UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Colloquial forms in Tokyo Japanese Katsura, Yasuko 1994

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

Item Metadata


831-ubc_1994-0555.pdf [ 4.74MB ]
JSON: 831-1.0087410.json
JSON-LD: 831-1.0087410-ld.json
RDF/XML (Pretty): 831-1.0087410-rdf.xml
RDF/JSON: 831-1.0087410-rdf.json
Turtle: 831-1.0087410-turtle.txt
N-Triples: 831-1.0087410-rdf-ntriples.txt
Original Record: 831-1.0087410-source.json
Full Text

Full Text

COLLOQUIAL FORMS IN TOKYO JAPANESE by YASUKO KATSURA B.A., Sophia University, 1972 Diploma, The University of British Columbia, 1989  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS in THE FACULTY OF GRADUATE STUDIES (Department of Asian Studies) We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA September 1994 © Yasuko Katsura, 1994  In presentin g thi s thesi s i n partia l fulfilmen t o f th e requirement s fo r a n advance d degree a t th e Universit y o f Britis h Columbia , I agre e tha t th e Librar y shal l mak e i t freely availabl e fo r referenc e an d study . I furthe r agre e tha t permissio n fo r extensiv e copying o f thi s thesi s fo r scholarl y purpose s ma y b e grante d b y th e hea d o f m y department o r b y hi s o r he r representatives . I t i s understoo d tha t copyin g o r publication o f thi s thesi s fo r financia l gai n shal l no t b e allowe d withou t m y writte n permission.  (Signature)  Department of / % / > /  7 S-htPt'CS  The Universit y o f Britis h Columbi a Vancouver, Canad a  Date Oct /Cf  DE-6 (2/88 )  f^fCl  ii  ABSTRACT This thesis discusses the interaction of prosodic constituency with segmental and tonal analyses. Thi s is done through the examination of two phonological reduction processes, Nasal Spreading and Vowel Deletion, in colloquial Tokyo Japanese. This thesis consists of two chapters. Chapte r I analyzes Nasal Spreading, a nasal assimilation process, which occurs when the negative suffix /nai/ is attached to verb roots. I n order to explain this process, first, the segmental conditions for Nasal Spreading are examined. Amon g verbs, onl y /r/-final roots trigger assimilation. Thi s is explained in terms of Rhotic Underspecification (Meste r and Ito 1989, and others), Featur e Class Theory (Padget t 1994), and Feature Licensing (Ito , Mester, and Padgett 1994), demonstrating that the target /r/ lacks specification under the root node.  The prosodic account analyzes Nasal Spreading as a syllabification process. Nasa l Spreading adds one mora; whereas the standard form, a general form in /nai/-suffixation, epenthesizes /a/ to break up the consonant cluster, by adding a mora and a syllable to the output string. Th e prosodic account also deals with the interaction with tonal system. A  disyllabic  requirement on the derived base can be explained with respect to a violation of tonal agreement. Th e tone melodies in the colloquial and standard forms must agree; however, the  iii constraints on Accent Shift or Initial Lowering require minimally disyllabic strings [ @ @] in the base for the agreement.  Chapter II discusses Vowel Deletion. I n Vowel Deletion as well the interaction of the three types of structure is discussed. Segmentally , the type of compounds and vowels for deletion are specified. Thi s is examined under the Theory of Underspecification (Archangel i 1988), and/or under the sonority scale approach (Princ e and Smolensky 1993) . Th e prosodic analysis deals with domain specification. Th e lexical tone in the second constituent of the compound blocks deletion; because it blocks formation of the single tone domain which would be the site for Vowel Deletion. Onl y one tonal peak is allowed in a single tone domain. Th e blocker splits a string of compound into two; Vowel Deletion cannot apply between two domains.  In Nasal Spreading and Vowel Deletion, segmental, prosodic, and tonal principles interact to determine the optimal output strings. Th e intricate interplay is accounted for with the Optimality Theoretic approach.  iv TABLE OF CONTENTS Abstract i  i  Table of Contents i  v  Introduction 1 Chapter I: Nasal Spreading 6 Introduction 6 1. Preliminarie s 7 1.1. Inventorie s 8 1.2. Syllabl e Structure and Mora Counting 8 2. Segmenta l Analysis 1 2.1. Standar d Form and Colloquial Form 1 2.2. Ver b Classification 1 2.3. Rhoti c Underspecification 1 2.4. Decompositio n of Feature Geometry 1  1 1 3 7 9  3. Prosodi c Analysis 2 3.1. Tw o Types of Syllabification 2 3.2. Monomorai c Roots 2 3.3. Superheav y Syllables in Nasal Spreading 3  6 6 8 1  4. Ton e Melodies 3 4.1. Ton e Systems of Tokyo Japanese 3 4.2. Accen t Assignment 4 4.3. Lexica l Tones 4 4.4. Agreemen t in Tone Melodies 4 4.5. Accen t Shift 5 4.6. Initia l Lowering 5  5 6 0 3 7 0 3  5. Optimalit y Theory 5 5.1. Basi c Component 5 5.2. Nasa l Spreading In Optimality Theory 5  5 5 8  5.3. Syllabl e Templates 6  3  6. Conclusio n 7  2  Notes 7  4  Chapter II: Vowel Deletion 8  1  Introduction 8  1  V  1. Environmen t for Vowel Deletion 8 1.1. Compoun d Verbs 8 1.2. Vowel s for Deletion 9 1.3. Sonorit y Scale and Underspecification 9  3 3 0 1  2. Tona l Interaction 9 2.1. Accente d vs. Unaccented Words 9 2.2. Stres s Clash and the OCP 9 2.3. Intonationa l Curves 10 2.4. Gree k Vowel Degemination 10  5 5 7 3 5  3. Formatio n of Tone Groups 10 3.1. Accente d Auxiliary Verb /oru/ 10 3.2. Mino r Phrases and Minimal Minor Phrases 11  6 7 1  3.3. Prosodi c Domains vs. Morphological Domains 11  6  3.4. Lugand a Low Tone Deletion 11  8  4. Summar y 12  0  5. Conclusio n 12  3  Notes 12  4  Conclusion 13  0  References  132  1 INTRODUCTION This thesis introduces and analyzes colloquial forms of Tokyo Japanese in a current theoretical framework. Th e central hypothesis is that prosodic constituency influences both segmental and tonal processes. Tw o phonological changes, Nasal Spreading and Vowel Deletion, are examined to illustrate the interaction among the above three processes. Th e basic hypothesis is that in phonological formations, in order to derive correct forms, the interaction of the above three must be considered. Fo r example, segmental analysis alone cannot account for all the complexities of each derivation. Th e above three interact with each other and condition the phonological changes.  Chapter I analyzes Nasal Spreading. Nasa l Spreading occurs in the negative construction of verbs. I t is derived by suffixing the negative morpheme /nai/ to verb roots. Spreadin g takes place between the verb root and /nai/ when a consonant cluster is produced by the suffixation. /n / of /nai/ spreads to the left, but it targets only /r/-final roots. I f the root-final sound is other than /r/, spreading does not apply.  (1) root+/nai / colloquia kawar-nai kawaN-na kawak-nai *  l neg. i '(It  kawaN-nai '(It  ) does not change.' ) does not dry.'  2 This must be first analyzed and formulated in terms of segmental features, by adopting an approach such as Rhotic Underspecification (Meste r and Ito 1989), Feature Class Theory (Padgett 1994), or Feature Licensing (Ito , Mester, and Padgett 1994) .  The prosodic analysis is as follows: spreading adds an extra mora into the string as a mora augmentation process; this is done via syllabification (cf . Ito 1989). Especially , it is significant when we compare another form derived through /nai/suffixation, which is called the standard form. Th e standard form is derived by epenthesizing /a/ to break up the consonant cluster. Th e crucial point here is that spreading makes /n/ a coda nasal /N/; /N/ is syllabified with the preceding syllable. On the other hand, /a/ cannot be linked with any existing syllable; instead, /a/ must create a syllable, by taking /r/ as its onset. I n other words, Nasal Spreading maximizes the syllable size to bimoraic, while /a/-epenthesis derives an additional light syllable. Thi s cannot be explained without considering prosodic structure.  (2) /wakar+nai  / '(I ) do not understand'  /a/-epenthesis @ fl @  _..,.._..  ,__,..  |/l V I AT/  @ @  /m /m m  waka na i wakar  /mm  1 I / l!  A nai  3 Heavy monosyllabic root verbs derive superheavy syllables through Nasal Spreading. Thi s also should be analyzed in terms of prosodic constituency: /N/ must be syllabified with the preceding syllable, however, the long syllable already contains two moras--three moras cannot be linked under a single syllable node. Prosodi c analysis does not necessarily account for all the cases that are ruled out in Nasal Spreading. Fo r example, light monosyllabic root verbs satisfy all the requirements for Nasal Spreading in terms of segmental and prosodic conditions; however, spreading does not apply. W e need to consider the role of tone in this specific case. Constraint s in the tonal system, such as Accent Shift or Initial Lowering have effects in cases like this.  A three dimensional approach accounts for the intricate cases and specifies the right environments for the changes. Th e three classes of structural constraints are organized in terms of Optimality Theory (OT) . O T enables us to capture the whole picture of the constraints in the changes. Fo r example, in OT, the difference between Nasal Spreading and /a/-epenthesis is captured by a set of constraints, including for example, Recoverability of Mora (RecM ) and Recoverability of Syllable (Rec@) (Pulleyblan k 1993, Archangeli and Pulleyblank 1993 (b), Ito, Mester, and Padgett 1993, and others).  (3) Rec F (Recoverabilit y of F-element): If [.. .00 ••]output then [.. .ex. ..] input. A n F-element (featur e or node) that is present  4 in an output form is also present in the output. RecM and Rec@ belong to the faithfulness family of constraints that deals with tight input and output relations. Nasa l Spreading violates RecM by adding a mora to the output which is not included in the input. /a/-epenthesi s violates both RecM and Rec@, by inserting an additional syllable.  Vowel Deletion in Chapter II is analyzed on the same analytical grounds. Vowe l Deletion deletes one of W a t a morpheme boundary derived via compounding. Vowe l Deletion occurs in a very specific class of compound verbs; only the compounds that constructed of 'verb(GER)+aux. ' are subject to deletion. Vowels also must be specified in terms of sonority or features in 'sonority prominence' (Archangel i and Pulleyblank in press, Prince and Smolensky 1993) and/or the Theory of Underspecification (Archangel i 1988).  On the basis of segmental specifications, prosodic structure must be considered to account for deletion. Vowe l Deletion applies domain-internally, but not domain-externally. Domain s must be defined in terms of prosody. However , morphological domains and prosodic domains do not necessarily match in some cases (Inkela s 1989, Ishihara 1991). Vowe l Deletion is blocked if the second member of compound is accented. Th e lexical tone in the second member of a compound blocks the formation of a  5 single tone domain (although , morphologically, the two constituents form a single phrase). (4) a. masite-iru [masite-ru  ] 'b  e increasing'  H b. masite-aru [masite][aru  I  ] 'hav e been added in advance'  Vowel Deletion applies in (4a) , but does not apply between two domains as in (4b) ; the lexical high tone (H ) in /aru/ is the blocker for deletion.  The intricate outputs are shown to follow from the interaction of the constraints on each kind of structure. Th e three dimensional approach accounts for all the cases. Th e Optimality Theoretic approach is adopted in both cases, Nasal Spreading and Vowel Deletion, showing that constraints are violable, and must be selected and ranked.  6 I. Nasa l Spreading Introduction  This chapter focuses on Nasal Spreading (nasa l regressive assimilation), which is the colloquial form of the negative verbal construction (se e Alfonso 1966, Miyara 1988).l Th e purpose of this section is three fold: (i ) formulation of the exact condition on Nasal Spreading in terms of segmental structure, (ii ) illustrate that Nasal Spreading is a syllabification process which maximizes the syllable template to bimoraicity. (iii ) show that tonal constraints interact with Nasal Spreading. Nasa l Spreading applies when the negative suffix /nai/ is attached to verb roots; /n/ of /nai/ spreads backward and is syllabified as a coda nasal /N/ under the preceding syllable node. Onl y the verbs that have /r/-final roots are subject to spreading.  This chapter is organized into five sections: preliminaries, segmental analysis, prosodic analysis, tonal interaction, and Optimality Theory. A s a preliminary, I will provide some basic information on Japanese phonology. Preliminar y information will include vowel and consonant inventories, syllable types, mora counting processes, and the coda condition in Japanese. I n a segmental analysis, I will show that the verb root final /r/ is  7 the only target for spreading. Thi s will be examined by adopting Rhotic Underspecification (Meste r and Ito 1989), Feature Class Theory (Padget t 1994), and Feature Licensing (Ito , Mester, and Padgett 1994), claiming that /r/ has a special status among Japanese consonants and lacks a place node. I n a prosodic analysis, I will focus on syllabification and well-formedness conditions. Th e difference between Nasal Spreading and /a/epenthesis will be shown by comparing the two cases of syllabification: Nasal Spreading adds a mora under the preceding syllable while /a/-epenthesis creates a light syllable. Superheavy syllables cannot be derived, since it violates the well-formedness conditions of the language. Th e tonal account demonstrates that tonal constraints restrict the syllable number in the derived base to minimally disyllabic. Finally , these are compared and explained by adopting Optimality Theory.2  1. Preliminarie s  In this section, the basic phonology of Japanese will be introduced: vowel and consonant inventories, syllable structure, mora counting, and the Coda Condition are included in the discussion. Th e above information is necessary for the segmental, prosodic, and tonal analyses. Th e segmental analysis requires the above information, for example, in order to define the status of /r/ and /n/ in Nasal Spreading, or the status of /i/ in Vowel Deletion.  8 Prosodic structure is expressed in a prosodic hierarchy. Each prosodic level in the hierarchy is defined on the basis of the basic phonological processes of Japanese. Th e tonal account also requires the above information; for example, the Tone Bearing Unit (TBU ) is in Japanese is defined as a sonorant mora. The information presented here is minimum and basic to the analysis. Detaile d discussion will be added elsewhere in this thesis. 1.1. Inventorie s  The following vowel and consonant inventories are taken from Shibatani (1990) ; see also Grignon (1984) . (1) Vowe  l Phonemes  3  i u e o  (2)  Consonant Phonemes P b  m w  t d s z r n  k g  h  1.2. Syllabl e Structure and Mora Counting  Japanese syllable types and mora counting are shown in (3)  9 and (4) . Thes e items are necessary for determining the Tone Bearing Unit (TBU ) in Japanese. I n Nasal Spreading, the syllable size is maximized by including the coda /N/. I n (3), marked syllables and unmarked syllables are contrasted; unmarked syllables occur only exceptionally in manifestations such as superheavy syllables. Superheav y syllables are limited to the following cases: recent loan words, onomatopoeia, verb past and gerund forms, some colloquial (ver y casual) and emphatic expressions, and interjection. A s to the external evidence for the markedness (o r illicit) status of superheavy syllables, language games or speech errors often do not derive superheavy syllables (se e Kubozono 1989, Tateishi 1989, Katada 1990). Syllables in (3 ) may be formed without any onset.  (3) Japanes e Syllable Types (cf . Poser 1990, Ito and Mester 1993) a. Unmarked Syllables (o r licit syllables) (C)V m (C)W so  e 'eye o 'layer  (C)VC mat-t (C)VN ha  ( O W N too  ' o  a 'waited N 'seal  b. Marke d Syllables (C)WC koot-t  ' e  ' at-t ' o  4  'drawing o 'king  ' '  a 'met ' N 'sound  '  5  a 'froze ' oot-t N 'tone  '  a 'covered '  10 Japanese contrasts vowel length: one mora is assigned to a short vowel, while two moras are assigned to a long vowel. Consonant length is also contrasted; one mora is assigned to a coda consonant (Rim e Projection). Th e following words show mora counting (It o 1990, Ito, Kitagawa, and Mester 1992).  (4) Mor a Counting  6  a. short vowel /ma/ 'pause'  b. long vowel /too/ 'tower'  /m a/ m @  c. coda cons. /hoN/ 'book' /h o N/ \m m @  geminate cons, /koppu/ 'glass /k o p u/  'Mi  \m m\m @  A restriction on coda consonants is called the Coda (Place ) Condition or Filter, as has been extensively studied in various articles (It o 1988, Yip 1991, Ito and Mester 1993, Ito, Mester and Padgett 1993, Ito and Mester 1994). Th e following is a summary of their research findings. Amon g all the consonants presented in (2), only the following obstruents /p, t, c, k, s/ or nasal /N/ can occur in the coda position; a mora is assigned to the coda consonants. I n exceptional cases, such as recent loans, voiced obstruents /b, d, j, g, z/ may appear in the coda  11 position: for example, /beddo/ 'bed' or /baggu/ 'bag'. However , the voiced obstruents /d/ and /g/ tend to be merged with their voiceless counter parts /t/ and /k/ as /betto/ or /bakku/, respectively. Th e coda obstruents must be doubly linked with the following onset. Th e coda nasal /N/ can be geminated, or partially geminated, linked with the place node of the following onset sound. Th e nasal /N/ alone among the above coda consonants can occupy the word-final position without gemination.  2. Segmenta l Analysis  This section analyzes the segmental environment for Nasal Spreading. Th e distinction between standard and colloquial forms, and verb classification specify the type of verb roots that can undergo Nasal Spreading. I n order to formulate the environment, the following facts will be considered. Rhoti c Underspecification claims that /r/ in Japanese has no place node. By adopting Feature Class Theory and Feature Licensing, the target and the trigger of spreading are examined, and Nasal Spreading is formulated.  2.1. Standar d Form and Colloquial Form.  Two kinds of forms are derived through the suffixation of /nai/: the standard form and the colloquial form. Th e following set of data displays the two types of negative verbal  12 constructions; in the first column, input strings 'ver b root + /nai/' are presented without any modification. Th e second column represents the standard form, used for example, in the written style. Al l verbs allow the standard form with the /nai/ suffixation. Th e third column represents the colloquial style on which we focus; the colloquial form is derived by Nasal Spreading.7 Comple x suffixation (wit h a string of two or more suffixes; for example, /tabe-naka-tta/ '(I ) did not eat') or compounds (fo r example, /tachi-domara-nai/ '(I ) do not stop') are excluded from the data in order to simplify the analysis.  (5) Negativ e Construction with /nai/ root+/nai/ a . standard neg. b. colloquial neg.  gloss  8  kawar-nai  kawar-a-nai kawa-N-na  i  'change'  wakar-nai  wakar-a-nai waka-N-na  i  'know.'  hayar-nai  hayar-a-nai haya-N-na  i  'be popular1  tomar-nai  tomar-a-nai toma-N-na  i  'stop'  suwar-nai  suwar-a-nai suwa-N-na  i  'sit'  tukur-nai  tukur-a-nai tuku-N-na  i  1  kaer-nai  kaer-a-nai kae-N-na  sasar-nai  sasar-a-nai sasa-N-na  i  'stuck'  mawar-nai  mawar-a-nai mawa-N-na  i  'turn round'  karamar-nai  karamar-a-nai-- karama-N-nai  i  make'  'return'  'get entangled'  In the standard form (5a ) the vowel /a/ intervenes between the two consonants /r/ and /n/. I n the colloquial form (5b) , the  13 nasal /N/ appears instead of /r/. Bot h segments /a/ and /N/ appear between the root and the suffix. 2.2. Ver b Classification  In order to characterize the type of verbs which can undergo Nasal Spreading, we also need to consider the classification of Japanese verbs (McCawle y 1968, Makino and Tsutsui 1986, Vance 1987, and others). Th e classification of verbs is important, because conjugation varies according to the classes to which the verb belongs. O n the basis of the verb classification, both forms, the standard form and that with Nasal Spreading, will be examined.  