of each example ends in l\l which becomes [r] between vowels (42a, b) or between a vowel and a glide (42c, d): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 37 (42) ‘I/r’ alternation (expanded on Kim-Renaud, 1974:66-67) a. /tal/ [tal] ‘moon’ /tal + a/ — > [ta.ra] ‘Oh, moon!’ b. /kol + ta/ [kol.da] ‘to snore’ (Indicative) /kol + a/ [ko.ra] ‘to snore’ (Stative) c. /il + mjon/ — » [il.mjon] ‘one side’ /il + won/ — > [i.rwon] ‘one member’ d. /pal + ph o/ — » [pal.ph o] ‘firing’ /pal + juk/ — > [pa.rjuk] ‘growth’ [1 ] appears in coda position as shown in the first examples of (42a, b, c, d) while [r] appears allophonically in onset position, as illustrated in the second examples. She expresses this phenomenon as follows: (43) 1-Weakening (Kim-Renaud, 1974:69) : 1 — > r / [+syll] [-cons] On different grounds, Kim (1971) argues for a creation of a phoneme ‘r’ which must be distingished from T . With the V as a phoneme in the inventory of Korean, one can now distinguish the following three cases: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38 (44) l\l vs. /r/ vs. /I/ (after Kim, 1971:107)1 9 a. /mut-/ ‘to bury’-- no alternation, mut-ko. mut-j. mut-uim. etc. b. /mur-/ ‘to ask’ — alternates with t pre-consonantally, mut-ko. mur- o. mur-uim. etc. c. /mul-/ ‘to bite’ — alternates with r intervocalically, mul-ko. muro. mur-uim. etc. Kim’s (1971) proposal has the advantage of capturing an underlying stem-final constrast illustrated in (44). In what follows, adopting his view, I attempt to explain the pattern of /t/-irregular verbs within OT. §2.3.2 The analysis A question still remains as to why ‘r’ alternates with ‘t’ in (44b) which is a /t/-irregular case in (41b). Based on the fact that the only significant difference between ‘r’ and ‘t’ is the specification of the feature [sonorant], I will in argue in this section that OT can provide an explication of the alternation by regarding the pre-vocalic ‘r’ as an underlying segment. Assuming an underlying representation like /lor-/ ‘to walk’, the derivation of the Stative and Nominal forms in (41) is straightforward. They are an automatic consequence of syllabification without any alternation. Consider (45) as an example: (45) a. Stative iy Kim (1971) is the first to observe that there is a case where N and /r/ are distintive in Korean: [oli] ‘someone who will come’ vs. [ori] ‘a duck’. However, it is pointed out by Bernard Comrie (p.c.) that the constrast is not clear because at surface the ‘1 ’ in [oli] is not different from a geminate ‘1 ’ as in [pal.li] ‘fast’. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 39 /kor-/ ‘to walk’(stem) + lo l ‘Stative’ — » Syllabification — > [ko.ro] b. Nominal /kor-/ + /m/ ‘Nominal — > Syllabification — » [ko.ruim]2 0 Then, the problem is how to explain the [r]~[t] alternation in Indicative and Connective forms in (41). I claim that the operative constraint in the [r]~[t] alternation here be a requirement that [r] be disallowed in syllable coda — to be referred to as *r]0.2 1 This constraint is also plausible on cross-linguistic grounds; for example, in Japanese, surface forms do not allow [r] in the coda.2 2 Note that [r] and [t] are crucially different in the specification of the feature [sonorant]: [r] is [+sonorant] while [t] is [-sonorant]. I argue that the constraint *r]c selects a candidate with ‘t’ in the coda by dominating a faithfulness constraint iDENT-IO(sonorant): (46) IDENT-IO(F) (after McCarthy & Prince, 1995:16) : Output correspondents of an input [yF] segment are also [yF]. 2 0 [ui] is epenthesized to support the onset consonant ‘r’(cf. §2.1). 2 1 In Korean, T is the only continuant that is allowed in coda position. 2 2 It was pointed out to me by Keiko Miyagawa (p.c.) that in Japanese, there is no word ending with [r] nor any sequence of [r] followed by some other syllable. O f course, M can be at the end of some underlying stems; /tsukur/ ‘to make’, /kaer/ ‘to return’, /ar/ ‘to exist’. But when they are inflected, there is a vowel epenthesis to parse the stem-final [r] as an onset. The present forms are like [tsukuru], [kaeru], and [aru], respectively. And the past tense forms are [tsukutta], [kaetta], and [atta], respectively. Therefore, in Japanese, at least in surface [r] is not allowed in syllable coda. However, it is pointed out to me by Bernard Comrie (p.c.) that Japanese is not quite parallel to Korea in that Japanese allows only nasal consonants or the gemination of the following obstruent to occur in syllable-final position. Then, the similar behavior of ‘r’ in Japanese is due to a word- or syllable-final constraint that is much stricter than in Korean. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 40 The following tableau shows that the crucial domination of iDENT-IO(sonorant) by *r]c leads to phonological alternation between ‘r’ and ‘t’: (47) Input: /kor + ko (Connective)/ ‘to walk’ [+son] Candidates MAX-IO(segment) *r]0 iDENT-IO(sonorant) a. [ko.ko] *! b. [kor.ko] 1 [+son] *! •a * c. [kot.ko] 1 [-son] * (47a) crucially violates the high-ranked MAX-IO(segment) due to the deletion of ‘r \ The [r]~[t] alternation occurs, due to the compelled violation of iDENT-IO(sonorant): form (47b) fails the constraint against syllable-final ‘r \ while form (47c) satisfies it at the cost of mapping the input ‘r’ to ‘t’ in the output. Evidently, iDENT-IO(sonorant) is the lowest ranked constraint, and may be violated in the output to satisfy the higher ranked MAX- IO(segment) and *r]c. Consequently, given such a constraint system as (47), it follows that ‘r’ alternates with ‘t’ in the Indicative and Connective forms in (41).2 3 2 3 It is possible to conceive of [kol.ko] as one of the candidates: since the [1] is [+son], [kol.ko] does not violate the iDENT-IO(son). I assume that there is a higher-ranked faithfulness constraint, DEP-IO(Iateral) which prevents underlying /r/ from surfacing as [1], Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 41 §2.4 /ui/-irregular verbs Stems ending in /ui/ are called phonologically irregular, since the /ui/ is deleted before a vowel-initial suffix, as shown in (48b). However, the stem-final ‘u i’ is not deleted when followed by a consonant-initial suffix, as shown in (48a): (48) /koph ui-/ ‘be hungry’ a. Indicative Connective /koph ui- + ta/ /koph ui- + ko/ [ko.ph ui.ta] [ko.ph ui.ko] b. Stative /koph ui- + a/ 4 [ko.ph a:] c. Additional /ui/-irregular stems k’ui- ‘to extinguish’ tamkui- ‘to soak’ nap’ux- ‘bad’ pap’iu- ‘busy’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Conditional /koph ui- + mjon/ u [ko.ph iu.mjon] 42 An important question arises as to why the stem-final /u 1/ is deleted and the following vowel is lengthened in (48b). As a starting point, let us consider in the next section the nature of the lengthened vowel in question. §2.4.1 Compensatory lengthening The following examples are a case of compensatory lengthening in verb forms. When [ui] is deleted, the other vowel is automatically lengthened: (49) Compensatory lengthening (after Kim-Renaud, 1986:49) N orm al Casual /mauil/ [mauil] ~ [ma:l] ‘village’ /kauii/ [ka.uil] ~ [ka:l] ‘autumn’ /kui + e/ [kui.e] - [ke:] ‘that child’ Here, a question may arise as to why the stem-final /ui/ is deleted instead of the other adjacent vowel. According to Sohn (1987:127), the vowel [ui] in Korean is deleted when it is adjacent to another vowel across a morpheme boundary regardless of whether it is stem-final or suffix-initial: she claims that /ui/ should be represented as the least specified vowel in the underspecification system of Korean, based on the observation that it is typically the focus of epenthesis or deletion: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 43 (50) Nominative Effective Stative Indicative (-turn) (-uini) (o) (-ta) a. Stem: tfu- fu:-m tfu:-ni tfu-o tfu-ta b. Stem: apW - aph -ux:m aph -ui:ni aph o : ap\ii-ta c. Stem: tfuk- tfuk-uim Ifuk-uini tfuk-o tf uk-ta ‘to give’ ‘to die’ In (50a), the suffix-initial vowel /ui/ in Nominative and Effective is deleted when the stem ends in a vowel. The stem-final /ui/ is also deleted before a vowel-initial suffix, as shown in Nominative, Effective, and Stative of (50b). The suffix-initial vowel /ui/ is not deleted in Nominative and Effective of (50c) since it is not adjacent to any vowel. This asymmetrical behavior of /ui/ is further evidenced by the epenthetic process in loanwords: (51) ui-epenthesis in loanwords (expanedon Sohn, 1987:83) sports -» rsm.ph o.tfh m l arcade -> [a:.kh e.i.dui1 France -» Iph ui.ran.suil spring — » [sui.pW rig] street — > rsui.th ui.ri:.th ui) Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 44 strike — > Ism.tVi.ra.i.kVil Christmas tree — > [kbjji.ri.s’jji.ma. siji.tNji.ri:] The epenthetic vowel quality is consistent throughout the whole phonology of a language: [i] in Yawelmani, [e] in Spanish, [u] in Japanese, and [ui] in Korean. The consonat clusters in loanwords shown in (51) are broken by inserting the vowel, to satisfy ‘CV(C)’ syllable structure of Korean. In what follows, I argue that the special behavior of /iu/ results from its input representation. From a simple decision to represent /ui/ as an empty root node in underlying representation, it follows within OT framework that the alternation of /ui/- irregular verbs leads to compensatory lengthening (§2.4.2) whereas that of /l/-irregular verbs do not (§2.5). §2.4.2 The analysis As discussed in the previous section, with the deletion of the irregular stem-final /ui/, the remaining vowel h i is lengthened (i.e., /kophui- + a/ — > [ko.ph a:] *[ko.ph a] in (48b)). In fact, ‘u i’ is in fact a phonetically unstable vowel in Korean. In a serial derivation model, a phonological rule would delete /ui/ when it comes to stand next to another vowel. Then, a natural question one may pose is why /ui/ is deleted rather than the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45 other vowel (i.e., /koph ui- + a/ -» [ko.ph a:]2 4 *[ko.ph a] in (48b)). In this regard, I propose that the irregular stem-final /ui/ be represented as an empty root node in the input representation, as illustrated below: (52) Why only Im J is typically the target of deletion or coalescence? => b. underspecified representation of /ui/ R (empty root node) As to the lengthening of the vowel in [ko.ph a:] ‘hungry’, I propose a faithfulness constraint which prevents the derivation of new long vowels: (53) DEP-IO (p/Nucleus): Every nucleus mora in the output has a correspondent in the input. Let us now consider the important candidates produced by GEN: 241 observe that compensatory lengthening occurs only in open syllables: closed syllables do not undergo the lengthening. I suspect that it may be due to certain constraint which prohibits extra-heavy syllables. a. /ui/ -cons +son I Place I Dorsal I Velar Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 46 (54) Input: fx fj. fj. /k o ph ui + Candidates Max-IO (ji) O N S DEP-IO (|i/Nu) a. kop .ph ui.a *! b. koph .ph a *! L L * * • c. kop .p a: * a. c a a A A I / W / M - M - / I \ / I I k o ph ui a c. a c A A / H H / I \ / \ / k o ph a: (54a) is faithful to the input. It crucially violates ONS due to a onsetless syllable. (54b) satisfies the ONS, but the deletion of an underlying mora is fatal because of a high-ranked constraint Max-IO (p.). (54c ) is the optimal output: it avoids the violations of both b. o C T A A / n / v - / I \ / I k o ph a Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 47 constraints by lengthening the second vowel at the expense of violating the low-ranked Dep-IO (p/Nu). The interaction of the constraints in (54) reveals that there is nothing irregular about /ui/-irregular verbs. In a derivatioinal model, they look irregular because [ui] seems to delete accidently when adjacent to another vowel. In OT, however, it becomes clear that the so-called ‘irregularity’ is due to the satisfaction of MAX-IO (p.) at the expense of lengthening the following vowel: /a/ must be lengthened as a result of deleting /ui/. §2.5 /l/-irregular verbs ‘/l/-regular verbs’ are those verbs whose stems end in a consonant As can be seen in (55), they do not show any alternation before consonant- or vowel-initial suffixes, except for the intervocalic variant [r] in (55b): (55) /mol-/ ‘to drive’ (regular stem) a. Indicative Connective Conditional /mol + ta/ /mol + ko/ /mol + mpn/ J J . U U [mol.ta] [ mol.ko] [mol.mjon] b. Stative /mol + aI I [mo.ra] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48 c. Additonal /l/-regular stems ul- ‘to cry’ kil- ‘long’ gbl- ‘to doze o ff mal- ‘to roll (paper)’ tful- ‘to decrease’ kal- ‘to change’ In (56), we see examples of verbs which have traditionally been called /lA-irregular2 5 verbs: (56) /mol-/ a. Indicative /mol- + ta/ li [mo.rux.ta] b. Stative /mol- + a/ U [mol. la] 2 5 Martin (1992:238) describes these verbs as ‘1-doubling vowel stems’: when a vowel-initial suffix is attached, the stem-final ‘1 ’ geminates. ‘to be ignorant o f (irregular stem) Connective Conditional /mol- + ko/ /mol- + mjon/ J J . li [mo.rui.ko] [mo.rui.mpn] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 49 c. Additional /l/-irregular stems t|"ol- ‘to tighten’ huxl- ‘to flow’ kal- ‘to divide’ kul- ‘to roll (over)’ mal- ‘to dry’ p’al- ‘fast’ kil- ‘to raise’ As for the epenthesis of ‘u i’ in (56a), I claim that the vowel be inserted under the pressure of ‘CV(C)’ syllable structure of Korean (cf. §2.4.1). The irregular stem-final N does not alternate before consonant-initial suffixes. Compared with the /l/-regular stems in (55), the pattern in (56b) is surprising in that (i) the stem-final /1 / becomes a geminate before a vowel-initial suffix, rather than [r] as in (55b), and (ii) ‘u i’ is epenthesized in (56a) but not in (55a). This suggests that there be a constraint which asserts the markedness of long vowels: (57) *LONGVOWEL : Long vowels are prohibited. There are two reasons for positing such a constraint as * L o n g V o w e l . Firstly, long vowels are marked on the universal grounds that the presence of long vowels in an inventory implies that of short ones, but not vice versa. Secondly, in derivational terms, such a constraint would reflect the fact that the Korean long vowels are usually neutralized Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 50 to short ones, which means that long vowels are more marked than short ones.2 6 Therefore, the constraint captures the insight that some vowels are more basic than others. In fact, a prohibition against long vowels such as *LONGVOWEL has been proposed by many phonologists including Selkirk (1984), Paradis (1989, 90), and Kay (1989), etc. Since constraints are never turned off in OT, the *LONGVOWEL must be dominated in Korean to be violated. One thing to note here is that the *L o n g V o w e l must be ranked below MAX-IO(p) and DEP-IO((x/Nu) s o that it penalizes the derivation of long vowels but protects the underlying long vowels. As to the distribution of N in Korean, it only occurs in syllable-final position: an underlying N becomes [r] when it is syllabified as onset (see (42)). I propose an alignment constraint responsible for this distribution of N as follows: (58) ALIGN ([lateral], Coda): Align every [lateral] with syllable coda. Now, a simple solution to the /l/-gemination consists in employing the same constraint hierarchy motivated earlier in accounting for the /ui /-irregular verbs. The following tableau considers the important candidates: 2 6 In a rule-based system, neutralization rules (e.g., rules o f voicing, devoicing, deaspiration, deglottalization, palatalization, etc) eliminate marked values in favor of unmarked values: Markedness Constraint (after Houlihan & Iverson, 1979) : Phonologically-conditioned neutralization rules convert relatively marked segments into relatively unmarked segments. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 51 (59)/molui + a/ ‘ignorant’ (Stative) => [mol.la] Input: a a a I I I p p p I I I / m o l ux + a/ ‘ignorant’ (Stative) Candidates Max-IO(jx ) ONS DEP-IO(p/Nu) *LOVO ALIGN ([lateral], Coda) a. mol.a *! * b. mo.la: *! * c. mo.lui.a *! * * «■ d. mol.la * a. a a A I / H U H / I I I m o 1 a b. A / n / I m o I A / u u / \ / a: a a d. A A I / H / H H / I / I I m o 1 ui a A / / HU / H / I \ / I m o 1 a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52 All the candidates preserve the underlying moras, satisfying MAX-IO(p). However, (59a) and (59c) have onsetless syllables, thus crucially violating ONS. (59b) crucially violates the DEP-IO(|i/Nu) since the underlying short vowel /a/ is lengthened. (59d) satisfies the high-ranked three constraints at the expense of geminating [1 ] in violation of the alignment constrant. §2.6 /r/-irregular verbs There is a type of verb stems that end in /ui/. This class of verbs do not alternate before a consonant-initial suffix, as shown in (60a). They are irregular in that an unexpected ‘r’ appears before a vowel-initial suffix, as in (60b).2 7 2 8 (60) /p urui-/ ‘blue’ a. Indicative /p urui + ta/ Connective /p urui + ko/ [p u.rm.ta] [p u.rui.ko] Conditional /ph urui + mpn/ [pVrui.mjon] 2 7 Martin (1992:242) names this type of verbs ‘1-inserting stems’. But I term them as ‘/r/-irregular verbs’ since the name better reflects my analytic point of view. 2 8 This type o f verbs does not have a regular counterpart. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 53 b. Stative /ph urui + 0/ li [ph u.rui.ro] c. Additional /r/-irregular stems sarui- ‘to winnow’ nurui- ‘yellow’ urorui- ‘to respect’ kaph arui- ‘be steep’ maktarux- ‘(an alley) be closed at one end’ From the decision to recognize /r/ as a phoneme in Korean (cf. §2.3.1), the answer to this apparent irregularity seems to follow. As a first approximation, let us assume that the /r/-irregular stems have a shape which ends in /uir-/ in the underlying representation. Then, the appearance of ‘r’ in (60b) should not be surprising. That is, the surface ‘r’ is an automatic consequence of the faithful syllabification of the input: (61) /ph uruir- + 0/ ‘blue’ (Stative) Syllabification [pVrui.ro] Adopting this strategy, the most vexing question is why the stem-final /r/ does not become [t] before consonant-initial suffixes, as shown in (60a). In fact, the constraint Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54 hierarchy proposed for the /t/-irregular verbs in §2.3 would wrongly predict that the optimal output is (62c), but the actual output is (62a). The results are shown in the tableau below: (62) Input: /phuruir + ta/ ‘blue’ (Indicative) [+son] Candidates MAX-IO(segment) *r]„ iDENT-IO(sonorant) a. [pSi.rui.ta] *! b. [p^u.ruxr. ta] 1 [+son] *! c. [ph u.ruit.ta] 1 [-son] * (62a) could be correctly predicted to be optimal if MAX-IO(segment) is re-ranked below iDENT-IO(sonorant): *r]C T » iDENT-IO(sonorant)» MAX-IO(segment). To provide an explanation for this conflict of rankings, it is necessary to adopt a model of lexical phonology along the lines of Kiparsky (1982) and McCarthy & Prince (1993). We must posit at least two levels in Korean phonology, Level I and Level II: each level selects the candidate form that best satisfies its parochial constraint hierarchy. At Level I, MAX-IO(segment) is more highly ranked than *r]a and iDENT-IO(sonorant): MAX- IO(segment) » *r]0 » iDENT-IO(sonorant). At Level II, however, r]a is more highly ranked than the other two constraints: *r]G » iDENT-IO(sonorant)» MAX-IO(segment): Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 55 (63) Level-ordered phonology Level I: /t/-irregular verbs MAX-ICXsegment) » *r]0 » iDENT-IO(sonorant) Level II: /r/-irregular verbs *r]a » iDENT-IO(sonorant) » MAX-IQ(segment) The existence of surface MAX-IO(segment) violation in (62a) makes it clear that the constraint must be demoted in Level II in which the /r/-irregular verbs belong. §2.7 /s/-irregular verbs Certain stem-final ‘s’ is deleted before a vowel-initial suffix,2 9 while it stays before a consonant-initial suffix: (64) /is-/ ‘to connect’ (/s/-irregular) a. Indicative Connective /is + ta/ /is + ko/ Ji a [it=.t’a] [it=.k’o] 2 9 This is the reason why Martin (1992) describes those as ‘s-dropping stems’. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 56 b. Conditional Stative /is + uim pn/ /is + • 0/ U n [i.ui.mjon] [i.o] c. Additional /s/-irregular stems nas- ‘to cure’ tfis- ‘to build’ kuis- ‘to draw (a line)’ The /s/ in (64a) becoming [t] syllable-finally in Indicative and Connective forms is a result of neutralization (cf. Chapter 3). Positing an underlying /s/ rather than /t/ in (64) is based on the fact that the ‘s’ appears at surface in some instances, e.g., /is + uimjon/ -> [i.sui.mjon], /is + 0/ — > [i.so]. The question is why the ‘s’ is deleted before such vowel- initial suffixes as Conditional and Stative in (64c), at the expense of unfaithful syllabification. Kim (1972:107) observes that the ‘s’ is phonologically an exception to to an intervocalic voicing phenomenon which every other lax obstruent of Korean undergoes, i.e., /p, t, k, yy becomes [b, d, g, cfe], respectively, but /s/ does not become [z]. Based on his observation, I suggest that the deletion of the intervocalic ‘s’ result from the interaction of a constraint VOICE prohibiting a voiced consonant between voiced segment and MaX- IO(segment): Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 57 (65) Voice : *VCV, *NCV where C is a voiceless lenis consonant, V a vowel, and N a nasal. By ranking VOICE above MAX-IO(segment), we obtain a desired result. The following tableau delivers the effect: (66) Input: /is- + o/ ‘to connect’ (Stative) Candidates v o ic e MAX-IO(segment) a. [i.so] *! fir b. [i.o] * However, one additional complication should be mentioned: there are so-called 7s/- regular’ verbs which do not show such an irregular behavior as above. Let us consider the following: (67) /us-/ ‘to laugh’ a. Indicative /us + ta/ II [ut=.t’a] b. Conditional /us + uimjon/ II [u.sui.mjon] (/s/-regular) Connective /us + ko/ II [ut=.k’o] Stative /us + o/ II [u.so] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58 c. Additional /s/-irregular stems sos- ‘to rise’ pos- ‘to take off As shown in (67b), the stem-final ‘s’ is not deleted before vowel-initial suffixes. As discussed in §2.6, the apparent contradiction is resolved if it is assumed that there is a change in the constraint rankings between levels. For the /s/-regular verbs to be accountable, we have to posit a ranking opposite to the one in (66): (68) Input: /us- + o/ ‘to laugh’ (Stative) Candidates MAX-IO(segment) VOICE « ■ a. [u.sd] * b. [u.d] *! In this sense, it is plausible to assume that there are distinct constraint systems handling the two types of the ‘s’-ending stems in Korean, as in §2.6. §2.8 Conclusion It is argued in this chapter that the so-called ‘irregular’ verbs in Korean which have been regarded as mysterious by many of the traditional grammarians of Korean (Choy, 1959; Huh, 1965; Martin, 1992) are in fact not anomalous. From the perspective of OT, there is no mystery here. On the contrary, the irregular verbs turn out to be phonologically predictable, employing the language-specific ranking of universal constraints that is central to OT. It is shown that the description of grammar along the lines of the Correspondence Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 59 Theory in the sense of McCarthy & Prince (1995) makes the grammar of Korean simpler by elimininating from it a great deal of phonological and morphological exceptions. Also, it is pointed out in §2.6 and §2.7 that we need to posit distinct constraint systems in terms of a standard Lexical Phonology of the grammar. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60 Chapter 3: Neutralization §3.0 Introduction In Korean, neutralization occurs syllable-finaily as a process of weakening. The purpose of this chapter is to provide a constraint-based account of the phenomenon of neutralization in Korean within the framework of the Correspondence Theroy (McCarthy & Prince, 1995). It will be shown that the triggering of neutralization can be expressed by ranking a constraint forbiding a release feature above a faithfulness constraint. §3.1 The paradigm Obstruents in Korean are all voiceless and occur in three different series with the increasing degree of aspiration: (69) a. TENSE (unaspirated) : p \ t’, k’, tf\ s’ b. LAX (slightly aspirated) : p, t, k, tf, s c. (tense heavily) ASPIRATED : ph , th , k \ tfh , This distinction is maintained syllable-initially (or word-initially), as illustrated in (70): Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 61 (70) a. p’ul ‘unicorn’ pul ‘fire’ ph ul ‘grass’ b. t’am ‘sweat’ tam ‘fense’ th am ‘greed’ c. k’i ‘meal’ ki ‘energy’ kh i ‘height’ In syllable-final position, obstruents are neutralized into their unreleased lax counterparts. On the phonetic level, Korean allows only a limited number of consonants in syllable-final position, i.e., [p=, t=, k=, 1 , m, n, t)]: (71) a. /nat/ => [nat~] ’ a grain’ /natty => [nat=] 'a piece' c. /ip + to/ => [ip=.t’o] 'mouth also’ /iph + kwa/ => [ip=.k’wa] 'leaf and' /kuk + to/ => [kuk=.t’o] 'soup also' /puokh + kwa/ => [pu.ok=.k’wa] 'kitchen and' /k'ok' + ko/ => [k'ok= .k’o] 'to break and’ c. /nas/ => [nat=] 'a scythe' Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 62 /nas’-/ => [nat=] 'was bom' /natf/ => [nat=] 'a day' /natfty => [nat=] 'a face' The coronal obstruents, for example, are neutralized into unreleased [t=] regardless of whether they are stops, fricatives, or affricates, as shown in (71a, c). Since the neutralization affects the laryngeal features, the place features survive it; labials remain labial and velars are still velar after neutralization, as shown in (71b). Interestingly, fricatives and palatal affricates always change their points of articulation to anterior coronal [t=] after neutralization, as illustrated in (71c). §3.2 Earlier treatments In what follows, I review previous approaches to the syllable-final neutralization. I will show that they are not adequate to explain the neutralization pattern illustrated in (71). §3.2.1 Iverson & Kim (1987) Within the framework of Radical Underspecification (Archangeli, 1984), Iverson & Kim (1987) formulate a neutralization rule, as shown in (72): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 63 (72) Laryngeal neutralization (Iverson and Kim, 1987) a. C]a b. C]0 I I R R A = * A L SL L SL t A A [ ]M P M P t I I [ ] PP PP t [ ] (C: Consonant; R: Root node; L: Laryngeal node; SL: Supralaryngeal node; M: Manner node; P: Place node; PP: Primary Place node) This rule delinks all the terminal features in coda, including all the place, manner, and laryngeal features (72a). As a result of neutralization, the segment will be maximally underspecified as in (72b). Later, the redundancy rules will supply the relevant default feature and feature values to spell out the least marked segment of Korean, i.e., [t], A general model of feature organization has been proposed in which features that regularly function together as a unit in phonological rules are grouped into co n stitu en ts (Clements, 1985; Sagey, 1986, among others). Thus, an adequate theory of phonological features must account for natural groupings of features; i.e., distinguish features that tend to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 64 function together in phonological rules from those that do not. There is so much evidence in the literature in favor of this claim (Clements, 1985, 1991; Sagey, 1986; Kim, 1987; McCarthy, 1988; Gorecka, 1989; Hume, 1992, etc.) that it is very plausible as a principle of universal grammar. Therefore, the phonological description of any individual language must be consistent with it. It follows then that for a linguistic theory to have a cross- linguistic empirical predictions, such a widely-attested and natural phenomenon as neutralization must involve a simple operation in phonology. Along these lines, the problem with this formulation of neutralization rule in (72) is that the delinked nodes (i.e., Laryngeal, Manner, and Primary Place nodes) do not form a constituent in the hierarchical representation in question, and the formulation represents a strong empirical hypothesis regarding the class of possible phonological rules. §3.2.2 Clements & Hume (1995) Clements & Hume (1995) point out that the rule in (72) would require a relaxation of such principle as (73): (73) Phonological rules perform single operations only. According to this principle, phonological rules are supposed to perform a single operation on constituent in case of common processes. In other words, only a single node may undergo delinking. In this regard, the formulation by Clements & Hume seems to be more satisfactory. The following derivation illustrates their analysis of neutralization; note that parenthesized nodes are automatically interpolated to preserve wellformedness: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 (74) Neutalization (Clements & Hume, 1995) Delinking Automatic deletion C -» C Default insertion C root / [cont] place (root) [- cont] (place) [coronal] I [anterior] t, tf, s [coronal] I [+ anterior] t However, it is clear that the derivation in (74) does not predict the survival of place features after neutralization illustrated in (7 lb). In (74) the root node including the place node is deleted. Since the default insertion is always supposed to produce [t], Clements & Hume’s analysis cannot explain the fact that labials and velars do not change their place features after the neutralization. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 §3.3 Glottalization of unreleased consonants Cohn (1990:179) reports that glottalization is predicted when the English /t/ is syllable- final.3 0 She claims that there is a phonological rule of glottalization in English whose effect is to add a [+constricted glottis] specification when /t/ is in syllable-final position: (75) Glottalization (Cohn, 1990:180) a ooR oo R [+son] o o L ooSL / \ 1 [-voice] i o o p I V [+CG] Coronal In addition, an English sonorant following a glottalized /t/ may be glottalized as well,31 in a quite categorical fashion. To account for this glottalization of a following segment, Cohn proposes a rule of Glottal Spread (76) which must be ordered after the Glottalization (75): 3 0 Kahn (1976:49-51) makes a similar argument. He compares words like ‘ m at, map, and Mac'. According to his observation, /t/ is unreleased , /p/ is preferably unreleased, and /k/ is preferably released. Only the unreleased /t/ is glottalized and he states the result of the observation as a rule shown below: [-cons] t Pause X [+C.G.] (C.G. = constricted glottis) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 (76) Glottal Spread (applies optionally) a I R I C oo R ooR oo L °oSL,_ °°L [+sonorant] [+constricted glottis] Syllable-finally unreleased consonants in Korean and other languages3 2 also accompany glottalization (Baek, 1991). As in English, the glottalization caused by coda unreleasing affects the following segment in Korean, as will be documented in Chapter 4. But this tranfer of glottal feature is not observed in Thai (Ambramson, 1972). §3.4 Closure and release phase of a stop Steriade (1991) presents evidence that phonetic gestures such as closure and release must be represented in phonology as positions defined in terms of degrees of aperture. She represents plosives as sequences of closure plus some release, either that of the approximant or of the fricative. Her study is based on the hypothesis that the release of a stop is identical in structure to the aperture position carried by an approximant or a fricative. In other words, 3 1 This is true for /r, n, 1/ or vowel. The following is a list o f phrases used in Cohn’s experiments: (i) Examples of glottalized sonorants : repea(t) Elmer, repea(t) rednek, repea(t) neo 3 2 Thai, Taiwanese, Toba Batak, etc. (Bruce Hayes (p.c.)) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 68 according to her idea, released stops consist of two aperture positions while unreleased stops consist of one, as illustrated in (77): (77) plain, released stop = Ao Amax unreleased stop = Ao affricated stop = Ao Af approximant = Amax fricative = Af abbreviations : Ao - or zero aperture - stands for closure, Ao Amax - °r maximal aperture - stands for approximant release, Af - or aperture-cum-friction - stands for fricative release. The distinctive property of the representations in (77) is that a clear distinction between released and unreleased stops can now be made: (78) plain, released stop = Ao Amax unreleased stop = Ao As shown in (78), unreleased stops are contrastively represented as a simple closure Ao without release while plain, released stops have a closure Aq with a release feature Amax- Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 69 §3.5 The analysis Taking advantage of the theory of closure and release in Steriade (1991), I propose the following analysis of neutralization in Korean. The syllable-final neutralization consists of two components: (i) delinking of the release - either Amax or Af in syllable-final position, (ii) insertion of [CG]. The situation is represented in (79): (79) a. Neutralization of stops in coda C]C T C]C T A = > I Ao Amax Ao I [CG] b. Neutralization of affricates in coda C]c C]<j A = * I Ao A f Ao I [CG] Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 70 We can now account for the fact that we set out to explain. I propose that since released stops have two aperture positions, i.e., AoAmax* neutralization process in syllable-final position entail losing one of the two aperture positions with the insertion of [CG]. In a rule- based approach, for example, /nat/ in (71a) would receive the following derivation: (80) /nat/ ~> C]c A Ao Amax III I claim that this neutralization effect can receive a straightforward explanation through the constraint interaction in OT. First of all, the syllable-final neutralization can be expressed as an interaction between a family of UNRELEASE constraint (81) and that of faithfulness constraints (82): (81) Unrelease constraints a. ALlGN(Obs]a, Ao): Every obstruent in coda must be unreleased (i.e., aligned with Ao). b. Aq -[CG]: Unreleased consonants must dominate [CG]. [nat- ] ‘a grain’ C]G Ao [CG] [t=] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 71 (82) Faithfulness constraints a. MAX-IO(Aniax): Every Amax in the input has a correspondent in the output. b. MAX-IO(Af): Every Af in the input has a correspondent in the output. c. Dep-IO([CG]): Every [CG] in the output has a correspondent in the input. A family of Unrelease constraints can trigger a change to the input if it dominates the family of relevant faithfulness constraint. That is, the following ranking forces the loss of a release feature Amax and the insertion of [CG]: (83) UNRELEASE constraints » Faithfulness constraints Tableau (84) shows the evaluation of example ‘[nat=] a grain’ from (71 a), in which neutalization produces an unreleased homorganic stop in coda: (84) Input: n a t](j l \ Ao Amax Candidates ALIGN(Obs]a, Ao) A0-[CG] Max-IO( Amax) a. n a t]< y Ao Amax *! * b. n a t=](j 1 Ao 1 [CG] * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 Given the constraint ranking above, there is no way for the coda to surface without the deletion of Amax with the insertion of [CG]. The violation of MAX-IO(Amax) is compelled by the satisfaction of the higher ranking ALIGN(0bs]o , Ao). (84a) is faithful to the input, crucially violating ALIGN(Obs]a, Ao) and Ao-[CG] which are dominant in the constraint hierarchy. Therefore, the optimal output is less faithful one (84b) which satisfies the high- ranked constraints. Now, let us consider (85); only palatal affricates change their place feature after neutralization: (85) Neutralization of palatal affricates3 3 /naff/ => [nat=] ‘a day’ /natfh/ => [nat=] ‘a face’ I argue that the deletion of a release feature (Af) from palatal affricates (76a) results in the configuration of an unreleased palatal stop (76b) which does not exist in the underlying inventory of Korean: 3 3 I am grateful to Bernard Com rie for pointing out that Sanscrit has very sim ilar m orphophonem ic alternations. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Coronal I Palatal Coronal I Palatal Thus, GEN does not produce (76b). The following tableau shows that in coda, a surface anterior coronal stop is favored over a palatal affricate at the expense of faithfulness to the release feature (Af) and Site feature (Palatal): (87) Input: /natf/ ‘a day’ K Ao Af I Constriction I Coronal I Palatal Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 74 Candidates ALIGN(0bs]o, Ao) Ao-[CG] MAX-IO(Af) MAX-IO(Site) a- C]„ [ \ Ao Af 1 Constr. 1 Cor *! * b- C]0 1 Ao 1 Constr. 1 Cor *! * * «- c. C]B 1 Ao K Constr. [CG] 1 Cor * * As in (84), the UNRELEASE constraints dominate the faithfulness constraints. MAX-IO(Af) and MAX-IO(Site) are undominated with respect to each other. (87a) fatally violate ALIGN(Obs]0 , Ao) by failing to delete Af syllable-finally. (87b) crucially violates Ao~[CG]. (87c) is the only candidate that passes this first two constraints, and turns out to be optimal: it will be supplied with Anterior site by default later in the interpretive component, and realized as [t=]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 75 §3.6 Conclusion The generalization that neutralization in Korean has the effect of avoiding a release feature (Amax or Af) in coda can be captured in derivational models, as seen in §3.2.: they can be handled by a rule delinking all the laryngeal features in the laryngeal tier. The OT account can also capture this generalization by making the UNRELEASE constraints dominate the faithfulness constraints, thus triggering the deletion of a release feature with the insertrion of [CG] in coda. The change of place feature in case of the syllable-final palatals is expected by assuming that GEN cannot produce an output which does not exist in the underlying inventory of the language in question. Therefore, OT enables us to explain the neutralization pattern of Korean, using the language-particular ranking of universal constraints. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 76 Chapter 4: Tensification §4.0 Introduction Many researchers of Korean have taken it for granted that the tensification of the consonants in syllable-initial position is not related to the neutralization of the preceding syllable-final obstruents, and posited a separate rule for tensification. That has been a general assumption within the framework of both linear (Kim-Renaud, 1974, 1986) and non-linear phonology (Sohn, 1987; Kim, 1987). This chapter explores phonological properties clustered around neutralization and tensification phenomena in Korean within the framework of OT. I aim to explain the close relationship between the two by representing segments as positions defined in terms of oral aperture in the sense of Steriade (1991). I will argue that the close correlation between the two phenomena — neutralization and tensification -- becomes clear when we define neutralization as the loss of release feature with the insertion of [CG], as proposed in Chapter 3. Tensification will be analyzed as a consequence of the neutralization of the preceding consonant. §4.1 The paradigm Let us begin by considering the tensification paradigm which presumably results from the weakening of the preceding segment. The fact requiring explanation is illustrated below: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 (88) a. /tfap + ko/ [^ap=.k’o] ‘catches and’ /kiph + ta/ => [kip=.t’a] ‘It is deep.’ b. /tat + if a/ [tat=.if’a] ‘Let us close!’ /path + kwa/ = f> [pat=.k’wa] ‘field and’ c. /tfhatf + ki/ => hat=.k’i] ‘searching’ /k’otfh + to/ [k’ot=.t’o] ‘flowers also’ d. /ahak + kogpu/ => [a.hak=.k’or].bu] ‘linguistic studies’ /puakh + to/ => [pu.ak=.t’o] ‘kitchen also’ /k’ak’ +tfa/ => [k’ak=.fa] ‘Let us shear it!’ (C’: tense consonant, C^: aspirated consonant, C ~: unreleased consonant) It is clear that in Korean, all obstruents are susceptible to syllable-initial tensification whenever they are immediately preceded by another consonant which is unreleased syllable-finally. §4.2 Earlier treatments Earlier investigations of neutralization and tensification treat these two as separate processes. For example, Kim-Renaud (1974, 1986) posits two rules as in (89) and (90) to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 78 express the neutralization of the syllable-final obstruents and tensification of the syllable- initial obstruents, respectively, within the framework of linear phonology: (89) Obstruent Unreleasing (Kim-Renaud, 1986:11) [+cons, -son] => [-release] / . (One or more obstruents are unreleased in syllable-final position.) (90) Post-Unreleased Fortition (Kim-Renaud, 1986:20) [+rel, -son] => [+tense] / [-rel] ______ (A released non-sonorant segment becomes fortis when preceded by an unreleased segment.) More recently, Sohn (1987:243) observes that when obstruents are unreleased in the coda position and followed by another consonant, the closure in place of articulation is extended over to the following articulation, hence building up the pressure of the air in the pharynx for a two-consonant duration. Thus, she claims that the increase of the air pressure in the pharynx caused by the unrelease of the obstruent in the coda gives the same effect as the glottal closure by which the air in the pharynx is compressed. Therefore, she argues that the tensification after the unreleased consonant is accounted for by the phonological equivalence to the constriction of the glottis. She formulates the tensification rule, within the non-linear framework, as a rule of inserting an X with [+constricted glottis] between two X-slots with [-son], as in (91a). The inserted [+constricted glottis] is spread onto the following onset, as in (91b). According to her, syllable-final neutralization is formulated as delinking of laryngeal features, as shown in (92). Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 79 (91) Tensification rule (Sohn, 1987:243) a. 0 -> X / X ________ X I I I [+CG] [-son] [-son] b. X X L ^ I [+CG] [-son] (92) Neutralization rule(Sohn, 1987: 264) Nucleus I X t laryngeal tier: [aF] The most important problem with previous approaches to tensification is that obstruent unreleasing in syllable-final position is not taken into account and posited as a separate rule in both formulations, as illustrated in (90) and (91). Since the syllable-initial tensification of a lenis consonant is typically accompanied by the syllable-final neutralization of the immediately preceding consonant, we would clearly miss the generalization if we failed to relate those two phenomena. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 §4.3 Tense consonants as geminates Tense consonants exist in the inventory of Korean independently of the tensification in question. There is a three-way distinction in Korean stops: plain stops /p, t, k/, aspirated stops /p*1 , th, k*V , and tense stops /p’, t \ k’/. Korean has some phonetic evidence that tense consonants are longer enough than lenis counterparts in duration, from which I assume that they have geminate structure. Kim (1988:25) makes the recordings of buccal air pressure behind the point of closure and reveals that the average duration of the increased pressure in about 150 samples is 141 msec for/p’, t \ k’/, 94 msec for/p, t, k/, and 145 msec for /ph, t*\ kty. That is, tense or aspirated consonants have approximately one and a half times longer duration of the increased pressue than lax consonants. More recently, Yoon (1992) investigates the possibility to apply the closure duration as a cue to distinguish the three series of stops. Three experiments were performed and subjects were 10 Korean students (4 males and 6 females) of the University of Alberta. In ‘experiment I’, a minimal pair [kobi] ‘a hard situation’ and [kop’i] ‘shackles’ was compared. It turned out that the average duration for [b] in [kobi] was 50 msec, and that for [p’] in [kop’i] was 250 msec. In ‘experiment II’, a minimal pair [nobi] ‘servant’ vs. [nop^i] ‘height’ was compared. Closure duration of [b] in [nobi] was roughly the same as that in [kobi] ‘a hard situation’, but that of [p] in [nop^i] was about 180 msec. A three- way distinction is made in ‘experiment HI’: [jagan] ‘during the night’, [jak’an] ‘a little bit’, and [jak^an] ‘weak’. Closure duration were 50, 230, and 180 for [g], [k’], and [kh], respectively. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 81 Han (1992) conducts an experiment which measures the closure duration of bilabial and alveolar stops with respect to position (word-initial vs. intervocalic) and phonation type (tense vs. plain). The test words and frame sentences are given in (93a) and (93b), respectively: (93) Han (1992:216) a. test words position phonation tense plain v_v alveolar at’a **’ ata **’ bilabial ap’a ‘father’ apa **’ # _ alveolar t’a ‘to pick’ ta ‘to touch’ bilabial p’a ‘to mash’ pa ‘to see’ '*’ = nonsense words b. frame sentences V_V: ikon......eyo. ‘This is # : sakw a.....cuseyo. ‘please......the apple’ c. Closure Duration of Bilabial and Alveolar Stops in Korean (ms): phonation position intervocalic word-initial jjlain tense plain tense speaker A 52 172 55 84 speaker B 51 118 51 81 average 54 145 53 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 82 As shown in (93c), the intervocalic tense consonants are twice as long in closure duration as those in word-initial position. Based on this experiment, Han suggests the following three types of underlying contrasts: (94) a. b. c. word-initial tense C word-initial or intervocalic plain C intervocalic tense C c c c c I \ / [+CG] [+CG] In the next sections, I will consider the aspects of Korean phonology that force us to regard tense and aspirated consonants as geminates. I will show that phonological behaviors of tense and aspirated consonants in Korean are different from those of lax consonants in some crucial respects. §4.4 Onomatopoetic and mimetic words One argument for positing a geminate structure for tense and aspirated consonants is based on the onomatopoetic or mimetic words in which tense or aspirated onsets tend to have different connotations from lenis onsets: (95) a. /pig.pig/: a mimetic word denoting the manner a top spins b. /phig.phig/ and /p ’ig.p’ig/ denote faster, more intense, spinning. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 83 When occurring in a series of expressives, consonants of different intensity (i.e., plain- aspirated-tense) correspond to different levels of intensity of meaning. It is not really easy to pin down an exact English translation for each of them. Generally speaking, plain consonants tend to symbolize motions or states of heavier and bigger objects, while their connotations are more or less weaker and slower than those of aspirated or tense counterparts; aspirated ones are for more flexible, elastic, buoyant, crispy, and swift actions or states; and tense ones are for smaller, tighter, more solid, faster, and more intensive and urgent actions or states. Some representative examples follow: (96) Shade of meaning from consonant intensity (after You, 1992) Plain Aspirated Tense Basic Meaning ping-R phing-R p ’ing-R ‘in circles’ pelt’ek-R phelt’ek-R p ’elt’ek-R ‘pitapat’ tan-R t^an-R t’an-R ‘sturdy’ telkek-R t^elkek-R t’elkek ‘rattling’ kam-R k^am-R k ’am-R ‘hopeless’ tfelpek-R tfhelphek-R f e lp ’ek-R ‘splashing’ pan-R p^an-R p’an-R ‘flat’ pantf’ak-R p’antf’ak-R ‘twinkling’ komtfil-R k’omtfil-R ‘budging’ osak-R os’ak-R ‘chilly’ (R stands for Reduplicated stem: e.g., ping-R = ping-ping) Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 84 It is of special interest that we can derive the same connotations by making N geminate even though it does not have tense or aspirated counterparts (Jun, 1992): (97) a./alorj alog/ => [a.rog a.rog] a a I A M - / M - M - I /II a r o g b. /allog allog/ => [al.Iog al.Iog] 0 a l\ A HU / M 1 \ / I I a 1 o g In (97a), a single N is always realized as [r] in syllable-initial position. When we make the N geminate as in (97b), we get the intensive counterpart of (97a). This strongly suggests that tense consonants have the same phonological status as geminates. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. ‘mottled’ ‘densely mottled’ 85 §4.5 Initial geminates in Mokilese3 4 One of those languages which do not allow consonant clusters but allow geminates in onset is Mokilese, a Micronesian language of the eastern Caroline Islands. Harrison (1984) gives a chart of contrastive consonants in Mokilese as follows: (98) Stop Nasal Fricative Lateral Trill He notes that all consonants in (98) occur both single and geminate. The Trukic languages show true geminates in all positions. However, Mokilese permits true geminate consonants only medially (i.e., intervocalically), even though historically it had such consonants in all position. Goodenough (1963) notes that the initial geminates of Micronesian langusges seem to have arisen through the reduction of morphological reduplication. According to Harrison (1984), the history of Mokilese reflects moves towards the dissolution of initial and final geminate consonants. It is suggested (in footnote 28) that the reason why medial geminates tend to be stable might be attributed to the fact that the two ‘parts’ of these geminates are assigned to different syllables. 34 Luganda is a sim ilar language. Bilabial Dental Palatal Velar Velarized labial p d j k pw m n ng mw s Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 Similar remarks apply to Korean tense consonants in the onset or coda: since the syllable structure of Korean does not allow geminates in either position at surface, the underlying geminate structure is simplified (i.e., the underlying mora is deleted). §4.6 Tensification in OT There are three series of obstruents in Korean: (a) (slightly aspirated) 1 m ; (b) (unaspirated) tense and (c) (strongly) aspirated. I assume that the difference between ]m obstruents on the one hand and tense and aspirated obstruents on the other hand lies in that the latter has an underlying mora while the former does not. Tense and aspirated consonants are distinguished by the sprecification of [+constricted glottis] and [+spread glottis], repectively. In what follows, I slightly modify Han’s (1992) proposal in terms of Moraic Phonology (Hayes, 1989). (99) illustrates the underlying constrast of plain, tense, and aspirated consonants: (99) Underlying representations of stops a. plain C b. tense C c. aspirated C V - M - I I R R R I I [CG] [SG] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 87 Tense and aspirated consonants lose an underlying (i syllable-initially as well as syllable- finally under the pressure of the syllable structure of Korean. (100) compares the surface representations of plain and tense consonants: (100) Surface manifestations of stops a. word-initial tense C b. word-initial or intervocalic plain C c. intervocalic tense C R [CG] \ R [CG] The intervocalic tense consonants are ambisyllabic at surface, as shown in (100c): the underlying mora survives and links to the preceding syllable. This explains why the intervocalic tense consonants are longer in closure duration than those in other positions. It was proposed §3.5 that there be an alignment constraint responsible for the syllable- final neutralization of lax consonants: (101) UNRELEASE constraints (repeated from §3.5) a. ALIGN(Obs]c, Ao): Every obstruent in coda must be unreleased (i.e., aligned with Ao). b. Aq-[CG]: Unreleased consonants must dominate [CG]. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 88 I claim that the the syllable-initial tensification following neutralization of the preceding syllable-final obstruents is triggered by a constraint requiring the onset to license a laryngeal node. The following is the licensing constraint in which licensing means a single motherhood: (102) LlCENSE(laryngeal node) : Coda must not license a laryngeal node. If the UNRELEASE constraints dominate the LlCENSE(laryngeal node) as in (103), we would obtain the candidate with shared specification of [CG] as the optimal output: (103) UNRELEASE constraints » LlCENSE(laryngeal node) » ALIGN-L(laryngeal node, a) The ranking means that the syllable-final loss of a release feature with the insertion of [CG] must accompany the sharing of the [CG] with the preceding consonant at the expense of violating an alignment constraint. In what follows, I illustrate the various tensification effects. §4.6.1 Tensification after stops With the ranking given in (103), we can now account for the fact that we set out to explain. (104) is a typical example of the neutralization of obstruents and the tensification of the following consonants: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 89 (104) Input: /p/ D/J /tfap + ko/ => [tfap=.k’o] ‘catches and’ A A Ao Am ax A q Amax Candidates ALIGN(0bs]o, Ao) Ao-[CG] LlC(lary.) ALIGN-L(lary., a ) a - p]0 «[k A A Ao Amax Ao Amax *! b- P=]„ 0[k 1 A Ao Ao Amax 1 [CG] *! c. P=L c[k’ 1 A Ao Ao Amax 1 [CG] *! * * ■ d. p=]„ 0[k’ 1 A A q A o Am ax [CG] * (104a) is faithful to the input, crucially violating ALIGN(Obs]0, Ao) whose effect is to forbid the release feature in syllable-final position. ( 104c) satisfies the dominant ALIGN- R(ct, Ao), but it violates the Ao-[CG] since the syllable-final Ao is not aligned with [CG]. (104b) violates the LlCENSE(laryngeal node) since the [CG] is solely associated with the coda which is not a licenser of a laryngeal node. [tfap=.k’o] in (104d) emerges victorious from the evaluation procedure: the coda-onset cluster shares the [CG] at the expense of violating the ALIGN-L(laryngeal node, a). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 90 §4.6.2 Tensification after fricative In Korean, /s/ neutralizes to [t=] syllable-finally and induces tensification of the following onset as any other obstruents with two aperture positions. According to Steriade (1991), fricatives have only one aperture position (i.e., Af) unlike stops. It is of interest then that this special property of /s/ can receive the same treatment as any other stops within OT. (105) illustrates the situation: (105) /sos + ta/ ‘to rise’ => [sot-.t’a], *[sot=.ta] /us + ta/ ‘to laugh’ => [ut=.t’a], *[ut=.ta] /pis + ta/ ‘to comb’ => [pit=.t’a], *[pit=.ta] Let us consider the following tableau in (106): (106) Input: /s/ /t/ /sos + ta/ [sot-.t’a] ‘to rise’ Af A Ao Am ax Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 91 Candidates ALI(0]a)3 5 A0[CG] Lic(lary) ALI(lary) MAX(Af) DEP(Ao) a. s]e c[t 1 A Af Ao A max *! b. t]a B [t 1 A Ao Ao A max 1 [CGI *! * * c- t=]c 0[f 1 A Ao Ao A max 1 rcG] *! * * - d .t = ) 0 „[f 1 A A q Ao Amax [CG] * * I assume that MAX-IO(Af) and DEP-IO(Ao) are low-ranked in the hierarchy. (106b, c, d) align the codas with Ao in violation of the faithfulness constraints. Other than that, the evaluation procedure is exactly the same as the tableau in (104). Thus, (106d) emerges as the winner. As a result of this analysis, tensification turns out to be triggered by the constraint separate from neutralization, but the correlation beween the two is captured: the constraint ranking given in this section forces the [CG] in the coda to be shared by the following onset. This is a welcome result, for we are now able to attribute syllable-initial tensification to the preceding syllable-final neutalization. 35 I abbreviate the nam es o f the constraints: A L !(0]0) = AUGN(0bs]o, Ao), AolCG] = Ao-[CG], LlC(lary) = LICENSE (laryngeal node, a ) , A u(lary) = A u g n -L eft (laryngeal node, o ), MAX(Af) = M a x-IO (Af), Dep(Aq) = D ep-IO (A o ). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 §4.6.3 No tensification after nasals In Steriade’s (1991) aperture theory, the status of nasals is equivalent to stops: both of them are characterized as two aperture positions, AoAm ax. It is of interest that the obstruents following nasals undergo voicing rather than tensification. Lenis consonants become voiced between voiced segments while tense and aspirated consonants do not, as illustrated in (107). A fricative /s/ in (107c) is immune to voicing because the voiced counterpart [z] does not exist in Korean: (107) a. /sontfa/ ‘grandchild’ — » [son.cfea], *[son.tf’a] b. /sagpjog/ ‘corporal’ -» [sag.bpg], *[sag.p’ jog] c. /kuimsok/ ‘metal’ — > [kuim.sok=], *[kuim.s’ok=] I argue that this can be characterized by the constraint in (108): (108) VOICE : *VCV, *NCV where C is a voiceless lenis consonant, V a vowel, and N a nasal. If the VOICE is ranked higher than all the other constraints in the present hierarchy, then exactly the right predictions are made. Let us consider the following example: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 (109) Input: In i /tf/ /sontfa/ ‘grandchild’ -» [son.cfea] A A Ao Amax Aq Af [voice] Candidates Voice ALIGN(Obs]a, Ao) A0-[CG] a- [n]B Jtf] A A Ao Amax Ao Af 1 [voice] *! * * b- [n]0 0 [ t f ] 1 A Ao Ao Af [CG] [voice] *! c- [n]c 0[ f ] 1 A Ao Ao Af 1 1 [voice] [CG] *! * «*d. [n]c Jcfc] A A Ao Amax Ao Af [voice] * * Since VOICE is undominated in the constraint hierarchy, failing to satisfy it turns out to be fatal and renders all the other lower-ranked costraints inactive. Accordingly, ( 109d) which has a voiced onset between voiced segments is determined to be the optimal output. This captures the fact the lenis onset consonant after nasal are not tensified in an underived environment. It will be shown in the next section that the lenis consonants get tensified after nasals, which will be motivated by the morphology of Korean. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 94 §4.6.4 Tensification in compounds In this section we turn to some apparently problematic examples which the approach developed so far cannot handle. Certain complications arise when we extend the scope of our investigation to compounds: (110) Compounds a. /os + kam/ =* [ot=.k’am] ‘cloth’ /pap + kaps/ => [pap=.k’ap=] ‘meal price’ /samkak + kwankje/ => [sam.gak=.k’wan.gje] ‘love triangle’ b. /tf* 1 0 + pul/ [ ^ o .p ’ul] ‘candle light’ /pi + sori/ [pi.s’o.ri] ‘rain sound’ /k^o + tuig/ [kho.t’uig] ‘nose ridge’ /me + tol/ => [me.t’ol] ‘grinding stone’ /tfha + tf an/ => [tfha.tf’an] ‘tea cup’ /tf^o + 1 e/ [tfho.t'e] ‘candle stick’ c. /pi + os/ [pi:.ot=] ‘rain coat’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 /se + al/ = * • [se:.al] ‘bird’s egg’ /k^o + usuim/ => [kho:.u.suim] ‘sniff /p£ + ali/ [pe:.a.ri] ‘stomachache’ /£ + uluim/ => [£:.u.ruim] ‘baby cry’ d. /pi + mul/ =» [pim.mul] ‘rain drop’ /k^o + nal/ =» [k^on.nal] ‘nose spine’ In (110a, b), there seems to be a strong tendency for tensifying the syllable-initial obstruents in the second element of compounds even though the first element ends with a vowel (i.e., there is no source of tensification), as in (110b). In (1 lOd), gemination takes place when nasals are preceded by a vowel. If there is a sequence of vowels separated by a morpheme boundary as in (110c), either option is evidently unavailable. §4.6.4.1 Sub-compounds Compounding involves two free morphemes or words. When the two elements of compounding are in a relation of subordination, we call it su b -co m p o u n d in g . And if they are in a relation of coordination,3 6 we call it co -co m p o u n d in g , as shown in (111): 36 I am indebted to Bernard Com rie for the distinction between subordination and coordination. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 (111) a. Sub-compounding b. Co-compounding [[[o t]x ] x " * ■ [ P lx lx (M x " * ■ t P lx lx where a, ( 3 are stems, and X is a lexical category. In Korean, the compounds showing tensification effects are sub-compounds (111a); that is, their parts stand in a modifier-head relation. Here, I propose that an adnominal suffix be represented as a mora dominating [CG]3 7 and that it be responsible for the tensification in the sub-compounds (112). Such a conception of an adnominal suffix would afford a straightforward explanation for the behavior of the sub-compounds with respect to tensification: (112) [[[tfh o ] N (j) ] N + [p u l]N ]N = > [ t f h o . p ’u l] ‘candle’ I T ‘light’ ‘candle light’ I [CG] (adnominal suffix) From the above, it is clear that the so-called ‘neutralization’ is not involved in the tensification of sub-compounds. However, if we postulate an adnominal suffix as in 37 In an earlier stage of Korean, some compounds had a genitive suffix “ s’. The adnominal suffix I’m proposing corresponds reasonably to that historical ‘s’. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 97 (112), the array of facts in (1 10a, b) is explained. Let us consider the following tableau which evaluates candidates built on the input /tf^o + pul/: (113) Input: jj. p. |l l I I I /tfh o + p u 1/ ir n I [CG] Candidates MaX-IOQj.) Dep-IO(jj/N u) a. tfho; pU i *! b. tfho.p’ul a a b. a a A A A A / n n / M - M / M - M - / M - M - \ / / 1 1 / 1 \ / 1 1 o: p u 1 y o p’ u 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 98 The constraint ranking motivated in §2.5 is responsible for deciding (113b) as the optimal output. (113a) does not violate any constraint while (113a) crucially violates the Dep- IO(ji/Nu) by lengthening the first vowel /o/. I assume that this set of constraints is dominant over that in §2.5: ( 114) Ma x-IO(|X) » DEP-IO(ji/Nu) » .... » UNRELEASE constraints » LlCENSE(laryngeal node) » ALIGN-L(laryngeal node, a) From the above, it is clear that the tensification in (113) is not related to the neutalization discussed in Chapter 3. Vowel lengthening occurs when there is a sequence of vowels across two morphemes. Let us consider the input /pi + os/ ‘rain + coat’. To see how the constraints interact to ensure [pi:.ot=] as the correct output, consider the interaction of MAX-IO(p) and DEP- IO((i/Nu): (115) Input: p. pp. I I I /pi + os/ 1? £ I [CG] Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 99 Candidates Max-IO(ji) DEP-IO(|i/Nu) a. pi.ot= *! « * ■ b. pi:.ot= * a. c a b. c t a A l\ A K / n M - / M - M - I1 I1 / 1 1 I / \ / 1 1 P i 0 t= P i: 0 t= (115b) must be preferred to (115a) due to the given ranking. Note that GEN does not produce the following configuration since extra-heavy syllables are not allowed in Korean: (116) < 7 /I / n / ( V jj. |i p. (extra-heavy) \ / I o: t= The comparison of (115a) and (115b) supports the argument I made in §2.5 for ranking MAX-IO(H) above Dep-IO(M YN u) since sayisfying DEP-IO(M-/Nu) in (U5a) compels a violation of MaX-IO (M -). Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 100 The gemination of the input nasal consonants in (llOd) is also predicted by inserting the adnominal suffix. The following tableau explains why gemination occurs instead of vowel lengthening: (117) Input: (j. p. p . I I I /p i + m u 1 / It [CG] Candidates Max-IO(jx ) DEP-IO(|i/Nu) a. pi:.mul *! « ■ b. pim.mul a. a c b. c t a A A A A / / M - M - / H M - / i j. |i / \ / / 1 1 / 1 \ / 1 1 P i: m u 1 P i m u 1 Matters are slightly complicated when we consider the cases in which a syllable-final nasal is followed by a syllable-initial lenis obstruent, with the latter undergoing Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 101 tensification. This is a surprising result in that the onset consonant does not get voiced instead of tensified, as opposed to the data in §4.6.3. But if we posit the adnominal suffix, then the fact that the the first consonant of the second morpheme in sub-compounds in question undergoes tensification should come as no surprise: (118) /san + kil/ [san.k’il] ‘mountain path’ /pom + pal am/ /pam + tfam/ /palam + sori/ [pom.p’a.ram] ‘spring breeze’ [pam.tf’am] ‘night sleep’ [pa.ram.s’o.ri] ‘sound of wind’ Consider one of the examples in (118): (119) Input: p.(i pp. /san + k i 1 / ‘mountain’ + ‘path’ rr M - I [CG] Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 102 Given the input, ( 120a) is one of the ill-formed structures which GEN does not produce: ( 120) a. G (J b. C T C T A A A A / PP H / H H / H H / M / I I \ / I I / II / I I s a n k’ i 1 s a n k’ i 1 I I [CG] [CG] The mora in the adnominal suffix cannot be parsed appropriately. It must be parsed as a part of coda or onset neither of which is allowed in Korean. (120b) will be the only plausible syllabification of the input and thus the optimal output. §4.6.4.2 Co-compounds Tensification is not observed in the configuration of co-compounding. Given the view that I have taken of adnominal suffix, the reason why the suffix ‘X’ is not involved in co compounds in (121) is obvious: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 103 (121) Lack of tensification in co-compounds (expanded on Sohn, 1986: 231) a. [ [pom]N + [kauil]N ]N [ponxga. uil] ‘spring’ ‘fall’ ‘spring and fall’ b. [ [non]N + [pathk ]n => [non.bat=] ‘rice field’ ‘field’ ‘a plot of field’ c. [ [nun]N + [p H n In [nun.bi] ‘snow’ ‘rain’ ‘rain and snow’ d. [ [khog]N + [pap]N In => [khoQ.bap=] ‘bean’ f c , * 9 nee ‘boiled mixture of rice and bean’ e. [ [pori]N + [pap]N ]n [po.ri.bap=] ‘barley’ ‘rice’ ‘boiled mixture of rice and barley’ The criterion that serves to distinguish between sub-compounds and co-compounds is generally assumed to be based on the semantic interpretation of the compound. Thus, if the compound carries a modifier-head relation, then it is a sub-compound, and otherwise, it is regarded as a co-compound. Therfore, co-compounding does not involve the adnominal suffix. All the examples in (121) are co-compounds, so the lack of tensification is expected. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 104 §4.7 Conclusion The correlation between neutralization and tensification in Korean has never received an insightful treatment within the framework of non-linear autosegmental phonology (Sohn, 1987; Kim, 1987), as well as linear framework (Kim-Renaud, 1974, 1986): they attempt to formalize the neutralization and tensification rules separately. However, from the fact that syllable-initial obstruents undergo tensification only when preceded by syllable-final unreleased (i.e., neutralized) consonants, it seems natural to suppose that there should be a close correlation between syllable-final neutralization and syllable-initial tensification. I argued in this chapter that we should be able to get a clear grasp of the correlation between those two separate phenomena from the interaction of constraints in the sense of McCarthy & Prince (1995). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 Chapter 5: Place Assimilation §5.0 Introduction The goal of this chapter is to explore the nature of place assimilation in Korean and the general representation of place assimilations within OT. As will be substantiated in §5.2, many languages exhibit effects of place assimilation. One thing to note, among others, is the fact that the consonants that undergo place assimilation in relevant languages are typically coronal consonants. In a derivational model, the coronality of the undergoers must be specified as a part of the grammar of individual languages. However, the grammar of each language must reflect the fact that the [coronal] specification of the undergoer is a natural pattern. Previous phonological theories (e.g., Cho, 1990; Kim 1987; Sohn, 1987), including Radical Underspecification (Archangeli, 1984)3 8 have no provision for expressing this generalization. In OT, this cross-linguistic aspect of the behavior of the coronal consonants does not have to be stipulated in the individual languages. I argue that the Correspondence Theory of McCarthy & Prince (1995) provides the proper means of representing this common core in place assimilation, which is explained through the interaction of a licensing constraint of place features and faithfulness constraints. I begin with the Korean paradigm and develop a theory of place assimlation in general. Then, I show that the theory has the advantage of handling the variation among the place assimilation effects of various languages. For instance, in Korean, labials as well as 3 8 The most unmarked status of coronals in Radical Underspecification is problematic (McCarthy & Taub, 1992; Kaun, 1993; Jun, 1995, among others). They suggest that the special behavior of coronals require Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 106 coronals can be the target of place assimilation, but in English, only coronals can be targeted. As for the trigger places in Korean, labials trigger place assimilation of coronals while velars trigger that of coronals and labials (§5.5.1). In English, both velars and labials can trigger place assimilation of coronals,3 9 but velars do not trigger place assimilation of labials (§5.5.2). §5.1 The paradigm Korean has 19 consonants in its inventory. There are four points of articulation: bilabial, alveolar, palatal, and velar. In the discussion of place assimilation, I use the term coronal for the purpose of referring to both alveolar and palatal:4 0 In Korean, a coronal obstruent,4 1 whether oral or nasal, optionally4 2 assimilates in point of articulation to the following consonant (122a, b, c, d, e). Labials also optionally assimilate in point of articulation to a following velar consonant (122f): actual specification of [coronal] in underlying representation to support dependent features and to participate in assimilation. 391 am indebted to Bernard Comrie for the following English data: Careful speech Casual speech ‘in-’ + ‘possible’ -» [in.pa.sa.bal] or [im .pa.sa.bal] ‘in’ + ‘Paris’ -» [in.pae.ris] or [im.pae.ris] ‘in’ + ‘Canada’ -» [in.kae.na.da] or [iQ.kae.na.da] cf. He is studying at Cornell. [tet.ko:.nel] or [aek.ko:.nel] He is studying at Purdue, [aet.par.dju] or [aep.par.dju] Also, at least in moderately slow speech, many English speakers have contrast between [koQ.gres] ‘congress’ and [kan.gre.Ja.nal] ‘congressional’, depending on the stress. Korean pattern is not subject to such a variation. 4 0 In fact, alveolar as well as palatal consonants are executed by a coronal articulator (i.e., tongue tip or blade). 4 1 A term used in the distinctive feature approach of SPE to refer to plosives, fricatives, and affricates. 4 2 Place assimilation typically occurs in a colloquial speech style. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 (122) a. alveolar -» labial (i) /nunmul/ -» [num.mul] /sinpal/ [sJim.bal] (»i) /kispal/ — » /os + man/ -» b. alveolar — > (i) /kunki/ — > /pat + ko/ -» (ii) /oskam/ — > /kuskori/ — » c. palatal -> /matf + ta/ /tfotf + so/ [kip=.p’al] [om.man] velar [kug.gi] [pak=.k’o] [ok=.k’am] [kuk=.k’ori] alveolar -» [mat=.t’a] [tfot=.s’o] ‘tears’ ‘shoes’ ‘flag’ ‘clothes only’ ‘military discipline’ ‘to receive an d ...’ ‘texture’ ‘tunes for an exorcism’ ‘to be right’ ‘a dairy cow’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 108 d. palatal —> labial /naif + pota/ — > [nap-p’o.ta] ‘to despise’ /k’oif’ + mal/ -» [k’om.mal] ‘flower name’ e. palatal -» velar /k’oif’ + kwa/ — » [k’ok-.k’wa] ‘flower and...’ /k’otf’ + katf U -» [k’ok-.k’a.cfei] ‘a spray of flowers’ f. labial — > velar /kamki/ — » [kag.ki] ‘a cold’ /op + ko/ -> [ok-.k’o] ‘to take (a baby) on one’s back and ...’ However, there are a number of cases where place assmilation does not occur. No change of articulation occurs if the following segment is a coronal, as shown in (123a, b, c, d). Velars and liquid ‘1 ’ do not change their points of articulation at all, no matter what consonant follows (123e, f, g, h, i, j): (123) a. *labial —> alveolar /pap + to/ -» [pap=\t'o], *[pat=.t'o] ‘rice also’ /nop + ta/ —> [nop=.t’a], *[not=.t’a] ‘high’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 b. *velar — » alveolar /tfaktu/ -» [gak-.t’u], *[tfat=.t’u] /kakso/ -> [kak=.s’D ], *[kat=.s’o] c. *alveolar — > palatal /os + tfarak/ — > [ot=.tf’a.rak=] /us + tfa/ -» [ut= tf’a] d. *labial palatal /siptfaka/ -> [sjip=.tf’a.ka] /kamtfa/ -» [kam.cfea] e. *velar -4 alveolar /kakto/ — > [kak=.t’o] /kagnam/ — > [kag.nam] f. *velar — > palatal /kogtfu/ -> [kog.cfcu] /kaktfa/ -» [kak=.tf’a] g. *velar -» labial /kukmul/ — > [kug.mul], *[kum.mul] ‘a straw cutter’ ‘a memorandum’ ‘the skirt’ ‘Let’s laugh!’ ‘a cross’ ‘potato’ ‘angle’ ‘the south of a river’ ‘a princess’ ‘individually’ ‘soup’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 /kogpu/ — » [kog.pu], *[kom.pu] ‘study’ h. *alveolar — y labial /kalpi/ — » [kal.bi] ‘a rib’ /palmjog/ [pal.mjog] ‘invention’ i. *alveolar -» velar /tfulki/ — > [^ul.gi] ‘a stalk’ /sal + ki/ — > [sal.gi] ‘living’ j. *alveo!ar — » palatal /kalfhj/ — > [kal.bi] ‘a hairtail’ /kaltfuig/ [kal.^’uxg] ‘thirst’ The question then arises as to why only certain types of place assimilation are found in the language. In this chapter, I aim to answer this question by showing how the varied assimilatory effects are derived from the theory of constraint interaction (Prince & Smolensky, 1993; McCarthy & Prince, 1993, 1995). §5.2 The generalizations Place assimilation in natural languages exhibits considerable amount of variability in (i) the consonants that undergo assimilation, (ii) the consonants that trigger assimilation, and Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. i l l (iii) the domains where assimilation applies. However, this variability is not unconstrained. It is reported in the literature (Bailey, 1970; Kiparsky, 1985; Cho, 1990; Jun, 1995) that coronals are favorite target of place assimilation cross-linguistically. As concrete illustrations of this generalization, §5.2.1 and §5.2.2 spell out the cross-linguistic observations about the target and trigger place, which a theory of place assimilation must be able to explain. §5.2.1 Target place As illustrated by the survey in (124), it is observed that coronals typically undergo place assimilation, but there are no instances where only the noncoronals undergo assimilation. Also, it is of interest that Korean is the only language where labials are more likely to undergo place assimilation than velars: (124) Place of articulation of the target in place assimilation (expanded on Jun, 1995: 82) (O = ‘targeted’, X = ‘untargeted’, blank = ‘undetermined’) coronal labial velar Brussels Flemish 0 X Catalan 0 X X Chamorro 0 Diola Fogny 0 0 0 English 0 X X German 0 X X Hindi 0 0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 112 (O = ‘targeted’, X = ‘untargeted’, blank = ‘undeter coronal labial vela Inuktitut 0 0 0 Irish 0 Japanese o 0 0 Keley-I o X Korean 0 0 X Libyan Arabic o Lithuanian 0 X Malay 0 0 o Malayalam o 0 Nchufie 0 0 0 Pangasinan o Syrian Arabic 0 Toba Batak 0 X X Yakut 0 X X Yoruba 0 0 o Zoque 0 Based on the table above, the traditional rule-based approach would have to make a stipulation that the undergoer of place assimilation must be coronal consonants, and include such a statement in the grammar of each language.43 However, an adequate theory 4 3 Again, I assume that the Radical Underspecification view, which treats coronals as placeless is problematic . Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 113 of phonology must account for the fact that the coronality of the undergoer in place assimilation is a more natural pattern than the labiality or velarity of the undergoer. §5.2.2 Trigger place According to Jun (1995:83), the trigger place does not show any consistent patterns across the surveyed 17 languages.44 He reports that Korean is the only case which shows an asymmetry:45 labials do not assimilate to the following coronals (e.g., /pap + to/ -» [pap=.t’o], *[pat=.t’o] ‘rice also’) but they do to the following velars (e.g., /pap + kap/ -» [pak=.k’ap] ‘meal price’). Based on this observation, he assumes the following statement as a universal implication on place assimilation in the absence of counterexamples: (125) Trigger place (after Jun, 1995:91) : If coronals trigger place assimilation, so do velars. Mohanan (1993:76) makes a similar universal generalization: (126) If labials trigger assimilation, so do velars. While there are instances where only noncoronals trigger place assimilation (e.g., English, Korean, Libyan Arabic, etc.), there are none where only coronals are the triggers. 4 4 They are Brussels Flemish, Catalan, Diola Fogny, English, German. Hindi, Japanese, Keley-I, Korean, Lithuanian, Malay, Malayalam, Nchufie, Toba Batak, Yakut, Yoruba, and Zoque. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 114 For example, it is reported in Hamp (1985:391) that in Arctic Quebec, Greenland, and Labrador,46 velars assimilate to coronals (cf. tuttu < tuktu ‘caribou’). But in these languages also, there are cases where coronals assimilate in place to the following noncoronal (cf. tikimmat < tikinman ‘because he/she arrives’). This observation is in line with Jun’s generalization in (125), on which I will base my analysis. I claim that the Korean case should be viewed as exceptional to the general pattern of place assimilation. In what follows, I assume that there is just a coronal vs. noncoronal asymmetry in triggering the place assimilation in general; i.e., if coronals trigger place assimilation, so do noncoronals. But I will show in §5.5 that the apparently exceptional case of Korean can be explained through the interaction of the constraints that are universally motivated. §5.3 Theoretical problems in place assimilation In this section, I outline the basic facts that a theory of place assimilation must be able to account for. I discuss some theoretical problems that the traditional rule-based approach could not explain. By and large they revolve around three issues: formal simplicity and naturalness (§5.3.1), cross-linguistic variation (§5.3.2) and direction of place assimilation (§5.3.3). 4 5 Jun( 1995:91) defines an asymmetric pattern as follows: C,Cb — »C„Cb but C .Q . ->CaCc, where a, b, and c are indices representing different articulators. 4 6 Arctic Quebec, Greenland, and Labrador are languages of Canadian Eskimo (or Inuktitut). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 §5.3.1 Formal simplicity and naturalness Adopting a non-linear feature representation (Clements, 1985; Sagey, 1986), we obtain the following derivations with a coronal consonant being a target of place assimilation: (127) a. /kunki/ — > [kug.gi] Rio ‘military discipline’ Place Place cor vel b. /tfaktu/ — » [tfak=.t’u], *[tfat~.t’u] ‘a straw cutter’ * R lc 0 [ R Place Place vel cor In general, (127a) is a common process of coronals assimilating to velars in place feature, while (127b) with velars assimilating to coronals is rarely attested (cf. §5.2.2). If we assume a fully specified feature representation as in (127), it is not clear why only velarity should spread rather than other place features. However, a formulation like (127) does not Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 116 predict the unmarked direction of a phonological change: a comparison between the two representations (127a, b) predicts that they are equal in phonological naturalness. Therefore, the representations in question fail to capture the generalization that (127a) is more frequent and natural than (127b). In derivational approaches, it has to be stipulated that velars assimilate in place to the preceding coronals but the reverse is not the case. A separate stipulation would be needed here to handle this asymmetry in place assimilation observed across languages. There is considerable consensus among the theorists of underspecification (Archangeli, 1984; ICiparsky, 1985; Paradis & Prunet, 1991; Yip, 1991) that only one value of a feature, the unpredictable one, is present in underlying representation and the predictable value is filled in by a redundancy rule during derivation or in the interpretive component. In such a system, assimilation is predicted to affect only segments lacking values for the spreading feature. For example, Paradis & Prunet (1991:6) argue that coronal articulator is the most unmarked (i.e., predictable) one. In other words, labials have a labial articulator and velars have a dorsal articulator, but coronals have no coronal articulator in underlying representation, as illustrated in (128a). The redundancy rule in (128b) fills in the default place feature of coronals at a later stage: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 (128) a. Labials Velars RN RN I I SLN SLN I I PN PN I I Labial Dorsal (RN: Root Node, SLN: Supralaryngeal Node, PN: Place Node) b. [OPIace] — » Coronal In light of the preceding, we expect the derivation in (127a) to be re-analyzed as shown in (129a), which captures the generalization that coronals stand apart from labials and velars in that they are typically the target of place assimilation: if we assume that coronals are underspecified for the place feature, the assimilation rule will be easily explicable: -» [kug.gi] ‘military discipline’ I Place - J vel (129) a. /kunki/ R]a Place Coronals RN ! SLN Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 b. /tfaktu/ — > [tfak=.t’u], *[tfak=.k’u] ‘a straw cutter’ * Rla otR Place Place vel However, this approach predicts that coronals in onset can assimilate to the preceding velars as shown in (129b). To prevent this, another stipulation has to be made to the effect that only syllable onset is capable of triggering place assimilation. But even this stipulation could not solve the problem if both trigger and target are part of a syllable onset or coda. The expression of place assimilation in terms of rules and representation, once again, fails to mirror the naturalness of the recurrent patterns. §5.3.2 Cross-linguistic variation Cho (1990:96) argues that assimilation is best characterized as the spreading of marked features to the less-specified slot. She assumes that assimilation is due to the specifications on the target and the trigger. For example, Korean allows assimilation if one segment is more marked than the other, where ‘more marked’ is defined as a subset relation. She claims that velars are more marked than labials or palatals in terms of feature specification which determines the direction of assimilation: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 119 (130)/kamki/ -» [kag.ki] ‘a cold’ • • Place [-cor] [-cor, -ant] According to her formulation in (130), /m/ assimilates to the following DsJ in place because the target specification ([-cor]) is a subset of the trigger specifiction ([-cor, -ant]). Labials do not assimilate to palatals or vice versa. The following is the configuration o f the labial- palatal sequence: (131) Labial Palatal • • Place I I [-cor] [-ant] No subset relation holds in (131). The two segments are equal in terms of markedness and therefore assimilation does not occur. She lists all the parametric conditions for Korean place assimilation as shown below: (132) Parameters of Korean place assimilation (after Cho, 1990:97) a. Site: place node b. Target/Trigger Specification: target c trigger c. Locality Condition: skeletal adjacency Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 120 d. Rule Order: spread before all default rules e. Directionality: right to left Place assimilation occurs widely in natural languages, and therefore must be considered natural in linguistic theories. In rule-based systems, all patterns are stipulated by rules. In (132), Cho attempts to cut down on the stipulations, by taking advantage of the subset relation of phonological features. Her analysis is descriptively adequate, but it is not explanatorily adequate since it stipulates patterns that are widely attested in natural languages, and thus obscuring the naturalness. For example, her model predicts that the /n/ assimilates to the following /k/ due to the subset relation of feature specification: zero specification of /n/ is a subset of [-cor, -ant] of /k/, assuming that /t/ is the maximally unspecified segment: (133)/kunki/ -» [kug.ki] ‘military discipline’ • • Place [-cor, -ant] If the change from /nk/ to [0 k] across syllable in (133) is the result of a language-particular constraint or rule, we cannot explain why most surveyed languages in §5.2 do not allow an unassimilated sequence like [nk] across syllables. To capture this naturalness, we must therefore assume that the tendency to prohibit a coronal nasal followed by a noncoronal stop is somehow derived from Universal Grammar. I claim in §5.4 that the trigger of place Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 121 assimilation must be of universal nature: I will argue that a licensing constraint of place features plays a role in triggering place assimilation. §5.3.3 Direction of place assimilation There is a cross-linguistic generalization (Webb, 1983; Ohala, 1990; Jun, 1995) that syllable onsets are more likely trigger of place assimilation than codas, while codas are more likely target than onsets. As shown in the following schematic representation, the instances like (134a) with coda assimilating to onset are widely attested while those like (134b) with onset assimilating to coda are not:4 7 (134) a. Anticipatory assimilation R lc atR Place Place cor vel b. Progressive assimilation R]« oC R Place Place vel cor The common occurrence of anticipatory assimilation in (134a) may reflect the relative strength of syllable positions: since onset is phonologically stronger than coda,4 8 onset is 4 7 Jun (1995:85-86) reports that Kambata and Musey are the only two languages that have progressive place assimilation. 