Japanese verbs are classified into three types: vowel verbs, consonant verbs, and irregular verbs. Th e classification is based on root-final sounds and upon their conjugation. Thre e classes of verbs in their negative construction are presented below:  (6) Vowe l verbs: vowe l verb roots end in a vowel, either /i/ or /e/.9 root standar  d negative  tabe tabe-na  i 'eat  mi mi-na  i 'see  ki ki-na  i 'wear  ' ' '  ne  ne-nai  1  de  de-nai  'get out  ire  ire-nai  'put'  ake  ake-nai  'open'  oki  oki-nai  'get up'  osie  osie-nai  'teach'  sleep'  Consonant verbs: consonan t verb roots end in one of the consonants /k, g, t, s, r, m, b, w/ as in (7) .10  (7)  root  standard negative  kak  kak-a-nai  'write'  kag  kag-a-nai  'smell'  mat  mat-a-nai  'wait'  sas  sas-a-nai  'sting'  tor  tor-a-nai  'take'  yom  yom-a-nai  'read'  yob  yob-a-nai  'call'  moraw  moraw-a-nai  'receive'  Irregular verbs: bot h vowel verbs and consonant verbs conjugate regularly, but irregular verbs do not, as their roots demmonstrate.  (8) roo  t standar  ku ko-na  d negative i 'come  '  15 a(r) na  i 'exis  su si-na  i 'do  t (animat e subject)  11  '  Among the three classes of verb, only consonant verbs contain the intervening /a/. Vowe l verbs do not contain the CC string, and therefore, no segment intervenes at morpheme boundaries. I n (9) /a/ is epenthesized in order to break up the consonant cluster. This is illustrated in (9) as a syllabification process (se e Ito 1989).12 /a / is then syllabified with /r/ and forms a light syllable. Th e presence of the added mora (an d the syllable) is a by-product of syllabification.  (9) /a/-insertion: @ @  13  @  @  k a w a r ]  k a w a r a ]  Unlike the standard construction, Nasal Spreading does not necessarily apply to all the verbs. Conside r the following sets of verbs.  (10) Nasa l Spreading and Verb Roots a. IT  I fina l roots: root Nasa  l Spreading  kawar kawa-N-na  i 'change  mawar mawa-N-na  i 'tur  ' n around1  16 sasar  stick'  sasa-N-nai  other consonant roots: sasow *saso-N-na  i  'invite'  todok *todo-N-na  i  'reach'  mawas *mawas-N-na  i  kagam *kaga-N-na  i  'turn' 'stoop down1  c . vowe l verbs: tabe *tabe-N-na  i 'eat  mi *mi-N-na  i 'loo  ki *ki-N-na  i 'wear  osie *osie-N-na  i 'teach  ' k at' ' '  In (10a) all the roots which end in /r/ undergo Nasal Spreading, while none of the roots in (10b) that have consonant endings other than /r/ contain the nasal /N/. Vowe l verbs in (10c) do not create the consonant cluster which triggers Nasal Spreading. It is thus possible to make the following three deductions from the above observations.  (i) A  CC (C^N ) string at the morpheme boundary must be broken  up in the output form (unles s Nasal Spreading takes place). (ii) Al l consonant verbs contain the epenthetic vowel /a/, (iii) Onl y /r/-final consonant verbs are subject to Nasal Spreading.  Why, we might ask, can /r/-final roots only (an d not others) trigger Nasal Spreading? I n order to answer this question, we should begin by formulating the condition for this spreading. Nasal Spreading spreads the nasality of /n/ in /nai/ backward as a total regressive assimilation. (11) Nasa l Spreading (informal ) C V  r  \ \ \ \ \  ] nai  C V  N  ] nai  i.  2.3. Rhoti c Underspecification  According to Mester and Ito (1989) , and Archangeli and Pulleyblank (i n press), /r / has the special status in the Japanese inventory; /r/ lacks a place node, for example, in the formation of mimetics, whereares other consonants have place specifications. Th e following discussion, taken from Mester and Ito, includin g their data, provides sufficient evidence to support the claim that /r/ has special status among Japanese consonants.  Mester and Ito claim that /r/ is underspecified in the Japanese consonant inventory: /r/ is not marked for [coronal ] at the relevant stage, for example, mimetic formation.14 Th e two independent sources of evidence lend support to the hypothesis that the Japanese /r/ is underspecified:  18 (i) it  s position in the segment inventory,  (ii) it  s appearance as an epenthetic consonant.  The resonant inventories are presented as follows: (12) Japanes e Resonants  15  Liquids: r Glides: w  y  In (12) , place is distinctive in glides; the glides /y/ and /w/ carry specifications for the tongue body features, and the obstruent system in (13 ) displays contrasts in voicing and place.  (13) Japanes e Obstruents lab co r pa l ve l Plosive voiceless  : p  voiced: b Fricative voiceless  : f  t  c  d  g  s  h  k  viced: z As for the second source of evidence, appearance as an epenthetic consonant, w e can take an example from the present and passive forms; here the occurence of the epenthetic /r / is seen in the verbal paradigm.  (14) Appearanc e as an Epenthetic Consonant  19 a. C-fina  l roots:  V-final roots  Present kak-u nom-u  Passive kak-are nom-are  tabe-ru mi-ru  tabe-rare mi-rare  'write' 'drink' •eat'  'see'  Consonant-final verbs take /u/ and /are/ in the present /ru/ and /rare/. A  consonant epenthesis rule inserts /r/ to break up the  vowel hiatus.  2.4. Decompositio n of Feature Geometry  On the ground that /r/ in Japanese lacks feature specifications, Nasal Spreading will be examined under two recent issues: one is Feature Class Theory by Padgett (1994) , and the other is Feature Licensing by Ito, Mester and Padgett (1994) . I t could be said that Feature Classe Theory is an attempt to decompose Feature Geometry. First , the 'traditional ' Feature Geometry (Clement s 1985, and others assumed; taken from Padgett above) is illustrated in (15).  Root = [son]  (15)  Laryngeal [voice] [asp] [glot]  Place  [nasal]  Lab Co r Dor s / \  [ant] [dist] Feature Class Theory (Padgett ) claims that " 'placeness' and  20 'laryngeality' are simply properties of features. W e might say [voice] , [asp], , Lab , Cor . . . (For example, ) Spread (Place ) means to spead a feature iff it is a member of Place." Thi s is based on the following two assumptions.  (i) Constraint s like SPREAD are gradiently violable in the sense given with Optimality Tehory. Th e occasional result: partial class behavior, (ii) Featur e geometry does not naturally capture partial class behavior. Fo r this reason, it can even obscure evidence for certain feature classes, thus failing on its own terms.  Representations under this account will be as follows. "Tre e preserves featural dependencies."  (16)  Root = [son]  [voice] Lab [glot ] Cor [nasal ] Dors [asp] /\  [ant] [dist] By adopting Feature Class Theory, Nasal Spreading is formulated as spreading of the feature [nasal] , instead of spreading feature nodes as illustrated below:  21  (17) r  - n  I  [nas]  The other issue Feature Licensing is developed through the Theory of Underspecification (Archangel i 1984, Mester and Ito 1989) and Grounded Phonology (Archangel i and Pulleyblank in press), and also by adopting Optimality Theory (Princ e and Smolensky 1993, and others). Accordin g to Ito, Mester, and Padgett, the theory claims that "upon consider(ing) the connection, .. . between feature redundancy and feature underspecification, . . we link the two by means of the notion of licensing. Th e hypothesis formulated in (18) , as a principle of universal grammar, is an explicit statement of this connection.  (18) Licensin g Cancellation: I f the specification F implies the specification G, then it is not the case that F licenses G.  For example, given the redundancy implication [son ] r> [voice] , a sonorant segment does not license [voice] . Th e [voice ] feature is licensed when linked to obstruents (19a) , but not when linked to sonorants (19b). "  22 (19) a. license d [voice ] b [d] [n  I I  [voice] [voice  . unlicense d [voice ] ] ]  In Nasal Spreading , Feature Licensing i s modified a s follows: Feature Licensing i s expressed with linking of segments, even without discussing 'licensed ' or 'unlicensed' , i f it is not necessary. Fo r example, lie I i s not linked with Place. Sinc e /r/ is the only lateral among Japanese sonorants, and also latera l i s universally coronal; specification of coronal fo r lateral is redundant. Th e other consonants i n the verb-final position must be linke d with Place features, becuase there is place distinctio n among them . B y combining th e above two issues, Nasal Spreadin g will be captured and formulated with minimal redundancy .  (20) r  - n [nas]  As we have seen, /r / lacks place feature . Onl y the root of /r/ is present. Th e spreader /n / is linked with [nasal] , but i t has no place specification, either; this will be discussed later . The above representation eliminates other verb-final consonant s from the application of spreading.  A limite d number of consonants, /k, g, t , s, r, m, b, w/, can occupy th e verb-final position. Amon g the m only /k , t, s, m/  23 can get gemination in Japanese. Voice d obstruents may get geminated, however, it is limited to special cases such as recent loan words. I n the representations of the above four consonants /k, t, s, m/, Place must be linked; because, place is contrasted in them. (21) k  t  s  I I  dor co  I  r co  m  I  r la  b  Only /r/ is not linked with Place, while other consonantns must have place specifications (licensin g conditions).  (22) Plac  e is not linked: r  Place is linked : k  t  s  m  P P  P  P  I I I I  For example, if one of /k, t, s, m/ occupies the root-final position, spreading is blocked; becuase it must be linked with a place feature. I n (23b) the velar stop /k/ is linked with [dorsal]; spreading cannot apply.  (23) Nasa l Spreading a.  r -n b [has] do  . * k - n r [nas]  In the formulation of Nasal Spreading in (23a) , the spreader  24 /n/ is not linked with Place, but linked with [nasal] . Nasa l Spreading only spreads nasality, and Place is not involved in spreading. I n order to explain this, let us first look at a set of Japanese sonorants. (24) la  b  cor  lat  r  nas m  n  pal  vel  D j  glide w  Place and manner feature specifications for the above sonorants in the inventory are as follows:  (25) r  m cor la lat na  n b co s na  w j r do r la b do r s na s (approx. )  IrI an d /n/ share the same place feature [coronal] . Ther e is no Place spreading involved in Nasal Spreading, since both Ixl an d In/ shar e the [coronal ] specification; only nasality of /n/ spreads to /r/ (26a). Also , the specification of [coronal ] for Nasal Spreading is redundant (26b) , as only nasality spreads. A s for manner features, lateral of /r/ is universally underspecified (grounded) for coronality.  (26) a . r  -  n b [a^as]  . r  -  n  [cor]  [nas]  25 In order to achieve spreading, the spreader must not be linked with Place (27a) . Thi s representation eliminates other nasals from spreading. Spreadin g is blocked if the spreader is linked with Place (27b) .  (27) a. Plac e is not linked b r - n  . Plac e is linked  i /  \A  * r -  m *  [nas] la  b [nas ] do  r -  h r [nas ]  The root (RT ) node exists for all segments. Th e root of /r/ is the target for nasal spreading. I n the representations up to here, RT is omitted for simplification; however, the existence of RT is not ignored. Sinc e RT of /r/ receives nasality from /n/ in Nasal Spreading, RT of /r/ is not deleted from the string through Nasal Spreading. Thi s is captured only with the existence of the root node.  (28) r  r -  o root R  Ii  T R  n T [has]  As we have examined, by adopting the two issues Feature Class Theory and Feature Licensing, Nasal Spreading is captured with minimam redundancy. Also , /r/ and /n/ in Nasal Spreading can be compared with other possible verb-final consonants with the least specified features. Th e segmental analysis must be  26 included in order to specify the segments involved in spreading. 3. Prosodi c Analysis  In this section, a prosodic analysis is presented to account for Nasal Spreading. A prosodic account shows the difference between Nasal Spreading and /a/-epenthesis; Nasal Spreading maximizes the syllable size to bimoraic, while /a/-epenthesis derives a light syllable. A prosodic analysis also accounts for the verbs that derive superheavy syllables. Superheav y syllables are ruled out, because they are not allowed in the negative construction.  3.1. Tw o Types of Syllabification  For consonant verbs, the CC cluster in the input must be broken to an alternating VCVC string in the output; otherwise, an ill-formed string would be derived.  (29) *  kawar - nai C  C string is created  The consonant cluster is broken up by inserting /a/ for the standard form, or by spreading /n/ of /nai/ for the colloquial form. Fo r the colloquial form, a closed syllable is formed by making the /n/ a moraic coda as in (30) . Th e mora is syllabified under the preceding syllable node to create a closed syllable;  27 the syllable size is thus maximized to bimoraic.  (3 0) a  . Nasal Spreading @ J&  @  @  I  AA  m / m m /  b. /a/-epenthesis  m m  k a w a RT_ - RT a i  @ a  a  1/1 A m/m  t.  / m  a  A  a /m m  / 1/ 1 / I  k a w a r - A - n a i  [nas] In (25 ) the root node (RT ) receives specifications from the following nasal. Ther e is no replacement per se. Al l the consonant verbs are subject to /a/-epenthesis, because /a/epenthesis does not require any specification for the root-final segment except [consonantal ] status. Nasa l Spreading is different from /a/-epenthesis with regard to the syllable count; /a/-epenthesis adds one syllable to the derived string, whereas Nasal Spreading minimizes the syllable count by linking the moraic /N/ under the preceding syllable node. Thi s is why processes such as Nasal Spreading are called 'phonologica l reduction'.  I have been using the suffix form /nai/; however, there is controversy concerning the underlying form of the suffix (Vanc e 1987, Shibatani 1990) . Th e other position is to choose /anai/ as the underlying form, including /a/ in the morpheme. Thu s /a/epenthesis does not exist; instead, negative construction is performed by the deletion of /a/ when vowel hiatus is created:  28 for example, /tabe-anai/ -- /tabe-nai/. Actually , the majority of verbs are consonant verbs, and therefore it seems reasonable to include /a/ in the underlying form. However , the account that chooses /anai/ leads to the following problems. Th e problem occurs when Nasal Spreading applies; for example, /anai/ suffixes to /wakar/ 'understand ' to form /wakaranai/ without any modification. Nasa l Spreading loses its motivation, since /wakar-anai/ does not create a consonant cluster. I f we assume /ra/-deletion prior to spreading, what motivates such deletion, and why do no other consonant verbs undergo nasal assimilation? The /k/-final verb /kawak/ 'dry ' is compared with the /r/-final verb /kawar/ 'change ' as below:17  (31) kawar+ana i kawarana kawak+anai kawakana  i kawaNna  i 'change  i *kawaNna  i 'dry  ' '  Not only does the choice of /anai/ need more derivational steps, but it also is unable to explain Nasal Spreading. Th e account with /nai/ reduces the number of steps and provides a better explanation for Nasal Spreading.  3.2. Monomorai c Roots  This section examines the case of monosyllabic roots in which Nasal Spreading is blocked. Firs t we examine the monosyllabic roots that consist of a light syllable, then move on  29 to monosyllabic roots that consist of a long syllable. Th e light syllable roots in (32) contain /r/ root-finally; however, Nasal Spreading does not apply.18  root+/nai/  Nasal Spreading  tor-nai  ?toN-nai  'take'  sar-nai  *sa-N-nai  'leave'  mor-nai  *mo-N-nai  'leak'  ir-nai  *i-N-nai  'need'  fur-nai  ?fu-N-nai  'fall'  nar-nai  ?na-N-nai  'become'  tar-nai  ?ta-N-nai  'suffice  ter-nai  ?te-N-nai  'shine'  All the above monosyllabic /r/-fina l verbs can have the epenthetic /a/; however, Nasal Spreading does not consistently apply for the above group of words.  There is some disagreement among native speakers with regard to these forms (se e note 9) . Roughl y speaking, as for Nasal Spreading, the verbs that contain the (light ) polysyllabic roots are considered definitely better and more consistent than the ones that contain the monosyllabic roots. Thi s judgement is based on the expressions listed in this thesis, and therefore any affixes, or morphemes should not be added for the analysis here.  30 If any suffixes or morphemes are added to them, the whole analysis could be changed in terms of phonology, morphology, or syntax. Fo r example, a monosyllabic word /ter/ 'shine' may form a colloquial negative form ?/teNnai/. Fo r some speakers, this is acceptable, but not for others. Whe n this word is pronounced in a sentence such as /asita mo teNnai yo ne/ 'probably tommorow either, it won't shine', acceptability perhaps will increase, because of the syntactic change influencing intonational pattens, the application of fast speech rules, etc. Th e examination of these effects on complex phrases (sentences ) is beyond the scope of this thesis.  In order to clarify the points for analysis, we need to identify the three types of forms among the /r/-final verbs: (i) polysyllabic roots in which Nasal Spreading applies most consistently. Polysyllabi c roots usually consist of two to four (light) syllables. The y are limited to indigenous Japanese (Yamato) verbs, and Sino-Japanese verbs and compound verbs are not included. (ii ) monosyllabic root verbs, including the long syllable root verbs. Thes e undergo Nasal Spreading; and therefore the judgement of the forms by native speakers varies more than the type (i ) verbs. (iii ) verbs in which Nasal Spreading does not apply, because of their semantic or pragmatic properties. Th e words that belong to 'literary' expressions rather than 'colloquial ' expressions do not have the colloquial alternatives. Fo r example, /kaor/ 'b e fragrant' or /todomar/  31 'stay' are considered as 'literar y use', and are less often used in casual speech; it is because Nasal Spreading is a colloquial formation. Compound s or complex suffixation, including SinoJapanese words, are excluded from the analysis, because other linguistic constraints, for example, syntactic or pragmatic conditions, are involved (c.f . Poser 1992 for periphrastics; e.g. /benkyoo-suru/ 't o study' or /kenkyuu-suru/ 't o do research').  In the following monosyllabic vowel verbs and irregular verbs with vowel endings, no modification is seen after /nai/ is suffixed. A s long as there is no CC string, no modification takes place regardless of the syllable count in the verb root.  (33) Monosyllabi c Vowel Roots a. vowe l verbs: root+/nai/ standar  d form  mi-nai mi-na  i 'loo  ki-nai ki-na  i 'wear  i-nai i-na  k at' '  i 'exis  t (animate subject)'  b. irregular verbs: root+/nai/ standar  d form  su-nai si-na  i 'do  ku-nai ko-na  i 'come  ' '  3.3. Superheav y Syllables in Nasal Spreading  32 The monosyllabic consonant roots that consist of a heavy syllable do not undergo Nasal Spreading either; although they derive the standard form by inserting /a/ as a syllabification process (a s will be shown later). Th e consonant verbs that contain long vowel roots are not as consistently accepted as the (light) polysyllabic root verbs.  (34) Monosyllabi c Roots (Lon g Syllables) root-/nai/ N-Spread  .  koor-nai ? koo-N-na  i 'freeze  toor-nai ? too-N-na  i 'pass  hoor-nai ? hoo-N-na  i 'throw  ' ' '  In (34 ) the heavy syllable roots derive superheavy syllables; the colloquial construction with /nai/ does not allow superheavy syllables. A s we have discussed, monosyllabic roots do not consistently undergo nasal assimilation, while all the polysyllabic root verbs undergo assimilation. What , then is going on in the verbs that contain long syllables as root finals?  (35) Syllabl e Types in Roots a. b  n7  /m/m /  /J/I '  kawar-nai tor-na  . c  .  m  I  i  33 Among /r/-final monosyllabic roots, either (35b ) or (35c) are ruled out for Nasal Spreading. Betwee n the two types of monosyllabic roots, (35b) derives a superheavy syllable. Ther e is no rule which truncates the final vowel to Vhonnai/; because Japanese contrasts vowel length, short versus long. Fo r instance, /kor/ 'become stiff an d /koor/ 'freeze' are two different words, and vowel length must be distinguished. Th e length distinction is an underlying property of words, with moraic structure presumed in the lexicon (McCarth y and Prince 1993, p.21) . Th e long vowel in /koor/ is an integral part of the syllable, which cannot be broken down. I n Hewitt (1994 ) Syllable Integrity is also discussed: 'lon g vowels and diphthongs are never split by feet into separate constituents .. . the idea was formalized as the Syllable Integrity Principle (Princ e 1975)' . The vowel length contrast in Japanese also has been measured phonetically in an experimental study conducted by Beckman (1986) . Thus , in (35), the string that derives the unfavorable superheavy syllable does not undergo spreading consistently. Th e above word /hoor/ can have /a/-epenthesis: this does not violate the syllable condition. Thi s supports the crucial role of syllabification in the negative construction.  