4 8 Many phonologists (especially Vennemann, 1972; Hooper. 1976) have argued that syllable-initial position is universally stronger than syllable-final position, and that, accordingly, consonant strengthening Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 122 typically a trigger of place assimilation. A rule-based approach would have to state this directionality repeatedly in the grammar of each language, as can be seen in (132e). I propose in the next section that the directionality of place assimilation be a reflex of an underlying system of constraints. §5.4 The proposal All the patterns of place assimilation discussed in this chapter can be captured in a derivational model. In such a model, each assimilatory process can be expressed through a different autosegmental rule. The blocking of spread can be expressed either by adding conditions to the rule or the representation. However, what any such model misses is a clear generalization that all of those processes have the effect of avoiding a configuration in which two consonantal place nodes are adjacent across a syllable boundary, as illustrated in (135): (135) a. a a / K R R R | R, R R I I PL, PL, process takes place in syllable-initial position, while weakening process typically occurs in syllable-final position. b. c / 1 \ R R R, R2 R R PLo (PL: place node) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 123 In an early derivational model since SPE, it would be a coincidence that across languages there are changes that have the effect of modifying the structure of the form (135a) into the structure that does not have adjacent consonantal place nodes across syllable boundary (135b). Thus, it fails to capture the fact that the structure in (135b) is widely attested across languages. In a later development of derivational models, the same effect as in (135) can be attributed to a derivational output constraint in the sense of Kisseberth (1970). According to this model, the application of a phonlogical rule is blocked if it would create a violation of the (arguably universal) constraint. If we posit a licensing constraint to the effect that codas cannot license place features as in (135a), it predicts the effect of place assimiltion (135b). But such a model still cannot explain why there are abundant surface exceptions to the constraint: e.g., there are consonant clusters that do not undergo place assimilation, such consonant clusters of Korean as labial-coronal, velar-coronal, velar-labial, etc. One could conclude that the constraint is not universal. Then, it is difficult to capture the generalization that the configuration in (135b) is favored cross-linguistically. §5.4.1 Licensing place features Since Ito (1986), a negative condition on coda has been considered in the literature (Goldsmith, 1990; Lombardi, 1991, among others) for the purpose of ruling out a particular configuration syllable- finally. One version is given below in (136): all consonantal place is ruled out syllable-fmally (or remains unlicensed): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 124 (136) Coda Condition (Ito 1989: 224) * C]c 1 Place However, the Coda Condition does not apply to the languages which allow consonant clusters across syllables without place assimilation (English, Korean, etc). In order to account for the facts of those languages, I propose to move away from a negative condition to a licensing constraint in which licensing is defined as exhaustive domination: (137) LICENSE (consonantal place) : Coda must not license place features. The LICENSE (consonantal place) has the effect of prohibiting (138a) and (138b), but allowing (138c) and (I38d): (138) Input: R R (R = [+cons, -son]) I I aPlace pPiace U LICENSE (consonantal place) Applied Output: a. *Coda Onset b. *Coda Onset I I I I aPlace PPiace yPlace PPiace c. Coda Onset d. Coda Onset aPlace PPiace Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 125 (138a) is faithful to the input: it violates the LICENSE with the coda solely dominating the underlying aPlace. (138b) is also in violation of the LICENSE due to the default insertion of "/Place after a complete loss of the original aPlace. (138c) and (138d) satisfy the constraint because the place features are doubly-linked: onsets license place features. In addition, such a configuration as ( 139a) without any place feature in coda will be prohibited owing to the alignment constraint independendy motivated in Chapter 2: it prevents a laryngeal node from occuring in coda by itself, as repeated in ( 139b): (139) a. *Coda Onset Lary. pPlace b. Laryngeal neutralization ALIGN-LEFT (laryngeal node, a): Every laryngeal node stands at the left edge of a syllable. In (139a), the deletion of aPlace will leave a laryngeal node behind, which will be prohibited by the alignment constraint. My claim is that place assimilation is a strategy of avoiding the configurations in (138a, b). I argue that OT allows us to capture the various effects of place assimilation through the ranking of the LICENSE (consonantal place) with a set of faithfulness constraints. In OT, the LICENSE is not inviolable, which is how the cases of failed assimilations will be accounted for: they violate the LICENSE due to higher ranking constraints. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 126 As to the syllable onset clusters, they do not undergo place assimilation. I propose that the onsets be protected by certain constraint which ensures that the coda, not the onset loses the place feature. To express this insight in terms of a constrint, I propose a markedness constraint below, which will account for why syllable onsets are not subject to place assimilation: (140) *ONSET(place) * Onset I Root, Placej I assume that every underlying place node is coindexed with its mother node, i.e., root node. Then, the *ONSET(place) assigns ‘*’ when an onset root node dominates a place node with a different index (i.e., the root has ‘i \ but the place has ‘ j ’) as in (140). The following diagram explains how the constraint can be interpreted. Crucial candidates are illustrated below with their underlying forms: (141) Constraint (140) Applied a. b. c. Underlying Root, Root, Root, Root, Root, Root, II II II Place, Place, Place, Place, Place, Place, U li U Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 127 Onset ♦Onset ♦Onset Surface Candidates Root. Root. Root, Root, Root, Root, Place, Place. Place Place In case (141a), each of the two roots is coindexed with its own place both at surface and underlying representation. Therefore, the constraint *ONSET(place) is not violated. Cases (141b) and (141c) characterize a situation of place assimilation, and are exactly what the constraint *ONSET(place) looks for. Root, dominates Place, in (141b), and Root, dominates Place, in (141c). Those two cases violate the constraint in question since they have different indices. In addition, the ♦ONSET(place) provides an explanation for why the syllable onset is typically the trigger of place assimilation rather than syllable coda. Its effect is to allow structures like (142a, b) and forbid such surface structures as (142c):4 9 " w Jun (1995:151) claims that in a sequence VC,C,V, C,(i.e., coda) is acoustically weaker than C,(i.e., onset). The acoustic strength of onsets over codas leads him to the following universal ranking in the constraints which formulate the preservation of place cues for consonants in different syllable positions: (i) Universal ranking for positions: Pres(pl(onset)) » Pres(pl(coda)) This ranking indicates that place cues for onsets must be preserved in preferece to those for codas, which indirectly captures the generalization that the codas are more likely target of place assimilation than onsets. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 128 (142) The effect of *ONSET(pIace) a. b. c On Nu Co On Nu Co On Nu Co Root, Root; Root, PL: PL: c. * c On Nu Co On Nu Co Root; Root: j i PL. In (142a), the root and the onset place are coindexed, so it does not violate the *ONSET(pIace). The same explanation can be given for (142b). One thing to note here is that Rootj and PL, have different indices, but *ONSET(place) is not violated since Rootj is in coda. (142c) is the very configuration the constraint prohibits: the onset Root; dominates PLj- After all, if the *ONSET(place) is undominated in the constraint hierarchy, place assimilation will never occur within syllable onset. Besides, the generalization is captured Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 129 that syllable coda is typically assimilated in place to the following onset. These effects will be shown in detail in §5.5. Another fact requiring explanation is that the anticipatory place assimilation occurs within complex codas as well. I will discuss this issue in §5.5.4. §5.4.2 Faithfulness constraints and the LICENSE (consonantal place) From the OT perspective, deletion of a segment means that some constraint dominates MAX-IO(segment), a member of the MAX constraint family, as illustrated in ( 143a): (143) MAX-IO(segment): Every segment in the input has a correspondent in the output. Given this, we can account for the deletion of a place feature in the target of place assimilation by ranking the LICENSE over MAX-IO(place node) as shown in (144): (144) LICENSE » MAX-IO(place node) One fundamental idea of OT is that a higher-ranked constraint forces a violation of a lower-ranked constraint: when the dominating constraint, MAX-IO(vel), is not relevant, the LICENSE comes into play and triggers place assimilation. Let us consider again the following situation: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 130 (145) The role of the LICENSE (adaped from (127)) a. /kunki/ —> [kug.gi] ‘military discipline’ \ l PL - LICENSE is satisfied. - vel b. /tfaktu/ -» [tfak-.t’u], ‘a straw cutter’ PL PL - L i c e n s e is violated. - vel cor In ( 145a), the LICENSE forces the violation of MAX-IO(cor) since the former dominates the latter, which results in place assimilation. In ( 145b), owing to the ranking ‘MAX-IO(vel) » LICENSE’, the dominance of MAX-IO(vel) over LICENSE forces the violation of the latter: therefore, the LICENSE is violated and two consonantal place features are preserved. We can see the effect of this ranking in the following tableau, which illustrates how the candidate with one place feature deleted is chosen by the constraint hierarchy: (146) Input: /[+cons] [+cons]/ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Place Place 131 Candidates License MAX-IO(place node) a. [+cons] [+cons] 1 1 Place Place *! b. [+cons] [+cons] Place * Korean provides evidence that the LICENSE is violable, since there are abundant surface exceptions to it. One example is the velars that do not undergo place assimilation in the coda (123e, f, g). This implies that the LICENSE is violable and can be ranked below the faithfulness constraints: (147) MAX-IO(vel) » License Even if the LICENSE is ranked low, it can still have an effect. In Korean, coronal and labial consonants are typically the target of place assimilation (see (122)). The following constraint ranking forces place assimilation: (148) License » MAX-lO(lab) »M AX-lO(cor) It is then highly significant that even the dominated constraint, LICENSE in (147), may be active as in (148) when MAX-IO(vel) is not relevant. In this regard, OT is sharply distinguished from the derivational approaches. Any model based on parameters or rules sees any linguistic principle in all-or-nothing terms. From (147) and (148), we obtain the following ranking by transitivity: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 132 (149) MAX-IO(vel) » License »MAX-IO(lab) » MAX-IO(cor)50 Each of the faithfulness constraints in (149) assesses a violation for each input element that does not have an output counterpart: the hierarchy of those constraints means that velars are less likely targets than labials, which in turn are less likely targets than coronals. For our purposes, this ranking expresses the cross-linguistic generalization in §5.2.1 that coronals and labials are more likely target of place assimilation than velars. §5.4.3 Alignment constraints The result of place assimilation is a linked structure, as shown below: 5 0 Jun (1995:150) proposes the following ranking for constraints which preserve the perceptual cues for the place of unreleased coronals, labials, and velars. Universal ranking for target places : Pres(pl(dor"')) » Pres(pl(lab")) » Pres(pl(coD) His proposal is based on the observation that velars are acoustically stronger than labials, which are in turn stronger than coronals: e.g., the duration of dg is more overlapped than that of gd because the cues for the unreleased d are so weak that there is little motivation for the speaker to preserve them (Kuehn and Moll 1976). This ranking indicates that place cues of unreleased velars must be preserved in preference to those of unreleased labials, which are in turn preserved in preference to those of unreleased coronals. It indirectly captures the generalization that (i) if velars are a target of place assimilation, so are the labials and (ii) if labials are a target of place assimilation, so are the coronals. After all, he proposes a universal schema. But as shown in §5.2.1, there are languages that allow velars to be the target of place assimilation (e.g., Chamorro, Inuktitut (cf. tuttu < tuktu ‘caribou’), Japanese, Malay, Nchufie, and Yoruba), so the hierarchy should be subjet to crosslinguistic variation. His observation is captured by the hierarchy of faithfulness constraints that I propose: MAX-IO(vel) » MAX(lab) » MAX-IO(cor). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 133 (150) a. a / i \ R R R, I PL, I assume that there is a constraint in Universal Grammar that militates against the linked structure.51 The constraint is in conflict with the LICENSE which forces a linked structure, and the resolution of the conflict is determined by the hierarchical ranking of the constraints. For example, in case the constraint against the linked structure is ranked above the LICENSE as in (15 la), the underlying phonological material would be preserved in the output (150a). If the ranking is reversed as in (151b), a linked structure obtains (150b): 5 1 McCarthy & Prince (1993:39) argue that in Axininca Campa, multiple place-linking results in a violation of the constraint Align-Right (stem, a), which requires right stem edges to coincide with right syllable edges. This is taken as the reason why the input/kim + aanch i/ comes out as ki.maan.ch i. instead of *kim.paan.ch i : *kimpaanc/'i, With Linking a a I / k i m : P a a n c" i i / [labial] ‘P’ is a bare root node that is epenthesized to satisfy ONSET. Assuming that GEN supplies a candidate with an assimilated Place node, McCarthy & Prince claim that this form is mis-ALIGNed, because the Place node o f the m is a part of the representation of the Stem, but is also syllabified, via P, as the onset of the second syllable. Thus while A lign requires sharply-defined constituent edges, linking undoes the desired relation. b. a R, R R PL, R R R, R, R R Place Assimilation PL, (PL: place node) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 134 (151) a. No Linked Structure L ic e n s e « ■ CandidateL (150a) * Candidate, ( 150b) *! b. L ic e n s e No Linked Structure Candidate, (150a) *! «* Candidate, (150b) * I propose that the constraint against linked structure be represented as in (152), revising the BASIC Alignm ent (BA henceforth) in the sense of Cole & Kisseberth (1993):5 2 (152) BA(PLj; Root;) : Every place node must be aligned with its underlying mother node (i.e., root node). This alignment constraint explodes into two separate components: BA(Coronal place,; Rootj) and BA(Placej; Rootj). I claim that they are ranked in such a way as to express the generalization in §5.2.2 that coronals are the more unlikely trigger of place assimilation than noncoronals: 5: According to Cole & Kisseberth (1993:3), harmony occurs when a constraint that builds a wide harmony domain (i.e.. Wide Scope Alignment) is ranked above a constraint that builds a narrow domain. In a language with no harmony, the F-domain for a phonological feature [FI is properly aligned with the underlying F-bearing anchor by a set of constraints termed Basic Alignment, given below: (i) Basic Alignment (after Cole & Kisseberth 1993:4) BA-left Align(Anchor-s, L; F-domain, L) BA-right Align(Anchor-s, R; F-domain, R) cf. ‘Anchor-s’ indicates a sponsoring anchor: an anchor sponsors [F] if it is affiliated with [F] in underlying representation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 135 (153) BA(Coronal places Root;) » BA(Place;; Root;) To recapitulate, BA(Coronal placej; Rootj), being more specific, has priority over B A(Placej; R ootj) which is a general constraint: the latter plays a role only when the former is not relevant - in other words, ‘elsewhere’. A violation of BA(Coronal place;-, Rootj) is assessed for each coronal place linking in the output. Examples illustrating its effect are given in (154): (154) a. [pap'.t’o] ‘rice also’ p a Pj PL: PL: [lab] [cor] satisfies B A (C o ro n a l p la c e s R o o t ( ) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 136 b. [patf.t’o] ‘rice also’ violates BA(Coronal p la ceR o o tt ) a ! ct pa tj ; tj o [cor] (154a) fulfills the BA(Coronal place^ Rootj): the coronal place in the onset is coindexed with its underlying mother node (i.e., tj - PLj - [cor]). But the place assimilated form (154b) violates the BA because of the additional link to the coda of the preceding syllable: the onset place node with an index ‘ j ’ is linked to tj and ts at the same time. Taking all the constraints introduced so far, the overall constraint hierarchy will be shown as in (155): (155) Hierarchy of constraints for place assimilation in Korean : *ONSET(place) » BA(Coronal place^ Rootj) » MAX-IO(vel) » LICENSE » MAX-IO(Iab) » BA^lacej; Rootj) » MAX-IO(cor) The sub-ranking ‘MAX-IO(vel) » LICENSE’ implies that the input-output identity of velars (in coda) forces the violation of LICENSE, which accounts for the cross-linguistic pattern that velars are typically not targeted in place assimilation. Labials and coronals are Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 137 targeted in Korean through the sub-ranking, LICENSE » MAX-IO(lab) » MAX-IO(cor). In addition, the same ranking forces the coronals to assimilate to the labials. In fact, ‘MAX-IO(vel) » MAX-IO(lab)» MAX-IO(cor)’ is a universal ranking for the target place. If labials are the target of place assimilation, so are coronals. There are no languages where labials undergo place assimilation but coronals do not. On the other hand, ‘BA(Coronal place^ Rootj) » BA(Placej; Rootj)’ is a universally fixed constraint ranking to obtain the universal generalization that if coronals trigger place assimilation, so do noncoronals. In other words, there are no languages in which coronals trigger place assimilation but noncoronals do not. The directionality of place assimilation from onset to coda is encoded in the undominated status of the *ONSET(place) which protects onsets from losing their underlying place features. In those languages that have progressive assimilation, *ONSET(place) can be violated. §5.5 Case studies In this section, I will present a unified analysis of a range of place assimilation cases across languages with the primary focus on Korean. The surface forms in (122) and (123) will be accounted for by the constraint ranking established in (155). Continuing to adopt the framework outlined so far, I argue that the treatment of the various patterns across languages shows the generality of the proposal: they differ minimally in the ranking of a single constraint in the same constraint hierarchy. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 138 §5.5.1 Korean The purpose of this section is to provide a complete formal analysis of place assimilation pattern in Korean. I discuss heterorganic clusters (§5.5.1.1) and homorganic clusters (§5.5.1.2). §5.5.1.1 Heterorganic clusters We observe in Korean that coronals assimilate in place to the following labials or velars. The tableau in (156) illustrates how the proposed constraint hierarchy delivers this effect: (156) Input: /sin pal/ ‘shoes’ PL PL cor lab Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 139 Candidates ♦ONSET5 3 BA(c) MAX(v) License m a x(D BA(pl) Max(c) a. n]a 3[p 1 1 PL PL 1 1 cor lab *! b- n]„ a[t P L ^ ^ P L 1 1 cor lab *! * * « ■ c.m]c D [p plT ^ p l 1 1 cor lab * * Notice that (156b) crucially violates the undominated *ONSET(place) since the onset loses the original labial place node. This leaves two candidates: the faithful (156a) and the assimilated (156c). This is the case where the LICENSE comes into play in choosing the optimal output. The LICENSE assigns a for each instance of two consonantal places on adjacent syllables, as in (156a). The effect of this constraint depends on its interaction with other constraints in the same hierarchy. We can see that the LICENSE, while low-ranking, is still active. The faithful candidate (156a) loses out owing to the LICENSE. (156c) violates BA(Noncoronal place^ Rootj) and MAX-IO(cor), but both of them are low-ranked. Therefore, ( 156c) is more harmonic than (156a) or any other competitor. Another possible candidate would be the one in which one of the two segments is completely deleted. I assume that this will be handled by a high ranking MAX-IO(Root), which is not reflected on the tableau. 5 3 Due to the limitation o f space, I abbreviate the names o f the constraints: *On s e t = *ONSET(place), BA(c) = Basic AuGNMENT(Coronal placej; Rootj), M a x(v) = MAX-IO(velar place). L ic en se = L icense Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140 The same analysis as that in (156) holds for coronals assimilated to velars, as shown in (157): (157) Input: /kun ki/ ‘military discipline’ I I PL PL cor vel Candidates ♦ONSET BA(c) Max(v) License MAX(l) BA(pl) MAX(c) a. n]„ a[k 1 1 PL PL 1 1 cor vel *! b. n]a a[t PL PL 1 1 cor vel *! * * c- „[k P L ^ P L 1 1 cor vel * * As shown earlier, labials assimilate to velars. Again, the hierarchy in (158) delivers this effect: (consonantal place), M a x(1) = MAX-IO(labial place), BA(pl) = Basic ALlGNMENT(PIacei; Rootj), M a x (c) = MAX-IO(coronal place). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 141 (158) Input: /kam ki/ ‘a cold’ PL PL lab vel Candidates *ONSET BA(c) MAX(v) License MaX(1) BA(pl) Max(c) a. m]0 D [k 1 1 PL PL 1 1 lab vel *! b. m]B B [p PL PL lab vel *! * * « ■ c. g]0 0[k PL PL 1 1 lab vel * * Now, let us consider why labials do not assimilate to coronals. The constraint hierarchy in (159) correctly predicts the facts: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 142 (159) Input: /pap + to/ ‘rice also’ I I PL PL lab cor Candidates *ONSET BA(c) MAX(v) License MAX(l) BA(pl) MAX(c) * * ■ a- p ]< j oft 1 1 PL PL 1 1 lab cor * b- p]f f «[p PL PL 1 1 lab cor *! * * c- t]B B [t PL PL 1 1 lab cor *! * One thing to note about (159) is that the LICENSE is not active. (159b) and (159c) crucially violate two high-ranking constraints forbiding a linked structure: *ONSET(place) and BA(Coronal place;; Root;), repectively. By the same token, velars followed by coronals do not undergo place assimilation. In light of (159), we can expect to find the BA(Coronal place^ Root;) active again, which is shown in (160): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 143 (160) Input: /tfak tu/ ‘a straw cutter’ I I PL PL vel cor Candidates *ONSET BA(c) MAX(v) License max(1 ) BA(pl) Ma X(c) * * ■ a- k]0 0[t 1 1 PL PL 1 1 vel cor * b- k]a c[k PL PL 1 1 vel cor *! * * c. t]„ c[t PL PL 1 1 vel cor *! * That velars do not assimilate to the following labials is expected given the ranking of MAX-IO(vel) in the constraint hierarchy in question. This is shown (161): (161) Input: /kog pu/ ‘study’ I I PL PL I I vel lab Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 144 Candidates ♦Onset BA(c) MAx(v) License Max(1 ) BA(pl) Max(c) ® a .g ]0 B [p 1 1 PL PL 1 1 vel lab * b. g]« „[k PL PL 1 1 vel lab *! * * c- m]c „[p PL PL 1 1 vel lab *! * As can be seen in (161), MAX-IO(vel) plays a crucial role in deciding between [kog.pu] (161a) and *[kom.pu] (161c). Under present constraint hierarchy, other things being equal, it is more harmonic to leave the cluster faithful to the input (161a) than to have it assimilated (161c). As is shown in (161c), *[kom.pu] would induce the deletion of velar place, violating MAX-IO(vel). Therefore, (161a) is determined to be the optimal output. From these simple cases, we turn now to more complex ones. Palatal affricates assimilate to the following labials or velars. I argue that the alignment constraint proposed in Chapter 3 is responsible for the loss of release feature (i.e., neutralization) and place assimilation: (162) ALIGN(Obs]0 , Ao): Every obstruent in coda must be aligned with Ao- If the alignment constraint is ranked above *ONSET and dominates the faithfulness constraints of the aperture positions as shown in (163), we can correctly predict the palatals assimilated to the following consonant: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 145 (163) ALIGN(0bs]o, Ao) » MAX-IO(Af), DEF-IO(Ao) As the following tableau indicates, the optimal output must suffer from the loss of a release feature in coda position: (164) Input: /k’otf’ + kwa/ -» [k’ok=.k’wa] ‘flower and I I AoAf Aq A m ax I I PL PL I I cor vel Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 146 Candidate A L I(A o ) ♦ O n s e t L i c e n s e MAX-IO(Af) D E P - I O ( A q) a- H o J k 1 1 AtjAf AoAmax 1 1 P L P L 1 1 cor vel *! * M ’la . f t 1 1 A qA max P L P L 1 1 cor vel *! * c. m . J k i i Ao A qA max 1 1 P L P L 1 1 cor vel *! » d . k]0 c[k 1 1 Ag AoAm ax P L P L 1 1 cor vel * (164a) is faithful to the input. But it has a release feature (i.e., Af) in coda, violating the top-ranked ALIGN(Obs]0 , Ao). (164b) is a place-linked structure, but it also crucially violate the same constraint with a release feature in coda. The remaining two candidates (164c, d) purchase alignment in exchange for the unfaithfulness of a release feature.S 4 5 4 I om it B asic AUGNMENT(Coronal places Rootj) and MAX(vel) = MAX-IO(velar), which are not relevant in the present discussion. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 147 However, (164c) loses on the L IC E N S E . (164d) satisfies the L IC E N S E at the expense of violating MAX-IO(Af) which is low-ranked, and turns out to be optimal. GEN does not produce the following configuration in (165) since the palatal affricate does not have a velar counterpart in the inventory: (165) f l . cfc I I A0Af AoAmax PL PL cor dor I I pal vel We find a similar case in the clusters in which a fricative assimilates to the following consonant. It can be simply accounted for through the interaction of the same constraints as in (164): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 148 (166) Input: /kis pal/ ‘flag’ —> [kip-.p’al] A f Aq A max PL PL cor lab Candidate ALI(Ao) *ONSET ... L i c e n s e MAX-IO(Af) Dep-IO(Aq) a- S L otP 1 1 Af AqA max 1 1 PL PL 1 1 cor lab *! * b- s]0 a[t 1 1 Af AoAm ax PL PL 1 1 cor lab *! * c- s]a c[p 1 1 Af AqA m ax PL PL 1 1 cor lab *! d. p]0 0[p 1 1 Ao AqA m ax PL PL 1 1 cor lab * * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 149 (166a, b, c) crucially violate ALlGN(Obs]a, Ao), which renders the properly-aligned (166d) more harmonic than any other competitor. Lateral ‘1 ’ does not assimilate to any consonants whatsoever, as illustrated in (167): (167) /kalpi/ — » [kal.bi] ‘a rib’ /palmjog/ — » [pal.mjog] ‘invention’ /tfulki/ [tful.gi] ‘a stalk’ /sal + ki/ — » [sal.gi] ‘living’ /kaltfhi/ — ^ [kal.tfhi] ‘a hairtail’ /kaltfuig/ —> [kal.tf’uig] ‘thirst’ GEN does not produce any candidates with ‘I’ assimilated in place to the following consonant. Because Korean does not permit any lateral approximant but an alveolar one, there is no candidate to consider in the case of clusters with ‘1 ’ component. §5.5.1.2 Homorganic clusters It is necessary to ask whether the constraint hierarchy developed so far can explain the place assimilation in homorganic consonant clusters. I will show that the hormorganic clusters usually result in place-linked partial geminates, and receive a simple account on the basis of the analysis presented thus far. Let us begin by considering the following data: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 150 (168) /kat + ta/ — » [kat=.t’a] ‘to be identical’ /kuk + kun/ — » [kuk= .k’un] ‘the national army’ /pam + man/ — > [pam.man] ‘night only’ In (168), we have a sequence of identical obstruents across syllable. Of interest here is the fact that in a sequence of two obstruents the first one is neutralized while the second one is tensified, apparently in violation of G em inate Inalterability (Hayes, 1986). In a derivational approach, Cho (1990:104) attempts to solve this problem by making a language-particular speculation that Korean does not have a total geminate. She argues that Korean has only partial geminates of the form in (169): (169) Partial geminates (after Cho (1990: 104)) C C • Place Root Laryngeal (169) is a partial geminate structure where the two segments share only the Place node. As a result, the Laryngeal node can be manipulated without mentioning what is involved in the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 151 other nodes. Adopting her idea, I claim that a sequence of homorganic but not identical consonants receive the same treatment. The following examples have the same structure as in (169): (170) /kan + ta/ — > [kan.da] ‘to go’ /kamparj/ — » [kam.bag] ‘prison’ /kogkan/ — > [kog.gan] ‘space’ In the next section, I propose a markedness constraint responsible for the place-linked structures in (168) and (170). §5.5.1.2.1 Obligatory Contour Principle One of the central constraints on autosegmental representations is the Obligatory Contour Principle (Leben, 1973, 1978; McCarthy, 1986b): (171) Obligatory Contour Principle (OCP) : At the melodic level, adjacent identical elements are prohibited. In derivational models, the OCP has been interpreted as a rule trigger (Yip, 1988). According to this interpretation, if a violation of the OCP occurs in the derivation, some mle must intervene to repair this violation. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 152 As a first approximation to the facts in (168) and (170), I propose the place-linked structures in question result from the satisfaction of the OCP on the articulator tier: (172) OCP( Articulator) : *txArticulator otArticulator I assume that consonantal and vocalic articulator nodes are formally identical. Then, a question arises as to how they are formally distinguished in the output for the purpose of applying the OCP(Articulator). To solve this problem, I revise the OCP(Articulator) as in (173), assuming that the mother node ‘[+cons]’ restricts the class of cooccurring articulator nodes to consonant clusters. I propose that the place assimilation in hormorganic clusters is a strategy of avoiding this configuration in (173): (173) OCP( Articulator) -revised- *[+cons] [+cons] I I aArt. aArt. I follow McCarthy (1988) in assuming that such major class features as [son] and [cons] should form the root node: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 153 (174) [ ‘ son “| consj Laryngeal [nas] [cont] Place Labial Coronal Dorsal A [dist] [ant] There has been a suggestion (Dresher, 1989; McCarthy, 1988; Selkirk, 1991; Padgett, 1991) that the domain of the OCP can be limited by a particular feature in a particular context. For example, McCarthy (1986b, 1988) argues that OCP always checks for identical ariculator nodes (i.e., Placej Place;) in Semitic. He classifies consonants into groups that tend not to cooccur in a root: (175) Classification of consonants in Semitic (after McCarthy, 1988:20) Labials: f b m Coronal sonorants: 1 r n Coronal obstruents: 0 3 t d s z 5 t d s5 5 | Dorsal sonorants: w y Dorsal obstruents: k q Pharyngeal and laryngeals: \ K h ? h ? 5 5 Underlines in fi , i , d . and s indicate so-called emphasis, constriction in the mid pharynx that exists together with the coronal constriction (McCarthy, 1988:19). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 154 There is an interaction of the place and stricture features here that must be explained. We cannot simply allow the OCP(Articulator) to rule out adjacent coronal articulators in all cases. That is because the coronal sonorants and obstruents in (175) are permited to cooccur when the relevant segments differ in [son]: (176) (after Padgett, 1991: 179) [+son] [-son] I I Coronal Coronal Therefore, the computation of the OCP on the articulator tier must involve some extra procedure that takes the [son] values into consideration, as argued in Padgett (1991) and McCarthy (1994). This argument gives further support for the relevance of a stricture feature [+cons] to the OCP(Articulator) as formulated in (173). In this way, we can make the constraint scan the consonantal clusters only, excluding any intervening vowels. §5.5.1.2.2 The analysis As the following tableau shows, the OCP(Articulator) conflicts with and therefore crucially dominates the BA(Coronal place^ Rootj): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 155 (177) Input:/kan + ta/ ‘to go’ -» [kan.da] PL PL cor cor ant ant Candidates OCP(Articulator) BA(Coronal place;; Rootf) a. n]c c[t 1 1 PL PL 1 1 cor cor 1 1 ant ant *! b- n]B B [t plT ^ p l 1 1 cor cor 1 1 ant ant * The OCP(Articulator), in concert with the LICENSE, is responsible for determining a place-linked structure as the optimal output. This argument is laid out with a set of plausible candidates in the following tableau: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 156 (178) Input: /kan + ta/ ‘to go’ -» [kan.da] PL PL cor cor ant ant Candidates OCP(Art.) *ONSET(place) BA(c) Ma x(v) License .... a. n]w B [t 1 1 PL PL 1 1 cor cor 1 1 ant ant *! * b. n]c B [t PL PL 1 1 cor cor 1 1 ant ant *! * c. n]0 a[t PL PL 1 1 cor cor 1 1 ant ant Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 157 (178a) and (178b) crucially violate OCP(Articulator) and *ONSET(place), respectively. The decision has to be made between (178c) and (178d). (178c) satisfies the *ONSET(place) by forming a place-linked structure and is determined as the optimal output.5 6 Now, let us consider the place-linked structures in labial and velar clusters. The details of the argument on them are identical to that given in the discussion of tableau (178) above. The optimal form obtains the satisfaction of LICENSE as well as OCP(Articulator) at the expense of MAX-IO(labial) and MAX-IO(velar), whose violations are not fatal as illustrated in (179) and (180): (179) Input: /kam pag/ ‘prison’ -» [kam.bag] I I PL PL I I LL LL (LL = Lower Lip) I I lab lab 5 6 MAX-IO(coronal) does not appear in the tableau due to the limitation of space. But it is clear that it must be low-ranked since it is not relevant. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 158 Candidates OCP(Art.) *ONSET(place) BA(c) MAX(v) License .... a. m]0 0[p 1 1 PL PL 1 1 LL LL 1 1 lab lab *! * b- m]c a[p PL PL 1 1 LL LL 1 1 lab lab *! ^ c . m]B 0[p plT ' ^ p l 1 1 LL LL 1 1 lab lab (180) Input: /koQ kan/ ‘space’ -4 [kog.gan] PL PL dor dor vel vel Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 159 Candidates | OCP(Art.) *ONSET(place) BA(c) MAX(v) | LICENSE .... a. 0]o ff[k 1 1 PL PL 1 1 dor dor 1 1 vel vel *! * b . 0 ] o Jk PL PL 1 1 dor dor 1 1 vel vel *! “ ■ c. ij]b a [k PL PL 1 1 dor dor 1 1 vel vel * Up to this point, we have only considered the consequences of identical consonant clusters. One remaining type of underlying consonant cluster is a sequence of identical articulators with different sites. For example, let us consider the sequence ‘alveolar + palatal’ in the input: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 160 (181) a. Input: /kot tfai]/ ‘immediately’ -» [kot-.tf’ag] PL PL cor cor ant pal b. Impossible candidates: (i) t l . J# (ii) t]a « [« f (iii) tic (iv) t]a .[if A q AgAf A 0 A f lA f A q A g A f A q A g A f PL PL PL PL PL PL PL PL cor cor cor cor cor cor cor cor ant pal ant pal ant pal ant pal When the GEN is applied to the input in (181a), it does not generate such potential candidates as in (181b). (181bi), (18lbii), and (18lbiii) result in the configuration where the coda has a palatal stop which is not allowed in Korean. (18lbiv) have the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 161 configurations of an anterior coronal affricate in onset, which is prohibited in Korean either. In sum, the constraints established to account for the place assimilation in homorganic clusters can be added to the constraint hierarchy in (155) in which there is no crucial ranking between OCP(Articulator), and ALIGN(Arel; onset): (182) Revised hierarchy of constraints for place assimilation in Korean : ALIGN(Obs]g. An). OCP(Articulator) » *ONSET(place) » BA(Coronal place^ Rootj) » MAX-IO(vel) » LICENSE » MAX-IO(lab) » BA^lac^; Rootj) » Max- IO(cor), MAX-IO(Af), DEP-IO(Ao) §5.5.2 English The system developed so far has the advantage of handling the variation among the place assimilation effects of other languages. In English, only coronals can be the target of place assimilation. In this section, I will show that the English pattern relies on re-ranking of a single constraint in the hierarchy for Korean given in (155). The examples of interest are the following: (183) Place assimilation in English (expanded on Jun, 1995:166) a. hot cake /hot kejk/ -» [hDk kejk] b. ten pounds /ten pawndz/ — » [tem pawndz] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 162 c. leap quickly /lip kwikli/ [lip kwikli] d. jack pot /cfeaek pat/ -» [cfeaek pat] e. book dealer /buk di:lar/ -> [buk di:lor] f. job ticket /cfcab tikit/ [cfeab tikit] *[lik kwikli] *[cfeaep pat] The crucial difference between English and Korean place assimilation lies in the re- ranking of MAX-IO(lab) in the hierarchy. While the LICENSE dominates MAX-IO(lab) in Korean (i.e., LICENSE » MAX-IO(lab)), the ranking is reversed in English (i.e., M a x - IO(lab)» LICENSE. This will capture the fact that labials are not targeted in English. An illustrative tableau is shown below, where the surface candidates for a schematic underlying sequence of /...pk.../ are evaluated: (184) Input: /lip kwikli/ ‘leap quickly’ PL PL lab vel Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 163 Candidates ♦Onset BA(c) MAX(v) Max(1 ) License BA(pl) Max(c) * * ■ a- Plo B [k 1 1 PL PL 1 1 lab vel * b. PL B [P PL PL 1 1 lab vel *! * * c- t]8 B [k PL PL 1 1 lab vel *! * It is clear that (184b) loses out because it crucially violates the high-ranked constraint *ONSET(place). The decision has to be made between (184a) and (184c). (184c) crucially violates MAX-IO(lab). Therefore, (184a) is evidently more harmonic than (184c). It should be stressed that MAX-IO(lab) is not active in other cases of place assimilation. Consider the tableaux (185) through (189): (185) Input: / hot kejk/ ‘hot cake’ I I PL PL I I cor vel Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 164 Candidates *ONSET BA(c) MAX(V) MAX(l) License BA(pl) Max(c) a. t]f f c[k 1 1 PL PL 1 1 cor vel *! b- t]a a[t PL PL 1 1 cor vel *! * * c- k]a a[k plT ^ p l 1 1 cor vel * * (186) Input: /ten pawndz/ ‘ten pounds’ PL PL cor lab Candidates *ONSET BA(c) Max(v) MAX(l) LICENSE BA(pl) MAX(c) a- n]c c[p 1 1 PL PL 1 1 cor lab *! b- n]w 0[t PL PL 1 1 cor lab *! * * c. m]a „[p PL PL 1 1 cor lab 5 * = * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 165 (187) Input: /<£aek pat/ ‘ jack pot’ PL PL vel lab Candidates *ONSET BA(c) Max(v) Max(1) License BA(pl) | MAX(c) er a- n]a a[p 1 1 PL PL 1 1 vel lab * b- n]a c[p PL PL 1 1 vel lab *! * * c- n]9 3[p PL PL 1 1 vel lab *! * (188) Input: /buk di:lsr/ ‘book dealer’ PL PL vel cor Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 166 Candidates ♦Onset BA(c) Max(v) MAX(1) License BA(pl) MAX(c) f f a .k ]a a[ d 1 1 PL PL 1 1 vel cor * b- k]„ „[g PL PL 1 1 vel cor *! * * c- t]s B [d PL PL 1 1 vel cor *! * (189) Input: /cfcab tikit/ ‘ job ticket’ PL PL lab cor Candidates ONSET BA(c) Max(v) m a x(D License BA(pl) MAX(c) ® a. b]„ a[t 1 1 PL PL 1 1 lab cor * b- b]B 0[p PL PL 1 1 lab cor *! * * c- d]0 J t PL PL 1 1 lab cor *! * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 167 §5.5.3 Other languages According to the table shown in (124), there are languages displaying the pattern of place assimilation involving all places of articulation as a target: e.g., Diola Fogny, Japanese, Malay, Nchufie, Yoruba, etc. For example, in Malay, consonants /p t k m n g/ occuring in coda position assimilate in place to the following stops and nasals: (190) Malay place assimilation (after Jun, 1995:64) a. Nasals (trivially adapted from Lodge, 1992:42) [malam tadi] [malam kames] ~ [makan buah] [malan tadi] [malag kames] [makam buah] [hidog manfog] ~ [hidom manfog] [pasag topeg] - [pasan topeg] b. Stops (Lodge, 1992:16) [kuat^ balac&a] ~ [kuap^ balacfea] ~ [kuaf balacfea] ‘to study hard’ [higgap*-’di’atas] ~ [higgat^ di’atas] ~ [higgaT di’atas] ‘to land on top o f The faithfulness constraints assert that the surface form and the lexical form are identical. Therefore, the fact that all places of articulation can be targeted in Malay means that they are crucially dominated by the LICENSE: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 168 (191) Constraint ranking for Malay place assimilation: *ONSET(place) » License » MAX-IO(vel), MAX-IO(lab), MAX-IO(cor) Since the LICENSE dominates all the faithfulness constraints in the hierarchy, it favors a linked structure (i.e., results in place assimilation) in all cases of adjacent consonantal place nodes at the expense of losing one of the two adjacent ones. But onsets do not assimilate to the codas due to the high ranking of *ONSET(place) in the hierarchy. The meaning of the ranking in (191) is rendered clearer when the rankings of Korean and English are considered: (192) Ranking for Korean place assimilation: *ONSET(place) » MAX-IO(vel) » LICENSE » MAX-IOflab) » MAX-IO(cor) (193) Ranking for English place assimilation: *ONSET(place) » MAX-IO(vel) » MAX-IOdahi » License » MAX-IO(cor) From this comparison, we can see that the analysis of place assimilation across languages supports the OT interpretation of constraints as violable and ranked. §5.5.4 An unsolved problem Since the LICENSE refers to consonantal place features in phonology, the constraint would also be invoked in dealing with the place assimilation within a complex coda. According to Mohanan (1993:72), for example, the distributional generalization within the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 169 syllable in English may be stated as follows: In adjacent [+cons] seg m en ts w ith in a sylla b le, i f the fir s t is [+stop] and the se c o n d is [-cor], the tw o segm ents m u st h a ve the sam e p la c e o f articulation. (194) Consonant clusters within syllable in English (after Mohanan 1993:72) a. English allows homorganic codas like [nt], [mp], [gk], [mf], and [ns]: sen t [sent], lim p [limp], sin k [sigk], n ym p h [nimf], rinse [rins] b. English allows heterorganic codas like [mt], [m9], [pt], and [kt] where the second consonant is a coronal: dream t [dremt], e x e m p t [igzemt], w arm th [wDrmQ], act [aekt], apt [aept] c. English allows heterorganic codas like [Ik], [Ip], [sp], [sk], [rp], and [rp] where the first segment is [-stop]: silk [silk], help [help], lisp [lisp], risk [risk], harp [harp], h a rk [hark] d. English does not allow sequences like *[np], *[nk], *[mk], *[mc], *[pk],*[kp], *[tk], *[tp], *[nf] within syllables. In Mohanan (1993:73), the generalizations on syllable-internal regularities in place of articulation in English are stated as follows: (195) Place assimilation within syllables: English a. Domain: syllable b. The first segment assimilates to the second. c. [+stop] consonants undergo assimilation; [-stop] consonants do not. d. [-cor] consonants trigger assimilation; [+cor] consonants do not. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 170 This statement of syllable-internal distribution can be collapsed with the statement of the alternation across syllables which was shown in §5.5.2. Then, it is a welcome result that the LICENSE can be called on to affect equally coda-onset clusters and complex codas, in such a language as English. However, it is not clear yet why the place assimilation within complex codas is anticipatory just like that in coda-onset clusters. This question remains to be investigated further. §5.6 Conclusion Recapitulating the main proposal in this chapter, I proposed that the essential properties of place assimilation be determined by a conflict between faithfulness constraints and the LICENSE on the one hand, and alignment constraints on the other hand. The pattern of place assimilation in Korean shows that the LICENSE is violable but not completely inert when the conditions leading to violation are not present. In the OT analysis of place assimilation, the LICENSE plays a role in favoring a place-linked structure but is subject to many violations in surface forms in cases where the linked structure results in a violation of higher ranking constraints. Compared with earlier work (Kim-Renaud, 1974, 1986; Kim, 1987; Cho, 1990), the analysis presented here have some positive results. OT allows for language-specific variation in constraint rankings. The seven constraints in (155) are all independently motivated and arguably universal. It has been shown that the language-specific patterns follow from the permutation of the constraint ranking established for Korean. For example, the interlinguistic variation among Korean and other languages including English lies in the re-ranking of the LICENSE in the hierarchy. OT has further advantage: it allows us to capture a cross-lingustic tendency that universal constraints such as LICENSE are not Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 171 always surface true, but are violable under appropriate conditions. This interpretation of constraints is not possible in derivational models. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 172 Chapter 6: Vowel Change and Umlaut §6.0 Introduction The purpose of this chapter is to present a critical comparison of synchronic and diachronic analyses of the umlaut phenomenon in Korean. On the one hand, Kyungsang dialect of Korean is productive in umlaut. Within a non-linear model of constriction location (Gorecka, 1989), I claim that the umlaut phenomenon can be straightforwardly accounted for by recognizing that front vowels and palatals are members of the same natural class. On the other hand, Kyungsang dialect provides consistent exceptions to umlaut. Current phonological theories including OT cannnot adequately deal with those exceptions. I claim that they can receive a diachronic explanation by assuming that the historical changes involved in this dialect around 19th century have obscured the umlaut which took place around 17th century. Choong-nam dialect, which was not affected by the historical vowel shift, provides a strong support for a synchronic analysis of umlaut. §6.1 Umlaut in Kyungsang dialect Korean has a vowel-fronting phenomenon (or, umlaut), which refers to the sound change by which central (k , o, a) or back vowels (u, o) become front vowels (i, y, e, 0,e). In Standard dialect, umlaut is not productive.5 7 According to Suh (1984), the history of 5 7 Standard dialect is defined as the language that are currently used in Seoul, capital of Korea. In Standard dialect, when they pronounce words, older generation makes umlaut much more often than younger generation does. In most cases, if they make umlaut, they sound less educated and privileged. As far as I investigated, all the speakers who live in Seoul and make umlaut often came from other districts of the Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 173 umlaut is traced back to the 17th century. Overall, the pattern of umlaut in Kyungsang dialect is still a productive phenomenon except for some exceptions. It is notable in Kyungsang dialect that umlaut also takes place when there are one or two intervening consonants between the two vowels, but these intervening consonants must be either labial or velar, as illustrated in (196). In addition, umlaut is always optional in every dialect of Korean: the examples at both side of the arrow are possible surface forms: (196) a. a<=>£ aki apt < ^ > egi ‘baby’ ‘father’ ami < = > emi magita < = s > megita ‘mother’ ‘feed’ c. 7 C < = > i iltf’rcki < = > iltfiki taditmi < = > tadimi ‘early’ ‘fulling block’ country where umlaut is a productive phenomenon (e.g., Kyungsang, Choongnam, or Chela, etc.) This leads us to the assumption that Standard dialect had no umlaut in the First place. The social aspect of umlaut still remains to be investigated. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 174 As can be seen from (196), the intervening labials or velars typically do not block umlaut. The peripheral (i.e., labial or velar) vs. central (i.e., anterior or palatal) distinction seems to play a role in the process of umlaut. Consonant clusters allowing umlaut furnish additional evidence bearing on this point. Consider (197): (197) tagkita < = > teg.gi.ta ‘to pull’ nampi < = > nem.bi ‘kettle’ pagmagi pag.meg.i ‘a club’ Apparently, umlaut is prevented just in case the intervening consonant is a palatal or palatalized one, as shown in (198a) and ( 198b), respectively. However, anterior coronals either block umlaut (198c) or allow it (198d): what must be explained then is why there should be a disparity between (198c) and (I98d). (198) a. /katfi / => [ka.c^i], *[k£.d$i] ‘branch’ /s0pufhi/ => [s0.bu.tfhi], *[s0.by.tfhi] ‘metal’ b. (i) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 175 /masita/ [ma.sJi.ta], *[me.sJi.ta] ‘to drink’ /kasi/ (ii) /amani/ /tfumani/ (iii)5 8 /kut + i/ /kat*1 + \J c. (i) /ati/ /nJtti(namu)/ (ii)5 9 /tfontta/ /kanrta/ [ka.sji], *[ke.sJ i] ‘thorn’ [a.ma.ni], *[a.me.ni] ‘mother’ [tfu.ma.ni], *[fu.me.ni] ‘pocket’ [ku.dp], *[ky.(%i] ‘insistently’ [ka.tfhi], *[ka.tfhi] ‘together’ => [a.di], ? [e.di] ‘where’ [n7t.ti(.na.mu)], *[ni.ti(.na.mu)] ‘a zelkova (tree)’ [tfo.ri.ta], [ka.ri.ta], ■ [tf0.ri.ta] ‘to boil down’ *[k£.ri.ta] ‘to hide’ 5 8 In Korean, /ati/ /uttia/ /thinun/ /thik ’7tl/ ‘t, t"’ are not palatalized in an underived environment, as shown below: [a.di], [tn.di.a], [thi.nun], [thi.k ’rtl], "[a.cfei] *[ut.<%i.a] *[tfhi.nun] • [ f i . k ’Jtl] ‘where’ ‘at last’ ‘a com (on the foot)' ‘dust’ ‘it’ turns to ‘i’ in these examples. This point will be made clear in the next section. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 176 d. Umlaut occurs optionally. (i) /mati/ => [ma.di], [me.di] ‘node’ /puti/ => [pu.di], [pi.di] ‘by all means’ /tfanti/ => [tfandi], [tfendi] ‘lawn’ (ii) /kjtrita/ => [k7t.ri.ta], [ki.ri.ta] ‘to draw, paint’ /parita/ => [pa.ri.ta], [pe.ri.ta] ‘to spoil’ /s’ltrita/ => [s’Tt.ri.ta], [s’i.ri.ta] ‘to be painful’ /tarita/ => [ta.ri.ta], [te.ri.ta] ‘to iron’ From the same consideration, umlaut is blocked in case the intervening i cluster has coronal component. This is illustrated with the examples in (199) cluster has palatal (tfty or palatalized (J, ii) component: (199) /magsin/ => [mag.sjin] *[meg.sjin] ‘disgrace’ /okni/ => [ok.ni] *[0k.ni] ‘an intumed tooth’ /jamtfhi/ = t> 03m.^hi] *[jem.y'i1 i] ‘a sense of shame’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 177 (196), (197), (198a, b, d), and (199) together show that the intervening consonant (or cluster) should be noncoronal to allow umlaut. But the fact is not always straightforward, as we see from (198c) in which anterior coronals (i.e., /t, 1/) block umlaut. I’ll show in the next section that this blocking effect of anterior coronals in Kyungsang dialect yields an argument against Hume (1990) that analyzes the umluat in Kyungsang dialect within the framework treating front vowels as coronal. §6.2 Earlier treatments Earlier works in Korean phonology (Cho, 1988; Iverson & Kim, 1987; Sohn, 1987) have suggested that anterior coronals of Korean be not specified for place of articulation in underlying representation. They are the least specified consonants in Korean and receive the specification [+coronal, +anterior] by default. This underspecification is motivated by such phenomena as consonant place assimilation, neutralization and epenthesis, among others (Cho, 1988; Iverson & Kim, 1987; Sohn, 1987). Drawing on the idea that front vowels are coronals, as first proposed in Clements (1976), Hume (1990) claims that palatal affricates differ from anterior coronal stops by the presence of the secondary articulation feature [+coronal]. In her framework, both coronal consonants and front vowels are specified as [+coronal] under the place node. In order to capture the fact that anterior coronals are transparent to the spreading of the umlaut feature in Kyungsang dialect, she makes a language-specific speculation that anterior coronals in Korean lack the specification [+coronal] underlyingly. The proposed representations in Kyungsang dialect are as follows: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 178 (200) a. Ill I place C-place b. I% I I place / C-place V-place I [+coronal] N place V-place [+coronal] An obvious question that we have to ask is whether there is empirical support for such a proposal. The answer to be drawn is that the attempt to analyze front vowels as coronal in Kyungsang dialect runs up against a serious problem. Let us see why. As can be seen in (200b, c), palatals and front vowels in Hume’s system form a natural class by the specification of [+coronal] under the V-place. This is how to explain the observation that an anterior coronal is transparent to umlaut while a palatal coronal is opaque. As is clear from the data in (198b, c), however, this proposal of treating front vowels as coronal would fail to capture the fact that anterior coronals are also opaque to the spreading of the umlaut feature in some cases. By way of illustration, consider the following: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 179 (201) /ati/ => *[e.di], [a.di] ‘where’ /s/ fxl lil => *[e.di] place place [+coronal] As shown in (201), the intervening consonant ‘t’ completely lacks place specification in underlying representation. Thus, ‘t’ should be transparent to the spreading of the umlaut feature ([+coronal]) that the following ‘i’ has. Therefore, Hume’s model predicts that all anterior coronals do not block umlaut in Kyungsang dialect, which cannot explain (198c). §6.3 The proposal Unlike Hume’s model, the Site-Articulator model adopted throughout this thesis does not make a language-specific stipulation in eliminating the redundant information from underlying representations. By incorporating the independently motivated view that front vowels do not form a natural class with coronal consonants (Gorecka, 1989), the model provides an account of the tranparency effect of anterior coronals with repect to the spreading of umlaut feature. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 180 It is first necessary to present the representation of consonants and front vowel which I assume in the synchronic analysis of umlaut. Let us consider several underlying representations shown below: (202) Underlying Representations of Korean consonants a. Labial b. (Anterior) Coronal c. Palatal d. Velar e. /i/ Place Place Place Place Place i i i i i Lower Lip Cor Cor Dor Dor I l \ Pal Pal -back I eliminate all the redundant information from the underlying representations on universal grounds. For example, default site features for L o w er L ip , C oronal, and D orsal articulators are underlyingly not specified: the universal default site features are Labial, A nterior, and Velar, respectively. Front vowels have stronger affinity with palatal coronals than with anterior coronals. This affinity is underlyingly represented by the specification of ‘Palatal’ site both in palatal consonants and front vowels. I propose that umlaut be characterized as a spreading of the Palatal site of /i/: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 181 (203) Umlaut as a spreading of Palatal site of !M r v / l a / U I Constriction I Place Dor Pal -back Based on (202), labials, coronals and velars are transparent to the spreading of umlaut feature (i.e., Palatal site) because nothing blocks it. (204b) is a sample derivation which shows that anterior coronals do not block umlaut in Kyungsang dialect: (204) Umlaut across anterior coronals a. Umlaut occurs optionally (repeated from (198d)). /mati/ = t> [ma.di], [me.di] /puti/ => [pu.di], [pi.di] /tfanti/ => [tfandi], [tfendi] /kTtrita/ => [k7t.ri.ta], [ki.ri.ta] /parita/ => [pa.ri.ta], [pe.ri.ta] ‘node’ ‘by all means’ ‘lawn’ ‘to draw, paint’ ‘to spoil’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 182 /s’Tcrita/ => [s’Tt.ri.ta], [s’i.ri.ta] ‘to be painful’ /tarita/ =* [ta.ri.ta], [te.ri.ta] ‘to iron’ b. /mati/ => [me.di] ‘node’ Im l /a/ /t/ /i/ Place Place ^ Cor Dor Pal [-back] By the same token, labials and velars are transparent to the spreading of ‘Palatal’ site because they are also underlyingly unspecified for their site features just like anterior coronals, which explains (196) and (197): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 183 (205) a. Umlaut across labials /ami/ => [e.mi] /a/ /m/ C I Place I Lower Lip ♦ ----- b. Umlaut across velars /aki/ =£ [e.gi] /a/ N J I C I Place I Dor * — 'mother' I'll I C I Place I Dor Pal [-back] 'baby' /V I C I Place I Dor . . . J \ Pal [-back] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 184 In addition, we have cases where anterior coronals apparently block umlaut (206a). I claim that this is the result of a sound change that shifted /7t/to h i around 19C. There is evidence (Suh, 1984) that all the surface [i]’s in (206a) came from a historical sound change, M > /i/, as illustrated in (206b): (206) a. (repeated from (198c)) /ati/ => [a.di], *[e.di] ‘where’ / niti(namu)/ = t> [mt.ti(.na.mu)], *[njt.ti(.na.mu)] ‘a zelkova (tree)’ /tfontta/ => [tfo.ri.ta], *[^0.ri.ta] ‘to boil down’ /kantta/ => [ka.ri.ta], *[ke.ri.ta] ‘to hide’ b. Historical vowel change (after Suh, 1984). /3t7t/ > /a.di1, ‘where’ /nitTt(namu)/ > /n7t.ti(.na.mu)/ ‘a zelkova (tree)’ /tfontta/ > /fo.ri.ta/ ‘to boil down’ /kantta/ > /ka.ri.ta/ ‘to hide’ (206b) shows that /Til turned to /i/ after anterior coronals around 19C in Kyungsang dialect. Therefore, the lack of umlaut in these examples is due to the historical appearance of M instead of h i before 19C when umlaut was more productive. Compared with h i, /Tt/does Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 185 not have the triggering feature of umlaut (i.e., Palatal site) as shown below, which explains why the N in (206a) fails to trigger umlaut across ‘t’ or T that should not block the spreading of umlaut feature: (207) a. N C Place Place Dor l \ Pal -back Dor Vel -back In Korean, /t/ becomes palatal [tf] by assimilating to the following high palatal vowel /if. This palatalization occurs only in derived environment, as shown in (208c): (208) a. (repeated from ( 198d)) /mati/ => [ma.di], [me.di] ‘node’ /puti/ => [pu.di], [pi.di] ‘by all means’ /tfanti/ => [tfandi], [tfendi] ‘lawn’ Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 186 b. (repeated from (198a)) /katfi/ => [ka.dji], /s0putfhi/ c. (repeated from (198b(iii))) /kut + U => [ku.<%i], /kath + i/ [ka.fhi], '[ke.c^i] ' [ky-c^i] |e [ka.^i] ‘branch’ [s0.bu.tfhi], *[s0.by.tfhi] ‘metal’ ‘insistently’ ‘together’ (208a) is a normal application of umlaut: the representations given in (202) predict that the old ‘i’ (as opposed to the ‘i’ historically turned from ‘re’) does trigger umlaut without triggering the palatalization of the intervening /t/ in an underived environment: (209) (repeated from (204b)) /mati/ => [me.di] /m J /a/ /t/ C ‘node’ /i/ C Place Place Cor Dor Pal - back Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 187 Assuming that palatalization as well as umlaut is a process of spreading Palatal site, it should be noted in (209) that the anterior coronal /t/ is not palatalized, allowing itself to be transparent to the spreading of the Palatal site of /i/. From the representations given in (202), it becomes clear that not only anterior coronals but also labials and velars should not block umlaut. Along the same lines, it would be a straightforward matter to account for palatals blocking umlaut in (208b). Both palatals and front vowel are specified for Palatal site. In other words, since palatals are already characterized by the spreading node, the umlaut rule cannot reach the target across the opaque segment, as shown below: (210)/katfi/ => [kafi], *[ketfi] 'branch' /a/ l%l N C C I I Place Place I I Cor Dor i II*-—.J^ Pal Pal - back The application of umlaut to the output of palatalization is not possible, either (see (208c)). As shown in (211), the output of palatalization results in a configuration where Palatal site is multiply-linked to both [sj] and [i]. Now, the vowel [i] cannot trigger spreading umlaut: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 188 umlaut spreads only to one segment.6 0 Besides, [sj] cannot trigger umlaut: there are no cases where a palatal consonant triggers umlaut by itself: (211) /kasi/ [ka.sji], *[ke.sji] ‘thorn’ la/ /sV C lii c Place Place Cor Ih- Pal -back I claim that L in kin g C onstraint by Hayes( 1986:331) is responsible for the failure of umlaut in this case: (212) Linking Constraint : Association lines in structural descriptions are interpreted as exhaustive. In the formulation of umlalut in (203), there is one and only one association line from Palatal site to Dorsal articulator. Therefore, the structural description is only met by a Palatal site which is singly linked to the Dorsal articulator. But if we apply umlaut to the 6 0 In a similar vein, Alicja Gorecka (p.c.) suggested to me that umlaut might be characterized as a sharing of Palatal site between two segments, which would neatly explain why the spreading of umlaut occurs only Continued on the next page Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 189 output of palatalization as in (211), we are appealing to the double linking of the single Palatal site in violation of the Linking Constraint. In sum, there are three different issues with respect to umlaut in Kyungsang dialect. First, when the trigger vowel /i/ is historically derived most likely from M , there is an apparent blocking of anterior coronals (see (206)). This is evidenced by the vowel shift occured in 19C (see (221)). Second, we have a nomal application of umlaut in other cases (see (196), (197), (198d)), which include umlaut across labials, velars, and anterior coronals. Third, /t + i/ surfaces as [tfl because of the derived environment and the surface palatal consonant blocks umlaut, which is predicted by the representations given in (202). Within the Site-Articulator model (Gorecka, 1989), it makes perfect sense to claim that palatals block umlaut on the assumption that the palatal site of /tf/ in (213b) blocks the spreading of the same site of the front vowel h i in (213c). However, anterior coronals blocking umlaut is not expected. As can be seen from (213a), [t] can be underspecified for site (i.e., Anterior) because Anterior site is the default site for Coronal articulator, based on universal markedness. Assuming that umlaut is characterized as a spreading of Palatal site, the anterior coronals should not block the spreading of umlaut because nothing blocks the spreading of the Palatal site of [i]: once per word. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 190 (213) a. [t] R b. [tf] R c. [i] R Place Place Place Cor (Ant) Cor Pal Dor Pal -back Obviously, the blocking effect of anterior coronals in (198c) remains to be worked out. There seems to be no universal phonetic reason why anterior coronal consonants should prevent umlaut from taking place, for we find in German such example as M ann (singular)- M a n n er (plural) ‘man’, G ott (singular)-Gotfer (plural) ‘God’, m u s s (1st or 3rd person singular)-m iissen (1st or 3rd person plural) ‘must’, etc. What then is the relation between umlaut and coronal consonants in Korean? In what follows, I illustrate the difficulty of handling such assimilatory effects as palatalization and umlaut in OT (§6.5). Then, I attempt to show that the history of phonology in this dialect sheds some light on this dual aspect of anterior coronals with respect to umlaut (§6.6). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 191 §6.4 Assimilation in OT In OT, constraints describe the universally unmarked characteristics of the structures involved. A family of faithfulness constraints in OT play the crucial role of insuring that any deviation from underlying representation must be justified by the interaction with other constraints. It corresponds to the principle in an input-oriented theory that no change in the representation occurs unless it meets the structural description of some rule: the default is to do nothing. Consequently, the language-particular processes of palatalization or umlaut would result only if some constraint dominates one of the faithfulness constraints. In §6.5.1, I briefly review how the markedness of phonological processes are formulated in SPE. I will show in §6.5.2 and §6.5.3 that OT is ill-equipped with such a mechanism, considering why it is difficult to conceive of an OT-based universal constraint that would force such phenomena as palatalization and umlaut. §6.4.1 Markedness in SPE In SPE, the notion of markedness is determined syntagmatically.6 1 As an example, let us consider one statement of marking conventions: (214) Marking convention (Chomsky & Halle, 1968:404) / [+cont] / + [+cons] [u cont] — > v. [-cont] (a) (b) This point is brought to my attention by Bernard Comrie (p.c.). Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 192 The unmarked value of the featue [continuant] is contextually determined. In initial position before a consonant (214a), the consonant with [n continuant] is interpreted as [-(-continuant]. It is interpreted as [-continuant], elsewhere. In other words, the unmarked consonant in preconsonantal positon is a continuant, while it is a stop in other positions. Since the marking conventions are universal and not part of an individual phonology, they are not asssigned any cost. Likewise, the umlaut process is conditioned by environments in Kyungsang dialect: it is conditioned by the point of articulation of the intervening consonants. However, OT is not equipped with any mechanism corresponding to the marking convention in SPE. Therefore, it is difficult to express the unmarkedness of widely attested phenomena which are contextually determined. §6.4.2 Palatalization Palatalization refers to the various assimilations that consonants undergo in the context of front vowels. Palatalization may result in the addition of an [i]-like articulation to a consonant (i.e., secondary palatalization), e.g., /k + e/ -» [ide], or a complete change in the major place of articulation of the consonant target, e.g., /k + e/ — » [tfe], McCarthy & Prince (1995:93-94) propose that palatalization involve spreading of the feature complex [+coronal, -anterior] to a preceding consonant and posit a constraint ‘PAL’, considering palatalization a kind of CV-linkage (Ito & Mester, 1993) of the coronal feature. For the purpose of forcing palatalization, McCarthy & Prince rank the constraint ‘PAL’ above a faithfulness constraint: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 193 (215) Palatalization in Akan /ke/ Pal iDEN-IO(-cor) « ■ a. tee * b. ke *! The ranking, PAL » iDEN-IO(-cor), is necessary to account for the non-existence of surface *ke in Akan. By the same reasoning, the palatalization in Korean would be analyzed as follows: (216)/kut + i/ =» [ku.cfci] ‘insistently’ /kut + U Pal lDEN-IO(-cor, +ant) ® a. ku.cfei * b. ku.ti *! One serious problem with the constraint Pal is that it encodes a dynamic process in terms of a constraint which has to be static. When translated into OT, ‘PAL’ simply does not explain palatalization. Let us compare the faithful output (217a) with the palatalized output (217b). In order for the PAL to be qualified as a constraint in OT, it should express the unmarked state of a structure. But it is not clear what sort of constraint forces such representation as (217b): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 194 (217) a. [k u t i] b. [k u c fc i] Place Place Place Place Cor Dor I l \ Ant Pal [-back] Cor Dor Ant Pal [-back] On the other hand, derivational approach captures the insight into the assimilatory effect. Let us consider again the palatalization in a derivational approach, as shown below: (218)/kut + i/ => [ku.cfci] ‘insistently’ /k/ /u/ /t/ + /i/ I I Place Place I I Cor Dor Pal [-back] If it is assumed that redundant information (i.e., Anterior site of /t/) is left blank and later filled in by redundancy rules, (218) provides a clear account of why the /t/ is subject to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 195 palatalization and captures the fact that palatalization is an assimilatory process of palatality. Regarding why assimilation is such a surprising result in OT, I claim that it is due to the fact that it is difficult to understand why the speakers of Korean do not tolerate a sequence of [ki] but prefer a sequence of [cfei]. In other words, it is not clear why they have to replace the underlying sequence with the one where palatality is shared. This is the question that OT at this point cannot answer. §6.4.3 Umlaut McCarthy (1994:14) proposes that there be a constraint UMLAUT which forbids the sequence [aC0 ii] in Icelandic. He claims that UMLAUT must dominate a set of faithfulness constraints to trigger alternations: (219) UMLAUT » faithfulness constraints However, the constraint UMLAUT poses the same problem as PAL in palatalization: it does not provide any reasons for umlaut. (220) illustrates the umlaut (or vowel-fronting) process: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 196 (220) a. /a p i/ =* I I C C I I Radical Dor I f X Phar. Pal [-back] It is difficult to conceive of a constraint that would force umlaut. In other words, it is not clear in OT why Korean speakers do not tolerate the representation in (220a) but change it to (220b). It would be very hard in OT to imagine a universal phonetic constraint responsible for sharing frontness, but not others. This is a problem which OT cannot provide a decent solution at this point. In serial approach, however, example (220) is a perfectly natural example of assimilatory process. §6.5 Historical vowel change In Kyungsang dialect of Korean, there was a vowel shift that could provide an explanation for the opacity of anterior coronal consonants in umlaut process. As shown in (221), /a/ extended its allophonic range as close as to that of /7t/,in Kyungsang dialect around 19th century: b. [e b i] ‘father’ c c c I I I Radical Dor Dor I I X Phar Pal Pal [-back] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 197 (221) i <------- 7 t -------> u (Suh, 1984) T As a result of the vowel shift in (221), the unstable /:t/turned into /u/ or /i/, as illustrated in (222) and (223), respectively: (222) ji => u / labials or velars /tf amk7t-/ => [^am.gu-] ‘to lock’ /tamkrt-/ => [tam.gu-] ‘to soak’ /kophjc-/ => [ko.ph u-] ‘hungry’ /bap'rc-/ => [ba.p’u-] ‘busy’ /mo7t-/ => [mo.u-] ‘to collect’ /S7C lph 7 C -/ => [sjil.ph u-] ‘sad’ (223) k = t> i / coronals /masrtm/ => [ma.sjim] ‘servant’ /S7C lph T C -/ = t> [sjil.ph u-] ‘sad’ Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 198 /pus7trem/ [pu.sji.rem] ‘a tumor’ /s’Jdke/ => [s’il.ke] ‘gall’ /s’rcl-/ => [s’il-] ‘to cut’ /tfTtlksp-/ => [tfil.kop-] ‘happy’ /karrc-/ => [ka.ri-] ‘to choose’ /kanttfhi-/ [ka.ri.tfhi-] ‘to teach’ At this point it becomes clear that each of the two processes in (222) and (223) has a distributional pattern. In (222), M turns to [u] after labials and velars, which is predictable because the process is featurally motivated: velars give velar constriction, labials give labial constriction. Adopting the Site-Articulator model proposed by Gorecka (1989), I represent the situation as (224) [p] + R C I Lower Lip I Labial ; follows: [k] => [u] R R Dor Lower Lip Dor I I I Velar Labial Velar Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 199 In (224), the representation of [p] combined with that of [k] produces exactly that of [u]. Therefore, the historical process of M turning to [u] after labials or velars is well motivated.6 2 On the other hand, /tt/tums to IM after coronal consonants (i.e., after /s, s’, I, tf/), as shown in (223). I suspect that this [i] would be a default case of this phenomenon. There are some apparent exceptions to this generalization, as illustrated in (225) which shows that /it/ became [i] after velars too: (225) /kTurim/ => [ki.rim] ‘picture’ /k’rcrhi-/ => [k’i.ri-] ‘to boil’ /krcri-/ => [ki.ri-] ‘to draw’ /krcrip-/ => [ki.rip-] ‘to miss’ /tf’igkmi-/ => [tf’ig.ki.ri-] ‘to scowl’ 62 Consider the following examples, where labials in the onset are followed by a round vowel: (i) [mul] ‘water’, [pul] Tire’, [ph ul] ‘grass’, [p’ul] ‘com ’ According to Sohn (1991:203), the representations in Middle Korean (15C) corresponding to the words in (i) were [m7Cl], [pnl], [ph 7C l], and [p’ttl], respectively. Sohn’s observation supports my proposal on umlaut through the vowel shift in (221): ‘Jt’ turns to ‘u’ after labial consonants. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 200 Apparently, these examples are against the generalization captured in (222) and (223), since M in (225) turns into [i] after a velar consonant ‘k \ However, closer reflection reveals that this alternation may be an umlaut process triggered by the following vowel ‘i \ In sum, as M turned into N after coronals (most likely as a default), it looks as if coronals are blocking umlaut. Given this, it is plausible that this is the case that created coronal opacity, whereas the cases that do not create coronal opacity are simply spreading umlaut across coronals because nothing blocks it. In other words, what we observe is a productive pattern from the times when there were N and /7t/,with the latter not triggering umlaut, obviously. §6.6 Umlaut in Choong-nam dialect Particularly thoroughgoing examples of umlaut are found in Choong-nam dialect where umlaut is currently more productive than in Kyungsang dialect. In Choong-nam as well as Kyungsang dialect, underlying palatal affricates are blockers in the assimilatory process of umlaut. Anterior coronals in Choong-nam are typically transparent to umlaut. I will show that secondary articulations such as velarization as well as palatalization play a role in blocking umlaut. The purpose of this section is to bring to light data illuminating these points. As shown in (226a), the trigger and target of umlaut in Choong-nam dialect should be separated by labial or velar consonants, as in Kyungsang dialect: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 201 (226) a. /ami/ => [e.mi], [a.mi] ‘mother’ /api/ [e.bi], [a.bi] ‘father’ /makita/ => [me.gi.ta], [ma.gi.ta] ‘to feed’ /komphagi/ [kom.pher).i], [kom.p^ag.i] ‘mold’ b. /namphjan/ => [n£m.phjan], [nam.phjan] ‘husband’ /samkhja/ => [sag.kh ja], [sam.khj3] ‘to swallow’ /hapkjak/ => [hek= .k’jak], [hak=.k’ jak] ‘pass’ /hakkjo/ [h£k= .k’ jo], [hak=.k’ jo] ‘school’ Palatal (tf, tf’, tf ) or palatalized (ii, sj) consonants systematically block the application of umlaut, as illustrated in (227): (227) No umlaut across [tf, tf’ , tf\ h , or J] /katfi/ => [ka.tfi], /tasi/ => [ta. sJ i], *[t£.sJ i] /s’amtfi/ =» [s’am.tfi], *[s’£m.tfi] ‘branch’ ‘again’ ‘a tobacco pouch’ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 202 /jamtfh i/ => [jam.tfh i], *[j£m.tfh i] ‘a sense of shame’ /umtf’ip/ => [um.tf’ip], *[ym.fip] ‘a dugout’ /okni/ => [og.ni], *[0 g.ni] ‘an intumed tooth’ /maktfi-/ => [mak=.f ’i-], *[mek= .tf’i-] ‘to eat (negative)’ /samtf’ikha-/ => [sam .f’i.kh a-], *[sam.tf’i.kh a-] ‘frightening’ /samnjan/ => [sam.njan], *[sem.njan] ‘three years’ /magsin/ => [mag.sjin], *[meg.sjin] ‘disgrace’ The intervening anterior coronals (e.g., s, n, t) allow umlaut in Choong-nam dialect. (228) illustrates the fact: (228) /us + kja/ = s > [yt_.k’ ja] ‘causing a laugh’ /ankjag/ => [en.gjag] ‘glasses’ /kjantin-/ => [kjen.din-] ‘to endure’ Finally, consonant clusters consisting of lateral ‘1 ’ followed by any consonant block umlaut. It should be stressed that the data in (229) form a pattern: syllable-final ‘1 ’ is velarized in all cases: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 203 (229) a. -l.t- alt’iki pulth i k3lth i- kulktikulk- b. -l.s- nals’i kwals’i kftls’i mals’i c. -l.n-63 talnim th 7tlni [oi-.t’i.ki] [pui-.th i] [kaith i-] [nai.s’i] [kwai.s’i] [krci-.s’i] [mai-.s’i] [tai-.lim] [thi*.li] *[py*.th i] *[kei-.th i-] [kui-.di.gul-] *[kyi\di.gul-] *[nei.s’i] *[kwei\s’i] *[kii.s’i] ’ • ‘[mei-.s’i] *[tei-.Iim] *[thii-.li] ‘a fool’ ‘flame’ ‘to put up (against)’ ‘very thick’ ‘weather’ ‘contempt’ ‘a character’ ‘language’ ‘moon (Honorific)’ ‘fake teeth’ 63 An alveolar nasal /n/ becomes a lateral [1] whenever it is preceded or followed by a lateral /l/. Some examples are given below: (i) /tf’alna/ /konlan/ /manli/ [fa l.la ] [kol.lanj [mal.li] moment 'trouble' '10,000 miles' Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 204 koltfh i => [koT.yh i] *[k0 i.tfh i] kaltfh i => [kai.tfh i] *[kei.tfh i] k 'o lfi => [k’ o-t.fi] *[k'0 i .f i ] e. -l.p- kalbi [kai.bi] *[kei.bi] suilph i => [suii.ph i] *[sii.ph i] f. -l.k- tfulki => [tfu+.gi] *[tfy^-gi] kalki- => [kai.gi-] *[kei.gi-] ‘trouble’ ‘a hairtail’ ‘the last’ ‘rib’ ‘sadly’ ‘stalk’ ‘to slash’ With [1 ] in coda, velarization of it blocks the spreading of umlaut. The representation of velarization illustrated in (230) shows why applying umlaut to the output of velarization is not possible: the specification of Velar site blocks the spreading of Palatal site: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 205 [nai.s’i], ♦[nefr.s’i] ‘weather’ [s’] [i] R I Place I Dor H ............... Pal -back In sum, Choong-nam dialect provides a complete paradigm with respect to umlaut: (i) anterior coronals as well as labials and velars always allow the spreading of umlaut feature, (ii) palatals block umlaut, (iii) secondary articulations such as palatalization and velarization also block umlaut. It is shown that these are the expected results, given the underspecified Site-Articulator representations sketched in (202). §6.7 Conclusion In Kyungsang dialect, anterior coronals either allow or block umlaut. As discussed in §6.3 and §6.4, the behavior of anterior coronals blocking umlaut is not expected by current synchronic phonological theories. Rather, what we observe diachronically is a productive pattern from the times (i.e., prior to 19C) when there were both N and /7 t/in this dialect, with the latter not triggering umlaut. It is plausible then that the vowel changes in 19C (230) /nal.s’i/ m R I Place Cor Dor Vel Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 206 shown in (2 2 1 ) have obscured the umlaut process of this dialect: the umlaut process took place in 17C and is fossilized. The vowel that nowadays fails to trigger umlaut across coronals was historically /7t/not /if. It looks as if coronals are blocking umlaut in Kyungsang dialect, since M turned to h! after coronals, most likely as a default. The Kyungsang dialect of Korean doesn’t have a pristine look of umlaut. But through historical reconstruction we could get to the stage where there was a pristine look. A reasonable answer is suggested by the fact that vowel changes have taken place in this dialect around 19C, which made umlaut opaque. However, the problem with this diachronic approach to umlaut in Kyungsang dialect is that the overall picture of the umlaut patterns in this dialect strongly suggests that it should be synchronically productive. The dual behavior of anterior coronals with respect to umlaut in Kyungsang dialect is not attested in Choong-nam dialect which was not affected by the historical vowel shift in 19C, and is also productive in umlaut. As for the most phonological phenomena dealt with in previous chapters, I argued that OT provides more insights into them than other previous models. But the umlaut phenomenon in Kyungsang dialect defies satisfactory explanation from the viewpoint of current phonological theroies including OT, Feature Geometry, and so on. Therefore, umlaut in Kyungsang dialect shows the significance of historical evidence in synchronic grammar. I conclude the chapter by pointing out that this is the problem which the phonological theories will address in the future. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 207 Appendix A: Site-Articulator Model Since the mid 1980’s most phonologists in the autosegmental tradition have assumed a version of feature geometry closely resembling that of Sagey (1986) illustrated below: (231) [constricted] [spread] [stiff vocal cord] [slack vocal cord] Root [continuant] [consonantal] Supralaryngeal Laryngeal Soft Palate [nasal] Place Labial Coronal Dorsal [round] ' x [high] [anterior] [distributed] [low] [back] Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 208 Sagey’s model has a place node under which there are three major active articulators of the vocal tract: Labial (the lips), Coronal (the tongue tip and blade) and Dorsal (the tongue body). Sageyan models have assumed the existence of those three articulators, but have not taken site features seriously. Where the simple partition into major active articulators is not sufficient, the finer distinctions axe made by adding dependent features under those major articulator nodes, which is an ad hoc choice. For instance, in case dental and palatal coronals such as III and Is/ contrast in certain language, this will be represented by the secondary specification of [±anterior] under the Coronal node. The problem is that the SPE system is fundamentally active-articulator-oriented, but the articulatory correlate of [±anterior] is the location of the constriction: forward of or behind the alveolar ridge. One fundamental insight of Gorecka (1989) is that the passive articulator (i.e., site) as well as the active articulator needs to be specified, which enables many analyses clearer. Gorecka (1989:112) argues for a phonological constituent called the Constriction Node, which represents the constriction gesture. The hierarchical representation of a segment including the Constriction Node is represented as follows: (232) Root Node Laryngeal Features Manner Features (Constriction) Constriction Site Articulator Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 209 The constriction node in (232) is formally equivalent to the articulator node in the Sagey’s model shown in (231). The Site-Articulator model reconciles two traditions of segment characterization: the IPA tradition where the site of a constriction is one of the most important principles, and the SPE tradition where the features based on active articulator are given priority. There are ten Site-Articulator combinations which represent physically possible gestures, after getting rid of the impossible ones:64 (233) Site-Articulator combinations (after Gorecka 1989:11) a. Libial/Lower Lip (labial sounds): p, b, m, u, o b. Anterior/Lower Lip (labiodental sounds): f, v, it) c. Anterior/Tongue Blade (anterior coronals): t, d, s, z, 0, 5 d. Palatal/Tongue Blade (palatal coronals): f, cfc , J, 3 e. Palatal/Tongue Body (non-low front vowels): c, j, q, i, e f. Velar/Tongue Body (velar sounds): k, g, x, q, = , k , u , o g. Pharyngeal/Tongu Body: ae, a h. Pharyngeal/Tongue Root: fr, < \, a To see how this model generates representations of simple or complex segments, let us consider some examples: 6 4 *Labial/Lower Lip, *Labial/Tongue Blade, *Labial/Tongue Body, *Labial/Tongue Root, *Anterior/Lower Lip, *Anterior/Tongue Blade, * Anterior/Tongue Body, *Anterior/Tongue Root, *PaIataI/Lower Lip, *Palatal/Tongue Body, *Palatal/Tongue Body, *PaIatal/Tongue Root, *VeIar/Lower Lip, *Velar/Tongue Blade, *VeIar/Tongue Body, *Velar/Tongue Root, *PharyngeaL/Lower Lip, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 210 (234) a. Examples of consonants: ftl Ih J Place Place /p/ Place Ikl Place Ant Cor Phar Radical Labial Lower Lip Velar Dorsal /Jv / /tV fk w l Place Place Place c c c r \ Pal Cor Vel Dorsal Ant Cor b. Examples of vowels: I d !\J C C [ \ ^ 1 Pal Dorsal Vel Dorsal Place Place /a/ Place loJ Place C Lab Lower Lip Phar Dorsal Pal Dorsal Phar Radical Phar Dorsal -high -back +high -back -high +back *Pharyngeal/Tongue Blade, *Pharyngeal/Tongue Body, and ^Pharyngeal/Tongue Root are physically impossible Site/Articulator combinations. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 211 lul Place I d Place ly f Place c c c c ( \ r \ Vel Dorsal Lab Lower Lip Pal Dorsal Phar Radical C C ^ 1 f \ Pal Dorsal Lab Lower Lip Site-Articulator model generates representations that are phonetically plausible and well supported by the natural class effects. For example, the two vowels, ly l and /II/, constrast in Swedish. These two vowels pose a problem for an SPE-based feature system, since they are both front, rounded, high and tense. The two representations in the Site- Articulator model are shown in (235): (235) a. /n/ Place C C Vel Dorsal Lab Lower Lip -back round b. ly l Place C C Pal Dorsal Lab Lower Lip -back round Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 212 It is clear that l \ \ l and ly l contrast by virtue of being specifed for different sites: the former is Velar, the latter is Palatal. In other words, /II/ is the fronted velar and ly l is the rounded palatal. In addition, the representations in (235) account for why velars are fronted, but not palatalized before/FI/in this language: /FI/triggers velar fronting because it it [-back]. On the other hand, ly l is a palatal vowel, therefore it triggers palatalization along with other palatal vowels in the language. But /FI/is a fronted velar, which explains why it fails to trigger palatalization. The major advantages of the constriction model over the preceding ones can be summarized as follows. First, the model generates all the representations necessary for capturing the existing contrasts, by slightly expanding the inventory of features. Second, it makes all articulation features equally available to consonants and vowels, thus accounts for the vowel/consonant interaction (e.g., mutual vowel/consonant palatalization, velarization, and pharyngealization, among others). Gorecka (1989:120-123) suggests the possibility that the Articulator dominates the Site, as illustrated below: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 213 (236) Root Node Laryngeal Features Manner Features (Constriction) Constriction I Articulator I Site Compared with the hierarchical structure in (232), the one in (236) implies that phonological processes can make access to the Site features independently of the Articulator features. I claim that the umlaut phenomenon investigated in Chapter 6 be consistent with the geometry in (236), which I adopt throughout this thesis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 214 Appendix B: Optimality Theory OT is a system of constraint interaction in phonology, originally laid out in Prince & Smolensky (1993). The emphasis of OT breaks with previous work by emphasizing the centrality of surface constraints in a theory of language. It is greatly revised and expanded in McCarthy & Prince (1995). In what follows, I briefly sketch these two models. Part 1: Prince & Smolensky (1993) OT focuses not on the step by step derivation, but on the constraints that underlie phonological changes. It provides the means to derive the varied effects of phonological patterns through the ranking of a limited set of universal constraints. Cross-linguistic variation is attributed to different rankings of constraints. McCarthy & Prince (1993:5) distinguish four major properties of OT: (237) Principles of Optimality Theory a. Violability: Constraints are violable; but violation is minimal. b. Ranking: Constraints are ranked on a language-particular basis; the notion of minimal violation (or best-satisfaction) is defined in terms of this ranking. c. Inclusiveness: The candidate analyses, which are evaluated by the constraint hierarchy, are admitted by very general considerations of structural well-formedness; there are no specific rules or repair strategies with specific structural descriptions or structural changes or with connections to specific constraints. d. Parallelism: Best-satisfaction of the constraint hierarchy is computed over the whole hierarchy and the whole candidate set. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 215 The input-output pairs are produced not by rules, but by an evaluation procedure that checks all possible outputs for some input against a set of universal constraints, and determines which output best satisfies those constraints. In OT, a grammar consists of three modules, GEN, CON and EVAL: (238) a. GEN: Produces an (infinite) set of possible outputs for any given input. b. CON: A set of UG constraints, ranked for each language. c. EVAL: A procedure for selecting the optimal output given a set of outputs and a ranked CON. An underlying form is fed into a function called GEN(ERATOR). A set of potential surface forms (called candidates) is generated by the function GEN applied to an underlying input. These surface forms (or candidates) are then evaluated against a ranked set of universal constraints. The surface form, that is, the preferred candidate, is chosen by this evaluation procedure: (239) GEN(inputj) => (cand,, candj, cand3 , candn ) EVAL(cand,, cand,, cand3 , candn ) => [output,] The fact that there is a conflict of constraints where candidate, satisfies constraint A but violates constraint B while candidate, violates constraint A but satisfies constraint B, is sufficient to determine the ranking of the two constraints, A and B. This complementary constraint satisfaction is resolved by the hierarchical ranking of the constraints in question: Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 216 (240) Candidates Constraint A Constraint B Candidate, * Candidate, *! The tableau illustrates the crucial ranking graphically in terms of columns ordered from left to right. Evaluation of the candidates proceeds recursively by constraint, until all candidates besides the winning candidate are eliminated. In tableaux, a **’ in row /, column j indicates that candidate i violates constraint j . Candidate,, for example, violates Constraint A. In the simplest case as in (240), the decision is already reached in the first round. **!’ denotes the crucial failure of candidate,. points to the optimal output. Once a candidate is out of the running, whether or not it violates other low-ranked constraints is irrevelant, a fact reflected on the tableau by having those lower cells shaded. In case of (240), the constraint A is dominant in the constraint hierarchy and decides which candidate is optimal and the violation of the constraint B is irrelevant. The violation of the constraint A by candidate, is fatal so long as candidate, cannot be the optimal surface form. Most sounds in any given form are conservative in that they do not diverge from their underlying forms; they tend not to alternate and do not vary in category. In derivational phonology, this is expressed by the convention that nothing in the underlying representation is changed unless it meets the structural description of a rule. According to OT (Prince & Smolensky, 1993), constraints of fa ith fu ln ess demand that the output be as close as possible to the input. OT sees phonological alternation in terms of the interaction between phonological constraints and faithfulness constraints. Thus, if the relevant faithfulness constraints are top-ranked (i.e., faithfulness constraints » phonological constraints), the phonological constaints typically do not show any activity. With the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 217 opposite ranking (i.e., phonological constraints » faithfulness constraints), the phonological constraints are triumphant, and faithfulness of the surface form to the underlying representation must suffer, which leads to phonological alternation. Part 2: McCarthy & Prince (1995) Expanding on Prince & Smolensky (1993) and McCarthy & Prince (1993), McCarthy & Prince (1995) propose a model of constraints on faithfulness: it extends the domain of reduplicant-base identity to that of phonological output-input faithfulness, under the notion of correspondence. The Correspondence Theory involves correspondence between stem and base, between base and reduplicant, and between stem and reduplicant, as portrayed in the following diagram: (241) Full Model (McCarthy & Prince, 1995:4) Input: /AfR E D + Stem/ I-R F a ith fu ln ess 7 1 1 ^ /h4' I-B F aithfulness Output: R -> B <r B -R Identity Correspondence refers to a relation between an element of one representation and an element of another representation: two elements are in the relation of input-output correspondence only if they are in a one-to-one relation. The definition of correspondence is given in (235): Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 218 (242) Correspondence (McCarthy & Prince, 1995:14) : Given two strings S, and S2 , correspondence is a relation 91 from the elements of S, to those of S2 . Elements a e S , and (3€S2are referred to as correspondents of one another when ocSJtp. Faithfulness constraints assess each candidate pair with its 9S, considering such aspects of faithfulness as the completeness of correspondence in S ( or S2 , the featural identity of correspondent elements in S, and S2 , and so on. The following is a hypothetical illustration in which some input-output correspondences are provided: (243) Hypothetical illustration (McCarthy, 1995: 15) S o m e 1 -0 C o rresp o n d en ts: Input = /p, a, u3 k4 ts a6 / p, Zj u3 k4 ts a6 A fully faithful analysis — perfect 1-0 correspondence p, a, - s - u3 k4 1 5 a6 Hiatus prohibited (by high-ranking ONS), so epenthetic - * • in O has ▲ no correspondent in I. p, u3 k4 1 5 a6 Hiatus prohibited, leading to V-deletion. The segment a in I has no ▲ correspondent in O. p, ^ u31 4 1 5 a6 The k4 in I has a non-identical correspondent in O, for phonological A reasons, b 1 u r k No element of O stands in correspondence with any element in I. A A A A A Typically fatal. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 219 This notion of correspondence is contrary to the earlier OT principle of C ontainm ent, under which all operations in GEN should be structure-building, not structure-changing: (244) (McCarthy & Prince, 1993:20) Containment: No element may be literally removed from the input form. The input is thus contained in every candidate form. Under the assumption of C ontainm ent, phonologically deleted segments are present in the output, but unparsed syiiabically, employing the notion of S tra y E rasure in Steriade (1982). McCarthy & Prince (1995:20) attribute the consequences of Correspondence Theory to the observation that deleted elements simply cannot play a role in determining the performance of output structures. It follows then that there is no need to restrict constraints to seeing only p a rse d segments. Three types of input-output faithfulness constraints can be formulated under Correspondence Theory, as shown in (245): (245) Faithfulness constraints (McCarthy & Prince, 1995:16) a. The Max-IO Constraint Family : Every segment in the input has a correspondent in the output. (I.e., there is no phonological deletion.) b. The DEP-IO Constraint Family: : Every segment in the output has a correspondent in the input. (I.e., there is no phonological epenthesis.) c. The lDEN(F)-IO Constraint Family Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 220 : Let a be a segment in the input and ( 3 be a correspondent in the output. If a is [yF], then P is [yF]. (I.e., underlying [yF] cannot change to [-yF], assuming full specification.) The constraint Max-IO family reformulates PARSE-segment in Prince & Smolensky (1993) and other OT work. A violation of Max-IO is incurred when each input element does not have an output counterpart. Therefore, MAX-IO will prohibit phonological deletion. The DEP-IO constraint approximates the function of FILL in Prince & Smolensky (1993). When each output element does not have an input element, DEP-IO assesses a violation for it. So, DEP-IO will prohibit phonological epenthesis. To take an example, MAX-IO(segment) forbids the deletion of segments. DEP-IO(segment) forbids the epenthesis of segments. When faithfulness is not violated, the output representation is identical to the input in all respects. However, when higher constraints override faithfulness, alternations take place. Epenthesis occurs when DEP-IO is crucially dominated by certain higher constraint, and deletion arises in case MAX-IO is crucially dominated by some higher constraint. The lDEN(F)-IO constraints require that correspondent segments be featurally identical to one another. Crucial domination of one or more Iden constraints leads to featural disparity, and phonological alternation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 221 References Ahn, S.-C. (1985) The Interplay o f P h o n o lo g y a n d M orphology in K o rea n . Ph.D. dissertation, University of Illinois. Abramson, A. S. (1972) “Word-final stops in Thai” In J.G. Harris & R.B. Noss (Eds.), Thai P honetics and P h onolo g y, 1-7. Bangkok: Central Institute of English Language. Archangeli, D. (1984) Underspecification in Y aw elm ani P ho n o lo g y a n d M o rp h o lo g y. Ph.D. dissertation, MIT. ____________ & D. Pulleyblank (1994) G ro u n d ed P honology. Cambridge, MA: MIT Press. Baek, E.-J. (1991) “Unreleasing in Korean: a phonetic explanation”, H arvard S tu d ie s in K o rea n L inguistics IV , 29-35. Seoul: Hanshin. Bailey, C.-J. N. (1970). “Toward specifying constraints on phonological metathesis”, L in g u istic Inquiry I, 347-349. Butcher, A. (1981) “Aspects of speech pause: phonetic correlates and communicative functions”, Institut fur Phonetic, Universitat Kiel. Cho, Y.-M. Y. (1988) “Korean assimilation”, P ro ceed in g s fro m W C C F L 7,41-52. ___________ (1990) Parameters o f C onsonantal A ssim ilation. Ph.D. dissertation, Stanford University. Chomsky N. & M. Halle (1968) The S o u n d Pattern o f English. New York: Harper and Row. Choy, Hyen-Pay. (1959) W uli-m alpon (A grammar of our language). Rev. ed., Cengum- sa, Seoul. Clements, G.N. (1976) “Palatalization: linking or assimilation?”, C L S 12, 96-109. ____________ (1985) ‘The Geometry of Phonological Features”, P h o n o lo g y Y ea rb o o k 2 , 225-252. ____________ (1991) “Place of articulation in consonants and vowels: a unified theory”. In W orking Papers o f the Cornell P h o n etics Laboratory, vol. 5, 77-123. Ithaca: Cornell University. ____________ & E. Hume (1995) “The internal organization of speech sounds”, In J. Goldsmith (eds.) The H andbo o k o f P honological Theory, 245-306. Oxford: Basil Blackwell. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 222 Cohn, A. C. (1990) “Phonetic and Phonological Rules of Nasalization”, U C L A W o rkin g Papers in P honetics 76. Cole, J. S. & C. W. Kisseberth. (1993) “An optimal domains theory of harmony”, Rutgers Optimality Archaive 22. Dresher, B. E. (1989). “Comments on guttural phonology”, Presented at the MIT Conference on Feature Geometry. Goldsmith, J. (1976) A u to se g m en ta l P h o n o lo g y. Ph.D. dissertation, MIT. Goodenough, W. H. (1963) “The long or double consonants of Trukese”, The P roceedings o f the N in th P acific S cien ce C ongress, vol. 3, 77-86. Gorecka, A. (1989) P h o n o lo g y o f A rticu la tio n . Ph.D. dissertation, MIT. Hamp, E. P. (1985) “Regressive consonant assimilation in Eastern Inuktitut: toward a sociolinguistic approach”, International Jo u rn a l o f A m erican L in g u istic s, vol. 51:4, Chicago, Illinois. Han, J.-I. (1992) “On the Korean tense consonants and tensification”, C L S 28, 206-223. Harrison, P. T. (1984) “Segmental quantity in Mokilese: a synchronic and diachronic study”, In Byron W. Bender (eds.) S tu d ies in M icronesian L in g u istic s, 375-401. P acific L inguistics, C-80. Hayes, B. (1986) “Inalterability in CV phonology”, Language 62(2), 321-351. ________ (1989) “Compensatory lengthening in moraic phonology”, L in g u istic In q u iry 20,253-306. Huh, Woong. (1965) K uke U m w unhak (K o rea n P honology). Seoul: Cengkumsa. Hooper, J. (1976). A n Introduction to N atural Generative P h o n o lo g y, New York: Academic Press. Houlihan, K. & G. Iverson (1979) “Functionally-constrained phonology”, in Daniel Dinnsen (ed.) C urrent A pproaches to Phonological T h eo ry. Bloomington and London: Indiana University Press. Hume, E. (1990) “Front vowels, palatal consonants and the rule of umlaut in Korean”, N E L S 20,230-243. ________ (1992) F ront V ow els, Coronal C onsonants and their Interaction in N o n lin ea r P honology. Ph.D. dissertation, Cornell University. Inkelas, S. (1994) “The consequences of optimization for underspecification”, Rutgers Optimality Archive 40. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 223 Ito, J. (1986) Syllable T heory in P rosodic P h o n o lo g y. PhD Dissertation, University of Massachusetts, Amherst. (1989) “A prosodic theory of epenthesis”, N a tu ra l Language a n d L in g u istic T heory 7, 217-259. & A. Mester. (1994) “Reflections on CodaCond and Alignment”, Ms. (linguistics department) UCSC. Iverson, G. & K.-H Kim (1987) “Underspecification and hierarchical feature representation in Korean consonantal phonology. In A. Bosch, B.Need, and E. Schiller (eds.), C L S 23: Parasession on A u to seg m en ta l and M etrical P h o n o lo g y , 182-198. Chicago: CLS. Jun, Jonho. (1992) “Metrical weight consistency in Korean and padal reduplication”, Ms. (linguistics department) UCLA. _________ (1995) Perceptual a n d A rticulatory F actors in P lace Assim ilation: A n O ptim ality T heoretic A pproach. Ph.D. dissertation, UCLA. Kahn, D. (1976) Syllable-B ased G eneralizations in English P h o n o lo g y, Ph.D. dissertation, University of Massachusetts. Kaun, A. R. (1993). “Coronal underspecification and theories of underspecification”, In D. Silverman and R. Kirchner (eds.), U C LA O ccasional Papers in L in g u istics N u m b e r 13; P apers in P honology. UCLA, Los Angeles, California. Kaye. J. (1989) P honology: A C ognitive View. Hillsdale: LEA Publishers. Kim, C. -K. (1991) “Pukhanei munwhae umun jenkuei sungkwa” (Report on the study of North Korean cultural words), U rimalei U m un (Korean P h o n o lo g y), Seoul: Taehaksa. Kim, K. -H. (1987) The P honological R epresentation o f D istinctive Features: K orean C onsonantal P honology. Ph.D. dissertation, University of Iowa. Kim, Chin-W. (1988) "On the Autonomy of the Tensity Feature in Stop Classification," S o jo u rn s in Language 1, 12-33. Seoul: Tower Press. Kim, Chin-Wu. (1971) “Regularity of the so-called irregular verbs in Korean”, In C.W. Kisseberth, (ed.), Studies in G enerative P h o n o lo g y, 104-113. Edmonton, Canada, Linguistic Research Inc. Kim, Sun-Hee. (1989) “The behaviour of /h i in Korean”, H arvard S tu d ies in K orean L in g u istics 111, 117-126. Seoul: Hanshin. Kim-Renaud, Y. K. (1974) K orean C onsonantal P h o n o lo g y. Ph.D. dissertation, University of Hawaii. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 224 . ____________ (1986) S tu d ies in K orean L inguistics. Seoul: Hanshin Publishing Co. Kiparsky, P. (1982) “Lexical phonology and morphology”, In I.S. Yang (ed.), L inguistics in the M o rn in g C alm , 3-91. Seoul: Hanshin. __________ (1985) “Some consequences of lexical phonology”, P h o n o lo g y Y ea rb o o k 2, 85-138. Kisseberth, C. (1970) “On the functional unity of phonological rules”, L in g u istic Inquiry 1, 291-306. Kuehn, D. P. and K. L. Moll (1976) “A Cineradiographic study of VC and CV articulatory velocities”, Jo u rn a l o f P honetics 4,303-320. Leben, W. (1973) Sup ra seg m en ta l P honology, Ph.D. dissertation, MIT. (1978) “The representation of tone”, In V. Fromkin (ed.), T one: A L inguistic S u rvey, 177-220. New York: Academy. Lee, B. -G. (1972) U nderlying S eg m en ts in Korean P h o n o lo g y, Ph.D. dissertation, Indiana University, Bloomington, Indiana. __________(1982) “A well-formedness condition on syllable structure”, In the Linguistic Society of Korea (eds.) L inguistics in the M orning Calm 1, 489-506. Seoul: Hanshin. Levin, J. (1985) A M etrica l T heroy o f Syllabicity. Ph.D. dissertation, MIT. Lieber, R. (1981) O n the O rganization o f the Lexicon. Ph.D. dissertation, MIT. Lodge, K. (1992) “Assimilation, deletion paths and underspecification”, Journal o f L in g u istic s 28, 13-52. Lombardi, L. (1991) L aryngeal F eatures a n d Laryngeal N eutralization. Ph.D. dissertation, Univeristy of Massachusetts at Amherst. ___________ (1994) “Laryngeal neutralization and alignment”. Ms. (linguistics department) University of Maryland, College Park. Maddieson, I. (1984) Patterns o f S o u n d s. Cambridge University Press, Cambridge: Cambridge. Martin, Samuel E. (1992) A R eference G ram m ar o f K orean, Charles E. Turtle Co.: Tokyo. McCarthy, J. (1986a) “Prosodic morphology”, Ms. (linguistics department) University of Massachusetts at Amherst and Brandeis University. ____________ (1986b) “OCP effects: gemination and antigemination”, L in g u istic Inquiry 17.2, 207-263. Reproduced with permission of the copyright owner. Further reproduction prohibited w ithout permission. 225 ____________ (1988) “Feature geometry and dependency: a review”, Phonetica 4 3 , 84- 108. ____________ (1994) “The phonetics and phonology of Semitic pharyngeals”, In P. Keating (ed.), P honological Structure and Phonetic F o rm . Papers in Laboratory P h o n o lo g y III., 191-233. Cambridge University Press. McCarthy, J. and A. Prince (1993) P rosodic M orphology I: C onstraint Interaction and S atisfaction. Ms. (linguistics department) University of Massachusetts at Amherst and Rutgers University. ________________________ (1995) “Faithfulness and reduplicative identity”, Rutgers Optimality Archive 60. McCarthy, J. and A. Taub (1992). Review of C. Paradis and J-F Prunet, (eds.) (1991). P h o n o lo g y 9.2, 363-370. Mohanan, K.P. (1993) “Fields of attraction in phonology”, In J. Goldsmith (ed.) T he L ast P honological Rule: R eflectio n s on Constraints a nd D erivations, 61-116. The Universirty of Chicago Press: Chicago & London. Myers, S. (1995) “OCP effects in optimality theory”, Ms. (linguistics department) University of Texas, Austin. Odden, D. (1986) “On the Obligatory Contour Principle”, L a n g u a g e 62, 353-383. Ohala, J. J. (1990) “The phonetics and phonology of aspects of assimilation”, In John Kingston and Mary Beckman (eds.) Papers in Laboratory P h o n o lo g y I: B e tw een the G ram m ar a n d P h y sics o f S p eech , 258-338, Cambridge University Press: Cambridge. Padgett, J. (1991) Stricture in Feature G eom etry. Ph.D. dissertation, University of Massachusetts at Amherst. Paradis, C. (1989) “On constraints and repair strategies”, The L in g u istic R eview 6, 71-97. __________(1990) “Focus”, in Gere Configurational Constraints. C urrent A p p ro a ch es to A frican L in g u istics vol. 7, in John P. Hutchison and Victor Manfredi (ed.), Dordrecht: Foris. Paradis, C. and J-F Prunet (eds.) (1991) Phonetics and P honology: The Special S ta tu s o f Coronals: Internal and External Evidence. Phonetics a n d P h o n o lo g y 2 , Academic Press. Prince, A. & P. Smolensky (1993) O ptim ality Theory: C onstraint Interaction in G enerative G ram m ar. Ms. (linguistics department) Rutgers University & University of Colorado at Boulder. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 226 Pulleyblank, D. (1988) “Vocalic underspecification in Yoruba’, L in g u istic In q u iry 19, 233- 270. Rosenthall, S. (1994) V ow el/G lide A ltern a tio n in a Theory o f C on stra in t In tera ctio n, Ph.D. disseration, University of Massachusetts at Amherst. Sagey, E. (1986) The R epresentation o f Features and R elations in N o n lin ea r P h o n o lo g y. Ph.D. dissertation, MIT. Schein, B. & D. Steriade (1986) “On geminates”, L inguistic In q u iry 17: 691-744. Selkirk, E. O. (1984) “On the major class features and sylllable theory”, Language S o u n d Structure, in Mark Aronoff and Richard T. Oerhle (eds.), Cambridge: MIT. Selkirk, E. O. (1991) “A two-root theory of length”, In Dunlap, E. and J. Padgett (ed.), P a p e rs in Phonology. University of Massachusetts Occasional Papers in Linguistics 14, GLSA, University of Massachusetts, Amherst. Sherer, T. (1993) P rosodic P honotactics. Ph.D. dissertation, University of Massachusetts at Amherst. Shimizu, K. (1971) Com parative Jukunoid: A n Introductory S u r v e y . Ph.D. dissertation, University of Ibadan. Sohn, H. S. (1987) U nderspecification in the Korean P h o n o lo g y. Ph.D. dissertation, University of Illinois. __________ (1991) “Aspects of feature geometry and underspecification in Korean phonology”, H a rva rd S tu d ies in K o rea n L inguistics TV, 197-213. Seoul: Hanshin. Steriade, D. (1982) G reek P rosodies a n d the N ature o f Syllabification, Ph.D. dissertation, MIT. __________ (1987) “Redundant values” , In A. Bosch, B. Need, and E. Schiller (eds.), C L S 23: Parasession on A u to seg m en ta l and M etrical P h o n o lo g y, 339-362. Chicago: CLS. __________ (1991) “Closure, release and nasal contours”. Ms. (linguistics department) UCLA. Suh, P. W. (1984) “On the vowel system of southeastern dialects”, Literature a n d L a n g u a g e, vol. 3, Seoul. To, S.-H. (1983) “Umlaut in Choong-nam dialect”, K orean L in g u istics, vol. 1, Seoul. Vennemann, T. (1972) "On the theory of syllabic phonology," L in g u istisch e B erichte 18.1-18. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 227 Webb, C. (1982) “A constraint on progressive consonantal assimilation”, L in g u istics 20, 309-321. Williams, E. (1981) “On the notions 'Lexically Related’ and 'Head of a Word”, Linguistic Inquiry 12. Yip, M. (1988) The Obligatory Contour Principle and phonological rules: a loss of identity”, L inguistic In q u iry 19.1, 65-100. (1989) “Feature geometry and coocurrence restrictions. P h o n o lo g y 6.2, 349-374. ________ (1991) “Coronals, consonant clusters and the coda condition”, in Paradis and Prunet (eds.), The S p ecia l S ta tu s o f C oronals, 61-78. Academic Press. Yoon, Y.-B. (1992) “Closure duration and the three series of Korean stops”, E onehakkw a Inchi (L inguistics a n d C ognition). Seoul: Hankuk Munwhasa. You, S.-H. (1992) “Expressives and ideophones in Korean”, W o rkin g Papers in Linguistics, University of Hawaii. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Causativity in Korean: Syntactic causative, control, and purpose