The other problem is that superheavy syllables are not allowed in Nasal Spreading; however, the past-tense construction derives superheavy syllables. Th e difference between the two formations is shown below:19  34 (36) Superheav  y Syllables  root negativ  e pas  t  ? hoor hooN-na  i hoot-t  a 'throw  ? toor tooN-na  i toot-t  a 'pass  ' '  The difference between the two constructions can be explained as follows: colloquial formations such as Nasal Spreading are more recent ones and must respect syllable conditions. O n the other hand, the past tense construction has a longer history. Durin g a long period of time, some phonological changes had taken place and the forms, such as in (36), became frozen.  Optimality Theory (e.g . Prince and Smolensky 1993) may explain the difference between Nasal Spreading and the past-tense forms. I n Nasal Spreading /hooranai/ is definitely better than /hoonnai/ in terms of adherence to the syllable type [ m (m)] . There is a choice in /nai/-suffixation: Nasal Spreading and/or /a/-epenthesis. However , the past-tense form /hootta/ is the only output form of /ta/ suffixation; no alternative, such as Vhotta/, Vhoota/, */hota/, or */hoorata/ is allowed. Unlik e /nai/-suffixation, /a/-epenthesis does not apply in /ta/suffixation. I n Optimality Theory, constraints are violable and hierarchically organized for each candidate set in order to evaluate candidates and identify the optimal form. Th e constraint on syllable size that limits the mora count to two should be ranked lower in the past tense formation, compared with  35 that of Nasal Spreading, in order to secure the correct result for superheavy syllables in the output forms. However , this implies that ranking is specific to particular morphemes. W e have examined long syllable roots; however, it is still puzzling why the monosyllabic roots, which do not derive superheavy syllables cannot undergo Nasal Spreading. T o answer this puzzle, disyllabic base [ @ @] for /nai/-suffixation will be proposed in the following section. 4. Ton e Melodies  In this section the influence of tone melody assignment on Nasal spreading will be discussed. A  tonal account solves the  problem with monosyllabic roots. I t will be shown that constraints on tone assignment restrict the syllable number in the base string in order to rule out monosyllabic root verbs. First, the tone system and accent assignment in Tokyo Japanese are introduced followed by the sample derivation of a phrase. The preliminary description is limited to general characteristics. A detailed discussion of the tone system will be added elsewhere. Ton e melodies indicated here are all based on the Tokyo accent. Differen t dialects have different tonal behaviours. Th e second part introduces accent assignment specific to the negative construction. Then , the monosyllabic verbs are re-examined in terms of agreement of tone melodies in the standard and the colloquial forms; the tone melodies in both  36 forms must agree, otherwise, Nasal Spreading does not apply. I t will be shown that two accent assignment processes, Accent Shift and Initial Lowering, are the major factors causing a disagreement in the tone melodies. 4.1. Ton e Systems of Tokyo Japanese  Japanese has a pitch accent system. Japanes e does not show any systematic stress patterns in terns of intensity (loudness ) of sounds in general. Althoug h there is weight distinction of light syllables versus heavy syllables, the distinction is not metrically governed (Beckma n 1982, 1986). Ther e is a theoretical question concerning the following two distinctions: (a ) stress languages, such as English or French, versus pitch accent systems; (b ) 'pure tone languages', such as Mandarin or Cantonese, versus 'pitc h accent languages', such as Japanese or New Shanghai ( a dialect which is spoken among young people in Shanghai). Th e following description is abstracted from Pulleyblank (1986 ; chapter 5) and clarifies some points of the above question: the difference between stress languages and pitch accent systems will be discussed first, then the position that discriminate 'pitc h accent languages' from 'pur e tone languages' will be re-examined.  In a stress language, prominence relations are read off of metrical trees or grids. Suc h metrical structure is  37 created by general rules that determine types of feet, direction of foot assignment, etc. ... but in accent languages .. . syllable structure typically plays no role in assigning tone. Also , in contrast with a typical stress system, a tonal accent system will typically not exhibit accent subordination. Tha t is, the tonal realization of accents does not vary depending on whether the accent is primary, secondary, tertiary, etc.  The position that distinguishes 'pitc h accent languages' from 'pure tone systems' has been argued for (e.g . Haraguchi 1977); however, Pulleyblank maintains that the properties presented are not all unique to accentual systems (cf . Clark 1986, Tateishi 1993) :  (i) (I t has been argued that) in 'pur e tone  languages', there are commonly as many (o r more) tones as there are tone-bearing units in a morpheme. I n accentual systems, on the other hand, there will generally only be one primary accentual unit (Hyma n 1978). However , there is no formal reason to expect one diacritic (=markin g of lexical tones) per accentual unit.  38 (ii) (i t has been argued that) in tone languages, tones behave in a relatively symmetric fashion, whereas in accent languages, one tone (generall y the H-tone) has some special status (Hyma n above). However, the important distinction between the asymmetries observed in languages so-called 'pur e tone languages' is observed in accentual languages. Asymmetrie s result because rules of spreading, etc. will treat H-tones differently from L-tones since rules affecting H-tones may apply prior to the assignment of L-tones, while rules affecting L-tones may only apply to a fully specified string.  (iii) (i t has been argued that) there is a stage in the derivation of an accentual language where the tonal representation 'consist s strictly of an integral number of copies of a fixed language-specific Basic Tone Melody' (Goldsmit h 1982). However , it is only the result of applying rule and convention to underlying representations that include prelinked tones (=tone s that are linked with TBUs underlyingly).  In conclusion, accentual diacritics can be replaced by prelinked tones; tonal melodies can be derived by  39 regular phonological rules  Tokyo Japanese has been analyzed as a two-tone system with high tones and low tones (Haraguch i 1977, 1991, Poser 1984). According to Pulleyblank (1986) , only high tones are present in the representation of tones (tona l asymmetry); low tones are assigned as default in fully specified strings. Bot h content words and function words are classified into two types for tone melody assignment: 'accente d words' and 'unaccente d words' (Pierrehumbert and Beckman 1988). Mos t function words are unaccented. Th e difference between the two lies in the existence of the lexical high tone, which is also called 'accent' . I n (37) the presence or absence of the lexical tone (H ) is represented as part of morpheme: the accented word /wakar/ 'understand ' has the lexical tone, while the unaccented word /kawar/ 'change ' does not. Th e underlying representations of verbs do not involve prelinking of tones; the lexical tones are linked by morphological operations.  (37) accente  d  unaccented  H wakar  kawar  One of the characteristics of the lexical tones is that they are expressed as a sudden drop in FO contour (Kubozon o 1986) . I t is considered that the tone bearing unit (TBU ) in Japanese is the  40 sonorant mora. Obstruents , including coda obstruents, are transparent to tone assignment. The syllable itself can be the site for tone assignment in some cases (Haraguch i 1977, Clark 1986, Kubozono 1989, also see chapter II) .20  Initial Lowering (IL ) has been discussed extensively among phonologists (Haraguch i 1977, 1990, 1991, Poser 1984, Selkirk and Tateishi 1988, 1991, Pierrehumbert and Beckman 1988). I L is considered as a low boundary tone insertion, which creates a LH rise word/phrase-initially. Ton e melodies may vary slightly in the actual speech of individuals, because pragmatic factors are involved in actual speech production.21  4.2. Accen t Assignment  This section demonstrates actual tone melody assignment. The standard modern transcription of tonal melodies in Tokyo Jpanese is presented in (38) (based on Poser 1984, p. 23; converted from Sakuma 1919 into the standard modern transcription),22 Th  e tone melodies with the unaccented subject  marker /ga/ are added in the right column.  (38)  hafsi  LH  'edge'  hajsi g a  LHH  hajsi  LH  'bridge'  hajsTj qa  LHL  hafsi  HL  'chopstick'  Eafsi ga  HLL  sajkura  LHH  'cherry'  sajkura ga  LHHH  41 Elbachi HL  L 'hibachi ' EHbachi  ofEalma LH  L '  ajtama LH  H 'head ' a/tama  soup ladle' dEaim  i qa HLL  L  a qa LHL  L  | qa LHH  L  These tone melodies are conditioned in part by the existence of the lexical tone. Th e lexical tone is part of the morpheme and prelinking of the tone is included in the input. Th e prelinking is only for nouns, since verbs behave quite differently. Fo r a two-syllable accented word, contrast is made among an initial-syllable accented noun, a second-syllable accented word, and an unaccented word. Fo r three-syllable words, the contrast involves one more type of word: the third-syllable accented word. Whe n a word, for example /Ea\s_i/ 'chopstick', is pronounced in isolation, the initial syllable is linked with the high tone. I n /ha^i/ 'bridge ' the second syllable is linked with H. Th e unaccented /ha'sl/ 'edge' does not have the lexical tone. In this context, there is no difference between /hasi/ 'bridge 1 and /hajsi/ 'edge'; however, when /ga/ is attached, the distinction is clear as shown in (38): the distinction among /ha/si/ 'bridge', /hajsi/ 'edge', and /Ha]gi/ 'chopstic' will be /ha/s~I\ga/, /hajsiqa/, and /najsiga/, respectively.  The unaccented words are subject to phrasal high tone assignment. Th e phrasal high tone is linked with the word-final Tone Bearing Unit (TBU) . Bot h lexical and phrasal high tones spread to the left. Finally , low boundary tones are associated  for all words at phrasal ends. Initia l Lowering only applies to the initial syllable if there is no lexical tone linked with the initial syllable, while the boundary tone associated at phrasal ends may spread to the left multiply.23 The following shows the derivation of /halsi/ 'bridge' followed by the particle /ga/; /ga/ does not affect the tone assignment of /hajiT/.  (39) Derivatio  n  Underlying Representation:  f  hasi 'bridge H  1  I  Phrasing:  hasi ga L% H L %  Low Boundary Tone Insertion:  , )'  •'  hasi ga  This is ordered as follows:  (i) phrasing (ii) Lo  : the unaccnted particle /ga/ follows /ha^i/.  w Baoundary Tone Insertion: (a) Insert L% to the  initial (light ) syllable, if it is not accented (Initia l Lowering). (b ) Insert L% to the final TBU and spread it to the left multiply until H blocks it.  4.3. Lexica l Tone s  Now we turn to Nasal Spreading. Th e tone melody assignmen t specific t o Nasal Spreading will be presented i n this section. For accented verbs th e lexical ton e is attracted t o the mora preceding /nai/ ; /nai / is an unaccented functio n word. I n (40 ) the citation (present-tense ) for m is presented first , then the standard negative form , followe d by the colloquial negative form .  (40) Accente d Stem s citation standar  d neg. colloquia  l neg.  wa|ka|r-u wafkar-afna  i wa'ka-N-jna  i 'understand  sajsajr-u sa/sar-al.na  i sa/sa-N-lna  i 'stick  '  hafk~ajr-u ha/kar-a4na  i ha/ka-N-jria  i 'weigh  '  kajramajr-u kaf  r ama r - a-\ na i ka/rama-N-foa  i '  '  entangle '  For th e accented words (40) , the accent i s attracted t o the TBU immediately preceding /nai/ ; fo r example, /wakar / 'understand ' gets a high tone on the TBU immediately preceding /nai / as /wa/karainai/. Thi s shift is detected, because /nai / shows sudde n pitch drop in FO contour.  Ishihara (1991 ) includes prelinked tone s in underlying representations o f verbs; however, the question i s how to determine th e TBU for prelinking. Fo r example, an accented verb /tabe/ 'eat ' gets the second syllable accented i n present-tense  44 and negative forms, /talbeiru/ and /tglbefriai/, while the initial syllable is accented in past-tense and gerund forms, /titoeta/ and /Ealbete/. I f we take the present-tense form /taj5e\ru/ for the underlying representation, since it is the citation form, what kind of phonological evidence can support this choice? Furthermore, if we take the present-tense form for the underlying form, in order to derive the past-tense and gerund forms, we have to delink the already linked tone, and link it again with the other TBU according to morphological conditions. Wha t I propose here is to include the lexical high tone as part of the morpheme for verbs in (41) , presented earlier in (37) ; however, the lexical tone is still not linked with any TBU.  (41) accente  d verb  unaccented verb  " H " wakar _  kawar  Linking takes place when morphological operations, such as compounding, suffixation, etc. are involved. Fo r instance, the underlying representation of /wakar/ includes a lexical tone, but it is not linked yet. /nai / suffixation determines the TBU for linking, that is, the TBU immediately preceding /nai/ . I t is possible to start writing derivations with prelinked tones by skipping the process of linking. Fo r example, in the compound verb /tibete-iru/ 'eatin g PROGRESSIVE (se e Chapter II)' , th e tone melody of /Obete/ is determined prior to compounding. Th e  45 attachment of /.ijFu/ does not affect the tone melody of /tHbete/ at all. Thi s is the same for both accented and unaccented verbs in this type of compounding. I n this case, the process of linking of the lexical tone in /ta\bete/ may be skipped.  (42) and (43 ) show the linking of the lexical tone and the derivation; the lexical tone is linked after suffixation (afte r /a/-epenthesis) takes place. Th e high tone spreads to the left multiply. Initial Lowering lowers the unaccented initial syllable. Th e initial light syllable is the site for Initial Loweing; this is marked as extratonal for the representational purpose.  (42) Linkin g of the Lexical Tone: H H [ ]  [  .. m] nai  (43) Accen t Assignment in Negative Forms: derivation H Input: waka  r H  Suffixation: wakar  a na i H  Linking of accent: wakara-na  i H  Extratonal syllable: (wa)kara-na  i .i  46 H High tone spreading:  , , v A • (wa)kara-nai  L% H Low boundary tone insertions  L  \, A A  %  wakara-nai /wafoarahai/ Unaccented words (44 ) do not have lexical tones. A phrasal high tone (no t the lexical tone) is assigned to the final TBU in the derived domain.  (44)  Unaccented Words s t a n d a r d neg.  c o l l o q u i a l neg.  kaVar-a-nai  kdwa-N-nai  'change  fcojmar-a-nai  tdma-N-nai  ' stop1  safwar-a-nai  saVa-N-nai  'touch'  majwar-a-nai  mafwa-N-nai  'turn'  The phrasal high tone cannot be assigned for accented words, because already one (lexical ) high tone exists in the same domain; to do so would violate the Obligatory Contour Principle (OCP).24 Suffixatio n must precede phrasal high tone assignment, since the phrasal high tone is assigned with the final TBU in the output domain. Hig h tones, either lexical or phrasal high tones, can spread forward multiply. Thu s the derivation of tone melodies in /nai/-suffixation is ordered as follows:  47 (i) Suffixatio (ii) Linkin  n  g of lexical tones with the TBU immediately  preceding /nai/. (iii) Hig h tone assignment: assign a high tone to the final TBU in the derived string, if there is no other high tone in the same domain. (iv) Hig h tone spreading: spread H to the left, multiply (not e that this is backward spreading by convention). (v) Lo  w boundary tone insertion.  In the unaccented words (44 ) the phrasal high tone is assigned to the final TBU of the derived domain and spreads to the left. Unaccented words are pronounced with level tone, except for the initial syllable, which is lowered by Initial Lowering.  (45) L  % H  \  Initial Lowering: kawara-na  ^ hi  'change  '  /k^waranai/  4.4. Agreemen t in Tone Melodies  The other characteristic of tone melodies in the negative construction is that tone melodies of the standard form and the colloquial form must agree. I n the following table, both accented (46a ) and unaccented (46b ) words demonstrate the corresponding tone melodies in the standard and the colloquial  48 forms.  (46) a  . accente d stems: citation standar  colloquial neg.  d neg.  kajwar-u k^war^a-na  ka^a-N-nai 'change  i  sa/war-u  sawar-a-nai  safoa-N-nai '  tofrnar-u  tqinar^a-nai  to/na-N-nai 'sto  '  touch' p  b. unaccente d stems: citation standar  colloquial neg.  d neg.  wg|klr-u wa(kar-a-\na  i  wa/ka-N-\naj. 'understand  '  kdma|r-u kqmar-a-\na  i  kqma-N-\nai 'b  e troubled'  ha^alr-u hayar-a-]na  i  ha^a-N-inai '  spread'  The next set of data represents monosyllabic roots in which tone melodies of the standard form and the colloquial form do not agree in the negative construction.  (47) a . accented stems: citation standar d NS(melod  y 1) NS(melod y 2)  rcytr-u  tdr-a-mai  ?to-N4nai TtolN-na  -u  safr-a-jnai  *'sa-N4nai  ir-a-jnai  ^I^N^nai  Ifc-u  :  sa\-N-nai  r  b. unaccented stems:  i '  I%z n a i  2 5  take' leave' roast(beans)  49 t atrial  ta/r-a-nai  ?ta-N-na\i_  ?ta-N-nai  1  ha^r^7!!  hafr-a-nai  ?ha-JM-naU.  ?na-JM-nai  'paste'  suffice'  Tone melodies ar e morphological properties o f the negative for m with /nai/ . However , the colloquial form , sharing th e same input structure with the standard form , must follo w the fixed melodies assigned t o the standard for m which is more general. Additiona l examples of 'fixed ' tone melodies are shown in (48) , in which present-tense verbs are compared with past-tense verbs.  (48) Ton e Melodies fo r Present and Past Tense Verbs a. accente  26  d cons , verbs  root pres  . pas  t  kak kiik-  u k"aU-t  a 'write  yom ydm-  u ycffl-d  a •read  tor 'fcbr-  u Ticfc-t  a '  ' '  take'  b. unaccente d cons , verbs maw m a  m ma/t-t  hak hs/k-  u ha/-i-t  kas k a ^  u kate-i-t  a ' a 'wear a 'lend  dance' ' '  Both present-tense an d past-tense form s agree in tone melodies, unless other elements are involved. 27  50 4.5. Accen t Shift  Accent assignment will be re-examined in this section. I t will be demonstrated that Accent Shift causes the mismatch between tone melodies in monosyllabic root verbs. I n noun compounds, whose second member consists of more than three moras, the accent is placed on the antepenultimate mora. Th e following nouns have an accent on the antepenultimate mora.  (49) Accen t Assignment H H yuki-geshiki 'sno  w scenery' fuyu-yasum  i 'winte r vacation'  H  I  e water1  koori-mizu 'ic  There is a morphological phenomenon called 'Accen t Shift' (Poser 1984, Kubozono 1986, Ishihara 1991}. Fo r unaccented recent loan words that contain the antepenultimate mora within a heavy syllable, the accent is shifted to the mora within the same syllable. Th e following recent loan words (fro m English and French) contain the antepenultimate mora in the heavy syllable.  (50) H  H  I I  wa[siN]ton 'Washington ' ko[roN]bi  a 'Columbia '  H H koohii[tee]buru 'coffe e table kuro[was]san'croissant  '  51 In (50 ) the accent does not fall on the antepenultimate mora, but on the initial mora in the heavy syllable which contains the antepenultimate mora.  (51) Accen t Shift H ... [m m] m m #  However, Accent Shift is not applicable in some loan words, as shown below:  (52) H !  a. yuubiisi H  UBC  i  i  b. brekkufaasut H c. puropaNgas  I  o  u  breakfast  propane gas  In the above words, tone assignment is determined either by the original English accent patterns (52a,b) , or for compound nouns by making two domains for accentuation; e.g. /propaN/ and /gasu/ in (52c) . Thu s antipenultimate accent assignment or Accent Shift is not applicable in the above cases. Fo r the application of Accent Shift (an d accent assignment in general as well), other factors such as Initial Lowering, must be considered (se e 4.6).  52 In some cases, it is possible to place the accent on either the first or the second mora within the same syllable. Th e effect of Accent Shift is due to the interaction of two constraints, one requiring an accent on the first mora of a syllable, and the second requiring an accent on the antepenultimate mora, with the syllable constraint outranking the mora constraint. Kubozon o (above ) proposes that this is so because the syllable, not the mora, is the unit for tone assignment. Pose r (above ) also points out that the syllable is the TBU and the mora is the postlexical unit for tone assignment. The system of Accent Shift can be adopted for the tone assignment in the negative construction.  (53) a  . H wa[kaN]nai  b. H wa[kaN]nai  L% H  L  %  \ / / wa[kaN]nai  L% H  \I  L  A  %  wa[kaN]nai  In (53a) , the accent is on the antepenultimate mora. In (53b) , the accent is shifted within the same syllable. Not e that high tones only spread to the left. The two intonational curves (54a ) and (54b ) show the low-high-low sequence. However , the next verb /tor/ 'take' displays a serious problem in terms of tone melody agreement:  53 (54) a  . b  L% H ? L %  \ I  /  . c  / I  .  ? H L %  tor-a-nai [toN]-na /to^aVnai/ /to-N-\nai  NI  HL  i [toN]-na / /to4N-nai  %  ^-7 \  i /  For accent assignment, only the above three melodies are logically possible; among them only the standard form /tc/rafciai/ (54), with the antipenultimate mora accented, can surface. I n (54b) the colloquial form has the antepenultimate accent; the high tone spreads to the left (I L does not apply--this will be explained in the next section). (55c ) has the initial accent because of 'Accen t Shift'. Bot h of these melodies for /toNnai/ do not agree with the melody of /toranai/.  4.6. Initia l Lowering (IL )  Another tonal process that explains the gap between the two types of tone melodies is the restriction on Initial Lowering. Haraguchi (1977 ) points out that words that contain heavy initial syllables tend to be pronounced without IL. 28 Unaccente d nouns that contain light initial syllables, and unaccented nouns that contain heavy syllables initially are contrasted as follows:29  surface  IL  koobaN  ? kgpbaN  'police station'  koNdaN  ?JsoNdaN  'discussion'  54  b.  yoohuku  ?yo/ohuku  'clothes'  soodaN  ?sp/odaN  'counselling'  ryootei  ?rvolotei  'Japanese style restaurant  ifcigogari  'strawberry picking'  kajsiori  'box of sweets'  ujbigawa  ' inside'  tpjrTgoya  'bird cage'  In (55a) heavy syllables tend to be pronounced with a level tone. On the other hand, IL applies consistently for all forms with light initial syllables. Th e following words containing heavy initial syllables follow the above system.30 The y are all unaccented words.  (56)  a.  b.  UR  standard neg. c o l l o q u i a l n e g  •  yor-nai  yopr-a-nai  ?yo/-N-nai  'drop by  lr-nai  jyr-a-nai  *i_fN-nai  'need'  nar-nai  na/r-a-nai  ?na/-N-nai  'sound'  tar-nai  ta/r-a-nai  ?tahN-nai  'suffice  kar-nai  ka/r-a-nai  *kaf-N-nai  1  c.  crop'  In (56c) , Initial Lowering does not apply; the forms with /N/ are not acceptable, because their tone melodies do not agree with the ones on the standard form (56b)-- a disyllabic base [ @ @]-nai is  55 required for /nai/-suffixation. 5. Optimalit y Theory  The above interaction of the processes can be captured in an approach like Optimality Theory (OT) . However , this thesis is not fundamentally based upon Optimality Theory, so the application of the theory is limited. O T integrates all the different sorts of environments or constraints discussed in the previous analysis, by ranking the constraints and by evaluating the candidate sets. First, I will outline Optimality Theory, based on McCarthy and Prince (1993) , Prince and Smolensky (1993) , Archangeli and Pulleyblank (1993b , in press), Pulleyblank (1993) , Ito, Mester, and Padgett (1993) , Ito (1994) , and Hewitt (1994) . In the second section, I will present an analysis of Nasal Spreading, within Optimality Theory.  5.1. Basi c Component  The central analytical proposals of Optimality Theory are, (i) , that attested phonological forms are determined by evaluation against a set of constraints, and (ii) , constraints are ranked in a hierarchy. I n OT constraints are violable; constraints that may be violated in optimal forms are lowerranked, so that the optimal output forms can be secured. Universal Grammar specifies the constraint sets. Individua l  56 grammars are set by parameters, or other kinds of arguments in general terms. Th e input candidate sets are produced by the function Gen. Thre e principles underlie the theory of Gen, as listed below (Princ e and Smolensky 1993, McCarthy and Prince 1993) .  (57) Ge n a. Freedo m of Analysis: Any amount of structure may be posited. b. Containment : No element may be literally removed from the input form. Th e input is thus contained in every candidate form. c. Consistenc y of Exponence: No changes in the exponence of a phonologically-specified morpheme are permitted.  By selecting and ranking the constraints differently, interlinguistic variation can be explained. Fou r properties of constraints interaction in Optimality Theory are:  (i) Violability . Constraint s are violable; but violation is minimal. (ii) Ranking . Constraint s are ranked on a languageparticular bases; the notion of minimal violation (or best-satisfaction) is defined in terms of this ranking.  57 (iii) Inclusiveness . Th e candidate analyses, which are evaluated by the constraint hierarchy, are admitted by very general considerations of structural wellformedness; there are no specific rules or repair strategies with specific structural descriptions or structural changes or with connections to specific constraints (McCarth y and Prince 1993).  The candidate sets are evaluated by the well-formedness constraints; the best-satisfied form is optimal. Whe n the constraints conflict in a candidate set, the relation in the constraints will be ranked as A dominates B (A>>B) , as illustrated below:  (58) Constrain t Tableau, A »B (Princ e and Smolensky 1993) Candidate « can  A  *  2  cand 2  B  *!  The basic conventions used here are:  -Left-to-right column order mirrors the domination order of the constraints. -Violation of a constraint is marked by *. -Satisfaction is indicated by a blank cell. -The sign ! draws attention to a fatal violation, the  58 one that is responsible for a candidate's nonoptimality. -The symbol « draws attention to the optimal candidate. -Shading emphasizes the irrelevance of the constraints to the fate of the candidate.  Violation becomes fatal when other candidates pass the constraint. Th e constraints, for example, PARSE and Rec (see 5.2) belong to ' a family of faithfulness constraints', that expresses a tight relation between input and output forms. I n this thesis, violations of PARSE are marked with <>. Th e segments that are not included in input forms are expressed with capitalized letters.  5.2. Nasa l Spreading in Optimality Theory  In Nasal Spreading, Optimality Theory integrates segmental, prosodic, and tonal constraints into a minimal number of constraints for a certain set of candidates. Fo r example, the monosyllabic root /tor/ 'take' is expressed in the following table. The capitalized letters /N/ and /A/ indicate that they are not shown in the input. Eac h syllable is bracketed.  (59) Nasa l Spreading: Monomoraic Roots Parse [to]<r>-nai  Disyll  RecM  Rec@  * !  [toN]-nai  * !  « [to][rA]-na i [to][rAN]-nai  * *  *  * * i  *  There are four possible output strings for /tor/ as in (59); Nasal Spreading does not apply to any candidate. In /to<r>-nai/, the root final segment IT I i s deleted, violating PARSE root. PARSE(root): underlying segments(roots) must be parsed into syllable structure.31  PARSE(root) violations are prohibited anywhere in the negative construction. Th e violation of this constraint is fatal, therefore it must be ordered first. /toN-nai / does not violate PARSE(root), since the root of /r/ exists in /toN-nai/; nasality of /n/ spreads to the root node of /r/ without any deletion.  (60) t  o RT - RT a i [nas]  In /toN-nai/ a mora is added to the output. IT  I i n the  input does not get any mora assigned; however, /N/ as a coda nasal, a mora is assigned to it. Thi s is formulated as the  60 constraint 'Recoverabilit y of Mora' (RecM) . RecM (Recoverabilit y of mora): If [...m...]output then [...m...]input. A mora that is present in a n output form is also present in the input (Pulleyblan k 1993, Archangeli an d Pulleyblank 1993 (b), Ito, Mester, and Padgett 1993, and Hewitt 1994) .  In /torA-nai/ a mora is added onto the output through /a/epenthesis. Thi s is also a violation of RecM. Th e difference between /toN-nai/ and /torA-nai/ is that /torA-nai/ adds an extra syllable onto the output, since /a/ cannot be syllabified anywhere except creating a new syllable node in the string. Thi s is formulated as a violation of Recoverability of Syllable (Rec@).  Rec@ (Recoverabilit y of syllable): If [...©...] output then [...@...] input. A syllable that is present in an output is also present in the input (sam e as the above).  /toN-nai/, on the other hand, maximizes the syllable size to bimoraic; the coda /N/ is syllabified with the already existing syllable without adding any extra syllable. /toN-nai / violates RecM, but does not violate Rec@, while /torA-nai/ violates both RecM and Rec@. /torAN-nai  / violates Rec@, and doubly violates  RecM, by adding two moras into the output.  61 (61) a .  rn  'm/m  [l  torA-nai The above three constraints PARSE, RecM, and Rec@ are the members of the faithfulness family of constraints that govern the inputoutput relations in the grammar.  So why is /toN-nai/ ruled out? Th e only difference, as far as superficial representation is concerned, is that /toN-nai/ contains only one syllable before /nai/, while other candidates in (61 ) consist of two syllables. A s we have discussed in the previous sections, the tone melodies in the standard form and the colloquial form must agree. Fo r the agreement, the input (base ) string for spreading cannot be monosyllabic, since different sorts of restrictions, Accent Shift or Initial Lowering, restrict the agreement. Fo r example, in order to apply Initial Lowering, there must be a string of minimally two light syllables (see 4.6). W e can also compare NS with /a/-epenthesis, which adds an extra syllable into the string by /a/-epenthesis. I n both forms the base must contain at least two light syllables in order to meet the condition of agreement in tone melodies. Th e disyllabic requirement is formulated as follows:  DISYLL: the derived base string must be minimally disyllabic  62 for /nai/-suffixation: [ @ @]-nai (i n order to meet the tonal condition).  This constraint also rules out cases such as /to<r>-nai/. Thi s candidate violates PARSE(root) and DISYLL. Thu s these two constraints also indicate that the negative construction is a (mora) augmentation process and not the one involving deletion. The constraints for the candidate set are ranked as follows: PARSE»DISYLL»RecM»Rec@. Rec M and Rec@ must be ranked lower in order to secure the best-satisfied /torA-nai/; of which, Rec@ is the final one for /a/-epenthesis, since only in /a/-epenthesis, Rec@ is violable.  The following table displays the case of a heavy syllable root /hoor/ 'throw'.  (62) Superheav y Syllables Parse  *3m@  Disyll  RecM  [hoo]<r>-nai  * !  *  [ho<o>N]-nai  * !  *  *  *  *  [hooN]-nai  * !  Rec@  »[hoo][rA]-nai  *  *  [hoo][rAN]-nai  ** i  *  Nasal Spreading does not apply in /hoor/. Th e long syllable contained in /hoor/ triggers an unfavourable superheavy syllable.  63 Unlike the past-tense formation, the /nai/ suffixation does not allow superheavy syllables. W e need to include the constraint on syllable size.  *Tri-moraic Syllables (*3m@) : avoid syllables with three moras: @<2m (Hewit t 1994).  The ranking of the faithfulness family of constraints, PARSE, RecM, and Rec@, is the same as the case of /tor/. /ho<o>N-nai/ violates PARSE by shortening the long vowel. Eve n if /hooN/ were shortened into /hoN/, it would not form a two syllable string. /hoo<r>-nai/ , /ho<o>N-nai/, and /hooN-nai/ violate the constraint DISYLL; the tone melodies of /hooN-nai/ or /ho<o>N-nai/ do not agree with the tone melody of the standard form. Violation s of PARSE and *3m@ are prohibited in /nai/suffixation, because /nai/-suffixation is achieved by augmentation, and not by deletion. DISYLL , a constraint for the base template which interacts with tonal condition follows the two constraints PARSE and *3m@. 32 Th e five constraints are ranked as follows: PARSE»*3m@»DISYLL»RecM»Rec@.  5.3. Syllabl e Templates  Syllable templates in outputs are compared in the following list. I n (56 ) the possible output syllable templates for Nasal Spreading are listed on the right of each word according to  64 syllabification and the mora count. A  templat e marked with #  indicates tha t it is in the word initial position. Withou t #, by adding some more syllables, it would derive different type s of templates; fo r example, /tor / must be distinguished fro m /karitor/ 'harvest ' which has a trimoraic inpu t string.  input  output(NS)  a.  tor-nai  *toN-nai  *#[mm]  'take'  b.  tor-nai  *torAN-nai  * [m] [mm]  'take'  c.  hoor-nai - -  *hooN-nai  *[mmm]  'throw'  d.  hoor-nai --  *ho<o>N-nai  * [mm]  'throw'  e.  kawar-nai -  kawaN-nai  [m] [mm]  'change  f.  kawar-nai -  kawa<r>-nai  *[m] [m]  'change  Nasal Spreadin g does not apply in monosyllabic root s (63a,b,c,d) , as we have discussed, /hoor / (63c,d ) represents heavy syllabl e (final) roots. I t violates *3m @ by deriving a superheav y syllable. I n /ho<o>N-nai/ (63c ) the long vowel is shortened; shortening violates th e integrity of the syllable. PARS E demands the segment t o be properly linke d with the appropriate node. (63f) deletes th e root fina l /r/ , violating PARSE(root) . Onl y in (63e) does Nasal Spreading apply.  Instead of using th e possible templat e patterns presented with moras and syllables, I use the actual words i n (64) , since it is easier t o see the comparison. Th e above six candidates are  65 laid out in the following table.  (64) Nasa l Spreading: Conditio n and Template Parse [ka][wa]<r>-nai  * !  [ho<o>N]-nai  * !  [hooN]-nai [toN]-nai [to][rAN]-nai « [ka][waN]-na i  *3M@  * !  Disyll  RecM  *  *  •  *  * !  *  Rec@  * * i  *  *  This representation is somewhat unusual since types of forms are compared rather than candidate outputs for a single input. This is because the evaluation of each set of candidates is done separately; for example, a candidate set of monomoraic roots, etc. Th e candidate set in (64) is to compare and to evaluate possible syllable types (=templates ) for Nasal Spreading. Amon g the six candidates in (64) , PARSE violations are prohibited anywhere in the negative construction; the violation of this general constraint is serious, and thus ranked first. Th e violation of *3m@ as well must be serious, because superheavy syllables are prohibited; this constraint applies only for heavy syllable roots (i n root final position). PARS E and *3m@ are related to syllable structure. Nasa l Spreading is a mora augmentation process which maximizes the syllable template to [mm] ; at the same time, the output strings must meet the well-  66 formedness syllable constraints. /nai/-suffixatio n is accomplished by violating RecM and/or Rec@--adding a mora to the output. Nasa l Spreading specifies the template [mM]^ (M=inserted), while /a/-epenthesis specifies the template [m]^ [M]. Th e difference between Nasal Spreading and /a/-epenthesis is made by Rec@. I n (64) RecM is ranked before Rec@, because the optimal form /kawaN-nai/ violates RecM, but not seriously as the case of /torAN-nai/. DISYL L rules out the monosyllabic stems in the input. DISYL L does not concern syllable internal structure. DISYLL governs the tonal agreement in the standard and the colloquial forms. Thus the constraints are ordered as follows: PARSE»*3m@»DISYLL»RecM»Rec@ fo r Nasal Spreading.  Vowel verbs do not involve any modification in the negative construction. The y are not subject to either /a/-epenthesis or Nasal Spreading. The y do not create a consonant cluster by suffixation. Next , in the following table, consonant verbs which do not have the root final /r/, for example, /kak/ 'write 1, are examined. /a/-epenthesi s applies, but Nasal Spreading cannot apply.  67 (65) /a / Epenthesis: Consonan t Verbs Parse [ka]<k>-nai [ka<k>N]-nai  Disyll  *!  *  * !  *  RecM  Rec@  *  [ka][kAN]-nai  ** i  *  *[ka][kA]-nai  *  *  The consonant cluster derived by suffixation must be broken up by /a/-epenthesis; /a/-epenthesis is treated as a syllabification process. Non e of the segments contained in the input can be deleted; PARSE is ranked first. Th e Ixl fina l root can undergo Nasal Spreading, however, the other consonant final roots cannot undergo spreading.  In (65 ) , the deletion of the root-final /k/ results in the fatal violation of PARSE(root). I n the /r/-final root, the root of /r/ is targeted for spreading. The root of /r/ receives its feature specification from the following /n/ by spreading, without deleting the root; obviously, this is different from the case of the /k/-final verb. Nasa l Spreading does not violate PARSE(root) in this regard. DISYL L is included in the set of conditions. Th e result of DISYLL overlaps with the result of PARSE; however, the distinction between the two constraints will be clear when we consider longer consonant verbs, for example,  68 /saso<w>/ 'invite' . /sasow/ , consisting of a disyllabic root, can pass DISYLL, but violates PARSE, if there is any deletion. /kakAN-nai/ violates RecM seriously, because the two moras for /a/ and /N/ are not included in the input. Rec @ is ranked last for /a/-epenthesis. Th e three constraints ranked as PARSE»DISYLL»RecM»Rec@ ca n account for the verbs which have consonants other than In I root-finally . Finally, the target /r/ and the spreader /n/, analyzed in 'the segmental analysis' (Chapter 1-2), will be re-examined under the Optimality Theoretic approach. I n this section, /r/ and /n/ are analyzed separately.  As we have examined, /r/ is not linked with any feature. Only the root node is present in the representation of /r/ in Nasal Spreading. Therefore , in (66a ) /r/ fulfills both licnsing of place and manner feature specifications. O n the other hand, (66b) violates licensing of the place feature, by linking [coronal], and (66c ) violates licensing of the manner feature, by linking [lateral] .  (66) a. r  b  . *  rc  . *  [cor] [lit  r ]  The two licensing violations are as follows: License Place (P) : the place feature(s) must be licensed (cf.  Ito, Mester, and Padgett 1994).  License Manner Feature (M) : the manner feature(s) must be licensed (sam e as above).  (67) a. Licens e Place is fulfilled: r  b. Licens e M is fulfilled: r  , t, w, s...  II  I  P P  P  , t, w, s...  This is illustrated in the following table.  (68) Th e Target /r/: Input  r  License P  License M  *  *  / \  lat co r r  f  *  cor r  1  *  lat S" x  [r-cor] violates License P, and [r-lat ] violates License M. [  70 lat, cor] violates both License P and License M. [r ] with no specification is optimized. The spreader /n/ can be captured in the same manner. /n / is licensed as [n-nasal] . Ther e is no place specification for /n/, but [nasal ] is specified in Nasal Spreading.  (69) a . Licens e Place: n  , m , h  I I  lab do r  b. Licens e Manner F: n  , m  i I  , h  I  nas na s na  s  Among the three nasals /n, m, n /, /n/ does not have a place specification. Th e other two have labial or dorsal specifications. A s for manner feature(s), all of the nasals are licensed by [nasal] . Th e candidates for the spreader /n/ are evaluated as follows:  71 (70) Th e Spreader In/: Inpu t  License P *  n  License M •  / \  nas co  r  n  *  1  cor « n  1  nas n  *  [n-cor] violates License P and License M. [n-nas , cor] violates License P. [n ] does not get any specification; License M is violated. [n-nas ] is optimized.  Feature Class Theory and Feature Licensing account for the examination of segments in Nasal Spreading. Unde r the Optimality Theoretic approach, possible candidate segments at morpheme boundaries are evaluated with the two constraints, License Place and License Manner Feature.  72 6. Conclusio n  A small set of constraints, PARSE, RecM, and Rec@ (members of the faithfulness family of constraints), *3m@, and DISYLL covers various types of conditions, and accounts for the candidates discussed in the previous sections. Differen t combinations and ranking of the above constraints evaluate different sets of candidates for Nasal Spreading and /a/epenthesis. Fo r example, /a/-epenthesis derives an additional light syllable, while in Nasal Spreading /N/ maximizes the syllable template to bimoraic. Th e difference between the two is that /a/-epenthesis adds one syllable to the string, violating Rec@. Bot h types of modifications must align with the morpheme boundary before /nai/.  In the final section, the segmental analysis has been reexamined under the Optimality Theoretic approach, adopting two issues, Feature Class Theory and Feature Licensing. Tw o set of constraints, License Place and License Manner Feature account for two set of candidates for the verb-final position and the suffixinitial position.  In the analysis of Nasal Spreading, I  have not included the  constraint ALIGN discussed in McCarthy and Prince (1993) , Pulleyblank (1993 ) and others; for instance, the right edge of the derived base [ @ @] in question must align with the morpheme  73 boundary before /nai/ ([@ @]-nai). Th e evaluation of candidates can be done with the above mentioned constraints without ALIGN, and in this way we can minimize redundancy in terms of exposition. Or analytically, in either Nasal Spreading or /a/epenthesis, ALIGN is underspecified for all the candidate sets by specifying segmental conditions prior to prosodic analysis as the premise: segmental analysis specifies the root-final segment to be targeted for change. Ther e is no contrast in terms of locus among the candidates in the presented tableaux. Probably , there is much more to say with regard to Optimality Theory; however, for a trial analysis, I would like to limit the discussion to the points described in this chapter, and leave it for future research.  74 NOTES  1. Ther e are some more suffixes for negative formation, such as /maseN/, /mai/, or /zu/. /nai / is called the informal negative suffix. /nai / is a present-tense form of the root /nak/. A part of the /nak/ conjugation is given below. nak-atta pas t tense /ta / = past tense suffix nak-ute gerun d /te / = gerundive suffix nak-ereba conditiona l /reba / = conditional suffix These forms are predictable foem the underlying form /nak/, and they are not allomorphs. Conjugatio n is complex, including intervening segments, such as /a/, /u/ or /e/. A s seen in the above data, /k/ of /nak/ is present in other forms; however, the intervocalic /k / is dropped in the citation form as /nai/ (Vance 1987, Shibatani 1990). /nai / can be compared with the adjective /nai/; however, the conjugation of adjective /nai/ is slightly different from the suffix /nai/. I will concentrate on the suffix /nai/ which is attached to verb roots. In this thesis, the citation form (present-tens e form) /nai/ is used. Th e present-tense form is also called 'non-past', because the present-tense and future-tense forms are the same in Japanese, as seen in the following verbs (Makin o and Tsutusi 1986). Root Presen tabe taber mat mat  t Futur u taber u mat  e u 'eat u 'wait  ' '  2. I interviewed four native speakers of Tokyo Japanese for the judgement of the data in this thesis. Al l of them were born in Tokyo and grew up there. Th e total number of years they lived outside Tokyo is less than three years. Th e age range is between late twenties and early forties. Thus, it is possible to state that the four speakers share the same linguistic background; however, there are still some differences in pronunciation and in judgements for some expressions. Ther e may be pragmatic factors, such as fast speech effect, etc; however, the pragmatic differences are not the objective of the analysis. The difference in judgements cannot be avoided when we analyze actual speech. 3. I n the vowel inventory in Japanese, /i/ is phonetically characterized as the central vowel /i/, and /u/ is the unrounded / W / . I n this thesis, in the most cases, segments are expressed phonemically, or even in the Romanized characters, unless there  75 is need for expressing phonetic constrast. 4. Th e string of (C) W is not invariably grouped together under a single syllable node; for example, a Chinese-origin word /iiN/ 'committe e member' is split into two syllables as /i.iN/ (Haraguchi 1977) . I t is mainly because this word is considered as a sort of a compound derived from two separate words, /i/ and /iN/. Phonetically , the two /i/s in /i.iN/ are not distinct from a long vowel /ii/, e.g. in /siiN/ 'scene' . Probably , this segregation is psychological, since phoneticaly there is no significance so far. Thi s is the same for diphthongs. Whethe r (Oyv; (Y4VJ constitue s one syllable or two separate syllables is an open question. 5. (C)W N syllables are very rare. Especially , WN seldo m shows up in Japanese. However , this syllable type is still possible, as in an emphatic expression, e.g. the word like /uuNto/ 'ver y much' is formed, but it is used in a very casual context. 6. Th e account that links the onset directly with the syllable node is based on Archangeli (1991) , McCarthy and Prince (1993) and others.  /\ /  C V C  n  \ m  Some phonologists (Kubozon o 1989, Katada 1990, Ishihara 1991) take the different position that links the onset under the following mora node with the evidence from their research on language games and error analyses. However , it doesn't make any difference in terms of mora counting at least in this thesis. I n the above representations, m expresses a mora, and @ is a syllable, either light or heavy. 7. I n Nasal Spreading, the phrase that I take for analysis consists of a verb root and /nai/. N o other segment or morpheme precedes or follows the phrase for the analysis. Fo r instance, in Nasal Spreading, I use the form of /toranai/ for the verb root /tor/ 'take' . Th e past-tense form or the gerund form /toranai/ is not included in the analysis: e.g. /toranakatta/ PAST, or /toranaide/ GERUND are not examined. Because they are subject to different sorts of structural categories, and the constraints on nasal assimilation could be different. This is.beyond our focus. Pragmatic factors, such as speech rate, may affect the judgement of words presented in this thesis. However , I minimize the involvement of such factors (bu t they are not totally ignored). I focus on the phonological, morphological, and prosodic aspect of colloquial formations. Th e analytical focus is on the underlying principles rather than the superficial  76 variations.  8. Th e gloss is added for each word/expression (o r for each verb root), but the tense or other functions (morpheme s or function words) are not included in translation. Fo r example, the translation of /kawaranai/ is '(It ) won't change'; however, only the meaning of the verb root /kawar/ 'change ' is presented in most cases in this thesis. Th e meaning of function words are added and explained where necessary. 9. /i / and /e/ are lexical properties of the verbs. Onl y /i/ and /e/ can appear in the root-final position. Th e reason why only the two vowels can occupy the the root-final position is an open question. Consonant verbs are restricted to the particular subset of the full range of consonants. I n the course of history, quite a few phonological changes took place, and some of the sounds in the current consonant inventory were derived through the merger of two sounds, etc. and only the limited number of consonants occupy the root final position in modern Japanese. Fo r example, /utaw/ 'sing ' which conjugates as /utaw-a-nai/ NEGATIVE, /utatte/ GERUND, in modern Japanese, has the following inflectional forms in Classical Japanese. citation continuitiv e forms (mor e like the gerundive form) utahu utaf  i utafit  e (Saek i and Yamada 1993; p. 60)  The above inflectional forms are found, for example, in the literature during the Heian period (794-1192 ) and the Kamakura period (1119-1131) . Thes e changes are part of the study of Classical Japanese (Saek i and Yamada, above). 10. /a / shows up when a consonant hiatus is created by suffixation. I n other suffixation, the vowel /i/ is also inserted when the consonant hiatus is created; for example, /kak/ 'write' takes /i/ for the suffixation of the 'polite ' morpheme /masu/, as /kak-i-masu/, or the reduplicative form /kaki-kaki/ 'while writing.' /i / may be considered as a suffix, since it can occupy the word final position; e.g. the nominalized verb /yomi/ 'reading1 end s in /i/. On the other hand, /a/ cannot occupy the word final position as */kaka/; it must be bound with some suffixes such as /nai/ as /kakanai/. Thi s is one of the reasons why /a/ is considered 'epenthetic'.  77 11. Accordin g to McCawley (1968) , the verb /aru/ 'exist' is inflected as follows: pres. ar u past att a neg. na i polite arimasu prov. areb a tent. aro o 12. I t is not clear why /a/ is chosen for the epenthetic vowel. (cf . Grignon 1984). Probabl y /a/ has asymmetrical status in the vowel inventory; /a/ is the only low vowel, and lacks its counterpart in terms of backness, as the only [+low ] vowel. 13. Th e two vowels /a/ and /i/ of /nai/ can be linked under one syllable node as a diphthong. However , I leave it as two syllables. Whether they are treated under one syllable or not doesn't make any difference in the analysis here. 14. "Th e mimetic vocaburary of Japanese consists of a large number of sound-initating and manner-symbolic roots traditionally classified into Giseigo (onomatopoetic ) and Gitaigo (mor e broadly ideophonic)." 15. Liquid s and glides are not classified together as sonorants, with distinctions between them based on place. Thi s is because of different behaviours between them. Som e phonologists argue that liquids lack place feature in the feature geometry (Sha w 1991, Paradis and Prunet 1991, and others). Fo r example, /r/ in Japanese lacks place feature, on the other hand, /w/ and /y/ are specified with place features. Fo r example, the palatal glide /y/ can be a part of a complex onset, but the labial /w/ cannot form a complex onset. 16. I n this context, whether /r/ is epenthesized or deleted is controversial (se e Shibatani 1990 for this issue); however, it is not our major focus here. I leave it as an open question. 17. /a / cannot be a part of the verb root; for example, */kasa/ 'lend 1 instea d of /kas/. I n the Japanese inflectional paradigm, it is reasonable to choose /kas/ for the verb root, because other vowel initial suffixes can follow the consonant ending root /kas/, creating various inflectional forms. Th e following examples demonstrate some other inflectional forms of /kas/ in contrast with the vowel verb /tabe/ 'eat'. kas-oo tabe-yo kas-eba tabe-reb kas-u tabe-r kas-i-ta tabe-t  o VOLITIONA a CONDITIONA u NON-PAS a PAS T  T  L L  78 kas-i-te tabe-t kas-i-masu tabe-mas kas-eru tabe-(ra)rer  e GERUN D u POLIT E u CAUSATIV  E  18. Th e distinction between 'literary ' words and 'colloquial' may conditions Nasal Spreading. However, the distinction cannot be made straightforwardly. Nasa l Spreadin g applies t o the words used frequently i n daily conversation, suc h as /kawar / 'change ' or /wakar/ 'understand' . Thi s tendency is general, and seen in other colloquial formation s as well. 19. Th e cases in (36 ) cannot be the sequences of two syllables *[ho][oN] , rather than the superheavy syllable [hooN] . This i s because vowel length is a lexical property and must be treated as an integrated unit. Superheav y syllables do exist i n the world's languages ; for example in Arabic, the final mora i n superheavy syllables i s marked as extraprosodic (McCarth y and Prince 1990 , pp.252-253). 20. Obstruent s are transparent fo r tone assignment. Only Sonorant moras are TBUs (Haraguch i 1977) . 21. Ton e melodies i n Tokyo Japanese may vary because of the following factors: (i) dialecta l backgroun d (ii) idiosyncrati c matters (iii) age/generatio n difference s (iv) syntacti c focu s In many cases, the differences i n tone melodies occur at phrasal boundaries; some people tend to insert boundary tones , but other s don't. Fo r instance, the insertion of a low boundary tone in the end of a minor phrase is shown below: melody 1: kalwarana melody 2: kj/waranaU  i '(It  ) doesn't change'  .  There i s no difference i n meaning i n the above two. Som e of the other types of tonal variations will be introduced elsewhere. 22. Originally , ton e melodies were transcribed i n the three-tone system : high tones, mid tones, and low tones. Pose r has converted i t into the two tone system fo r the modern Tokyo Japanese. 23. Initia l Lowering i s considered as the continuation of a lo w boundary tone from the preceding phrase (Pierrehumber t an d Beckman 1988) . Se e also note 22 in Chapter II. 24. Th e OCP will be discussed i n Chapter II.  79 25. Se e (4.6 ) for IL. 26. Vowe l verbs have more complex ton e assignment system s Tone melodies chang e when suffixes, such as the polite suffi x /masu/, are attached to verb roots. 27. Ton e melodies of the present-tense for m and the pasttense form in some accented vowel verbs do not agree.  root tabe  present taBe\ru  past tabe-ta  oki  dki-\ru  ~o\ki-ta  sime  sj/me-\ru  sT\me-ta  This disagrement i n tone melodies can be solved by making the initial syllable in the present-tense for m extratonal fo r Initial Lowering. I n longer words as well Initia l Lowering applies. present pas  t  (ta)'bej-ru Ealbe-t  a  (o) k~H-ru "oki-t  a  (spmej-ru sUme-t  a  28. I n the case of the CVC string in the word initia l position, Initial Lowering may apply. Becaus e [-sonorant ] segments are transparent t o tone melody assignment; the TBU is the sonorant mora. e.g. ki/ssate n 'coffe te/ttee '  e shop'  thorough'  naff: toku 'understanding saftrtuu 'rushing  '  '  There are some variation in tone melodies bacause of individual backgrounds, or idiosyncratic differences . Th e above words ten d to be pronounced in level tone especially among young people in Tokyo (Asah i Newspaper 1992) . 29. Especiall y young people in Tokyo tend to pronounce the words suc h as presented i n (55 ) in a level ton e (Asah i Newspaper, above) .  80 30. Although , there are some differences in tonal melodies as described in (not e 21), i n the most cases tonal melodies are fixed. Th e example in (56c) is not favoured in Tokyo Japanese. 31. PARSE(root ) is simplified as PARSE. 32. Fo r example, /ikidoor/ 'b e indignant' contains a CVCW root. However , this is a compound verb which consists of two morphemes /iki/ and /door/. 33. Th e roots that have the string 'heav y syllable followed by light syllable ([mmym] y ma y undergo Nasal Spreading. However , this is not consistent as the following examples show. No t many words contain this type of string. root mookar - - mookar-a-na i - - mooka-N-na i 'mak e a profit' hoobar - - hoobar-a-na i - - ?hooba-N-nai 'cra m (foo d into one's mouth)'  81 II. Vowe l Deletion Introduction  This chapter analyzes Vowel Deletion, a different type of colloquial formation in Tokyo Japanese. Bot h Nasal Spreading and Vowel Deletion exhibit superficial similarity as phonological reduction processes in colloquial formation. Th e difference between the two formations is that Nasal Spreading is treated as a syllabification-driven process, while in Vowel Deletion the formation of a single prosodic domain is crucial. Th e focus is on the interplay between a segmental change and tone melody assignment under the model in which prosodic constituency influences both segmental and tonal analyses. I t will be shown that tonal constraints influence the application of Vowel Deletion. Vowe l Deletion applies optionally in casual speech (cf. Miyara 1980). I t is different from a fast speech effect, because it may also apply in deliberate speech (cf . Hasegawa 1979) .  Vowel Deletion occurs in a specific type of compound verbs. The first member of the compound must be a verb in the gerund form, and the second member must be an auxiliary verb in any inflectional form (som e people call it 'quasi-auxiliary verb' (Poser 1984). The compound verb in question is composed of two  82 phrasal parts (tw o minimal minor phrases); they are unified into one single tone group (mino r phrase) after Vowel Deletion is applied.1 Th e number of tonal peaks is one of the major indicators of phrasing; basically, one tonal peak is assigned per domain (se e Poser 1984, Selkirk and Tateishi 1988, 1991 for the definition of these phrases). Afte r Vowel Deletion applies in a compound verb, the accent in the auxiliary verb is entirely suppressed, so that only one tonal peak can surface. I f the auxiliary verb is linked with a lexical high tone, Vowel Deletion is blocked; the auxiliary verb must be accentless for Vowel Deletion to apply.  The organization of this chapter is as follows. First , the environment in which Vowel Deletion applies will be discussed, including the definitions of compound verbs and auxilary verbs in Japanese. Second , the effects of tone melody assignment on the applicability of Vowel Deletion will be examined. Tw o types of compound verbs are discussed: one takes a single minor phrase as its tonal domain, and the other takes a minimal minor phrase. The lexical tone in the auxiliary verb plays an important role in forming a tone domain. Third , the distinction between the minor phrase and the minimal minor phrase will be defined in order to clarify the interface between tone melodies and Vowel Deletion. Finally, the processes of deletion and the tone domain formation will be formalized.  83 1. Environmen t for Vowel Deletion This section examines the environment for Vowel Deletion. There are several types of compound verbs in Japanese. Amon g them, only a specific type of compound verbs is subject to Vowel Deletion. Th e morpho-syntactic structure of compound verbs will be analyzed first, in order to specify the type of compound verbs subject to deletion. Second , vowels subject to deletion will be examined in light of the sonority scale (Princ e and Smolensky 1993) and with respect to the Theory of Underspecification (Archangeli 1988).  1.1. Compoun d Verbs  Vowel Deletion deletes one vowel at the boundary between the verb and the auxiliary verb, when a vowel hiatus is created by compounding. I n the list of compound verbs below, expressions in the left column indicate the standard form, which is used here as the base for both written and spoken styles. Th e expressions in the right column are the variants of the standard form which are produced in colloquial usage.2  (1) Vowe l Deletion in Compound Verbs standard form colloquia  l form  tabe-te-iku tabe-te-k VERB GER AU X eat g o  u '(I  ) eat and go'  84 tabe-te-iru tabe-te-r PROG  u '(I  ) am eating'  tabe-te-oku tabe-t-ok in advance  u '(I  ) eat in advance'  tabe-te-ageru tabe-t-ager to do (fo r you) you tabe-te-oru tabe-t-or PROG  u '(I u '(I  ) will eat for ' ) am eating'3  Vowel Deletion applies only in those compound verbs which are composed of a sequence of a main verb followed by an auxiliary verb. Th e root of the main verb must be followed by the gerundive suffix /te/. Th e auxiliary verb may be in any inflectional form and follows the main verb as in (1).  Auxiliary verbs have many meanings, including aspectual meanings (duration , completive, perfective, inceptive, etc.), as well as other 'modal ' meanings (Makin o and Tsutsui 1989). Th e auxiliary verbs originate as regular verbs, sharing the same roots and inflectional behaviours. However , when these words are used as auxiliary verbs as a part of a compound, often they lose their original semantic contents and function with the above features. Th e contrast between their meanings as main verbs and their meanings as auxiliary verbs is presented in (2) , using the word /oku/. Th e conjugation of /oku/ is exactly the same in (2a) and (2b).  85 (2) a . Ver b /oku/ 'put' hoN-o ok  u '(I  book-OBJ. put-VER  ) put the book' B  b. Aux . /oku/ 'do in advance' tabete - ok u '(I eat-GER i n advance-AUX.  ) eat in advance'  The number of auxiliary verbs is quite limited as listed in (3) (Makino and Tsutsui above).  (3) Lis t of Auxiliary Verbs  4  original meaning meanin  g as AUX.  iru  exist (animat e subj ec:t)  aru  exist (inanimat e subject) have been done  iku  go  do..and go, keep doing..  kuru  come  do., and come, com e to do..  oku  put, place  do..in advance  ageru  give  do..for s.o.  miru  look, see  do..and see, try to do. .  morau  receive  have s.o. do for me  oru  exist (animat e subj ect)  be doing.., have done  kureru give  be doing..., have done  s.o. does..for me  simau  put away  have done.., finish  hosii  want  want s.o. to do  The following compounds in (4) are structually distinguished from the compounds in (1). Althoug h the first member in the  86 compound is in the gerund form, no deletion takes place, even though a W strin g is created (se e Poser 1984).5  (4) compound s  Vowel Deletion  tabe-te-aruku — eat wal k  *tabe-t-aruku '(I  yoN-de-miru read se e  *yoN-d-miru  kai-te-asobu write pla y  *kai-t-asobu '(I  ) walk and eat'  '(I) read and see' ) write and play'  naosi-te-oeru -- - *naosi-t-oeru '(I  ) correct and finish1  correct finis h kait-te-au return se e  *kaet-t-au  '(I) return and see  The compound verbs in (4) have the following syntactic structure: they are composed of a 'ver b plus verb' string (exocentric) , while the others (1 ) are composed of a 'ver b plus auxiliary verb' string (endocentric) . Th e second part of compound verbs in (4) maintains the original semantic contents and plays a role as a regular verb; for example, a noun such as /sosite/ 'and ' may intervene as a conjunction between the two verbs, resulting in /tabete-sosite-aruku/ 'ea t and walk around'. O n the other hand, no word may intervene between the main verb and the auxiliary verb in the case of (1) . Th e difference between the two usages is illustrated in (5).  87 (5) Syntact c Differenc e a. Verb + Verb: Vowe l Deletion does not apply. tabe-te-aruku *tabe-t-aruk u 'wal k around and eat' eat wal k VERB VER B tabete sosite aruku eat an d wal k VERB NOU N VER B b. Ver b + Aux.: Vowe l Deletion is applicable tabe-te-iru tabe-te-r eat -in g VERB AUX .  u 'b  e eating'  *tabete sosite iru eat an d -in g VERB NOU N AUX .  In Lexical Phonology, Mohanan (1990 ) deals with tw o kinds of compounds i n Malayalam and Sanskrit: one type is sub(ordinate ) and th e other type is co(ordinate). Sub-compound s have the common 'hea d + modifier' structure, for example, the 'ver b + aux.' construction /tabete-iru/ , while co-compounds have th e structure 'hea d + head1 , suc h as the 'ver b and verb' constructio n /tabete + aruku/. Th e two kinds of compounds belong t o different derivational stages . Th e application of phonological an d morphological rule s to compounds i s determined according t o what stratum th e compounds belong to . I n the case of Japanese compounds, Vowel Deletion belongs t o the strutum where 'ver b + aux.' compoundin g take s place.  (6) sub-compound  : tabete-ir  co-compound: tabete-aruk  u V u V  + Aux. + V  88 In the following compound verbs (7) , Vowel Deletion does not apply. Althoug h the compound is composed of two parts, a regular verb and an (auxiliary ) verb, it has a different morphological structure. The first member is not in the gerundive form, but in the conjunctive form with the suffix /i/ at the root-final position, or just the root (fo r vowel ending roots) without any modification.  deletion (7) compound s *ari-eru/uru ari - eru/uru exist be possible  'possible to be'  oci - au drop mee t  *oc-au  'arrange meeting'  tabe - aruku eat wal k  *tab-aruku  'walk around and eat'  kaki - oeru write finis h  *kak-oeru  'finish eating'  tabe eat  *tab-naosu  'eat again'  tori - kaeru take chang e  *tor-kaeru  'change with something'  hasiri - mawaru run tur n  *hasir-mawaru  'run around'  tabe - sugiru eat pas s  *tab-sugiru  1  over eat'  *toor-kakaru  1  come about'  kaki - haj imeru write star t  *kak-hajimeru  'start writing'  mi - tsuzukeru look continu e  *m-tsuzukeru  'keep looking'  moosi - komu say tak e  *moos-komu  'apply for'  toon pass  naosu fix  kakaru start  89 The structure here belongs to the co-compounding, 'hea d + head' type, since most of the cases the second member of compounds maintains its original semantic content. Th e difference between the two types of co-compounding is that one is 'lexical ' and the other is 'syntactic' : as discussed earlier, in the syntactic compound, a word such as /sosite/ 'and' may be inserted, however it does not apply to the lexical compound. Fo r instance, the construction of /tabe-naosu/ 'ea t again' in (7) is a lexical compounding, on the other hand, /tabete-aruku/ 'ea t and walk' in (4) is a phrasal concatenation.  (8) co-compounding : tabe-naos co-compounding: tabete-aruk  u lexica u phrasa  l l concatanation  The number of compound verbs in (7) is large. The y are produced rather freely (les s restricted compared with the compounds in (1)). Som e of them have become idiomatic expressions (Pose r 1984, Ishihara 1991).  In the above compound verbs (7) , Vowel Deletion does not apply even when the vowel hiatus is created: for example, /tabearuku/ 'wal k around and eat' does not form */tabe-ruku/ or */tabaruku/. Th e morphological difference between the two types of compounds is explained as follows. Th e compound verbs in which deletion applies consist of two free morphemes, and the other compounds consist of a string of a bound morpheme and a free  90 morpheme. Fo r instance, the gerund for m /tabete / 'eat ' can be used i n isolation such as /tabete / IMPERATIVE; on the other hand, the conjunctive for m /yomi / 'read ' cannot be used alone, unless it is reduplicated as /yomiyomi / 'whil e reading'. Vowe l Deletio n applies only between two free morphemes.  (9) a . sub-compounding b. co-compounding  : yoNd e - iru 'b [free] [free ]  e reading'  : yom i - oeru 'finis [bound] [free ]  h reading'  The string consisting of the 'verb(GER ) + aux.' must contain a vowel hiatus fo r Vowel Deletion to apply. Vowe l Deletion occur s in the lexical sub-compounds which consist of two free morphemes.  (10) sub : tabete-ir u free-fre co: tabe-naos co: tabete-aruk  u bound-fre u free-fre  e lexica e lexica e phrasa  l Vowe  l Deletio n  l * l *  1.2. Vowel s fo r Deletion  The structure of the compound verbs with respect t o Vowel Deletion was defined i n the previous section . Thi s sectio n examines which vowels are subject t o deletion and why. Wha t vowel i s deleted fro m the string? Wha t determines th e selectio n of the vowel? I n the string of W create d between the verb and the auxiliary verb, either one of them can be dropped.  91 (11) [ [ [ t a b e ] t e ] i r u  ]  I  word boundary  There is inconsistency about the site of deletion. Wh y is /i/ deleted in (11), and /e/ dropped in other cases as in (1) ? T o answer this, we need to consider the nature of vowels and formulate the conditions for deletion, because the site of Vowel Deletion can be either the verb final or the auxiliary initial position. Th e environment for /i/ or /e/ deletion is listed as in (12 ) :6  (12) a. /i / deletion: tabe-te-iru  tabe-te-ru 'eating  kai-te-iku  kai-te-ku 'writ  ' e and go'  b. /e / deletion: tabe-te-oku  tabe-t-oku 'ea  kai-te-ageru  kai-t-ageru  Deletion in V2:  VI V 2 /e/ + /i/  Deletion in VI:  /e/ + /a/ /e/ + /o/  t in advance'  /i/ deletes first in V2; if there is no /i/, /e/ of the gerundive suffix /te/ deletes in V2.  1.3. Sonorit y Scale and Underspecification  92 By comparing the above vowels, we can relate the position of each vowel in the vowel inventor y to deletion; /i / and /e/, [-low], [-back ] vowels, are subject to deletion. I n the following triangula r schema , the [+high] , [-back ] vowel /i/ is most deletable, and the [+low] vowel /a/ is placed at the bottom as most stable . However , there is no evidence here distinguishing /o / and /a/, or even /e/ from /o/ and /a/. /e / drops because it is part of the gerundive suffi x /te/ . I f the auxiliary verb has the initial /i/ , the final segment in the main verb drops (cf . Pulleyblank 1988) . /o / and /a/ can be grouped together in terms of deletion, if other elements, such as tones, are not involved.  The phenomenon under consideration can be related to the sonority scale , if we interpret sonorit y in terms of height and backness features ; the more sonorous a vowel is, the greater the chance the vowel can survive.7  (13) Sonorit  y Scale Expressed in the Vowel Features  (cf. Ladforged 1982 , p. 204) [high] deletable ^  i (u  ) .  -e o  [back a, • [low]  stable  ]  8  93 Sonority prominence has been studied extensively in previous literature (e.g . Goldsmith 1990). I n Optimality Theory (Princ e and Smolensky 1993) segments are expressed in terms of P (peak) and M (margin ) in a syllable internal structure, consisting of nuclei, onset and coda; that is, more sonorous segments make better syllable nuclei. Prominenc e scales are expressed as follows:  (14) Syllabl e Position Prominence: P  > M (P=peak , M=margin)  Segmental Sonority Prominence: a  > i > ... > t  The definition of /t/ and /a/ is that in a given language, /t/ denotes a segment of minimal sonority and /a/ a segment of maximal sonority. Base d on the above scales, Margin and Peak Hierarchies are expressed as follows:  (15) Pea k Harmony: P/ Margin Harmony: M/ Margin and Peak Hierarchy: *M/  a > P/i >...> P/ t t >...> M/i > M/a a » *M/ i » *P/ i » *P/ a (= *M/i » *P/i )  In the above hierarchy, *P/\ indicate s that X must not be parsed as a syllable Peak (i.e. , associated to Nuc(lei)), and in *M/X. mus  t not be parsed as a syllable Margin (i.e. , associated  to Ons(set) or Cod(a)).  94 The sonority scale approach can be compared with the Theory of Underspecification. Accordin g to Archangeli (1988) , /i/ is an unspecified vowel in Japanese.9 (16) Underlyin g Representation of Japanese Vowels i e a o u H L + B + +  This corresponds in part to the sonority scale. I n the Theory of Underspecification, the asymmetrical segment is in part determined by the frequency of being used for epenthesis (or deletion). I t is reasonable to state that /i/ is deleted unconditionaly in Vowel Deletion, because it has unspecified status. However , Underspecification does not necessarily always correspond with the sonority scale. Th e unspecified vowel is determined by language specific featural properties. Therefor e underspecified vowel can be even more sonorous segments, such as /a/ or /e/. Fo r example, /a/ is unspecified in Latvian, and /e/ in Spanish (Archangel i above, pp. 28-37). Furthermore , if we assume, for example, that /a/ is more sonorous than all other vowels, this is not encoded in terms of underspecification. I n a sonority scale approach, either (13 , 14, 15), segments (vowels ) can be ordered hierachically; on the other hand, Underspecification does not provide any explanation why other vowels delete as well. However , both sonority prominence and Underspecification account for Vowel Deletion; since only /i/  95 deletes from V2, otherwise /e/ deletes from VI. 2. Tona l Interaction  This part discusses tonal interaction in Vowel Deletion. I t will be shown that a lexical high tone blocks Vowel Deletion. There are two types of auxiliary verbs with regard to the accent (denoted as H): accente d auxiliary verbs and unaccented auxiliary verbs.10 Whethe r an auxiliary verb has the tone or not is determined by pronouncing it in isolation.11 A n accented auxiliary verb shows the HL tonal curve. Fo r example, /aru/, 'have done' is pronounced with the HL contour, whereas the unaccented auxiliary verb /ageru/ 'do for you' with the LHH melody does not show the HL falling contour in the citation form.12 Th e distinction between the two plays an important role in deletion. I  f there is a lexical tone in the auxiliary verb,  deletion is blocked. I n order to explain this blocking effect, Stress Clash and the Obligatory Contour Principle (OCP ) will be introduced first, then the high tone deletion will be formulated.  2.1. Accente d vs. Unaccented Words  A compound verb that contains an accented auxiliary verb /a^ru/ does not undergo Vowel Deletion, while an /a/-initial unaccented auxiliary verb like /ageru/ 'do for someone' is subject to deletion.13 I t is thus possible to say that the  96 lexical high tone in /a\ru/ blocks deletion, because Vowel Deletion does apply in /afcjeru/ as shown in (14). Th e construction of each word is as follows: for example, /EajbeteaVu/ is broken down into two parts, /taibete/ and /a\ru/, of which /taibete/ is a gerund verb, and /ajru/ is an auxiliary verb, /tapete/ consists of two morphemes: the verb root /tabe/ and the gerundive suffix /te/. /aru / consists of the root /ar/ and the present tense suffix /u/. I n the analysis of compounds, I only distinguish the first member and the second member of compounds, /taibete/ and /a^ru/. (17) a . H  I  /aru/ accented main verbs compounds Vowe  l Deletion  EaJbe-tepaBru  *tift)e-t-aru  'have been e a t e n '  yopsr-defajru  *y5pNf-d-aru  'have been r e a d '  mTVte-jajru  *ml|-t-aru  'have been seen 1  u n a c c e n t e d main v e r b s mafsi-te-ajru  *majsi^t-ajru  'have been added'  kati-te-alru  *kajri-t-alru  'have been borrowed'  sij-te-ajru  *sij-t-atru  'have beem done'  b. /ageru  /  accented main verbs  97 tUbe-te-ageru tabe-t-ager  u 'ea t (i t for you)'  yoflST-de-ageru  yojN-d-ageru  ' r e a d ( i t f o r you) '  "miL-te-ageru  mi^-t-ageru  ' l o o k a t ( i t f o r you) '  unaccented main verbs maisi-te-ageru masi-t-ager  u 'increas e (i t for you) '  ka'ri-te-ageru ka]ri-t-ager  u 'borro w (i t for you) '  z  sif-te-ageru si/  t-ageru 'd  o (it for you) '  We can summarise the conditions for deletion, including the case of the accented words, in the next following statements.  (i) Vowel s for deletion: /i/ is deleted from V2. /e/ in VI is deleted, if there is no /i/ in V2. (ii) Th e lexical high tone is the blocker for deletion.14  These observations are illustrated in the following table. 15  H  (18)  1  e+i  e+o  e+a  e+a  /i/ deleted  yes  N.A.  N.A.  N.A.  /e/ deleted  N.A.  yes  yes  --  2.2. Stres s Clash and OCP  This section explores why the accented auxiliary verb blocks  98 Vowel Deletion, by referring to the principles of Stress Clash and the Obligatory Contour Principle (OCP ) (McCarthy 1986, and others) . Th e following accented word /tajbete/aVu/ 'have already been eaten' has lexical tones in the input; Vowel Deletion does not apply.  (19) inpu  t  H H [tabe-te] [aru]  output Vowe H L  I AI  H  l Deletion  L  I  [tabete-aru]  * [t^be/t-ajru]  The sequence above does not create Stress (accent ) Clash, since the juxtaposed high tones are not adjacent to each other in the output form. Stres s Clash is formulated by Myers (1991 , p. 318): "Th e Clash Filter forbids adjacent entries on the metrical grid (o r *xx, where x is a grid entry)." Th e domain for Stress Clash in (19) is derived by compounding. I n /Eabeteja^ru/ the two high tones are not adjacent to each other, because low boundary tones are assigned, creating an alternating HLLHL string in the output; Stress Clash does not account for this. Instead , this may be considered as avoiding having two lexical high tones in a single domain prior to the assignment of low tones, since compounding precedes the low boundary tone insertion. Th e string with two peaks, such as /taJbetejalru/, is tentatively considered as an OCP violation, in which two identical adjacent elements are prohibited in the melodic level; since morphology and phonology may freely create OCP violations (Yi p 1988, p. 66).  99 (20) Obligator y Contour Principle (OCP) At the melodic level, adjacent identical elements are prohibited (McCarth y 1986, Prince 1983, Hyman and Pulleyblank 1988, Yip 1988, 1991, Myers 1991, Archangeli and Pullleyblank (i n press), and others).  In the word (19) , the adjacent identical high tones are prohibited within the same domain.  (21) *  H H [tabetearu]  This must be avoided, and therefore the two constituents /tatoete/ and /afru/ in (22b) cannot be put together in the same domain. Instea d they form two separate tonal domains. Ton e melodies are mapped on each domain as in (22a)--here is no OCP violation. Vowe l Deletion cannot apply between the two domains. So what about the compounding? Morphologically , we still consider /"taibetef ra]ru/ to be a single compound. However , tonal domains are different from morphological domains. Ther e must be two tonal domains in a single compound--there is a gap between two types of phrasing.  (22) a  . * H H  II I  b. H  H  I  [tabetel [aru ] [tabetearu  ]  100 There are two possible domain assignments as in (22 ) . I n (22b ) we have th e domain of the OCP (an d of deletion) marked withoc; then because th e OCP prevents analysis (22b) , i t blocks deletion. In (22a ) Vowel Deletion does not apply between two tonal domains, even the OCP is not violated. ( I will come back t o this point later.)  Another problem arises at this point; why does th e lexical high tone in /aru/ blocks Vowel Deletion in the string which consists of an unaccented verb followed by an accented auxiliar y verb?  16  (23) maisite-a\r  u - - *ma/sit-a\r u 'hav  yoNde-ajru -  - *yoNd-a]r u 'hav  e been increased ' e been invited'  ka/rite-afrru - - *kjjrTt>liin i 'hav  e been borrowed'  sj/te-ajru -  - *sj/t-atr u 'hav  e been done1  kiite-alru -  - *kiit-a!p: u 'hav  e been heard'  In (23) , there is only one peak (plateau ) derived by compounding. There i s no OCP violation as in (24) , because the main verb is not accented.  (24) H  i  [masite-aru], must be th e domain of the OCP and of deletion; however, deletion  101 does not apply. Thi s is configured as follows: (25) Derivation : H Input:  [masite][aru] H  Compounding:  [masite-aru]  High tone spreading:  ftna)site-aru] L% H  Low tone insertion:  L%  [masite-aru] /m^siteatru/  Although the string above does not violate the OCP, the high tone in the auxiliaray verb plays a role in phrasing and in mapping of tone melodies. Th e Obligatory Contour Principle (OCP ) cannot provide a complete explanation for the above case; that is, why Vowel Deletion is blocked in the above string even when it does not violate the OCP--only one lexical tone exists within a single domain in (25). Compar e the words in (23) with the following compounds that consist of a string of unaccented constituents.  (26) compound s deletio mateite-iru ma'sit-er yoNde-iru yoJMd-er  n u 'b u 'b  e increasing' e calling'  102 kajrite-iru  ka/rit-eru  'be borrowed'  sijte-iru  si/te-ru  'be doing'  kiite-iru "kiite-r  u 'b  e listening'  In (26) , neither the first or the second component contains a lexical tone. Vowe l Deletion applies in all strings in (26) . The lexical tone in the second member of the compound must be the blocker of deletion. I n the course of derivation, the unaccented strings in (26) are only subject to the Phrasal High Tone Assignment and Initial Lowering:  (27) Hig h Tone Assignment: assign a phrasal high tone to the final TBU in the compound, if there is no lexical tone in the string.  If the lexical tone is in the same domain, the phrasal high tone is not assigned because of the OCP. I L creates a LH rise, otherwise, it will be pronounced with a level tone. (28)  Derivation:  Input:  [masite]  [iru]  Compounding:  [masite-iru]  H i n s e r t i o n and spreading:  |ma)site-iru] L% H  Initial Lowering: [masite-iru  \ ^?\  ]  103 2.3. Intonationa l Curve  The above problematic cases may be explained from a different point of view. Th e difference in tone melodies lies in the existence of lexical tones and phrasing. I n order for Vowel Deletion to apply, the second member of the compound must be accentless. Thi s can be seen by classifying possible tone melodies of compound verbs which undergo deletion. The y are classified into four types, depending on the existence of the underlying lexical high tones in both verbs and auxiliary verbs. The classification is not based on surface tonal properties. I n the list below, H stands for an accented word and o for an unaccented word.17 Th e classification defines the type of the compounds that are subject to deletion.  (29) a. H  +p  compounds  deletion  Eajbete-iru  T^ajbe-te-ru  'be eating'  yoTN-de-iku  yoHSf-de-ku  'read and go'  TcIVte-ageru  k~ij-t-ageru  'come (fo r you)'  tojt-te-oku  tojt-t-oku  'take it in advance'  b. 0  + 0  kafci-te-iru  kajri-te-ru  'be borrowing'  ma/si-te-iku  majsi-te-ku  'have been increased'  yoN-de-ageru  yoN-d-ageru  ' i n v i t e (you)'  104 sij'-t-oku  do it in advance  taSbe-tefabru  *"Eaibe-t-aru  'have eaten'  yoBST-defajru  *yoPJ-d-aru  'have read'  Tca\i-tefa\iru  * Icajj-t-aru  'have written'  tb^t-tefafrru  *tojt-t-aru  'have taken'  siy-te-oJcu  c. H  d.  + H  0 +H  kajri-te-ajru  ^kgfri-t-atru  'have been borrowed'  ma/si-te^alru  ^ma/si-t-airu  'have been increased 1  yoN-de-a|ru  *yoN-d-alru  'have been called'  sJ/-te-airu  *si/-t-airu  1  have been done'  Among all the compounds above, Vowel Deletion applies only in (29a,b). Th e sequence of the two constituents must be [H ] + [0] or [0]+[0] for Vowel Deletion to apply, but it cannot be [H]+[H ] or [0]+[H]. [H]+[H ] violates the OCP, but [0][H ] does not. Th e compounds whose second member contains the lexical ton e (29c,d ) cannot undergo Vowel Deletion.  deletion  (30)  OCP  0+9  H+0 0+H  *  H+H  *  *  105 2.4. Gree k Vowel Degemination  A similar situation is found in Nespor (1979 , 1987), Nespor and Vogel (1986) . The y observe many cases of vowel deletion and vowel degemination in numerous languages. I n their research, for example in Greek, some surface strings have a left-headed stress prominence system.  (31) Gree k Vowel Degemination aj'orase ena a^orasen monuzo oti nomizot  a '(he i '(I  ^arxomuna ama "^arxomunam  a '(I  ) bought a' ) think that' ) come (condit ) if  In Greek, the secondary stress is suppressed in a word after Vowel Degemination applied. Onl y one (stress ) peak is allowed to surface.  Prince (1983 , pp. 91-95) discusses the correlation between stress systems and tonal systems in regard to the grid notation, by referring to Japanese data. Th e analysis of the stress system in Greek can be compared with the analysis of the Japanese tonal phenomenon, as we concern the prominence systems of sounds. Th e following discussion of the Greek Vowel Degemination in Nespor (1987) should be appropriate in this context for discussing the case of tonal behaviour in Japanese.  106 All these initially stressed words, however, are socalled function or grammatical words. Thi s class of words, in Greek, as in any other languages, is characterised by the possiblity of undergoing a destressing rule, or, alternatively, of having stress assigned by sentence stress assignment (cf . Kiparsky 1982), possibly after Vowel Degemination. And , in fact, in (31 ) the function words /ena/, /oti/ and /ama/ do not bear primary word stress.  The above input strings show a sequence of two stressed words; after Vowel Degemination applies, the stress of the second constituent is lost. I n the case of Japanese Vowel Deletion as well, the strings show a left-headed prominence system, [o+o ] or [H+o], which are subject to deletion.  3 . Formatio n of Tone Groups  18  This section examines the following two areas: (i) the lexical tone, the blocker of deletion--this will be examined by analyzing the case of the accented auxiliary verb /oru/. (ii ) the domains for deletion. First , the minor phrase and the minimal minor phrase will be defined in terms of the phrasal hierarchy to show the exact morphological conditions for compounds. Second , the discrepancy between morphological domains and prosodic domains will be discussed. Finally , the case of  107 Luganda Low Tone Deletion is introduced as an example of tone domain formation. 3.1. Accente d Auxiliary Verb /oru/  The accented auxiliary verb /ojtru/ will be compared with the other accented auxiliary verb /^ru/, and with the unaccented case in this section. /o>ru / has an initial accent, however, deletion can apply. Sinc e the lexical tone blocks deletion, in order for deletion to apply, the accent on /o\ru/ has to be delinked. Thi s operation is assumed to be fairly late, and applies only for /o~jru/. Th e auxiliary verb /ojru/ originated /o]ri/ 'exist' in Classical Japanese.19 I n modern Japanese, /olru/, as an auxiliary verb, has the function of progressive, completion, continuative state, etc. /oVu / is used less often, compared to other auxiliary verbs in Tokyo Japanese. Whe n /olru/ 'exist' is pronounced in isolation as a regular verb, it sounds 'old ' or 'classical' and is not favoured among young people in Tokyo. This is the same when it is used as an auxiliary verb as a part of a compound.  (32) accente d main verbs standard Vowe taibe-te/:roTru tBbe-t-or  l Deletion u 'b  e eating'  yotN-depotru yo|Sf-d-or  u '  be reading'  kaji-teFolru Tcaji-t-or  u 'b  e writing'  108 mU-t-oru  'mil-tefo]iru  'be looking'  unaccented main verbs ma/si-te-oru  ma/si-t-oru  'be increasing1  ka<ri-te-6|ru  ka/ri-t-oru  'be borrowing'  siFEe^ojru  sij-t-oru  'be doing'  In the compounds (32 ) th e verb final vowel /e/ is dropped. A s discussed earlier, the lexical tone in auxiliary berbs blocks deletion. Thi s discrepancy may be solved by stipulating that the lexical tone delinks only for /oru/. First , we take a look at only the case with accented main verbs.  (33) Lexica l Tone Delinking H -- 0 I [  H ] + oru  The above delinking is related at least to the OCP. Fo r example, delinking applies for the string /'tBbetePo|ru/ 'be eating' avoiding having two high tones in a single domain as shown below:  (34) H  I  H  I  Delinkin  g H  I I  Deletio  [tabete][oru] [tabete-oru  n H  ] [tabet-oru  ]  Delinking must precede Vowel Deletion, since it provides the domain for Vowel Deletion. I f the order were reversed, they would be in a bleeding order as below.  109 (35)  H  ii  H  i  Deletion  * H  [tabete][oru] [tabet-oru  i  H  I  Delinking  H  ] [tabet-oru  ]  In (35) , */ta]bejto(ru/ is ill-formed because of the OCP. Th e optional delinking for /oru/ must be ordered as follows:  (i) Delinkin g (ii) Vowe l Deletion  However, Delinking formulated in (33) does not apply to the case with unaccented main verbs. Fo r example, when the unaccented verb /masite/ 'increase' precedes /oru/, delinking does not apply. Th e string o+H does not meet the condition for delinking in (33).  (36) H  *  [masite] [oru ] *  delinking Vowe  l Deletion  [ma/site-oru] [ma/sit-oru  ]  The string */ma/si teoru/ cannot surface without Vowel Deletion; delinking cannot apply. Notic e that the input [0]+[H ] string can form a single tone domain without delinking, since there is only one high tone; the OCP is not violated. Delinkin g applies in order to provide a single tonal domain for deletion. Thus , Vowel Deletion must immediately follow compounding.  110 (37) H  compoundin g L  I\  % H  L% Vowe l Del. L H  ^\ \ \  [masite] [oru ] [masite-oru  L  A I  ] [masitoru  ]  delinking I I L H  L L  H  I A *i  [masitoru] [masitoru  / ^  ]  The fina l lo w tone may be optionally delinke d i n the fully specified strin g afte r Vowe l Deletio n take s place ; however , thi s is superficia l reorganizatio n o f the tone melody, triggere d by the los s of the syllable. Delinkin g doe s not apply i n the string, sinc e onl y on e lexical ton e exist s i n the domain. Ther e is no motivation fo r delinking, sinc e th e OCP is not violated. These operation s ar e ordered a s follows: (38) (i) Lexica  l Ton e Delinkin g  (i i) Compoundin  g  (iii) Ton e Melody Assignmen t (iv) Vowe  l Deletio n  (v) Phrasa (vi) Lo  l High Tone Assignment  w Tone Delinking  These operations an d possible output strings are illustrated i n the following table.  Ill (39) Delinkin g And Its Application, H delinking  input H H tabete oru  N.A. H [tabete][oru]  H masite oru  compounding  Vowel Del. L delinking  H i H L [tabete][oru]  N.A.  N.A.  [tabelfeo'ru]  [tabefofu]  N.A.  [masiteoru]  [masitoru]  N.A.  i_  N.A.  L ^^7&  [masitdru] Note that delinking (I,II ) and Vowel Deletion are postlexical operations in casual speech. Al l the tone melodies illustrated above are superficial variations. I n order for Vowel Deletion to apply, a single tone domain must be formulated. A  question here  is what the two constituents H and 0 exactl y mean in terms of phrasing? Wha t kinds of characteristics do they have? W e will find answers to these questions by identifying the appropriate domains for each string in the next section.  3.2. Mino r Phrases and Minimal Minor Phrases  In order to analyze the accented auxiliary verbs in a phrasal context, we need to consider the phrasal hierarchy in tonology, and define the phrasal terms necessary for the discussion. Especially , the definitions of the minor phrase and the minimal minor phrases in the hierarchy are rather confusing and must be explicated in order to demarcate each string into the  112 appropriate tonal domain. (40) Phrasal Hierarchy20 Utterance Major Phrase Intermediat  e Phrase  Minor Phrase Accentua  l Phrase  Minimal Minor Phrase Prosodi  c Word Syllable Mora  (Poser 1984) (Pierrehumber  t and Beckman 1988)  In this thesis, phrasal terms are based on Poser's, as shown in the above hierarchy. Pierrehumber t and Beckman (1988 , pp. 118-120) illustrate the Phrasal Hierarchy with a sample utterance /ape-no a|kai seefcaa-wa dojjko desu-ka/ 'Wher e is my big sister's red sweater?'. Originally , the term 'Intermediat e phrase' was used for 'majo r phrase', 'accentual phrase' for 'mino r phrase', and 'word ' was used to indicate 'minima l minor phrase'. I n (41) and (42 ) the above terms, as used in their discussion, are converted into our terms for convenience.  113  utterance (u)  (41)  major phrase (i) a  ^a  ^A  @@ @@  ii l I  m1 m  1 @ @@  III  m mm 1  minor phrase (a)  a ~w @  ,A  \A %  N  w  @@  @ @  I i I  minimal minor phrase(w)  w  syllable  @  | I  i  ' ' U - IL' I i „ U J I J m  mm mm  m  mm  m m  m  ane-no aka i seetaa-w a dok o desu-k a big sis. red sweate r wher e is (42)  w  w  A  @ e s  @ @@  mm m  m mm  / I  L% H  ane-no  w  /I  L% H  mora phoneme tier gloss.  minimal minor phrase syllable  mm mm m  mora  HL  tone tier  I  seetaa-wa  phoneme iter  akai In (41) , the prosodic structure for a complete sentence is presented, and (42 ) represents associations between morae and the tone tier for a portion of the utterence. Th e low boundary tone is associated with the initial mora in each minor phrase (=accentual phrase). /ajcai / 'red' and /seeitaa-wa/ 'about sweater1 ar e linked together under the same minor phrase node. Initial Lowering indicates the boundary between two minor phrases, except for the words that have a lexical high tone on their initial vowels. The following descriptions are the summary of Selkirk and Tateishi (1988 , pp. 322-323), discussing the distinction between the minor phrase and the minimal minor  114 phrase.  The minor phrase is comprised of the units of prosodic structure that are demarcated by instances of Initial Lowering. The term "minimal minor phrase", which is approximately the same as "prosodic word (use d by them)", designates a basic lexical item plus function word sequence. A minor phrase may consist of just one minimal minor phrase, or of more than one (McCawle y 1968, Poser 1984, Pierrehumbert and Beckman 1988, and Kubozono 1988). Th e following two examples (adde d here) show a case where two miminal minor phrases form a single minor phrase through compounding.  (43) Example s of Phrasing: Noun Compounding mp m  ii  p m  /  mmp mm p mm kajki fulra i compoundin oyster deep fried dee NOUN NOU N  p  \  p mm p g kaTci-fujra i p fried oyster  yuTci Icejsik i -- compoundin g yuTci-ge^sik i 21 snow scener y snow y scenery NOUN NOU N  The original tone melodies are mapped onto the words in the left column in (43). Th e forms in the right column show that the two minimal minor phrases are put together under a single minor phrase node. Whe n each word undergoes compounding or forming a larger unit, different tone melodies are mapped. Initia l  115 Lowering inserts a low boundary tone at the beginning of minor phrases, and the original tonal properties are suppressed in the derived domains. Th e compound verbs in (43) as well, can be organized under the above phrasal nodes; for instance, /tajbeteiru/ 'b e eating' is devided into two minimal minor phrases, /tajbete/ and /j£ru/, under one minor phrase. /ma'siteajru / 'have been increased in advance' also shows a single minor phrase after compounding.  (44) a .  mp  mp  J ,  I  d p  p m  p  1  b.  -  comPou„di„g  ^  41  tabete lr u tapete-ir c. m  I I  mmp mm  p  masite aru  mp  d.  \  P  u  mp mmp  J.  mmp  mdsite-a\ru  Selkirk and Takeishi (above ) further describe the distinction between minor phrases and minimal minor phrases, (summarized) as follows. Ther e are reasons independent from the tonology for distinguishing between the minor phrase and the minimal minor phrase. The y are subject to different sorts of conditions regarding accent. Th e mapping of the sentence onto the minimal minor phrase ignores accent. O n the other hand, the minor phrase is sensitive to accent; one tonal peak is assigned per minor phrase. Pose r (1984 , p. 101) describes this fact as  116 1  the smallest unit to have independent intonational  assignment' .22 3.3. Prosodi c Domains vs. Morphological Domains  There is a problem with the distinction between the minor phrase and the minimal minor phrase. Th e minor phrase is 'the smallest unit to have independent intonational status' (Poser above); however, there is a discrepancy between the morphological domain and the tone domain ((TD ) or prosodic domain; for this analysis, cf. Inkelas 1989, Ishihara 1991). Conside r the following words:  (45) [tabete  ] [iru ] a mmp mm p m  . [tajbeteiru] p  b.  [tiibeteiru] TD  [ t a b e t e ] [aru] mmp mmp  c.  [tajbetq[alru] mp  d. [tBbete l [ajru ] TD T D Each bracketed word in the left column refers to minimal minor phrases. Th e two minimal minor phrases are, morphologically, united into a single minor phrase by compounding. Fo r tone melody assignment, each bracketed word in (45b,d ) refers to a single tone group, since it can be the smallest site for tone melody assignment. I n (45a,b) the demarcation of minor phrases and TDs coincides; however, in (45c,d) minor phrases and TDs  117 disagree. /ta^bet^Hru/ has two tonal peaks which must be split into two tone groups. Th e discrepancy between the two types of domains, morphological domains and tonal (o r prosodic) domains, can be solved by using the term "tone domains" alone for the demarcation of strings. A "ton e domain" contains only one tonal peak. I f there are two peaks in the string, regardless of morphological conditions, it must be split into two tone domains. This means that in [tabeteiru] (45a,b) , the minor phrase is the TD, and in [tcjbetejtlru] , the minimal minor phrase is the TD. Thi s can be characterized in terms of a 'Rankin g Schema' in Optimality Theory (McCarth y and Prince 1993, pp. 102-103).  (46) Rankin g Schema: P »  M (P=prosody  , M=morphology)  If some morphological domain is to be prosodically conditioned, then in that domain P must dominate M. Prosodic Circumscription can be seen as fixing a dominance relation between the well-formedness requirements of two different domains, prosody (P ) and morphology (M) . .. . I n templatic and circumscriptional morphology, the constraints on prosodic structure .. . take precedence over the constraints on morphological structure.  The prosodic domains derived after Vowel Deletion are illustrated in the following table.  118 (47) Vowe l deletion and Tone Domains Delinking e+<i> <e>+a <e>+o H  1  e+a  <H>  1  <e>+o  23  Vowel Deletion  mp=TD  --  yes  yes  --  --  --  yes  yes  yes  Compounding derives one TD for the strings that contain unaccented auxiliary verbs. However , for the strings that contain accented auxiliary verbs, the lexical tone blocks formation of a single TD. Th e optional delinking provides a single domain for deletion. Afte r Vowel Deletion applies, only a single TD appears.  3.4. Lugand a Low Tone Deletion  Vowel Deletion in Japanese is not the only case where a segmental change interacts with tone melody assignment. Th e next example, taken from Hyman, Katamba, and Walusimbi (1987) , including data, discusses the formation of the Tone Domain (TD) in Luganda, one of the Bantu Languages spoken in Uganda. Lugand a exhibits some similarity to Japanese in terms of its pitch accent system (MacCawle y 1979). On e of the morphological rules in Luganda deletes a low tone when surrounded by high tones in order  119 to form a H-plateau. Th e (postlexical ) domain in which Low Tone Deletion applies is called the Tone Domain.  (48) L  Tone Deletion (LTD) L -- <t> I  H H  Any form within the same clause may join the verb to form a TD; however, when the clause begins with the so-called initial vowel (IV) morpheme, LTD is blocked and the TD is not formed.  (49) * L -- 0 / I  V  With few exceptions, all common nouns can take an IV. Th e sentences in (50 ) lack the IV. Th e sentences in (51) illustrate the present-tense with noun objects having an IV.  (50) tu-labH 0  a bi-kopo 'w H L  e see cups'  tu-lab-a mu-kazi 'w  e see a woman'  tu-lab-a ma-futa 'w H 0 H L  e see oil'  U0  (51) tu-labH L  H  L  a e-bi-kopo 'w H L  e see cups'  tu-lab-a o-mu-kazi 'w H L H L  e see a woman'  tu-lab-a a-ma-futa 'w H L H L  e see oil'  120 Depending on the noun class, the IV is either /e-/ , /o-/, or /a-/, as seen. Accordin g to Hymn, Katamba, and Walusimbi (above ) the semantic difference between (50 ) and (51 ) is one of focus: the sentences with IV have relatively even focus, while those without IV focuses on the object. Th e sentences without IV form a single tone group in which LTD applies. O n the other hand, IV blocks LTD from applying, and the two separate melodies are left on the strings. A s the conditions of LTD are rather complex (Hyman and Katamba 1989, 1993), the difference between the two types of domains is focused: the domains in which LTD applies and forms a TD, and the domains in which LTD does not apply. Formation of a single TD is illustrated in the following table. (52) T D Formation IV  one TD  LTD  H <L> H  --  yes  yes  H L IV H  yes  --  --  In (52) , the string without IV, /H <L> H/, and the string with IV /H L IV H/ are contrasted. I n /H <L> H/, LTD applies, and a single TD is formed. I n /H L IV H/, IV blocks deletion; a single TD cannot be formed. <L > indicates the low tone to be deleted.  4. Summar y  The blocking effect on phrasing in Vowel Deletion has been  121 examined; a single tone domain is formed after deletion applies. The next table displays the application of Vowel Deletion in different types of segment strings, adopting an 'Optimalit y Theoretic' approach. Th e following table displays vowels and conditions for Vowel Deletion. Segment s or tones which are deleted from the input strings are marked with <>. (53) Vowe l Deletion: Conditions for Application sonority  lexical H  dekinking  mmp=TD  Vowel Deletion  e+<i>  --  --  --  --  yes  <e>+o <e>+a  yes  --  --  --  yes  e + a  yes  yes  --  yes  --  yes  yes  yes  <¥>  <e>+o  yes  In (53 ) the conditions including constraints are listed in the horizontal line. Unlik e Optimality Theory, there is no hierachical order among them. A set of vowel strings at the morpheme boundary in compounds is presented in the vertical line. The first three strings refer to the compounds whose second members are unaccented; among them /<e>+o/ and /<e>+a/ are put together, because they behave exactly in the same way with regard to deletion. Th e fourth and the fifth strings represent the compounds that have accented auxiliary verbs. Fo r all the vowel strings, the first constituent in a compound can be either accented or unaccented.  122 Among the conditions in (53) , SONORITY and LEXICAL H are input properties. SONORIT Y indicates the special status of /i/ in the Japanese vowel inventory (cf . Underspecification in 1.3). The least sonorous vowel /i/ is deleted in the second constituent of the compound; no other vowels delete in V2. Unde r the issue of sonority scale, the more sonorou s vowel creates the better nucleus. Th e distinction between /i/ and the other vowels in terms of sonority indicates the special status of /i/ in deletion (although the relation between the two issues is an open question). LEXICA L H, an underlying blocker for deletion, is marked. Th e DELINKING delinks the lexical tone in the auxiliary verb in order to form a single domain for deletion; and therefore delinking must be done prior to compounding. Th e other delinking operations do not influence Vowel Deletion, because they apply after deletion takes place. mmp=T D indicates that the minimal minor phrase is the TD; that is, the compound consists of two separate tone domains. Th e compound must form a single domain in order for Vowel Deletion to apply. Finally , the strings that Vowel Deletion can apply to are marked.  As we have seen in both Luganda and Japanese, the blockers, whether IV or lexical H, play a role inhibiting forming a single tone domain. Throug h phonological changes, whether Low Tone Deletion or Vowel Deletion, only one single tone domain is derived.  123 5. Conclusio n  Vowel Deletion exhibits an interesting property regarding the interplay between the segmental, prosodic, and tonal domains: for instance, in Vowel Deletion, phrasing exhibits the discrepancy between the morphological and tonal analyses. Tw o types of phrasing are possible for the compound verbs: one is to form a single tone domain (= a minor phrase), and the other is to form two tone domains (=tw o minimal minor phrases). Thi s is due to the existence of the underlying lexical tone on the second member of the compound. Th e OCP accounts for the cases of /aru/ or /oru/; the OCP prevents the derivation of two high tones in a single domain--if there are two high tones in a string, the string must be split into two tone domains, or delinking of one of the high tones must apply. Th e derived phrase for Vowel Deletion is one type, involving a single tone domain in which only one tonal peak is assigned. I n other words, Vowel Deletion is applicable domain internally, but never applies at the phrasal boundary. Th e Vowel Deletion facts in Tokyo Japanese are not merely a language specific phenomenon when we consider the interaction with assignment of tone melodies and its domain for application. Rather , it is seen in languages such as Greek, Luganda, or Yoruba (Pulleyblan k 1988), especially in colloquial speech forms.  124 NOTES  1. Th e minor phrase is a smallest domain for intonational assignment. Th e minimal minor phrase is equivalent to the prosodic word. Th e minimal minor phrase (usually ) consists of one content word plus a function word (se e Poser 1984, and section 3.2 in this chapter). 2. Se e Soga, Koyama, and Ohso (1987 ) for Vowel Deletion in compound verbs. 3. /oru / is not often used in the Tokyo dialect compared with the other auxiliary verbs. /oru / will be analyzed later in this chapter. 4. Honorifi c auxiliary verbs, such as /kudasaru/ 'give', /idadaku/ 'receive' , or /sasiageru/, 'give' , are not included in the discussion. Ther e are some differences between honorific auxiliary verbs and regular auxiliary verbs in terms of pragmatic and semantic restrictions. 5. Th e gerund /te/ becomes the voiced /de/ as in /yoNdemiru/ when preceded by /N/. Thi s voicing assimilation is discussed in Ito and Mester (1986) , Ito, Mester, and Padgett (1993), and others. 6. N o other combination of vowels is found. Se e the list of auxiliary verbs in (3). later.  7. Th e difference between /o/ and /a/ will be discussed  8. I n the vowel inventory in Japanese /uV is unrounded. AM/ does not appear in the initial position in the auxiliary verbs in question. 9. /a / is also considered as an unspecified vowel in Japanese (Grigno n 1984). 10. Kubozon o (1986 ) characterizes the lexical high tone as 'sudden drop in FO contour' (se e Chapter I). Th e lexical high tone is immediately followed by a low tone (cf . Pierrehumbert and Beckman 1988). 11. Th e tonal representation here skips the process of linking of the lexical tone. I use only the gerund and the citation forms.  125 12. /aru / originally means 'exist' ; however, when it is used as an auxiliary verb, the original meaning is lost. Fo r example, /tabete-aru/ means 'hav e been eaten already'. Whe n /aru/ i s attached to typical action verbs, such as /tabe/ 'eat' or /yom/ 'read' , it functions with the meaning of 'hav e been done in advance'. Whe n /aru/ is attached to /ake/ 'open' or /shime/ 'close' (transitiv e verbs), i t indicates 'continuou s state' (cf. Soga and Matsumoto 1990). The LHH melody is derived as follows. /ageru / is an unaccented word, and therefore it is pronounced without HL fall (technically, it is done by assigning a high tone to the final TBU; th e high tone spreads to the left). Initia l Lowering lowers the initial syllable, deriving the LHH melody. Se e Haraguchi 1977, 1991 , Selkirk and Tateishi 1988, 1991, Pierrehumbert and Beckman 1988 for Initial Lowering. 13. /ageru / is subject to another type of deletion. Som e colloquial forms have the contraction as follows. The y are derived through phonological deletion. kiite-ageru kiitar 0 0 0  u '(  I will) ask (fo r you)'  shite-ageru shitar 0 0 0  u '(  I will) do (i t for you)'  14. Th e role of lexical tones will be discussed later. The lexical tone itself blocks deletion; consequently, the lexical tone blocks forming a single tone domain. 15. Onl y /a/ and /o/ are accented in this context. Th e case of /oru/ will be discussed later. 16. Initia l Lowering (IL ) does not apply to CW an d CVN syllables (Haraguch i 1977). 17. A s far as I observe, there is no case that places accent on the non-initial positions in the vowel-initial auxiliary verb in the Tokyo dialect in regard to Vowel Deletion. Other dialects, such as the Kansai dialects, place accent on the second syllable: e.g. in Tokyo /aru/ is initial-accented, whereas in the Kansai region, /aru/ is pronounced with the second syllable accented. In addition to the four types of melodies in (29), there are two other optional tone melodies as listed below. Th e extra melodies are enclosed. compounds deletio H+0  tajbe-te-i/ru  --  n  tUbe-teFru  'be e a t i n g '  126  H+H  yotN-de-dicu  ydN-de-fTcu  'go and read'  klili-te-age/ru  kalj-t-ageFu  'write (i t for you)'  tdjt-te-ofku  tdjt-t-c/ku  'keep (i t in advance)'  tiibe-te-aru  *tajbe-t-aru  'have eaten'  ycftJ-de-aru  *yo|fJ-d-aru  'have read'  iali-te-aru  *Tcaji-t-aru  'have written'  £ofc-te-aru  *rat-t-aru  'have kept'  As for the above tone melodies, there is no difference meaning compared with the basic tone melodies in (29 ) . Th e tone melodies are derived in the following manner. First, in pattern H+0, tone melodies are assigned to the first and the second constituents prior to compounding. Th e tone melodies both constituents are not affected by compounding. e.g,  input  tone melody assign. H L  H  L  H  compound. H L  H  in above the of  Vowel Del. H L  H  IA I A [tabete][iru] -- [tabete][iru ] -- [tabetei iru] -- [tabeteru ] It seems that this operation is a recent trend as Ishihara (1991 , p. 147) mentions. As mentioned in Poser (1984) , the younger generation has a compound accent pattern different from the one cited in this thesis. Fo r them, every compound is accented regardless of the accentedness of its constituents. In the above configuration, Vowel Deletion occurs strings as a late postlexical operation. Fo r 0+0 is no optional melody, since the two constituents lexical tones are put together by compounding and melody is mapped onto them. e.g. inpu  t compoundin  g  in the output strings, there without any a single tone  127 L H  I ^ [masite][lru] [masiteiru  ]  Second, for the pattern H+H with /aru/, delinking optionally applies to the output strings. Thi s delinking applies in fast speech; in deliberate speech, the initial /a/ tends to be hightoned. (fast speech): H  -- 0 / " H +  compounding delinkin H  H  II  H L  I  A I  H  aru - a-  g singl  L H  e TD  L H L  I 1/1*  1I  H  /^ 1  [tabete][aru]--[tabete][aru]--[tabete][aru]--[tabetearu] Note that delinking applies only for H+H strings. Th e OCP would be violated if there were two Hs in a single domain. I n o+H strings, for example in /masitearu/, delinking does not apply; delinking would result in an ill-formed melody as shown below: H [ma/sitearu]  [masite] [aru]  This fast speech rule and its application are illustrated as follows. Vowe l Deletion cannot apply for the strings with /aru/; delinking is ordered after Vowel Deletion. delinking  input H H  compounding H L  1 1  tabete aru  N.A.  •MM  H  V.Del.  delinking  L  [tabete][aru]  N.A.  N.A. H L  1 ^/\  [tabetearu] L H  H  1  masite aru  N.A.  L  1M 1  [masitearu]  N.A.  N.A.  For this thesis, only the four types in (29 ) are considered for analysis, because the existence of lexical tones is projected straightforwardly in them. Th e extra melodies are derived by  128 additional operations, such as fast speech rules, etc. The y are superficial variations and do not affect Vowel Deletion. 18. Ther e are a variety of terms used to describe phrasal domains, for example, tone groups, prosodic domains, morphological domains, tone domains, accentual phrase, clitic phrases, association domains, and so forth. Th e definitions for some of them are cited in the literature as follows: Tone Group: th e part of a sentence over which a particular intonation pattern extends (Ladeforged , 1982). Prosodic Domain: th e phonological word and the phonological phrase are proposed as the domains of postlexical rule application (Inkela s 1989; cf. Selkirk 1986). Prosodi c Domains are expressed in the prosodic hierarchy. Tonological Domains: postlexica l domains, made up of words concatenated by the rules of syntax (Selkir k and Shen 1990) . Morphological Domain: th e domain consists of stems, clitics, affixes, and roots. I t indicates morpheme combination, compounds, and bound morpheme types, and so forth (se e Inkelas above). Tone Domain: th e postlexical domains in which phonological rules, such as Low Tone Deletion in Luganda, apply (cf . Hyman, Katamba, and Walusimbi 1987; see 3.4. of this chapter). Accentual Phrase: use d in Pierrehumbert and Beckman (1988 ) to indicate the minor phrase (o r approximately equivalent to the minor phrase). Clitic Group: use d in Nespor and Vogel (1986) . Th e term indicates syntactic constituency. Association Domain: a domain in which tones may shift and spread, if the domain has more than one syllable. Th e domain is sensitive to syntactic relations; the domain is determined by a positional parameter (Duanm u 1992).  19. I n Classical Japanese, /ori/ is considered pronounced with the onset /w/. 20. Compar e with the Prosodic Hierarchy below: Prosodic Hierarchy (McCarth y and Prince 1993, Ito and Mester 1992, and others)  129 PrWd I @  m 21. Sequentia l Voicing changes the initial /k/ of /kesiki/ to the voiced counter part /g/. Se e Ito and Mester (1986) , Ito, Mester and Padgget (1993) , and others. 22. I n the illustration (41 ) and (42) , the two minimal minor phrases /a/kai/ 'red' and /see|taa-wa/ 'sweater-TOPIC for m a single minor phrase. /seeltaa / 'sweater' is an initial accented noun, but no low boundary tone is linked to the initial syllable. The experimental study conducted by Pierrehumbert and Beckman (1988) suggests that the initial lowering effect is shown on the initial accented syllable (phrase-internally) ; however, this phonetic phenomenon is not considered for phonological analysis, as they ignore it in (42). 23. Thi s table does not include fast speech effects and their rules.  130 Concluding Remarks Nasal Spreading and Vowel Deletion are analyzed in terms of three domains: segmental, prosodic, and tonel domains. I n order to specify the environments for their formations, segmental analysis must be discussed first. A s for segmental analysis, in Nasal Spreading, Rhotic Underspecification, Feature Class Theory, and Feature Licensing specify the status of /r/ and /n/ sounds in Japanese. I n Vowel Deletion, sonority prominence and the Theory of Underspecification support the claim that /i/ has special status in the Japanese vowel inventory. Prosodi c analysis deals with the output syllable templates in Nasal Spreading: Nasal Spreading is analyzed as a mora augmentation and contrasted with /a/-epenthesis, which is another type of mora augmentation in /nai/-suffixation. I n Vowel Deletion, prosodic domains are discussed, focusing on the mismatch between morphological domains and prosodic domains. Tona l processes are involved in both formations. I n Nasal Spreading, the agreement in tone melodies between the standard and the colloquial forms are expressed as the disyllabic requirement on the derived base: the constraints on Accent Shift or Initial Lowering require such a structure. I n Vowel Deletion, the lexical high tone is treated as an underlying blocker for deletion. Thes e complex environments are sorted out by adopting the Optimality Theoretic approach. A minimal number of constraints are selected and ranked, and each candidate set is evaluated. Fo r example, RecM and Rec@ makes a distinction between Nasal Spreading and /a/-epenthesis.  131 The forms in this thesis are drawn from modern Tokyo Japanese. Fe w analyses deal with such forms, especially, in a current theoretical framework. On e of the reasons is that there are quite a few variations in surface representations, due to ongoing language change. Th e judgements of the data by native speakers vary inevitably. I the analysis. I  included these surface variations in  also classified the words for both literary and  colloquial forms and specified the type of words in which the colloquial formation applies most consistently. Thi s contrasts forms that undergo the changes and with the words that are not subject to change. I n most cases, the expressions are based on the Tokyo accent spoken among young people, so that the present trend or 'th e evolution' of the language can be seen.  132  Works Cited Akinlabi, Akinbiyi. "Underspecificatio n and the Phonology of Yoruba /r/." Linguisti  c Inquiry (LI ) Vol. 24, 1993.  Alfonso, Anthony. Japanes e Language Patterns. Vol . II. Tokyo: Sophia U, 1966. Archangeli, Diana B. Underspecificatio n of Yawelmani Phonology and Morphology. Ne w York: Garland, 1988. , "Syllabification and Prosodic Templates in Yawelmani." Natural Language and Linguistic Theory. (NLLT ) Vol. 9. 1991: 231-283. Archangeli, Diana B. and Douglas Pulleyblank (a) . "Tw o Rules or One...or None? [ATR ] in Yorba." Proceeding  s of the  Nineteenth Annual Meeting of the Berkeley Linguistics Society. Genera l Session and Parasession on Semantic Typology and Semantic Universals. CA : UC Berkeley, 1993. , (b ) Qptimality, Grounding Theory, and Rule Parameters. handout. Rutger s Optimality Workshop I. Nov . 1993. , Grounded Phonology. (i n press) "Ima Tokyo-go Wa. (Toky o Dialect Now)". Asah i News Papaer. Tokyo. 14 , 15 Oct. 2 Nov. 1992. Beckman, Mary E. "Segmen t Duration and the 'Mora ' in Japanese." Phonetica 39 (1982) : 113-115. , Stress and Non-Stress Accent. Dordrecht : Foris , 1986. Clark, Mary. "Japanes e as a Tone Language." Issue s in Japanese  133 Linguistics. Eds . Imai , Takashi and Mamoru Saito. Dordrecht: Foris, 1986. Clements, G.N. "Th e Geometry of Phonological Features." Phonology Yearbook Vol. 2., 1985. Duanmu, San. Rim e Length, Stress, and Association Domains, ms . U of Michigan, 1992. Goldsmith, John A. Accent Systems, ms . Indian a University. 1982. , Autosegmental and Metrical Phonology. Ne w York: Basil Blackwell, 1990. Grignon, Anne-Marie. Phonologi  e Lexicale Tri-dimensionnelle du  Japonais. Diss . Universit e de Montreal, 1984. Haraguchi, Shosuke. Th e Tone Pattern of Japanese: A n Autosegmental Theory of Tonology. Tokyo : Kaitakusha, 1977. , "The Multi-dimentional Grammatical Theory." Japanes e Linguistics. Eds . Imai , Takashi and Mamoru Saito. Dordrecht: Foris Publications, 1986. , A Theory of Stress and Accent. Dordrecht : Foris, 1991. Hasegawa, Nobuo. "Casua l Speech vs. Fast Speech." Paper s from the Fifteenth Regional Meeting. 19-2 0 Apr. 1979. Chicago : Chicago Linguistic Society, 1979. Hewitt, mark S. Deconstructin g Foot Binarity in Koniag Alutiiq. ms. U  of British Columbia, 1994.  Hyman, Larry M. "Ton e and/or Accent." Element s of Tone, Stress and Intonation. Ed . Napoli, D.J. Washington: Georgetown UP, 1978.  134 Hyman, Larry M. and Francis X. Katamba. Th e Augment in Luaanda Tonology. ms . UC , Berkeley; U of Lancaster, 1989. , "A New Approach to Tone in Luganda." Languag e Vol. 69, 1993. Hyman, Larry M., Francis Katanba, and Livingstone Walusimbi. "Luganda and the Strict Layer Hypothesis." Phonolog y Year Book Vol. 4, 1987. Hyman, Larry M. and Douglas Pulleyblank. "On Feature Copying: Parameters of Tone Rules." Language , Speech and Mind. Eds. L.M. Hyman and C.N. Li. London : Routledge, 1988. Inkelas, Sharon. "Prosodi c Constituency in the Lexicon". Diss . Stanford U, 1989. Ishihara, Masahide. " Verbs." Diss . U  Tiie^ijejdc^J^jQ^o&i^ of Arizona, 1991.  Ito, Junko. Syllabl e Theory in Prosodic Phonology. Ne w York: Garland, 1988. , "A Prosodic Theory of Epenthesis." NLLT Vol 7, 1989. , Prosodic Minimality in Japanese, ms . UC , Santa Cruz, 1990. Ito, Junko, Yoshihisa Kitagawa, and R. Armin Mester. Prosodi c Type Preservation in Japanese: Evidenc e from zuuia-go. ms. UC, Santa Cruz, 1992. Ito, Junko and R. Armin Mester. "Th e Phonology of Voicing in Japanese: Theoretical Consequences for Morphological Accessibility." L I Vol. 17, 1986. , Weak Layering and Word Binarity. ms . UC , Santa Cruz, 1992. / Licensed Segments and Safe Paths. ms . UC, santa Cruz, 1993.  135 , Reflections on Coda Cond and Alignment, ms . UC , Santa Cruz, 1994. Ito, Junko, Armin Mester, and Jaye Padgett. Licensing and Redundancy: Underspecification in Qptimality Theory. ms . UC, Santa Cruz, 1993. Katada, Fusa. "O n the Presentation of Moras: Evidenc e from a Language Game." L I Vol. 21, 1990. Kiparsky, Paul. "Fro m Cyclic Phonology to Lexical Phonology." The Structure of Phonological Representations Vol . 1. Eds. H . van der Hulst and N. Smith. Dordrecht : Foris, 1982. Kubozono, Haruo. "Th e Organization of Japanese Prosody." Diss . U of Edinburgh, 1986. , Prosodic Compound Formation in Japanese, ms . U of Edinburgh, 1988. , "The Mora and Sylable Structure in Japanese: Evidenc e from Speech Errors." Langua e and Speech Vol . 32(3), 1989. Ladefoged, Peter. A Course in Phonetics. 2n d ed. Ne w York: Harcourt Brace Jovanovich, 1982. Makino, Seiichi and Tsutsui Michio. A Dictionary of Basic Japanese Grammar. Tokyo : Japan Times, 1986. McCarthy, John J. and Alan S. Prince. Prosodi c Morphology, ms. U of Massachusetts, Amherst; Brandeis U, 1986. , "Foot and Word in Prosodic Morphology: Th e Arabic Broken Plural." NLL T Vol . 8 (1990) : 209-282. , Prosodic Morphology I: Constrain t Interaction and  136 Satisfaction, ms . U of Massachusetts, Amherst; Rtgers U, 1993. McCawley, James D. Th e Phonological Component of a Grammar of Japanese. Th e Hague: Mouton, 1968. , "What is a Tone Language?" Tone : A Linguistic Survey. Ed . Fromkin, Victoria A. Ne w York: Academic Press, 1979. Mester, R. Armin and Junko Ito. "Featur e Predictability and Underspecification: Palata l Prosody in Japanese Mimetics." Language Vol . 65, 1989. Miyara, Shinsho. "Phonologica l Phrase and Phonological Reduction." Studie s in Japanese Linguistics Vol. 7 (1988): 79-121. Mohanan, K.P. Th e Theory of Lexical Phonology. Boston : Dordrecht, 1986. Myers, Scott. "Persisten t Rules." L I Vol . 22, 1991. Nespor, Marina. "Clas h Avoidance in Italian." I L Vol. 10, 1979. , "Vowel Degemination and Fast Speech Rules." Phonolog y Yearbook Vol. 4 (1987) : 61-85. Nespor, Marina and Irene Vogel. Prosodi c Phonology. Dordrecht : Foris, 1986. Nishikawa, Makoto. "Japanese Lexical Phonology and Morphology." Diss. U  of Southern California, 1987.  Ono, Kooji. "/Q/-Insertio n Rule in English Loan Words in Japanese." Geng o Kenkyu No . 100, 1991. Padgett, Jaye. Featur e Classes, handout . U of British Columbia, Vancouver, 1994.  137 Paradis, Carol and Jean-Francois Prunet. "Introduction : Asymmetry and Visibility in Consonant Articulations." Phonetics and Phonology Vol . 2. Th e Special Status of Coronalsj_ Interna l and External Evidence. Eds . Paradis, Carol and Jean-Francois Prunet. Ne w York: Academic Press, 1991. Pierrehumbert, Janet and Mary Bechman. Japanes e Tone Structure. Cambridge: MIT Press, 1988. Poser, William John. "Th e Phonetics and Phonology of Tone and Intonation in Japanese." Diss . MIT, 1984. , "Evidence for Foot Structure in Japanese." Languag e Vol. 66, 1990. , "Blocking of Prasal Constructions by Lexical Items." Lexical Matters. Eds . Sag , Ivan A. and Anna Szabolcsi. Stanford U, 1992. Prince, S. Alan. Th e Phonology and Morphology of Tiberian Hebrew. diss . MIT . 1975 . , Stress? or Applying Stress? ms . 1976. , "Relating to the Grid." L I Vol . 14, 1983. Prince, S. Alan and Paul Smolensky. Explanator y Typology through Optimality Theory, handout . Stanfor d Phonology Workshop. 11 Feb. 1993. Pulleyblank, Douglas. Ton e in Lexical Phonology. Dordrecht : Reidel, 1986. , "Vowe l Deletion in Yoruba." Journa l of African Languages and Linguistics 10 (1988) : 117-136.  138 / Vowel Harmony and Optimality Theory, ms . U of British Columbia, 1993. , Neutral Vowels in Optimality Theory: A Comparison of Yoruba and Wolof. ms . U of British Columbia, 1994. Rice, Keren and Peter Avery. "O n the Relationship Between Laterality and Coronality." Phonetic s and Phonology Vol 2. Special Status of Coronals: Internal and External Evidnece. Ne w York: Academic Press, 1991. Saeki Kenzoo and Yamada Toshio. (Eds.) Kog o Jiten (Dictionar y of Classical Japanese). Tokyo : Kadokawa, 1993 (edition) . Sakuma, Kanae. Kokug o to Akusento (Japanes e and Accent). Tokyo : Doobunkan, 1919. Selkirk, Elisabeth. "O n Derived Domains in Sentence Phonology." Phonology Yearbook Vol. 3, 1986. Selkirk, Elisabeth and Tong Shen. "Prosodi c Domains in Shanghai Chinese." Th e Phonology-Syntax Connection. Eds . Inkelas , Sharon and Draga Zee. U of Chicago Press, 1990. Selkirk, Elisabeth and Koichi Tateishi. "Constraint s on Minor Phrase Formation in Japanese." Paper  s from the General  Session at the Twenty-fourth Regional Meeting. Eds . MacLeod, Lynn, Gary Laron, and Diane Brentari. Chicago : Chicago Linguistic Society, 1988. , "Syntax and Downstep in Japanese." Interdisciplinar y Approaches to Langauge: Essay s in Honor of S.-Y. Kuroda. Eds . Georgopoulos , Carol and Roberta Ishihara. Dordrecht: Kluwer Academia, 1991.  139 Shaw, Patricia A. "Consonan t Harmony Systems: Th e Special Status of Coronal Harmony." Phonetic  s and Phonology  Vo 1,. ,2.._. The Special Status of Coronals: Interna l and External Evidence. Ne w York: Academic Press, 1991. Shibatani, Masayoshi. Th e Languages of Japan. Ne w York: Cambridge UP, 1990. Soga, Matsuo and Matsumoto Noriko. Foundation s of Japanese Language. Taishukan , 1990. Soga, Matsuo, Koyama Yoko, and Ohso Mieko. Standar d Japanese. Tokyo: Taishukan, 1987. Tateishi, Koichi. "Theoretica l Implications of the Japanese Musician's Language." Proceeding s of the West Coast Conference on Formal Linguistics. 8. CA : Stanford U, 1989. , "A Review: A Theory of Stress and Accent by Shosuke Haraguchi." Englis h Linguistics. Journa l of the English Linguistics Society of Japan. Tokyo : Kaitakusha, 1993. Vance, Timothy J. A n Introduction to Japanese Phonology. Ne w York: State U of New York Press , 1987. Yip, Moira. "Th e Obligatory Contour Principle and Phonological Rules: A  Loss of Identity." L I Vol . 19 (1988) : 65-100.  , "Coronals , Consonant Clusters, and the Coda Condition." Phonetics and Phonology Vol. 2. Th e Special Status of Coronals; Interna l and External Evidence. Eds . Paradis, Carol and Jean-Franci s Prunet. Ne w York: Academic Press, 1991.  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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


Related Items