PDF

Borrowing and borrowability

PDF

Aspects of questions in Japanese and their theoretical implications.

PDF

Beyond words and phrases: A unified theory of predicate composition

PDF

Andrej Belyj's "Petersburg" and James Joyce's "Ulysses": A comparative study

PDF

Conditions on null objects in Basque Spanish and their relation to leismo and clitic doubling

PDF

'Beowulf' and 'the hobbit': elegy into fantasy in j. R. R. Tolkien's creative technique

PDF

Arthur Machen’s treatment of the occult and a consideration of its reception in England and America

PDF

A rhetoric of the short story: A study of the realistic narratives of Flaubert, Maupassant, Joyce, and Hyon Chin'gon

PDF

A case grammar of the parker manuscript of the "Anglo-Saxon chronicle" from 734 to 891

PDF

A comparative study of focus constructions

PDF

Justus Georg Schottel and linguistic theory

PDF

Antigone: A study in critical method

PDF

Intonation Of Single-Word Utterances In Japanese: An Acoustic Phonetic Study

PDF

"Emilia Galotti" and its aesthetic response: A case study for the reader-response criticism

PDF

A critical edition of the Escorial manuscript of 'historia de los indios de la neuva espana' of fray toribio de benavente (motolinia) (spanish text)

PDF

"A broken bundle of mirrors": Identity in the work of John Barth

PDF

A historical study of the stage directing theory and practice of David Belasco

PDF

Controlling myths of the new religious right: A narrative analysis of Pat Robertson's 1988 presidential campaign

PDF

Communicative rights and responsibilities in an East Los Angeles barrio: An analysis of epistemic modal use

Asset Metadata

Creator Lee, Sechang (author)

Core Title Correspondence and faithfulness constraints in optimality theory: A study of Korean consonantal phonology

Degree Doctor of Philosophy

Degree Program Linguistics

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag language, linguistics,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Gorecka, Alicja (committee chair), Comrie, Bernard (committee member), Schein, Barry (committee member), Woodard, Roger (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c17-267215

Unique identifier UC11353544

Identifier 9733085.pdf (filename),usctheses-c17-267215 (legacy record id)

Legacy Identifier 9733085.pdf

Dmrecord 267215

Document Type Dissertation

Rights Lee, Sechang

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA