UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An optimality account of tone-vowel interaction in Northern Min Jiang-King, Ping 1996

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

Item Metadata

Download

Media
831-ubc_1996-091015.pdf [ 12.29MB ]
Metadata
JSON: 831-1.0087288.json
JSON-LD: 831-1.0087288-ld.json
RDF/XML (Pretty): 831-1.0087288-rdf.xml
RDF/JSON: 831-1.0087288-rdf.json
Turtle: 831-1.0087288-turtle.txt
N-Triples: 831-1.0087288-rdf-ntriples.txt
Original Record: 831-1.0087288-source.json
Full Text
831-1.0087288-fulltext.txt
Citation
831-1.0087288.ris

Full Text

AN OPTIMALITY ACCOUNT OF TONE-VOWEL INTERACTION IN NORTHERN MIN  By PING JIANG-KING B.A., Huazhong Normal University, China, 1975 M.A., Huazhong University of Science and Technology, China, 1984 M.A., The Ohio State University, U S A 1987  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Department of Linguistics) We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA March 1996 © Ping Jiang-King, 1996  In presenting this  thesis in partial  degree at the University of  fulfilment of  the  requirements  for  an advanced  British Columbia, I agree that the Library shall make it  freely available for reference and study. I further agree that permission for extensive copying of this thesis for department  or  by  his  or  scholarly purposes may be granted by the head of her  representatives.  It  is  understood  that  copying  my or  publication of this thesis for financial gain shall not be allowed without my written permission.  Department The University of British Columbia Vancouver, Canada Date  DE-6 (2/88)  ABSTRACT This dissertation aims at constructing a fully articulated theory of tone-vowel interaction within the framework of Optimality Theory (OT). It examines the nature of this phenomenon in Northern Min languages, as well as various Southeast Asian languages. The questions addressed are (i) what is the nature of tone-vowel interaction? (ii) how do they relate to each other? Two important findings emerge from the investigation. First, tonal types and syllable types are closely related to each other. That is, different groups of tones occur only in a certain kind of syllables. These cooccurrence restrictions are identified as a correlation between tonal contour and syllable weight. Second, tone does not directly affect vowel distributions and alternations. Rather, it is the relative syllable positions in which a vowel occurs and the number of segments present in a syllable that trigger vowel distributions and alternations. These findings lead to the conclusion that tone and vowel do not interact directly and that there is no feature-to-feature correlation between them. Their interaction lies in the prosodic anchor mediating between them. To account for the correlation between tonal contour and syllable weight and the close relationship between syllable structures and vowel features, I propose a prosodic anchor hypothesis which attributes the tone-vowel interaction to the mora and its function as an anchor for both tone and vowel The theory proposed in this thesis contains two sets of constraints. The first set governs linking of tones to moras. The examination of moras as tone-bearing units shows that moras differ with respect to how many and what kind of tones they may bear. Thus, Head Binarity (i.e. a nuclear mora must bear two tones) accounts for the quantitative distinction between tonal contours and syllable weight in Fuzhou. whereas the tonal sonority hierarchy (i.e. | +UPPER \  >  \ -RAISED  moraic structures (i.e. the constraint ranking  |) and their harmonic association to the  *NUC/[-RSD] »  *NUC/[+UPR]) explain  the  phenomenon of a L tone restricted to the non-nuclear mora in Fuqing. I further show that ii  the interaction of the constraints is capable of deriving the unmarkedness of tonal systems, as well as the cross-linguistic variation of tonal distributions. The second set of constraints regulates the relation between syllable structures and vowel features. It has been observed that linking of vowel features to prosodic anchor in tight syllables is more restrictive than in loose syllables. This asymmetry is expected under the prosodic anchor hypothesis since the tight syllables are argued to contain one less mora than the loose ones. I further demonstrate that the interaction of the basic syllable structure constraints with the faithfulness constraints can automatically derive the vowel distribution patterns. Two kinds of stress effects on tone-vowel interaction are identified. First, stressed syllables always preserve their lexical specifications (either tonal or segmental). Second, the vocalic changes in unstressed syllables (i.e. the non-final syllable of a domain) involve reduction of syllable weight. These stress effects can be captured by the constraints Prominence Alignment and Prominence Reduction, respectively. The former assigns a metrical grid to arightmostsyllable, hence preserving its lexical properties, while the latter prohibits a stressed syllable from having two moras. I show that these constraints, interacting with the constraints on syllabification, can successfully derive vowel alternating pairs in disyllabic words  iii  T A B L E OF CONTENTS Abstract  u  Table of contents  iv  Acknowledgments  viii  CHAPTER 1  INTRODUCTION  1.1 Overview  1  1.2 Theoreticalframework:Optimality Theory  9  1.2.1 Basic principles of OT  10  1.2.2 The basic types of constraints  15  CHAPTER 2  T H E NATURE OF TONE-VOWEL INTERACTION: PROSODIC OR FEATURAL?  2.0 Introduction  19  2.1 Phonetic correlates  20  2.2 Tone-vowel interaction in phonological theory  21  2.2.1 The "intrinsic height correlation" hypothesis (Wang 1968, Yip 1980) ...22 2.2.2 The "metrical duration" hypothesis (Wright 1983)  27  2.2.3 The "metrical-tonal contour" hypothesis (Chan 1985)  31  2.3 Fuzhou tone-vowel interaction  33  2.3.1  Tone-vowel interaction in monosyllabic words  36  2.3.2  Tone-vowel interaction in reduplications  41  iv  2.3 3  Tone-vowel interaction in "cutting foot words"  2.4 Fuqing tone-vowel interaction  45 49  2.4.1  Tone-vowel interaction in monosyllabic words  52  2.4.2  Tone-vowel interaction in disyllabic words  57  2.4.3  Tone-vowel interaction in reduplications  60  2.5 The nature of the tight-loose distinction  62  2.5.1  Tone and duration  65  2.5.2  Segment and duration  67  2.5.3  Further evidence from Southeast Asian languages  69  2.6 Conclusion  CHAPTER 3  71  CORRELATION BETWEEN TONAL CONTOUR AND S Y L L A B L E WEIGHT  3.0 Introduction  72  3.1 The prosodic anchor hypothesis  77  3.2 Mora: the prosodic anchor for tone-vowel interaction  82  3.3 Head mora vs. nonhead mora  90  3.3.1  Head Binarity: Fuzhou tonal distributions and prosodic structure  93  3.3.2  Head Prominence: Fuqing tonal distributions and prosodic structure97  3.3.3  A summary  102  3.4 Hyman's typology of tonal distributions and syllable structures  103  3.5 Conclusion  106  v  CHAPTER 4  LINKING BETWEEN PROSODIC STRUCTURE AND VOWEL FEATURES  4.0 Introduction  1 0 8  4.1 Syllable theory in OT  109  4.1.1 Basic syllable structures  110  4.1.2 Segmental sonority constraints  113  4.2 Fuzhou syllabification  115  4.2.1 Fuzhou syllable structure  116  4 2.2 The contrast between monophthongs and diphthongs  120  4.2.3 The tense/lax distinction  128  4.2.4 The harmonic restrictions on the tight syllables  130  . 4.2.5 The asymmetric behavior of the high vowels  134  4.2.6 A summary  141  4.3 Fuqing syllabification  142  4.3.1 The high/mid contrast  143  4.3.2 The tense/lax distinction  147  4.3.3 The harmonic restrictions on the tight syllables  149  4.3.4 The asymmetry of the high vowels  152  4.3.5 The monophthongs vs. diphthongs  154  4.3.6 A summary  157  4.4 Conclusion  158  vi  CHAPTER 5  STRESS EFFECTS ON TONE-VOWEL INTERACTION  5.0 Introduction  160  5.1 How does stress affect tone-vowel interaction?  161  5.1.1  Identifying the stress effects  162  5.1.2  Asymmetric behavior of syllables in Fuzhou  170  5.1.3  Asymmetric behavior of syllables in Fuqing  174  5.1.4  A summary.,  180  5.2 Vowel alternations in disyllabic compounds  181  5.2.1  Fuzhou disyllabic words  183  5.2.2  Fuqing disyllabic words  186  5.2.3  A summary  ..189  5.3 Vowel alternations in reduplications  190  5.3.1  Fuzhou reduplications  191  5.3.2  Fuqing reduplications  198  5.3.3  A summary  204  5.4 Vowel alternations in Fuzhou Fanqie words  204  5.5 Conclusion  215  LIST OF REFERENCES  216  APPENDIX  243  vii  ACKNOWLEDGMENTS There are many people who have helped me greatly while writing this thesis. Without their constant support and guidance, this work would never have been completed. I would especially like to thank my committee members Doug Pulleyblank, Pat Shaw and Mark Hewitt for their teaching and comments, and for sharing with me a number of insights that have been incorporated into this thesis. My thesis advisor Doug Pulleyblank taught me how to challenge every claim I made, and justify them one by one. Pat showed me how to make arguments consistent, and she has been a model for me, as both a professor and a mother. Mark was always around to cheer me up when I was depressed. Their encouragement and confidence in me have been a constant inspiration for writing this thesis and for the years to come in my future academic career. I am also grateful to Moira Yip for her detailed comments and suggestions on my generals paper which formed the core of this dissertation. Bruce Bagemihl and Laura Downing introduced me to phonological theory and so set the stage for my excursions into the depths of prosodic structure. Dr. Kinkade showed me valuable techniques in field research methodology. The syntacticians in our department: Hamida Demirdache, Henry Davis, Michael Rochemont and Rose-Marie Dechaine, taught me the fundamentals ofrigorousscholarly thinking. Graduate advisors Dr. Ingram and Dr. Carden not only guided me through the administrative complexities of my life as a graduate student, but also gave me much emotional support. My professors in China trained me in the techniques of exhaustive descriptive analysis of linguistic phenomena. Ms. Carmen DeSilva offered me her warmest compassion and help when I needed it. My fellow students Eleanor Blain, Nike Ola, Myles Leitch, Kimary Shahin, Akd Uechi, William Turkel, Taylor Roberts, Lisa Matthewson, Monica Sanchez, Susan Blake, Yanfeng Qu and William Thompson shared "studentship", companionship, and friendship. My special thanks goes to William Turkel for his careful proofreading of this thesis. I am grateful to Doug Pulleyblank for providing me with financial support from the two research grants awarded to him by the Social Sciences and Humanities Research Council of Canada (SSHRCC No 410-91-0204 and 410-94-0035). Without the support from the SSHRC grants this research would never have been possible My daughters Mona and Sara were forever tolerant of mom's "homework". They gave up lots of time with mom in their earlier childhood. Their sacrifice is one of the important factors that helped mefinishthis dissertation. Finally, or rather first, I would like to thank my husband Brian King for his support and help. During my years at UBC, he has served as an editor (both unpaid and on-call) for my school work. It is he who could see the unhappiness deep inside me while I was working for my CGA (Certified General Accountant) in an effort to give myself marketable skills. Without his constant encouragement, I would never have started my Ph.D. in linguistics in the first place. As the Taoists say: "Gain and loss go hand-in-hand". I dedicate this thesis to my parents, as well as my brother and sisters.  viii  CHAPTER 1  Introduction  1.1 Overview  The goal of this dissertation is to develop a fully articulated theory of tone-vowel interaction. Contrary to previous theoretical attempts that concentrate on the correlation between tone and vowel quality (PilszczikowarChodak 1972, 1975, Newman 1975 for Hausa; Cheung 1973 for Omei dialect of Mandarin; Wang 1968, Maddieson 1976, Yip 1980, Chan 1985 for Fuzhou), the present study focuses on the prosodic anchor that mediates between tone and vowels and its role in triggering their interactions. The central questions addressed are how tone and vowels relate to each other and the nature of their interaction. The core of the theory proposed is the prosodic anchor hypothesis It attributes the phenomenon of tone-vowel interaction to the combination of a prosodic representation in which the mora serves as a prosodic anchor for both tone and vowels, and of a set of universal wellformedness constraints on linking of the mora to tone/vowel The languages being investigated are mainly Fuzhou and Fuqing. the two Northern M M languages spoken on the southern coast of China. Attention is also drawn to the tone-vowel interaction reported in various Southeast Asian languages, such as Sre (a Mon-Khmer language spoken in the South Vietnamese city of Di Linh and the surrounding area), Hu (a MonKhmer language spoken in the Xiao Mengyang area in Jinghong county, Sipsong Panna, Yunnan province of China), Siamese (the standard Thai language spoken in Bangkok) and Red Tai (spoken in North Vietnam and the north part of the Sam Nuea province in Lao). In the rest of this section, I will summarize the principal claims of each chapter. 1  Chapter One  Jiang-King, 1996  In chapter 2,1 address the question of the nature of tone-vowel interaction and argue that tones and vowels do not interact directly (ie., feature-to-feature); rather, their interaction lies in the prosodic anchor mediating between them. These arguments are based on the findings from a thorough investigation of tone-vowel interaction in Fuzhou and Fuqing. It shows that tonal distributions are closely related to syllable types and that there is a correlation between tonal contour and syllable weight. More complex tones are found in bimoraic syllables than in monomoraic syllables. This type of correlation is further supported by evidence from cooccurrence restrictions between tonal contours and vowel length found in various Southeast Asian languages '(Le., Sre. Hu., Siamese and Red Tai). This finding is compatible with the findings that rime length affects stress and tonal patterns (Duanmu 1990, 1993). Where Duanmu argues that different Chinese languages have different morale structures, this thesis argues that different syllables within the same language have different morale structures in Northern Min. Furthermore, an examination of vowel distributions and alternations in both Fuzhou and Fuqing reveals that tone does not directly affect vowel distributions and alternations. It is the relative syllable position in which a vowel occurs and the number of segments present in a syllable that are the direct factors triggering surface realizations of vowels. The new findings in chapter 2 lead to the prosodic anchor hypothesis proposed in chapter 3. It is stated in (1):  2  Chapter One  Jiang-King, 1996  (1) Prosodic anchor hypothesis of tone-vowel interaction  a.  Representational Requirement Both tone and vowel must directly link to the lowest prosodic anchor on the prosodic hierarchy, that is, the mora.  b.  Constraint Satisfaction Optimal linking between the prosodic anchor and tone or vowel is determined by a set of universal output constraints.  This hypothesis captures the direct relationship between syllable structure and tone, and the direct relationship between syllable structure and vowels. The condition (la) states that tone and vowel must link to the mora in order for them to interact directly with the syllable structure. This is encoded in the representation in (2d) but not in (2a), (2b) and (2c), since (2d) is the only representation that satisfies the condition (la).  d. T | |I | V  0= segmental root T = tone V = vowe LI = mora o = syllable  Condition (lb) states that the linking of tone or vowel features to the prosodic anchor must be regulated by well-formedness constraints. The basic constraints governing tonal distributions are these known previously as the well-formedness conditions (WFC) (Goldsmith 1976). They can be incorporated into Optimality Theory as a set of  3  Chapter One  Jiang-King, 1996  faithfulness constraints (McCarthy and Prince 1993a, b, 1994, 1995, hereafter M & P; 1  Prince and Smolensky 1993, hereafter P & S; Pulleyblank 1994, McCarthy 1995, among others).  (3) Faithfulness constraints on tonal distributions  a.  PARSETONE:  A tone must be incorporated into prosodic structure.  b.  .Fiix-u.: A mora must be filled by a tone.  c.  LINEARITY:  d.  UNIFORMITY:  e.  LEXTONE:  String, reflects the precedence structure of String , and vice versa. 2  No element of Strihgj has multiple correspondents in String,.  A tone that is present in an output must be present in an input.  The PARSETONE constraint (3a) requires every tone to be parsed onto a prosodic anchor. The FILL-P constraint (3b) demands all moras to be filled by a tone. The LINEARITY constraint (3c) prevents association lines from crossing. That is, the linear precedence relations of tones of the inputs must be preserved by their outputs. The UNIFORMITY constraint (3d) prohibits a multiple linking between tones and tone-bearing units. To prevent a complex tonal contour (i.e., M L M and MHM) in Fuzhou from giving a trimoraic structure, which is unattested in this language, I appeal to the notion of the nuclear mora, first introduced by Shaw (1992, 1993). The nuclear mora is defined as a syllable head; I propose a constraint that requires a nuclear mora to bear two tones.  (4) Head Binarity (HDBIN) A mora must bear two tones x and y, iff it is a syllable head (Le., a nuclear mora).  1  The faithfulness family of constraints has later been incorporated into a generalized theory of  correspondence in the works of M &.P 1995 and McCarthy 1995.  4  Chapter One  Jiang-King, 1996  The interaction of the faithfulness constraint  UNIFORMITV  and Head Binarity defines two  types of tonal systems: ranking HDBIN above  UNIFORMITY allows  contour tones while the  reverse ranking gives level tones only. The square brackets without a subscripted "a" sign indicate a syllable head (i.e., a nuclear mora), while the ones with a subscripted "c" sign indicate syllable boundaries.  (5) Defining tonal systems by constraint ranking t i n ! nl„ Input: a.  UNIFORMITY  HDBIN  Remarks  *  V  contour tone allowed  *  level tone only  T T  !{*•]„ f\ T T  B  1.1 T T  The tonal system defined in (5a) is found in both Fuzhou and Fuqing since both of these languages allow contour tones However, the correlation between tonal contour and syllable weight in Fuqing differs from that in Fuzhou in that it is not the tonal quantity but the tonal quality that gives rise to the distinctive syllable weight. Specifically, the tones in the light syllables are non-L while the ones in the heavy syllables always contain a L tone. To account for the asymmetrical behavior of L tone with respect to the syllable weight, I extend the segmental sonority hierarchy (Zee 1988) and propose that tones are also differentiated in terms of their intrinsic sonority. Assuming that a H tone is more sonorous than a L tone, their sonority difference can be encoded into the hierarchy in (6), given that they are represented by the feature [+UPPER] and [-RAISED] respectively (Yip 1980, Pulleyblank 1986):  5  Chapter One  Jiang-King, 1996  (6) Tonal Sonority Hierarchy I +UPPER I > | -RAISED |  In Optimality Theory, this tonal sonority hierarchy can be formulated in terms of the constraint ranking in (7), which says that parsing a L tone to a nuclear mora is less favorable than parsing a H tone to a nuclear mora.  (7) Tonal Alignment Constraint (TAC) *NUCW[-RSD] »  *NUC1V[+UPR]  Adding the constraint *NUC1-V[-RSD] to the ranking schema in (5) gives the two types of tonal system in (8a) and (8b). Ranking HDBIN above both UNIFORMITY and *NUC^/[-RSD], as shown in (8a), forces a L tone to be linked to a head mora, givingriseto a tonal system like Fuzhou. The reverse ranking in (8b) requires a L tone to be linked to a nonhead mora, resulting in a heavy syllable, as attested in Fuqing.  (8) Defining tonal systems containing a L tone  Input:  UNIFORMITY  *NUCW[-RSD]  *  *  HDBIN  T L  Mo T  1  b  1  -  I  possible in Fuzhou  L  Un]nl  u  0  I  *  V  Remarks  possible in Fuqing  T L  It will be shown that the correlation between tonal contour and syllable weight in both Fuzhou and Fuqing can be derived from the reverse rankings of the same set of constraints on tonal distributions.  6  Chapter One  Jiang-King, 1996  Chapter 4 deals with the linking between prosodic structures and vowel features. It is shown that vowel distributions in Fuzhou and Fuqing are related to syllable positions in which they occupy. It is also demonstrated that syllabification governed by the interaction of the syllable structure constraints (i.e., Nuc,  ONS, -CODA) (M  & P 1993a, b, P & S 1993)  and the harmonic alignment schema *Nuc/ct, (which encodes the intrinsic sonority of segments suitable for certain syllable positions,) can automatically derive the attested vowel distribution pairs of the monosyllabic words in both of these languages. Stress effects on tone-vowel interaction are examined in chapter 5. Three findings emerge from this examination. First, a final syllable within a disyllabic domain preserves all of its lexical specifications (tonal and segmental). Neither tone sandhi nor vocalic change takes place in this position. This kind of featural stability in the final syllable is argued to be due to stress In particular, stress is assigned to the final syllable, so it secures the featural content in that syllable. Second, a non-final syllable within a domain changes either tone (in tight syllables) or both tone and vowels (in loose syllables): In contrast with the feature stability in the final syllable, the change of tone and vowel iri non-final syllables is argued to be another stress effect, namely,  PROMINENCE REDUCTION  which requires a  stressed syllable to be monomoraic. The tonal changes observed in this position are of two types. One is contour simplification in loose syllables (i.e, a complex tonal contour gets simplified, hence no complex contour tone appears in a non-final syllable), and the other is the change of tonal features in tight syllables. Although the conditioning factor for the second type of tonal change is not clear so far, the factor that triggers tonal simplification in the loose syllables is argued to be stress, that is, an unstressed syllable cannot have two moras, hence cannot have a complex tonal contour. Third, vowel changes in non-final syllables are the same as the vowel distribution patterns that are found in monosyllabic words. Namely, vowels in loose syllables become the ones in their corresponding light syllables. The similarities between vowel alternations and vowel distributions are not coincidental. Rather, they are argued to follow from the moraic distinction (i.e., 7  Chapter One  Jiang-King, 1996  monomoraic vs. bimoraic) between the two types of syllables. If a non-final syllable is unstressed it can only be monomoraic, and hence will have an identical moraic structure to the tight syllables. Therefore, both the non-final syllables and the tight syllables in a final position share the same pattern of vowel features. These findings further support the prosodic anchor hypothesis in that the tonal sandhi and vowel changes must be mediated by their prosodic anchor. They interact with each other only when their prosodic anchor gets affected. This is indeed the case as far as stress effects are concerned. Following Wright's (1983) insight, I propose that the stability of the feature content in the final syllable and the changes of tone and vowels in the non-final syllable are subject to different constraints. Prominence Alignment and Prominence Reduction. The former ensures featural stability, while the latter reduces morale structure. Thus, the tonal contour simplification and the vowel changes parallel to the vowel distributions in tight syllables are accounted for. These two constraints on stress are given in (9) and (10) respectively:  (9) Prominence Alignment (PROMALIGN) Given a domain x, a metrical grid mark must be aligned to the right edge of x. x = a morphological word.  (10) Prominence Reduction (PROMREDUC) If a is not assigned a metrical grid mark, a cannot be bimoraic.  PROMALIGN  ensures the final syllable is stressed, while  PROMREDUC  requires unstressed  syllables to be monomoraic, that is, to make the weak weaker (Prince 1983). The interaction of the constraints governing the stability in the final syllable and the changes in the non-final syllable are shown in (11):  8  Chapter One  Jiang-King, 1996  (11) Tonal and vocalic changes triggered by stress Input  Outputs T  a  (i)  T  1 1 V  T  T  T  II  II  II V  T  II  V  V  V  .  V  b b  T  T  V  b  V T  1 V  V  PROMREDUC  PARSER  *  *  V T  V  1 i  '  ' ' V  V  T T  T  V  \  V  V  T T  T  V T  V  v  i  V  *  T  IM1 IM.] 0  V  T  T  1  V  a  (ii) T T  T  T  B  V  The tableau above shows that when a disyllabic word contains two heavy syllables (i.e., is bimoraic), shown in (Hi), only the non-final syllable changes its tone and vowel, since a bimoraic syllable is disallowed in a non-final position. On the other hand, if there is a light syllable (i.e., monomoraic) in the non-final position, shown in (llii), there is no vowel change since the vocalic change involves reducing a mora. Thus, the different stress effects on tone-vowel interaction can be achieved by the same set of constraints.  1.2 The theoretical framework: Optimality Theory (OT)  The theory of tone-vowel interaction being developed in this dissertation is formulated within the constraint-based grammar of Optimality Theory (P & S 1991, 1993; M & P 1993a, b, 1994, 1995; Pulleyblank 1994, 1995, McCarthy 1995, among others). Contrary to rule-based ordering theory, Optimality Theory assumes that Universal Grammar (UG) contains two functions Gen zndEval. The former freely generates any amount of output  9  Chapter One  Jiang-King, 1996  candidates (featural or prosodic) for a given input, while the latter evaluates all members of a candidate set and chooses an optimal output based on a set of ranked constraints. This approach greatly reduces the complexity of determining underlying representations in standard generative phonology, since the grammar focuses its attention on outputs rather than inputs. A lexical entry in OT is assumed to contain both melodic information (i.e., Felements) and moraic information (i.e., prosody) (M & P 1993a, Pulleyblank 1994, etc.). In the following, I first summarize the basic principles of OT and the linguistic generalizations it captures, then follow up with an outline of basic families of constraints and their empirical coverage of phonological and morphological phenomena.  1.2.1 Basic principles of Optimality Theory  Four basic principles are crucial for Optimality Theory. Thefirstis violability. That is, an optimal output may violate some constraints, but the violation is minimal. This amounts to asserting that any surface-true form is only optimal relative to other candidates, and that there is no absolute correctness or incorrectness. The notion "optimal" contrasts sharply with the notion "correctness" in its empirical consequences. An output form may be optimal in context A but not in context B. For instance, in Fuzhou. a syllable without an initial consonant is perfectly possible in word-initial position. However, in non-initial position within a disyllabic word, a vowel-initial syllable gets an onset from the coda in its preceding syllable. In the OT framework, this asymmetrical behavior of an onsetless syllable can be explained by constraint ranking, another basic OT principle we turn to: The second principle is constraint ranking. That is, there is a dominance relation among constraints Since all constraints are assumed to be universal and present in UG, language-particular behavior is accounted for by different ranking. In other words, whether a particular constraint is active or not depends on its relation to other constraints in a ranking schema. Even within a single language, a particular output may be acceptable 10  Chapter One  Jiang-King, 1996  in one case but not in another. As an illustration, an onsetless syllable in Fuzhou is acceptable in word-initial position, but not in non-word-initial position when there is a coda consonant preceding it. This is explained in the constraint ranking shown in (12) below. In particular, ranking LEXF (which prohibits insertion of any F-element that is not specified lexically) above ONS (which demands that syllables have an onset) permits an onsetless syllable in word-initial position, but forces an onsetless syllable in non-wordinitial position to get an onset from the coda in its preceding syllable if any.  (12) V-initial syllables in different positions within a word Ranking: LEXF Inputs  Candidate set  (i)  a.  /v+cv/  b.  (ii)  a.  CV.CV  /cvc+v/  b.  CVC.V  LEXF  CVCV  »  ONS,  ONS ONS  *•!  V.CV  Given the ranking LEXF  »  *  *  when an onsetless syllable is in a word-initial position, the  candidate (12-i-b) is better than (12-i-a), since the former satisfies the highly ranked constraint  LEXF,  even though it violates  ONS.  Oh the other hand, when an onsetless  syllableisin a non-word^initial position (12-ii-b), it is not acceptable since there is a coda consonant in the preceding syllable. (12-ii-a) satisfies all constraints and therefore is optimal. The third principle is inclusiveness. All candidate analyses must participate in the evaluation process. That is, the function Evdl must access an entire set of candidates, and there are no special rules or repair strategies for a particular candidate:  11  Chapter One  Jiang-King, 1996  The fourth principle is parallelism. This amounts to asserting that all constraints and all members of a candidate set are supplied to the function Eval in a parallel fashion. In other words, the relation between input and output is parallel and not derivational This principle sharply contrasts with the Ordering theory in that it eliminates all intermediate stages that are crucial for the rule-ordering theory. The basic properties of OT outlined above are summarized in (13).  (13)  Basic principles of Optimality Theory (M & P 1993a: 1)  a.  Violability. Constraints are violable; but violation is minimal.  b.  Ranking: Constraints are ranked on a language particular basis; the notion of minimal violation (or best satisfaction) is defined in terms of this ranking.  c  Inclusiveness: The constraint hierarchy evaluates a set of candidate analyses that are admitted by very general considerations of structural well-formedness. There are no specific rules or repair strategies.  d.  Parallelism: Best- satisfaction of the constraint hierarchy is computed over the whole hierarchy and the whole candidate set.  If we assume that a grammar has two constraints A, B; and that A dominates B, then the optimal candidate will be the one that satisfies both constraints or the higher ranked constraint A. This situation is illustrated schematically in the tableau below. The tableau implements Prince and Smolensky's interpretation conventions as follows: (i) A » B indicates that constraint A dominates constraint B; (ii) constraints are arranged in columns from left to right in order of ctomination; (iii) a "*'* marks constraint violation; (iv) a "*!" signifies a fatal violation (and the subsequent rejection of the candidate); (v) a blank cell indicates constraint satisfaction, and (vi) a pointingfinger^ or a thumbs up & marks the optimal or winning candidate.  12  Chapter One  Jiang-King, 1996  (14) Ranking: A » B  /input/  1  Candidate set  [  «- Cand,  [  Cand,  A  B *  *!'  Cand, in (14) violates A (the higher ranked constraint) and therefore loses out to Candj, which satisfies A. It is selected as the optimal candidate, since it obeys the higher ranked constraint A, although it violates the lower ranked constraint B. Where both candidates violate the higher ranked constraint, choice of the optimal candidate depends on the satisfaction of the lower ranked constraint. In this case, as we can see in tableau (15), Cand^ obeys constraint B, whereas Cand, violates it. Cand thus emerges as the winner. 2  (15) Ranking: A » B Candidate set /input/  A  B  Cand,  *  '*!  ®= Cand,  *  In situations where both candidates behave identically with respect to violating or obeying constraints, the choice of a winner is dependent on which candidate incurs fewer constraint violations Thefollowingtableaux (16) and (17) demonstrate this scenario:  13  Chapter One  Jiang-King, 1996  (16) Ranking: A » B Candidate set /input/  A  B  Cand,  *  **l  «° Cand,  *  *  (17) Ranking A » B Candidate set /input/  A  B  Cand, *  Cand,  Two constraints which are not crucially ranked in a grammar do not select one over the other. Candidates which satisfy or violate them can be equally selected as optimal, as illustrated in the following tableaux (18):  (18) Not Ranking: A, B  /input/  Candidate set  A  •*- Cand,  *  B  *  *• Cand,  Optimality Theory with the basic principles outlined above is successful in explaining a number of generalizations that are left unaccountable in generative phonology. First, it explains why an unmarked case can be both present and absent in a particular language.  14  Chapter One  Jiang-King, 1996  Certain output forms are allowed generally in a given language but are disallowed in the reduplicative and epenthetic structure of that very language (M & P 1994). This shows that unmarked forms, such as C V syllables, emerge in a given language even though marked forms also show up in that language; Second, OT explains cross-linguistic typology in a principled manner. That is, constraint ranking defines different types of languages. A particular ranking like A » B is true for a language x, the reverse ranking B » A is also true for a language y. This is attested, for example, in tongue root harmony of low vowels in various African languages (Pulleyblank et al 1995) The universality of U G is ultimately strengthened by this cross-linguistic typology.  1.2.2 The basic types of constraints  Constraints proposed in Optimality Theory are of three types. The first type is the faithfulness family. Since the function Gen freely generates any kind of output candidates, restrictions are needed to rule out any candidates that have no relation with a given input. The function of the faithfulness constraints is to demand an identity relation between input and output. Various faithfulness constraints proposed in OT (M & P 1993a, b, P & S 1993, Archangeli and Pulleyblank 1994a, b, thereafter A & P; Pulleyblank 1994, McCarthy 1995) are given below:  (19) PARSE family PARSE-©:: a (feature or node) must be incorporated into prosodic structure.  (20) FILL family A prosodic constituent must be filled by a F-element.  15  Chapter One  Jiang-King, 1996  (21)LEx-a The addition of feature F or an association to F is prohibited.  (22) LINEARITY  String, reflects the precedence structure of String , and vice versa. 2  (23) UNIFORMITY  No element of String;, has multiple correspondents in String,.  The function of the  PARSE  family is to ensure that all F-elements are incorporated into a  prosodic constituent so they may be phonetically realized. The prosodic anchor to be filled by a F-element  LEX-O.  FILL  family requires a  prohibits any insertion of an F-element  or a path that is not present in an input. LINEARITY prevents association lines from crossing. UNIFORMITY  disallows a multiple linking between F-elements and prosodic anchors  The second type of constraint is the structural family. The function of this type is to restrict certain prosodic constituents to certain structures; For example, a syllable must have a nucleus, it may have an onset at the leftmost or a coda at the rightmost (M & P 1993a, b, P & S 1993). These are given below:  (24) Syllable structure constraints  a.  NUCLEUS (Nuc):  Syllables must have nuclei,  b  ONSET (ONS):  Syllables must have onsets.  c.  NOCODA (-COD):  Syllables cannot have codas.  d.  *COMPLEX (*COMPX):  No more than one C or V links to one syllable position.  16  Chapter One  Jiang-King, 1996  The general function of the structural constraints above is to ensure an unmarked syllable type CV. It is possible for other types of syllables such as CVC or VC to emerge from interaction of the structural constraints with other types of constraints such as the faithfulness family presented above. The third type of constraint is the alignment family. This type captures harmonic processes such as tongue root harmony in African languages (A & P 1994, b; Akinlabi 1994; Pulleyblank 1994, 1995; Cole & Kisseberth 1995; Kirchner 1993; M & P 1993a, b, 1994; Pulleyblank & Turkel 1995; Pulleyblank et al 1995; etc.), nasal harmony, stress assignment (Hewitt 1994, 1995), and reduplication (M & P 1993a, b, M & P 1995, McCarthy 1995, Shaw 1995).  (25) Generalized Alignment (M & P 1993b)  ALIGN (Csi\,  Edge 1, Cat2, Edge 2) = def  V Catl 3 Cat2 such that Edge 1 of Cat 1 and Edge 2 of Cat 2 coincide, Where Catl, Cat2 e PCat w GCat Edgel, Edge2 e {Right, Left}  Notice that the function ALIGN in (25) takes four arguments, two of which are categories and the other two of which are edges of the categories The categories are denned in terms of either prosodic or grammatical constituents, while the edges are either left or right The theoretical framework outlined above offers a number of advantages First, it allows various phonological processes to interact with each other in a parallel fashion. For instance, the languages being investigated here exhibit different kinds of phenomena, such as vowel distributions, vowel alternations, tone sandhi, stress effects, etc.. All of them twist together so that it is hard to make out what's what. The OT framework provides 17  Chapter One  Jiang-King, 1996  leads for solving these problems by examining possible outputs. Second, it makes it possible to uncover certain unmarked phenomena which otherwise would be obscured. Take the syllable structures in Fuzhou, for example. Onsetless syllables are common in this language One might mistake this as a counterexample to the unmarked syllable type C V. However, disyllabic compounds in this language show that when two monosyllabic morphemes combine to form a disyllabic word, consonant gemination takes place between the two syllables This contradiction (i.e., onset is not necessary vs. onset is necessary) can be simply explained by constraint ranking in the current theoretical framework. That is, the unmarked syllable type CV is indeed respected in this language unless something else outranks it. In the Fuzhou ease,  LEX-F  may rank above  ONS  so that insertion of a  consonant in onset position is disallowed. Third, It captures cross-linguistic variation by constraint ranking: The constraint-based grammar of Optimality Theory, therefore, offers an explanatory power for various phonological and morphological phenomena.  18  CHAPTER 2  The Nature of Tone-Vowel Interaction: Direct or Indirect?  2.0 Introduction  This chapter examines the phonological nature of tone-vowel interaction. It focuses specifically on the problem of whether tonal features and vowel features interact directly (i.e., whether there is a feature-to-feature correlation) or indirectly (i.e., something else plays a role in their interaction). First, I will summarize the results of experimental studies which have found phonetic correlates for tones and vowels Second, I will review three representative phonological hypotheses that offer accounts of this phenomenon. Third, I will thoroughly investigate tone-vowel interaction in Fuzhou (a Northern Min language spoken principally in the Fujian province of China), the best documented case. Data not treated in earlier theoretical studies will be given serious consideration. To explore crosslinguistic variation, I will also investigate tone-vowel interactions in Fuqing. another Northern Min language spoken principally in the Fujian province of China, which has not previously been studied in any theoretical framework. Finally, I win examine the nature of the so-called "tight % ^ - loose #H|" distinction exhibited in both Fuzhou and Fuqing. and show that this distinction is best identified in terms of syllable weight, i.e., between light and heavy syllables. To support this conclusion, the correlation between tonal contour and vowel length exhibited in various Southeast Asian languages, such as Sre (a Mon-Khmer language spoken in the South Vietnamese city of Di Linh and the surrounding area), Hu (a Mon-Khmer language spoken in the Xiao Mengyang area in Jinghong county, Sipsong Panna, Yunnan province of China), Siamese (the standard Thai language spoken in Bangkok), and Red Tai (spoken in North Vietnam and the northern part of the Sam Nuea 19  Chapter Two  Jiang-King, 1996  province in Lao) are also examined. This identification leads to the conclusion that the nature of the tone-vowel interaction is indirect, namely, that tone does not directly affect vowels and vowels do not directly affect tone. Instead, it is the prosodic anchor that has a direct relation with both tone and vowel independently, triggering their interactions.  2.1 Phonetic correlates  A number of experimental studies have been conducted on the nature of the tone-vowel interaction. The studies yield two principal findings. First, there is a correlation between fundamental frequency (F ) and vowel height, indicated by the first formant (Fj), and 0  second, there is a correlation between vowel duration and pitch height or pitch contour. Several studies indicate a tendency for higher vowels to be uttered with a higher pitch than lower vowels in similar environments (Peterson and Barney 1952, House and Fairbanks 1953, Lehiste and Peterson 1961, Ladefoged 1964, Mohr 1969, Lea 1972, 1973, Hombert 1976) For instance, the intrinsic fundamental frequency (F ) of vowels in 0  American English is related to their height (Fj): high vowels ([i] and [uj) have a higher fundamental frequency than low vowels ([a]) (Black 1949, Peterson and Barney 1952, House and Fairbanks 1953, Lehiste and Peterson 1961, Lea 1972,1973, among others). A similar correlation between vowel height and F is also found in other languages, such as 0  Danish (Petersen 1976), French (Di Cristo and Chafcouloff 1976), Korean (Kim 1968), Serbo-Croatian (Ivic and Lehiste 1963), Mandarin (Chuang and Wang 1976, Tsay and Sawusch 1994) and Fuzhou (Tsay and Sawusch 1994) Although evidence for the intrinsic correlation between F and vowel height is 0  convincing, there are also other significantfindings.For instance, rounded vowels tend to have a higher F than their unrounded counterparts (Sundberg 1969, Ewan 1975, Reinholt 0  Petersen 1978, Iivonen 1987). It is also reported that the intrinsic pitch (F ) of the short 0  lax vowels [i] and [o] is equal to that of the long tense vowels [i:] and [u:] in German, 20  Chapter Two  Jiang-King, 1996  even though they are known to have different tongue height (Fischer-J0rgensen 1990). This suggests that tongue height (Fj) is not the only factor that influences the intrinsic F  0  for vowels, and that other factors such as duration should also be taken into consideration. The correlation between duration and pitch height (F ) or pitch contour is indeed found in 0  a number of experimental studies. For instance, a high tone is shorter than a mid or a low tone (Benedict 1948:186 for Cantonese; Pike 1974:171 for Chatino; etc.), and a rising tone or a concave tone is longer than a high level tone or a falling tone (Zee 1978 for Taiwanese; Abramson 1962 for Standard Thai; Dreher and Lee 1966, Chuang 1972, Howie 1974 for Mandarin; Langdon 1976 for Yuman languages, etc.). What these findings suggest is that higher F correlates with shorter duration and lower F with longer 0  0  duration. Therefore, both vowel height (F,) and duration are possible correlates for either fundamental frequency (F ) or pitch contour. 0  2.2 Tone and vowel interaction in phonological theory  Since tone, denned as linguistic use of pitch, is also primarily identified in terms of fundamental frequency (F ) (Gandour 1978), it is natural to ask whether the intrinsic 0  correlation between vowel height (F ) and F manifests itself phonologically in natural t  0  languages. In other words, the question is whether there is any empirical evidence suggesting a phonological correlation between tone and vowel height. The evidence from Hausa (an African language principally spoken in Nigeria), for example, is inconclusive: Data is offered both for (Pilszczikowa-Chodak 1972, 1975) and against (Newman 1975) this position. A highly controversial case is Fuzhou (a Northern Min language spoken on  21  Chapter Two  Jiang-King, 1996  the southern coast of China). In Fuzhou. a whole series of finals participate in vowel 1  alternations in accordance with their tonal environment. It has been claimed, on the one hand, that in a tone sandhi environment a vowel undergoes raising when the tone it occurs with increases its F (Wang 1968). This is characterized as a tone-induced vowel raising 0  process (Yip 1980). I refer to this claim as the "height-correlation" hypothesis On the other hand, it has been argued that the vowel alternations in Fuzhou involve not only differences in height, but also differences along other dimensions, such as a front/back axis, monophthongs versus diphthongs, etc. (Maddieson 1976, Chan 1985). The heightcorrelation hypothesis, therefore, is not sufficient to explain all instances of tone-related vowel alternation. The implicit assumption behind this debate is that tonal features and vocalic features may interact directly. This raises a more fundamental question as tb the nature of this interaction. In other words, whether the interaction between tone and vowels is direct (i.e., feature-to-feature) or indirect (i.e., mediated by something else). In the following, I will review three different hypotheses on tone-vowel interaction in Fuzhou Each of them cites a different phonetic correlates as evidence motivating the hypothesis. I will point out the insights and inadequacies of each.  2.2.1 The height correlation hypothesis (Wang 1968, Yip 1980)  The height-correlation hypothesis, proposed by Wang (1968) and supported by Yip (1980), claims that there is a correlation between high tones and high vowels, since both of them give rise to a higher F . The evidence for this hypothesis comes from Fuzhou 0  In traditional Chinese phonology, a syllable is divided into two parts: (i) the initial consonant and (ii) the rest of the syllable, called thefinal.For instance, thefinalin a syllable [biao] 'quick, rapid' is iio, which contains the on-glide i, and a diphthongal nucleus ao. 22  Chapter Two  Jiang-King, 1996  vowel alternations. Wang (1968) extracted data from Yuan (et al. 1960) and presented themih(l):  (1)  os - » 0 -> y o —> u  a -> e -> i  In his view, these alternations operate together with alternations of tone in such a way that "each time a tone with a lower F changes to one with a higher F the vowel also changes 0  0  to a higher vowel." Based on Wang's observation, Yip (1980) characterizes the conditioning tonal factor for the vowel raising as the [+upper] register, and gives the complete tonal inventory in (2) and the vowel alternating pairs in (3) respectively (column I = "tight" finals  column fl = "loose" finals feW)  (2) The complete tonal inventory in Fuzhou given by Yip (1980:341)  I. "tight"finals'M^  B. "loose" finals mi  [+upper] tones  [-upper] tones  44  HH  12  LH  52/4 HL  13  LL  242  LHL  22  LL  35  LH  23  Chapter Two  Jiang-King, 1996  (3) Vowel alternating pairs dealt with by Yip (1980:275)  I  n  a.  i  ei  b.  y  c.  u  I  n  d.  ei  ai  cey  e.  cey  osy  ou  f.  ou  ou  Yip treats the vowels in the lbose finals (in column II) as underlying forms and the ones in the tight finals (in column I) as derived. To derive the correct surface vowels for tight finals, the following vowel-raising rule is proposed (Yip, 1980:277):  (4)  V  ->  [a low]  r - low ] / — 1 - a hi i  [+upper]  This rule states that vowel-raising is triggered by the tonal feature [+upper] register. By applying this rule, a [+LO] vowel becomes mid and a [-LO] vowel becomes high in the environment of [+upper] register. The height-correlation hypothesis, however, faces difficulties concerning the entire system of vowel alternations. The first problem is its inadequacy in accounting for all vowel alternations in this language. As Maddieson (1976) and Chan (1985) point out, other features in addition to vowel height, such as roundness and backness, are also involved in the vowel changes. The alternations, as shown in (5), are the result of (i) modifying the first, or both elements of the diphthong (5.1), or (ii) deleting the non-high part of the diphthong (5; II). (5) is retabulated from Chen and Norman (1965) by Maddieson (1976:194):  24  Chapter Two (5)  I.  a)  Jiang-King, 1996 ai  -> ei  b) au -> ou c) oi -» 0y  II. a) ei  -» i  b) ou -> u c) 0i -> y  If only the feature [+upper] triggers the vowel alternations, changes along the parameters of roundness and backness are left unexplained; This indicates that an intrinsic connection between vowel height and tone can only partially account for these observations. The second problem for this hypothesis is its inadequacy in deriving the surface forms in (3 a-c), which involve simplification of the diphthongs (II) in (5). Applying Yip's rule (4) to (3 a-c) gives the long forms [ii], [yy] and [uu], respectively, but not the short forms [i], [y], and [u]. To derive these correct forms, a monophthongisation rule is also needed in addition to the vowel-raising rule in (4). Furthermore, the vowel-raising rule must precede the vowel-shortening one since the entire set of non-high vowels in loosefinalsin (3a-c) are dropped in the corresponding tight finals The vocalic change that occurs in (3 a-c), therefore, may not be best regarded as vowel raising, but as monophthongisation instead (or diphthongisation, depending on whether the tight finals or the loose finals are underlying). The third problem for the height correlation hypothesis is the implausible vowel inventory that it yields. As Chan (1985) notes, Yip's treatment of the vowel alternations in sandhi forms implies that monophthongal high vowels are not independent phonemes in Fuzhou. but only appear with non-high vowels in diphthongs. If high vowels are only present with a non-high vowel in the surface representations, we need to explain why they  25  Chapter Two  Jiang-King, 1996  cannot occur alone in surface representations, given that the feature [+HI] is an active Felement for lexical entries. This is rare, if not unattested, in vowel system typologies. The fourth problem for this hypothesis is its inability to capture the cross-linguistic variation in tone-vowel interaction. Some other closely related Northern Min languages, such as Fuqing. Fuan and Ningde. also have similar patterns of vowel alternations. Particularly, the tight-loose distinction of finals exists in all of these languages and vowels in the loose finals change into the corresponding forms in the tight finals under different tonal conditions. The tonal inventories of these languages are given in (6). The Fuzhou data is from Liang (1982, 1983, 1988, etc.), Fuqing data is from Feng (1993a), Fuan and Ningde data are from Norman (1977):  (6) Tonal inventories for the four Northeast Min dialects I. tight finals -Jfff  a. Fuzhou  mm  b. Fuqing  mm  c. Fuan  m  d. Ningde  11. loose finals  Yin Ping, Yang Ping YangRu PHA 44 H 53 H M 5  Shang  ±*  Yin Qu  YinRu  31 M L  213 M L M  23  Yang Qu IB* 242 M H M  53 H M  44 H  5  33 M  21 L  22  41 H L  43 H M  11 L  21  41 HL  35 M H  54  13 L M  33 M  11 L  5  41 H L  35 M H  33  41 HL  Notice that in table (6) above, the tones in the Yang Ping category (the shaded cells) vary from language to language. Namely, H M is found in Fuzhou (6a), H in Fuqing (6b), and L in both Fuan and Ningde (6c, d). If vowel changes in Fuzhou were triggered by the [+upper] register (Le., the tones in the tight finals in I), the same alternations triggered by the tones in the same Yang Ping category in Fuan and Ningde (i.e.. the cells in (6c, d)) are not expected.  26  Chapter Two  Jiang-King, 1996  Although the height correlation hypothesis cannot explain all instances of vowel changes related to tonal conditions, it is important to realize that some sort of relationship between tonal environment and vowel changes does exist. It is this observation that has attracted researchers to study the phenomenon of tone-vowel interaction: The insight of the criticism from Maddieson (1976) and Chan (1985) is that pitch height distinctions cannot be directly responsible for inducing the vowel alternations. An intrinsic association of vowel height and pitch therefore cannot be invoked as an explanation to account for the data in Fuzhou. the most often-cited example in support of this claim. Maddieson also raises the question whether vowel height differences can ever induce linguistically significant pitch variation. He points out that intrinsic variation in the pitch of vowels does not necessarily become phonologically significant  2.2.2  The "metrical duration" hypothesis (Wright 1983)  Wright's (1983) metrical-duration hypothesis is based on duration differences between vowels in different positions within a disyllabic word: She draws on spectrographic studies which show that the duration of thefirstsyllable of a disyllabic compound is reduced twothirds from its citation form, while the duration of the second syllable is only reduced less than one-third of its citation form. This evidence does not support the claim that there is a direct cause-and-effect relationship between tonal height and vowel alternations (Wang 1968, Yip 1980). Rather, it suggests that both tone sandhi and vowel quality change are independently related to stress. She attributes both tone sandhi and vowel alternations to an iambic stress pattern which reduces the duration of unstressed syllables. The vowel alternations Wright tries to account for are listed in (7) (I = the tight finals, H = the loose finals).  27  Chapter Two  Jiang-King, 1996  (7) Vowel alternating pairs dealt with by Wright (1983:28) I  n  d.  eirj  aiq  0y(o)  e.  0yn  oin  ou(n)  f.  OUT)  auq  I  II  a.  ifa)  ei(n)  b.  y(q)  c.  u(n)  .  To account for the vowel alternations in (7), Wright treats the vowels in the loose finals as underlying, and posits the constraint in (8) and the two vowel reduction rules in (9) to derive the vowels in the tight finals.  (8) Constraint against branching nuclei in an unstressed position (Wright 1983:47)  * a  w  N  The function of (8) is to prevent a syllable having a branching nucleus from occurring in an unstressed position, namely, the first position of a disyllabic domain.  (9) Fuzhou Vowel Reduction Rules (Wright 1983:52) a.  Deletion: If two segments of the branching nucleus agree on all features or all but one, delete the first segment.  b  Quasi-Deletion: If the two segments of the nucleus differ in more than one feature, a short diphthong will be formed, with distribution of [-hi][+hi] in that order, and [+rd] if either segment shows rounding. The other features will be drawn from the second vowel.  28  Chapter Two  Jiang-King, 1996  Rule (11a) derives i(rj), u(n), y(rj) from ei(rj), ou(rj), 0y(n), respectively, while rule (lib) derives 0y(rj), ou(n), ei(rj) from oi(rj), au(n), ai(rj), respectively. Here Wright treats 0y(rj), ou(rj), ei(n) in an unstressed position (i.e., non-final position of a disyllabic domain) as single^segment diphthongs , and their corresponding forms oi(n), au(n), ai(rj) in a stressed 2  position (i.e., final position of a disyllabic domain) as two-segment diphthongs . The 3  question is that if stress conditions vowel alternation, why do the otherfinals,given in (10) below, retain their citation forms when unstressed?  (lO)Otherfinals(Wright 1983:32)  ja(rj)  wa(n)  je(n) wai wei  The high vowels [i] and [u] preceding another non-high vowel are written as 0 ] and [w] in (10). They are defined as 'weak' vowels because of their low intrinsic sonority value. In order to resolve the dilemma, Wright attributes the difference between alternating finals and non-alternatingfinalsto the different structures in (11), in which the subscripts "s" and "w" are defined as "strong" and "weak" based on vowel sonority:  2  What a single-segment diphthong means in terms of moraic structures is that two sets of vocalic features  link to a single mora. 3  Similarly, a two-segment diphthong may be seen as two sets of vowel features linked to two different  moras (i.e., each linked to a separate mora).  29  Chapter Two (11)  Jiang-King, 1996  a. Alternating finals  A  Initial Final  b. Non-alternating finals  Initial Final  r\  'w Vs  The structure of the alternating finals in (11a) represents a falling diphthong with a s-w nucleus, while the structure of non-alternatingfinalsin (1 lb) represents a rising diphthong with a w-s nucleus. The constraint in (8) that reduces branching nuclei is claimed not to refer to the "w" left branch of the nucleus. Wright concludes that the vowel alternations in Fuzhou lie in the different prosodic structures of syllables. It has nothing to do with tones. First, in disyllabic words, the finals in an unstressed (i.e., non-final) position reduce in duration from those in stressed or isolation positions. This reduction can be expressed by a constraint forbidding branching nuclei in a weak position as in (8). Second; syllables that have branching nuclei in monosyllabic forms undergo Deletion and Quasi-Deletion. The insight of Wright's proposal is thatfirst,tone sandhi and vowel alternation in Fuzhou involve different syllable structures. In particular, tonal change and vocalic change are independent of one another; both of them are triggered by stress Diphthongs in an unstressed position cannot have a branching nucleus, hence reduce some feature content, becoming single-segment diphthongs. Second, tone sandhi involves the loss of a mora since the mora is the tonerbearing unit and unstressed syllables cannot have two moras. The problem for Wright's (1983) proposal is that the close relationship between tonal groups and vowel distributions in citation forms (i.e., monosyllabic words) is completely ignored Although the attribution of vowel reduction to stress (in weak position) without considering tonal factors apparently accounts for the vowel quality change in the disyllabic cases, the vowel distributions with the "tight/loose" distinction in a final syllable of disyllabic forms (the "strong" positions) remain unexplained.  30  Chapter Two 2.2.3  Jiang-King, 1996  The "metrical-tonal contour" hypothesis (Chan 1985)  Chan (1985:424-425) rejects the height-correlation hypothesis, citing evidence from neighboring dialects of Fuan and Ningde where the high tones 151 and 754/ co-occur with the lower set of vowels In the context of a correlation between tone and vowel height, such a cooccurrence would be unexpected. For Fuzhou. Chan characterizes a rising tonal contour as one of the conditioning factors for the vowel quality changes This is based on the phonetic finding that a rising contour is longer in duration than a falling contour or a level tone (Gandour 1977, Zee 1978): However, she observes that there is a rising sandhi tone [35] (MH) which does not trigger vowel quality changes. Therefore, a rising tonal contour alone is not a sufficient condition for vowel quality changes in Fuzhou. The main factor, Chan claims, that differentiates the rising contour in citation tones from the rising tone in sandhi contexts is precisely that of stress. Therefore, the conditions for vowel quality changes in Modern Fuzhou (Chan 1985:468) are stress ('strong' syllables) and a L H (rising) tonal contour, because both of them increase syllable length (only the CVC syllables undergo vowel quality change). The Fuzhou vowel system proposed by Chan contains three underlying vowels: / i u a/. Their combinations give 7 surface vowels, using the particle type of representations: IM = [i], lul - [u], /a/ = [a]; /iu/ = [y], /ia/ = [e], /ua/ = [o], /uai/ = [0]. Chan treats all diphthongs as single segments. The tonal representations she assigns are given below:  (12)  Underlying Tones in Fuzhou (Chan 1985:99-107):  Tones in "tight" finals  Tones in "loose" finals  Ying  /44/H  /32/LH  /213/LH  Yang  /51/HL  /5/HL  /131/LHL  31  /13/LH  Chapter Two  Jiang-King, 1996  To get the appropriate vowel quality, Chan posits three rules which derive the loose finals from the tight ones, listed below.  (13) Vowel-lowering rule (Chan 1985:478) [eiq]-> [aiq] and [0yq] -> [oyq] v  A /  a  c i  (14) i-loss rule (Chan 1985:479) [e] -»[a], [eu] -»[au], [0] -> [o] v  c  A / i  a  (15) Diphthongisation rule (Chan 1985:481) [iq] -> [eiq], [uq] -»[ouq], [yq] -> [0yq] v  0  c [+hi]  Chan's proposal differs from Wright's in that it takes tonal contour to be a factor in triggering vowel quality change. Even though Chan claims that both stress and a rising tonal contour are conditions for such changes in that they both increase the syllable length, she does not implement this idea formally. It is therefore never made clear how it actually works. The insightful idea in Chan's proposal is to relate tonal contour to syllable length. This is compatible with the phonetic correlation between pitch contour and duration that is found in a number of experimental studies (see section 2.1 in this chapter for details) What Chan's insight implies is that the intrinsic duration in pitch contour may leave a trace in the phonological behavior of tones.  32  Chapter Two  Jiang-King, 1996  2.3 Fuzhou tone-vowel interaction  This section investigates tone-vowel interaction in Fuzhou. the most controversial case which has been cited as evidence both for and against the intrinsic height correlation hypothesis. In particular, I will examine vowel distributions and alternations with respect to their tonal environment, and identify the exact factors that trigger the vowel distributions and alternations. The Fuzhou phonological system has been comprehensively described by a number of scholars, such as Tao (1931), Chao (1943), Chen (1967, 1969), Lan (1953), Wang (1965, 1968, 1969), Liang (1982, 1983, 1984), Zheng (1983, 1988), Chen and Zheng (1990) It contains the 15 consonants In (16), the 10 vowels in (17), and the 7 citation tones in (18):  (16) Consonants:  p  t  p*  t  ts ts  h  k k  h  s m  ? h  h  n  n  1 In traditional Fuzhou phonology, there is a zero initial consonant, represented by zero. What the zero initial consonant really represents in terms of syllable structure is Onset. Presumably, native speakers access the onset position even though it may be segmentally empty. E. G. Pulleyblank (1986) argues that the zero initial is actually a laryngeal glide:  (17) Vowels:  i  y  u  e  0  o  e  ce  o  a  33  Chapter Two  Jiang-King, 1996  There is a considerable amount of variation in the phonetic realization of vowels and tones. Take the low vowel [a] in the vowel inventory (17) for example: As can be seen in (19), it has three variant forms: [a], [a] and [o]. The vowel [o] only occurs with a following rounded high vowel [u] Whether this local assimilation is properly analyzed as phonetic or phonological, it has no bearing on the phonological issues examined in this thesis and will not be transcribed except in the chart in (19) On the other hand^ the variation between [a] and [a] is significant since both vowels occur in the same segmental environment. This can be seen in (19) and will be discussed at length below. The Fuzhou phonological system presented here is from Liang (1982, 1984), Chen & Zheng (1990) and Feng (1993b). The tone letters H , M , L are used here to facilitate exposition In particular, Ff represents the tonal number 5 or 4 on Chao's (1930) system of tone letters in which 5 indicates the highest pitch, while 1 indicates the lowest pitch. M represents 3 or 2; and L represents 1. (18) Tones :  Tones in "tight" finals  4  m Yin W Yang  ^•Ping 44 H 53 H M  Ji Shang 31 M L 5 ? HM  Tones in "loose" finals 213 M L M 242 M H M  KRu 23? M L M  The chart in (18) only contains citation tones. Sandhi tones are not included here, but will be discussed later (see chapter 5) A citation tone refers to a tone in a monosyllabic morpheme in isolation. Historically, the tones in the Yin category are developed from syllables with initial voiceless obstruents, while those in the Yang category are from syllables with initial voiced obstruents.  4  The tones 5 and 23 only occur in syllables with a glottal stop conda. They are assumed to have the tonal  contours HM and MLM since they behave identical to 53 and 213, respectively. The shortening is presumably due to the glottal stop coda whose effect on tonal contour is open for further research.  34  Chapter Two  Jiangr-King, 1996  Fuzhou speakers distinguish between two groups offinals:the "tight"finalsand the "loose"finals(the nature of this tight-loose distinction will be identified in section 2.5). The seven citation tones in (18) are divided into two groups corresponding to the groups of finals. In particular, the tightfinalscontain either a high level tone (H) or a simple falling tone (HM, ML), while the loose ones have either a concave tone (MLM) or a convex tone (MHM), as shown in (18). Segmentally, the tight finals differ from the loose ones in a number of ways, listed in (19) below.  (19) Alternatingfinals(I = "tight"finals;H = "loose" finals)  n  I  I  n  I  H  I  u  ou  iu  -  ieu  usi  -  uoi  uai  ~ uai  i i?  -  ei ei?  y  ~  0y  y?  ~  0y?  in  ~ eirj  yq  ~  0yq  ie ie? ien  ~ ie ~ ie?  0y 0y?  -  oy?  --  ayrj  5  ~  ;  u?  -  ou?  un  --  oun  yo -~ yo yo? yo? yorj -- yon  iet)  o  ~  0  uo  o?  ~  0?  uo?  ~ UD ~ UO?  uon  ~ uon  0yrj  ia ia? iarj  ~ ia  ua  ~ ia?  ua?  --  ua?  ~ iaq  uan  --  uarj  ai  ~ ai  au  ei?  ~ ai?  ou?  ein  ~ airj  a  -  a  a?  --  a?  arj  --  an  II  -  oy  ua  au  - DU? oun ~- oun  The i ~ ei corresponding pair occurs in Liang's (1982, 1983) descriptive works as i ~ ei Here and throughout, I write the i ~ ei pair as i ~ ei to be consistent with the representation of the u ~ ou and 0 ~ 0y alternations.  35  Chapter Two  Jiang-King, 1996  First, the table in (19) shows that monophthongal high vowels in the tight finals ([i], [u], [y]) correspond to diphthongs containing a high vowel component in the loose finals ([ei], [ou], [0y]) respectively. Second, tense non-high vowels in the tight finals ([e], [o], [a]) correspond to their lax counterparts in the loosefinals([e], [o], [a]) respectively. Third, diphthongs formed by a mid vowel and a high vowel in the tight finals correspond to that comprised by a low vowel and a high vowel in the loosefinals.Apart from the alternating finals, there are a few non-alternatingfinals,listed in (20) below:  (20)  Non-alternating finals a.  e  e?  oe  b.  m  n  q  ce?  eu  (20a) are found only in tight finals; there are no corresponding loose finals. (20b) are syllabic nasals, which are only found in the morpheme denoting negation.  2.3.1  Tone-vowel interaction in monosyllabic words  In this section, I investigate the vowel distributions in monosyllabic words (the citation forms), and explore the precise factors which trigger the observed distributional effects. Recall that Fuzhou distinguishes two types offinals:the tight finals and the loose finals, whose segmental properties are listed in columns I and II of (19), respectively. Tones in the tight finals are either H level or simple falling (HM, ML), whereas those in the loose finals are either concave or convex. (MHM, M L M ) These cooccurrence restrictions are illustrated in all sets of data below.  36  Chapter Two (21)  Jiang-King, 1996 I "tight"  a.  ts  b.  pirj  iML H  c.  (22)  I "tight" a. ku b;  H  tsuq ^ 1  c. u?™  (23)  Gloss  Gloss  Distribution i ~ ei  'only'  d.  tsei^^  'will'  'guest'  e.  peirj  'combine'  'reach'  f.  kei?^  Gloss  MLM  Gloss  Distribution  'old; reason'  u~ou  d.  k MLM  'permit'  e.  tsouq ^- ^  'don't'  f.  GU  Gloss  a.  gyHM  'must'  b  Sy?HM  c.  tyi]  'lucky'  BL "loose"  'alone'  I "tight"  HM  H "loose"  ou  1  1  JMLM  'handsome' 'house'  H "loose"  Gloss  Distribution  d.  <j0yMHM  'sequence'  y~0y  'continue'  e.  s0y?  'save up'  'repeat'  f.  t0yqMHM  MLM  'middle'  The data in (21), (22) and (23) show that a single high vowel in the tight finals ([i], [u], [y]) corresponds to a diphthong in the loose finals ([ei], [ou], [0y]), and that the diphthongs comprise a high vowel preceded by a mid vowel with the same feature value for roundness. However, non-high vowels behave differently from high vowels. The examples in (24), (25) and (26) demonstrate that underlying mid vowels and low vowels surface as tense vowels ([e], [o], [a]) in tight finals and as lax vowels ([e], [o], [a]) in loose finals.  37  Chapter Two (24)  Jiang-King, 1996  I "tight"  Gloss  a.  tsierj  'stick'  b.  tsierj ^  c.  H "loose"  Gloss  Distribution  d.  tsierj " ™  'fight'  e ~e  'felt'  e.  tsien ^  'occupy; capture'  tsierj  'exhibit'  f.  sierj " ™  land; be good at'  I "tight"  Gloss  H "loose"  Gloss  Distribution  a.  ko  'song'  f.  'individual'  o~0  b.  uorj™  'forget*  g  c.  p  'vigorous'  d:  kyo™  e.  (25)  (26)  H  111  ML  11  1  1  1 1  u 3 n  MHM  'flourishing'  h.  p  pMLM  'to peel; to shell'  •bridge'  i.  kyjrfMLM  'decisive'  y pHM  'jump'  j-  y^JMHM  'read (lit.)'  I "tight"  Gloss  U O  pHM  G  a. b.  1 1  tsia  H  c. d  kau  e.  Ijh ML ai  H  u 3  H "loose"  'false'  f.  'to cover'  g-  'few, scant'  h.  'suburbs'  i  'change'  j-  tsia  MLM  6  Gloss  Distribution  'holiday'  a~a  'sugarcane' 'hung up'  kau  MLM  'enough' 'approximate'  The patterns of correspondence in (24), (25), and (26) have not been treated in previous theoretical studies These same patterns are also included in descriptive studies by Liang (1983, 1984, 1990), Zheng (1983; 1988), and Chen and Zheng (1990). Notice that the mid vowels alternating along the tense/lax dimension must not be followed by a high  6  Notice that the corresponding pair e ~ e only occurs with a preceding high vowel [i|. No word is found  in the "loose" finals with a mid front unrounded vowel without a preceding [i]. The lack of e ~ e distribution without a preceding high vowel is assumed to be an accidental gap. 38  Chapter Two  Jiang-King, 1996  vowel. If a mid back vowel is followed by a high front rounded vowel, it appears as front in the tightfinalsand as back (0 ~ o) in the loosefinals,shown in (27):  (27)  I "tight"  Gloss  a.  tsVy  H  Tiurry;urge'  b.  ts 0y  ML  'marrow'  b  H "loose"  Gloss  Distribution  c.  ts oy  'break to pieces'  0y~oy  d  ts'toy ^  h  MLM  "intensifier adv.'  1  Comparing (27) with (25), it can be observed that in (25) the mid vowel surfaces in tense/lax fashion regardless of whether it is preceded by a high vowel. However, when the mid vowel is followed by the high vowel [y], as shown in (27), it appears as a front or back vowel respectively. This different distributional behavior clearly suggests that (i) the presence of another segment, and (ii) the relative syllabic position might also be factors that trigger the particular realization of vowel quality Further examples of the presence of coda consonant affecting vowel distributions are found in (28) and (29) below:  (28)  I "tight"  Gloss  a.  mei?™  'sfrange(lit)'  b  tein ^  'wait'  I "tight"  Gloss  a  tsoun  'stolen goods'  b.  pou?™  'thin (lit.)*  (29)  1  H  H "loose"  Gloss  Distribution  c.  mai? ^  'strange(colloq.)'  ei ~ ai  d.  tain*^  'surname'  1  II "loose"  Gloss  c.  tsaun**™  'to bury'  d.  pau? ^  'explode'  1  Distribution ou~au  The examples in (28) and (29) demonstrate that a low vowel [a] in the loosefinals,when followed by a high vowel plus a consonant, corresponds to the mid vowels [e] and [o] in the tightfinals,respectively. Notice that in (26), the low vowel [a] in the loosefinalsdoes not raise to mid in the corresponding tightfinals.Instead, it changesfromlax to tense. The 39  Chapter Two  Jiang-King, 1996  lack of vowel-raising effects cannot be attributed to the tonal environment since the tones in the two types of finals in (26) are the same as those in (28) and (29). What makes the low vowel in (28) and (29) different from that in (26) is the presence of both a post nuclear glide and a coda consonant. This suggests that the number of segments in post nucleus position becomes a conditioning factor for the realization of a low vowel to a mid vowel. Of more interest is the asymmetrical behavior of high vowels in different syllable positions. High vowels manifest the correspondence between single segments and diphthongs in (21), (22) and (23), but not in (24), (25) and (26). The only difference is the relative syllable positions in which they occur. Particularly, the high vowels in (21), (22) and (23) are the only vowels in the tight finals, hence they can be regarded as in nuclear positions under the nuclear moraic model (Shaw 1992), whereas the ones in (24), (25), and (26) are not necessarily treated as nuclei, since they cooccur with other non-high vowels in the tight finals. The syllable positions, therefore, may also be a factor influencing the realization of vowel quality: This effect of syllable position on vowel distribution is not expected in the height-correlation hypothesis, since the featural specification for a high vowelis the same regardless of its possible syllable positions. The data in this section clearly suggest that tone is not the only factor affecting vowel realization. Other factors, such as the presence of a coda consonant and the syllable position of vowels, can affect vowel quality change as well. Therefore, tone-vowel interaction in Fuzhou cannot be treated simply as "feature-to-feature correlation" (Le., certain tonal feature(s) correlate with certain vocalic feature(s)).  40  Chapter Two 2.3.2  Jiang-King, 1996  Tone-vowel interaction in reduplication  In Fuzhou. adjectives can be reduplicated to denote intensity: The simple case is the reduplication of monosyllabic adjectives in (30), (31), and (32) below. The tone and vowel changes are underlined. Data are from Zheng (1988), Chen and Zheng (1990)  (30) a. b. c.  (31)  Adj.  Gloss  JgjMHM  'sharp'  s  o  u  MLM  J0yPMLM  'colourless'  —>  SU SOU  'resentful'  ->  ty^y? ^  Adi  Gloss  a  noM L M  "big-headed'  b.  k V ^  'quick'  (32)  1  Adj.  Redup. Adj.  toyrj ^  b.  pairf -  c.  mnu?  1  D  1  M  MLM  MLM  1  Alternation  'very sharp'  ei -> i  'very colorless' 'very resentful'  Redup. Adj.  Gloss  Alternation  ->  rjoH^npMLM  'very big-headed'  o -» o  _>  k  Gloss  a.  lM  Gloss  h  a H % a  M L M  h  'very quick'  a  Redup. Adj.  Gloss  Alternation  t^^oynM™  Very heavy'  ay -> 0y  'heavy'  ->  'stubborn'  -> pein^pain™^  'very stubborn'  ai -» ei  'painful'  -»  Very painful'  ou -» ou  mou^iDU? ^ 1  The data in (30) (31) and (32) show that when a monosyllabic adjective reduplicates into a disyllabic adjective, both vowel and tone in the first syllable of the reduplicated form undergo changes. In particular, four kinds of change take place in the reduplications. First, a diphthong comprised of a mid vowel and a high vowel loses the mid vowel in (30): Second, a mid or a low lax vowel becomes its tense counterpart in (31). Third, a back mid vowel becomes front in (32a). Finally, a low vowel raises to a mid vowel in (32b-c). The  41  Chapter Two  Jiang-King, 1996  tonal Change, on the other hand, involves tonal simplification. A complex contour tone becomes a simple contour tone: Notice that the vocalic changes involved are the exact same patterns as the patterns of vowel distribution in monosyllabic words. That is, the vowels in the loose finals become the corresponding ones in the tight finals. However, if an original monosyllabic adjective contains a level or a simple contour tone, as the tones in the tight finals in the monosyllabic words, there is no vocalic change to accompany the tonal change in the reduplicated form. The data in (33) below illustrate this pattern.  (33)  Monosyl. Adj.  Gloss  a.  pa^  •'fair  b.  puai™  'fat'  Redup Adj.  Gloss  ->  pawpaw -  'very full'  -»  puai ^ pu9i -  1  ;  :HM  'very fat"  (33) shows that once a monosyllabic adjective reduplicates into a disyllabic adjective, the tone in the first syllable of the reduplicated form changes. In (33 a), M L in the original monosyllabic form becomes a M H in the first syllable of a reduplicated form, while in (33b), H M in the original morpheme changes into a M L in the reduplicated form. Unlike examples in (30) (31) and (32), the tonal change in these cases is not accompanied by a vocalic change, and these tones are simple falling contour tones The same patterns of tone-vowel interaction are also observed in verbal reduplication forms. Data are from Zheng (1983). The little squares in the column of Chinese characters indicate a lack of appropriate Chinese characters to represent these Fuzhou morphemes:  Verbs  -»  Redupl. Vs  a  M  tWei?MLM  |i  b.  %  PJittlpeiMHM  ^  gfc  s y i M sou ^  \  e.  S  O  U  MLM  Gloss  1  42  H  'just kick something'  j£H  'just comb hair'  Wi  "just count'  Alternations i -> ei  u -> ou  Chapter Two  Jiang-King, 1996  d  xou? ^  $  xuHxour^M  e.  ts^yMLM  •  ts'y^ts'^yMLM  f  l0yMHM  p  lyHM.tyyMHM  1  1  'just whisk the dust' 'just take a look'  • •  y -> 0y  'j St tfy tO Wade' U  The examples in (34) show that the diphthongs [ei], [ou], [0y] in the first syllable of a reduplicated verb lose the mid vowel part and become monophthongal high vowels [i], [u]> [y], respectively. The data in (35) and (36) show that the lax non-high vowels become their tense counterparts.  Verbs  (35)  -> Redupl. Redupl Vs  Gloss  a.  # sie sie?  b.  sie?  c.  pgMHM  d.  muDn  H  H  JS MLM  Verbs  PQHMpjMHM  (51 muon ^ muon 12  ->  a. k'io?  c.  sau  d.  k'unMLM  MLM  •  MLM  MLM  #$f  'just stand here and there'  -f|"&  'eat this and that'  m&  'just hold'  InJInJ  'just ask'  Redupl. Vs  Gloss  ' HMp' ?MLM  'just pat something'  k'ia^k'ia?™"  'just take a picture'  sau ^ sau  'just clear dust'  P  b.  H  a  Q  1  MLM  k'anHM k'aijMLM  Alternation  m  e —» e  o  Alternation a —> a  'just take a look'  (37) and (38) demonstrate that a low vowel becomes a mid one and a mid back vowel becomes front when they are followed by both a high vowel and a coda consonant.  (37)  Verbs  ->  Redupl. Vs  a.  tain* ™  ^  tein^tairi ^  b  k'airjMLM  ^  k'ein^k'ain^  11  1  43  Gloss  Alternation  gfg  'put sth under sth else'  airj -> eirj  MM  'just cover it'  Chapter Two  (38)  Jiang-King, 1996  Verbs  ->  a.  soy ^  b.  SDyqw™  1  1  Redupl. Vs  ^  S  jg|  0yiM  S 3  yMHM  s^yrj^soyrj^  Gloss  Alternation  j^ffi  'just sit*  oyn -> 0yrj  j££s  'just see somebody off  It is important to notice that all the instances of vowel change in reduplications involve the complex contour tones. The verbs with either a level tone or a simple contour tone do not have vocalic changes, as shown in (39), (40), and (41) below.  Verbs a.  lirj"  b  n  c. d.  ->  •  y?H  lirj lirj  'carry everywhere' 'step on everywhere'  p'urjMLp'urp M H  W  ts'oH  H  y M L l y y H  n  M L  Verbs  ->  4  M L  m  nieq  $  nienMen"  b.  sia  c.  t'ai  a  p  m s  H  d: xuan  8$g$(3(j)  1  M L p  a  i  a  a  M H j s  H M  a  M L  #  fai^'ai"  HE  lau -lau  Wa  xuarjHxuan  H  M1  11M  'keep robbing with hands' 'jump everywhere' 'pick up something'  Redupl. Vs  m  H M  'keep ladling out'  jig  p'uon^p'uon ^  p  m  ts'o^s'oH  jgg  a.  'hold with hands'  Gloss  p'uoqHM 11  WW  Redupl. Vs  g  Verbs  e.  Gloss  H  p'UjjHM  m  Redupl Vs  Gloss  mm  'climb everywhere'  nn  'write everywhere'  m  'kft by 2 persons'  WM  'keep flowing'  grown  'turn staff over'  11  44  Chapter Two  Jiang-King, 1996  The important difference between the ones that have vocalic changes and the ones that do not is their tonal categories. That is, the vowels with a level tone or a simple contour tone do not change, whereas the ones with complex contour tones do change. The patterns of vowel changes are exactly the same as the vowel distribution patterns in monosyllabic words.  2 3.3  Tone-vowel interaction in "cutting foot words"  The "cutting-foot" words (data are from Liang 1982) are disyllabic words formed from monosyllabic words by a process resembling partial reduplication In particular, the first syllable in the output shares the most sonorous vowel and any segmental material before that vowel with the original word, while the second syllable of the output retains the tone and all segmental material of the original word except the onset, which is replaced by a new onsetfl-].This is illustrated in (42), (43) and (44) below.  Original  "cutting foot"  Gloss  'iHM  ts'iMLljHM  'not smooth'  b. ku™  kuML 1 HM  'tieup*  c. min"  miML^irjH  'hide'  a.  (43)  t s  U  d  ki?H  e.  surj  f.  pu?  g-  ny?H  nyMLJyyH  'fill in; squeeze'  Original  "cutting foot"  Gloss  pie^-lieu  11  'spray*  tie^.lien  H  *be given to'  'tickle' su^.lurj  H  H  a. pieu  pU^-.M  H  b. tien"  H  'move shakily' 'put long thin stuff in a nostril'  1 1  45  Chapter Two  Jiang-King, 1996  c.  mo"  d  'rise'  |QH  M L  M  O  rjo?  n0  MLl 7H  e.  ts'UQH  tS'UOML.lUOH  'tighten up a screw'  f.  kuorjM -  J^QML 1 JML  'roll up'  8-  ts'uo?"  ts'uoML.luO?  'scare'  Original  "cutting foot"  Gloss  ta .la  'twist*  H  1  UOT  a  ML  b. hia  H  c. d. lau  'hold one's head high'  0  H  11  ML  hia^.lia"  'split open'  sa^.lai™  'seat there not move'  laML.lau  'frown'  11  e  san  H  sa^.lan"  'arrest'  f  ka?  H  ka^.la?"  'ward off  8-  k'uan™  k'ua^.luanHM  •bind'  h  hia?  hiaML.lia?"  *hang down; droop'  i.  qian"  H  n  i  a  M L J j  a  n  'upright'  H  (42) shows that when a syllable contains a single high vowel, the first syllable of the output keeps that vowel without any change. (43) and (44) show that when a monosyllabic word contains more than one vowel, the first syllable of the output always retains the non-high vowel and any segmental material before that vowel. If, for instance, an original word has a sequence like CVjV C, where V, is not high and V is high, the first 2  2  syllable of the output only keeps V, but not V . This is illustrated in (45) below. 2  46  Chapter Two  Jiang-King, 1996  original  "cutting foot"  Gloss  a.  teirj  te^.lein  'poke (sand in the shoes)'  b.  kei?  c.  H  H  ke^.lei?'  'press from both sides'  poun  po^-.lourj"  'roll over in sandy material'  d  rjou?  nc^.lou?  'look upward'  e.  m0yrj  f.  l0yrj  11  H  H  H  m0 .l0yrj  H  ML  'puffy'  H  Ventilate'  H  Of interest are the cases where the first syllable of the output changes its vowel from the original word Data in (46), (47) and (48 > illustrate these cases.  original  "cutting foot  Gloss  Alternation  a.  tie  tieL.lie^  'drip'  e -> e  b.  hieu  c.  hier) * ^  hie^lierj ^  'throughout'  d.  pie?^  pieJUie? ^  'roll up'  original  "cutting foot"  Gloss  a.  so?  spt.lo?^  'tie tightly'  b.  b?  c.  tuoi^  (47)  MLM  MLM  1  0  WLM  hie .lieu t  1  WLM  1  1  1  M t M  'listless'  Alternation  'put on' 1  tuot.luDi^M  'hold tightly*  In (46), (47) and (48) the nuclear vowel in the first syllable of each output changes from lax to tense. The changed vowel and tone are underlined The question is why do only the examples in (46), (47) and (48) undergo vocalic change but the ones in (42), (43), (44) and (45) do not? The difference between these two sets of data are the tonal categories. 47  Chapter Two  Jiang-King, 1996  Specifically, the tones in (42), (43), (44) and (45) belong to the tight finals (i.e., H, M , HM), while the ones in (46), (47) and (48) belong to the loose finals (i.e., MHM, MLM).  (48)  original a. hau  MLM  "cutting foot"  Gloss  Alternation  haklau*^  'stretch out'  a —> a  1  b.  'be controlled'  c.  taL.iaqMLM  'permeate; seep*  d  la .lairj  'stand on tiptoe'  e.  hiaL.lia?^  l!  MLM  'collapse'  The most interesting cases are those in (49) where the outputs have two alternative forms for the first syllable. This is Indicated by the parentheses. In particular, if the input has a sequence CV,V C, where V,. is mid and V is high, the first syllable of the output can 2  2  be either V, or V . Both forms are possible outputs 2  ( >  original  "cutting foot" word  Gloss  Alternation  a. tseipwi-M  tsi^tse^.lei? ^  'squeeze'  i~e  b.  ts'i'(ts'e ' ) . l c i ? -  49  ts'ei? -w M  1  1  vlLM  'spray'  c. hou^w  hu'-^o'O.lou^  d.  houP^LM  hu^hc^.lou? ^  'sweep past'  ©•  ts'ouqMHM  ts'u (ts'o ^.lounMHM  'wring out wet clothes'  f  l0y?MLM  ly (l0 ).l0y?MLM  'subside'  g.  t'0V?MLM  t'yL(t'0L);l0y?MLM  'drawback'  'pour liquid on'  1  1  L  L  L  o ~u  0~y  Comparing (49) with (45), the sequence of the segmental material is identical. But the alternative outputs observed in (49) do not exist in (45) The question, then, is why do only the examples in (49) have two alternative outputs while those in (45) do not. The 48  Chapter Two  Jiang-King, 1996  only difference between (45) and (49) is again the tonal categories. Specifically, the tones in (49) are complex contour tones (MHlvL MLM) while the ones in (45) are level tones (H, M) or simple contours tones (HM). It is the tonal category but not a particular tonal value that determines the output forms. To sum up, the investigation of Fuzhou tone-vowel interaction reveals that tonal category Is only one of the factors that affects vowel distributions and alternations. However, this factor does not affect vowels directly. The tones have a much closer relationship with the types of finals (i.e., the tight/loose distinction) than with vowels. This kind of close relationship will be identified as a correlation between tonal contour and syllable weight (see section 2.5 for details). On the other hand, the syllable position and the number of segments present within a syllable are the factors that directly affect the behavior of vowels (see chapter 3 and 4 for an account of the direct relation between syllable structures and vowels).  2.4 Fuqing tone-vowel interaction  Fuqing is another Northern Min language spoken principally in Fujian province on the south coast of China. There are over a million speakers in Fujian province alone. Other Fuqing speakers are found in Southeast Asia, Europe, as well as North America. The Fuqing phonological system contains the 17 consonants (50), 12 vowels (51), and"7 tones (52). The consonants within parentheses occur only in a non-initial position of a domain. Particularly, Q3] is a sandhi form for [p] and [p'l, while [3] is for [tj and ft'].  49  Chapter Two  Jiang-King, 1996  (50) 15 consonants (Feng 1993a:28)  p  t  ts  k  p'  t'  ts'  k'  0)  (3)  s  m  n  ?  h n  1  (51) 12 Vowels (Feng 1993a:31)  i  y  u  e  0  o  e  ce  o  a  a  (52) Tones (Feng 1993a:35)  I. The "tight" finals  D. The "loose" finals  PHA Yin Ping  Yang Ping  YangRu  53 H M  44 H  5?H  B8A Shang • 32 M  YangQu  YinQu  41 HL  21 M L  22? M L  As in Fuzhou, there is a tight/loose distinction for finals in Fuqing. The seven citation tones, listed in (52) are divided into two groups. Unlike Fuzhou. however, there are no complex contour tones in Fuqing. The ones in Group I (i.e., in tight finals)-are either level  50  Chapter Two  Jiang-King, 1996  tones (H or M) or high failing tone (HM), whereas the ones in Group II (i.e., in loose finals) are either HL or ML. In other words, the tones in the tight finals do not contain a L, while the ones in the loose finals must have a L. Accordingly, vowels are also divided into two groups, as shown in (53)  (53) Fuqing Alternating finals (1= "tight"finals;H = "loose" finals) (Feng 1993a:31) Nuclear V High Vs  Mid Vs  Low Vs  I i y u e ie 0 o uo yo a ia ua  n ~ e - 0 ~ o ~ e ~ ie ~ oe ~  0  ~ uo ~ yo ~ a ~ ia ~ ua  I in yn un en ierj 0n on uon yon an ian uan  --<•~ *----~---  n en 0rj on en ien oen on uon yon orj ion uan  I i? y? u? e? ie? 0? o? uo? yo? a? ia? ua?  n ------------  e? 0? o? e? ie? ce? 0?  uo? yo? a? ia? ua?  n  I  eu iu  ~ eu ~ ieu  oi ui  ~ oi ~ uoi  ai/au ieu uoi  -  ai/au ieu uei  The generalizations that can be abstracted from (53) are the following. First, high vowels [i], [u], [y] in the tight finals correspond to mid vowels [e], [o], [0] respectively in the loose finals. Second, the tense mid vowels [e], [o], [0] and low vowel [a] in the tight finals correspond to their lax counterparts [e], [o], [ee] and fa] in the loose finals, respectively. Third, a low vowel [e] in Group II (i.e., in loose finals) corresponds to mid vowel [e] and [o] in Group I (i.e., in tight finals) depending on its surrounding vowels. Finally, mid vowels [e] or [o] in loose finals disappear in the corresponding tight finals when they are both preceded and followed by a high vowel: This kind of vowel correspondence in tight/loose finals is comparable to that in Fuzhou even though the vowel inventory inFuqingis different from that in Fuzhou. In the following sections, I will examine tone-vowel interaction in Fuqing in detail. 51  Chapter Two  2.4.1  Jiang-King, 1996  Tone-vowel interactions in monosyllabic words  As in Fuzhou and other Chinese dialects, morphemes are usually monosyllabic, hence, a syllable can be a word. In this section, I investigate vowel distributions with respect to their tonal environment in monosyllabic words and show how tones and vowels interact with each other. The data are from Feng (1993a:33-35).  (54)  I "tight"  Gloss e.  b.  p'i  M  gg. tobb  f.  c.  tin™  'treas  g-  d.  ni?  (55)  g 'day'  H  I "tight"  H"loose"  Gloss  tse™-  'character  1  'air*  «  hi  ke?ML  Gloss  'small town' 'urgent*  H'loose"  Gloss  a  ts'uHM  j|a  'rough'  e. Vet*-  ft- 'rabbit'  b  p'urp'  9  *bee'  f.  (SJ 'stifling'  c.  sun  "boat*  g. p o ^  If  'wealthy'  d.  pu?"  'lie down'  h. po?^  H  'stomach'  H  #  a.  tsy  $t  b.  tsyrj  c.  hyrjH  ||  d.  ty?H  ^£  M  M  mon™-  Gloss  I "tight"  'boil'  f (» 'swell'  Distribution  II"loose" e. ts 0 ,  ML  f.  ts0nP-  1>ear'  g-  0  'each'  h  t  u~o  Gloss  Distribution  'place'  y~0  'crowd'  r,HL  m 'use'  0?ML  ft  52  Distribution  "bamboo'  Chapter Two  Jiang-King, 1996  The examples in (54), (55) and (56) above show that high vowels [i], [u], [y] in the tight finals correspond to mid vowels [e], [o], [0] in the loosefinals,respectively. The tones in the loosefinalshave two characteristics. First, they must be a contour tone. Second, they must involve a L part, hence, be falling (either HL, or ML). On the contrary, the tones in the tightfinalsdo not have these properties. They can be either a level tone (H or M) or a high falling tone (HM). They do not involve any L part. This seems to suggest that vowel height correlates with tonal height. A close examination of the data in (54), (55) and (56), however, reveals that the phonological correlation between tonal height and vowel height cannot be established. First, the tonal height in the two groups offinalscannot be clearcut. On the one hand, tones in the tightfinalsare not always H; the M tone also occurs in the tightfinals,as shown in (54b) and (56a, b). If the vowel change from the tight group to the loose group were to involve vowel-raising under a tonal condition of H, the same vowel change with a M tone environment in (54b) and (56a, b) would be left unexplained. On the other hand, the tones in the loose group are not always L. The L tone is only a part of the falling contour The other part of the falling contour is either H or M . If the vowel change from the loose to the tight were to be characterized as vowel-lowering under the condition of L tone, we could not explained why only the L part of the tonal contour conditions the vowel change and the H or M part of the tonal contour in the loose finals cannot be part of the condition. Second, the data in (57), (58), (59) and (60) below reveal that the correspondence between the mid vowels and the high vowels in the two groups of finals only occurs in the nuclear vowel. Non-nuclear high elements in (57c-g), (5 8d-j) and (60d-k) in the tight finals do not change into mid. This suggests that tone is not a direct factor, or at least not the sole factor affecting vowel features. Rather, the relative possible positions influence vowel distributions directly.  53  Chapter Two  (57)  Jiang-King, 1996  I "tight" a.  s e  b.  ten™  c.  pe?  HM  Gloss IS  m.  H  d. e.  nierj  f.  sieF  g-  (58)  t  e  u  HM  Bfc m  t  'oM  b.  Gloss  m 'small'  'west'  h.  s e  lamp'  i.  ten™-  'pull'  j-  peyML  A  'chicken'  k.  laeML  m 'season'  1.  qie™-  'art*  'tongue'  m.  tsie?^  m 'connect'  'carve'  n.  k'euML  In  m 'dye'  M  I "tight" a.  ITloose" ML  Gloss  'eight'  •button?  II"loose"  Gloss  Distribution  'table'  o~a  it  'ask for'  k.  t 3  •  'oil lamp'  1.  koq ^  W slippery'  m.  m •bind'  n.  PUD ^  m 'turn!  o.  nuon™-  m 'willing'  P  kuo?^  a  ML  'steel'  1  # •bone*  ko?  d.  p  e.  tuon  f.  nuo?  n  g-  ypHM  m •medicine'  q-  y HL  m 'read'  h  ts'yoqP  $  'wall'  r.  t 'y ML  m 'sing'  i.  k'yo?  m •play'  s.  m 'pile'  t.  j-  (59)  U O  HM M  H  t o  H  jHM  I "tight" a.  'month'  Gloss  'cloth'  1  0  S  0 n  &  'country'  'decide' 'face'  ITloose"  Gloss  SOgML  'not familiar with'  'comb'  e.  'donkey'  f, •tee -  C. tyX}™  'winter'  g-  toen" -  'move'  d.  'eye'  h.  poeTML  'north'  b.  10  H  m0?  H  12  1  54  e ~e  "Deng (surname)'  c.  H  Distribution  'limonene'  Distribution  Chapter Two  (60)  Jiang-King, 1996  I "tight"  Gloss  IT'loose"  Gloss  Distribution  'hundred'  a ~a  a.  p HM  •6  'white'  1.  b.  p  m  'class'  m. par)™-  'sick'  c.  a?  ft  tjox'  n.  O?ML  'duck'  •a.  sia  'write'  0.  sia™-  'sound'  P-  'drag'  q  tua™-  a  A N  HM H  M  p ML  W  A  #  'thank'  m  'line'  e.  s  f.  t'uaHM  g-  puarj  'dish'  r.  puarj* -  'half  h.  uarj  •bowl'  s.  uarj™-  'change'  i.  pua?  m  'postscript'  t.  p  "big bowl'  sajHM  m  lion'  u  kaiML  j-  k.  j  a r  jHM  m  H  M  H  ir 'steal'  •big' 41  U a  pML  # •boundary' 'cover*  V.  Third, the vowel distributions in (57), (58), (59) and (60) take place among the nuclear non-high elements. In particular, the tense mid vowels [e], [o], [0] and low vowel [a] in the tightfinalscorrespond to their lax counterparts [e], [o], [os] and [a] in the loose finals, respectively. If vowel distributions were purely conditioned by tone height, the distributions with the tense/lax parameter in (57), (58), (59) and (60) would not be 7  expected. Notice that low vowels behave differently depending on their surrounding segments. Comparing the distributional behavior of low vowels in (60) with that in (61), it becomes clear that a low vowel varies long the tense/lax dimension only when it is not flanked by  7  It is possible to interpret the tense/lax distinction as a tongue-height distinction. This interpretation,  however, cannot be used to support the correlation hypothesis between tonal height and vowel height, since the tones in the two groups offinalsinvolve a H part in tonal contours, as mentioned above. 55  Chapter Two  Jiang-King, 1996  two high vowels. When it is both preceded and followed by a high vowel, it raises to mid: This kind of asymmetric behavior of the low vowels cannot be entirely attributed to their tonal environment since tones in the tight finals in both (60) and (61) are of the same categories. There must be some other factor(s) involved in triggering these different distributional effects. I "tight" a.  s  b.  pieu  c:  p  d  p'uoi  j  Gloss  HM  e u  ' f t  U O  M  k  Gloss  Distribution ieu ~ reu  e.  t  ML  m  'jump'  f.  kieu™-  m  'sedan'  'cup'  8-  p  j*.  'shell'  'skin'  h.  p'uei™-  & 'watch'  jHM H  •burn'  n"loose" 'i  B U  U B  iML  uoi ~ um  m 'quilt'  The data in (62) below are interesting because we see another asymmetry in the distributions of mid vowels. Comparing (62) with (57) and (58), we see that mid vowels vary along the tense/lax dimension in (57) and (58), but they do not in (62). Instead, a mid vowel is deleted ih (62) when it is flanked by two high vowels. The deletion of a mid vowel in (62) cannot be attributed to its tonal environment since the tones in the tight finals in both (57), (58) and (62) are of the same categories. Therefore, the direct factor that conditions vowel distributions cannot be tone alone.  I "tight" a.  pui  b.  tui  c.  ts'iuHM  d  kiu  H  H  H  Gloss  H "loose"  •  Gloss  Distributions  'rash'  iii ~ uoi  IB  'fat'  e.  p  H  'thump'  f.  tuoi -  'team'  'autumn'  g-  kieu" -  JS- 'uncle'  ^aU'  h.  kieuML  %  ^  U 0  jML 12  1  56  'study'  lu ~ ieu  Chapter Two  Jiang-King, 1996  To sum up, our investigation of tone-vowel interaction in monosyllabic words suggests that, on the one hand; there is a cooccurrence restriction on tone and vowel, and on the other hand, the vowel distributions in monosyllabic words cannot be entirely and directly governed by their tonal environment Three kinds of asymmetries are observed in this section. First, the high vowels in tight finals behave differently with respect to their relevant syllable positions They correspond to the mid vowels when they are the only vowel within a syllable. When they occur with another non-high vowel within the same syllable, they do not have such correspondence, and remain as high. Second, the mid lax vowels in the loose finals correspond to their tense counterparts only when they are not flanked by high vowels If they are surrounded by high vowels, they are deleted in the corresponding tight finals. Third, the low vowels also have two different characteristics. The lax low vowels in the loose finals correspond to their tense counterparts only when they are not surrounded by high vowels. If they do, they raise to mid rather than become lax. These asymmetries in vowel distributions cannot be attributed to their tonal environment since the tonal categories of all tight finals are the same. It seems like the tones relate only to the distinction between tight and loose finals, but do not directly trigger any vowel distribution effect. This suggests that tone does not interact with vowel features directly. Tonal effects on vowels at most can only be associated with the types of finals.  2.4.2  Tone-vowel interaction in disyllabic words  In this section, I investigate tone-vowel interaction in disyllabic words, and explore whether tonal changes relate to vocalic changes. As I mentioned before, morphemes in Fuqing are usually monosyllabic. When a monosyllabic word combines with another monosyllabic word to form a disyllabic word, the tone of the first syllable within a  57  Chapter Two  Jiang-King, 1996  disyllabic word changes. Sometimes, the tonal change is accompanied by vocalic change. This sort of co-variation is illustrated in (63) and (64). (63)  Morph  Gloss -»•  Disyl.  a.  ne *  'ear'  rjiHpa  b.  p' ML  #  'nose'  c.  pon  m  'excrement'  pujrja^t'0rj  d.  t s  iii  t o go out'  ts'uTttkaML  e.  ts^  ft  'pour'  [gyHM gML  f.  t 0  IT  •bamboo'  tyPHts'yo™  1  e  m  ' pML 0  yML  H  mi  **  Gloss  Alterm  'earpick'  e -> i  'nasal mucus' 'manure bucket'  M  m  '(a girl) to marry' 'pay attention to'  m  o —> u  0 •-> y  •bamboo mat'  The data in (63) show that the HL or M L tones in the monosyllabic words change into H or HM, respectively, when they occur as the first syllable within a disyllabic word. Meanwhile, the vowel undergoes change as well. In particular, the mid vowels [e], [o] and [0] raise and become [i], [u] and [y], respectively: This kind of co-variation might be viewed as evidence supporting the height correlation hypothesis since the vocalic change in these cases involves vowel raising, while the tonal change can also be characterized as register raising (i.e., from M L to HM), except the example in (63a) where the tonal change is best characterized as tonal simplification (i.e., from HL to H) rather than tonal raising. However, the height correlation hypothesis runs Into difficulties as far as the following data in (64) are concerned. The changed syllables are underlined.  (64)  Morph  Gloss -»  Disyl. words  Gloss  Alternations  seHne™-  'careful'  e -> e  tieuHny"  •to fish'  a.  se ^  $5  'thin'  b.  tieu^  #j  'to  c.  tien™-  tfe  'electricity' tien pieu  d.  nDr)  ft  'tender*  1  ML  fish'  H  M  non^muoi^  58  'meter' ftfcjc  'young sister* o -> o  Chapter Two  Jiang-King, 1996  e.  kai^-  f.  lau" -  g.  puan" -  ft  1  1  4&  'mustard'  kaP^ai™*  'mustard'  'old'  lauHn0rj  'old man'  'half  puan ni?  H  H  a-»a  'half a day'  H  The examples in (64) show that the tonal changes involved are the same as those in (63) (i.e., M L becomes HM, and HL becomes H). However, the vocalic changes are different. First, the non-high vowels [e], [o], [a] become [e], [o], [a], respectively Second, unlike the high vowels [i] and [u] in (63), the high vowels (i e., [i] in (64b, c, e) and [u] in (63) and (64b, f, g)) do not change at. all, even though the tonal environment in both (63) and (64) is the same. The only difference between (63) and (64) for the high vowels is their relevant syllable position. In particular, the high vowels in (63) are syllable nuclei while the ones in (64) are not. The effect of syllable position inducing vowel change cannot be explained by the height correlation hypothesis. We therefore need a theory that is capable of encoding both tonal and syllabic factors in affecting vowel change.  Monosyl  m m  doss  Disyl. words  'short'  eMikian™  'wash'  se^ts'iu™  'wash hands'  tonHqa™  'decision-maker*  a.  e  b.  se  c.  ton™  "become'  d  ho?  'to study' hoTHhau *  e.  t'au  f.  ha?  M  M  H  H  H  1  'head'  t'au^uo?  M  'combine' ha?MLpa?  M  Gloss  m  ¥®  'short guys'  'school' 'hair' 'purse'  Notice that the first syllables of disyllabic words in (65) changes their tones However, unlike the tone change in the previous data, there is no vocalic change occurring with the tonal change If vowel change were to be attributed to the tonal change, the lack of  59  Chapter Two  Jiang-King, 1996  vocalic change in (65) cannot be explained. Thus, tone is not a direct conditioning factor on vowel changes.  2.4.3  Tone-vowel interaction in reduplications :  In this section, I investigate tone-vowel interaction in reduplication and identify the exact factors that trigger their interaction. The data in (66) show that when a monosyllabic adjective reduplicates to form a disyllabic adjective, the first syllable of the disyllabic word undergoes both tonal and vocalic changes.  monosvl.  —> Redupl words  a.  s e  b.  tsen™-  #  c.  merj  d.  k'eML  e.  toi™-  •f.  huon™-  g-  ML  mm  Very thin and long  tsin^sen™-  WW  Very quiet'  it  merjHmen -  tili  'slowly'  &  k'ei&ik'e  m  HL  111  1  'littlebag'  huorjHhuon™-  'far away*  kau -  kauHkau"  1  'very thick'  h.  tsan  tsan tsan  ML  'how come'  i.  tsia ^ -  tsiaHtsia ^  j-  tsieu " -  ts'iuffl^ts'ieu  1  ML  1  1 11  m  H  e -> i  e -> e  'quickly'  ML  toi^Di™-  111  Alternations  doss  o -» o  a —> a  'just a beginning'  1  ML  'just smile'  ieu -» iu  There are three kinds of change taking place in (66). First, the tense mid vowel [e] becomes the high vowel [i] in (66a-b). Second, the lax non-high vowels [e], [o] and [a] in (66c-i) change into their tense counterparts [e], [o], and [a], respectively. Third, a mid vowel [e] is deleted in the output. Meanwhile, the tones in (66) also change; The tonal changes are of two types. First, low falling (i.e., ML) becomes a high falling (i.e., HM), as  60  Chapter Two  Jiang-King, 1996  shown in (66a, d, j). Second, a high falling becomes a high level (i.e., H), as shown in (66b, c, e, f, g,) However, the vocalic changes do not show up in the reduplicated forms in (67), (68) and (69) below.  monosvl  -»  Redupl words  Gloss  a.  hiHM  ffi  hi hi ts0?  'watery'  b.  ti?  1;  ti? ti?  c.  kin  %  Idn^nrrjHM  H  M  d. e.  H  H  H  ML  Very straight'  H  m  jjHyHM  tsy  H  ^  'closely' Very dark'  tsy^syHluo? ^  'calm and confident'  1  In (67), there are two kinds of tonal changes First, H M and H of the original words become H in the first syllable of the reduplicated forms, as shown in (67a, b, d, e). Second, a M tone of a original word becomes M L in the first syllable of the reduplicated form, as shown in (67c).  monosvl. ->  Redupl. words  a.  keq  kerj rjen  b.  teu  H  M  H  ft  c.  Gloss  H  teu^teu^sien^ 11  AB*/f3*  Very high'  mm  'shiver  p'ienftnien™  d.  hc^  e.  mon  f.  0rjH  8-  m0n  'just in right time'  ho^ho ^ 1  im  mon mon  H  H  1  H  'properly' 'frequently' Very red'  H  It  m0n m0n H  H  WM  'twilight'  The same kinds of tonal changes are also observed in (68) and (69). However, there is no vocalic change taking place in (67), (68) and (69). The question then, is why is the 61  Chapter Two  Jiang-King, 1996  tonal change accompanied by vocalic change in (66), but not in (67), (68) and (69)? A closer examination shows that the tones in (66) are the ones belonging to the loose finals while tones in (67), (68) and (69) are the ones in the tight finals. It seems that it is the types o ffinalsthat determines the cooccurrence of tonal and vocalic changes Tone itself does not have a direct effect on this choice.  (69)  monosyl. -> Redupl. words  Gloss  a.  ta™  Very dry'  b.  larj  c.  tsia?  d.  ua?  taHta™  $i  H  H  laq«lan  3g$i  H  tsiapHtsia?"  H  fc  ua? ua? H  Very-blue'  ( M ) ^ 'windy'  H  fcfc  'lively'  To sum up, the investigation of Fuqing tone-vowel interaction furnishes further support for the findings in Fuzhou. First, there is a cooccurrence restriction on tonal categories and types of finals Second, tone does not directly affect vowel change. Third, syllable positions and the number of segments present within a syllable are the direct factors inducing vowel change.  2.5 The nature of the tight-loose distinction  Tone-vowel interaction in Fuzhou (section 2.3) and Fuqing (section 2.4) reveals that the tight-loose distinction exists in both Northern Min languages. The notion of "final" refers to the subsyllabic constituent "rime", which includes all segmental and tonal features except onset. Since the onset consonant in these languages is not obligatory for forming a syllable (i.e., syllables may lack an onset in Fuzhou and Fuqing. see chapter 4 for details), the final alone may stand as a syllable. The tight-loose distinction offinalscan therefore be  62  Chapter Two  Jiang-King, 1996  viewed as a distinction of syllable types: The following table gives a summary of the segmental and tonal distinctions in the two types of syllables.  (70) Segmental and tonal distinctions of different types of syllables in Fuzhou and Fuqing Lg's I 0)  Fuzhou  (ii)  Fuqing  tight syllables  Segments a. monophthongs b. tense vowels c. round harmony required d. low-high vowel sequence disallowed with a coda C "a. two high vowels without a mid vowel in between b. tense vowels c. round harmony required d low-high vowel sequence disallowed e. high vowels  loose syllables  Tone  H, ML, HM  H,M, HM  Segments  Tone  a. diphthongs b. lax vowels MHM, c. lack of round harmony MLM d. low-high vowel sequence allowed with a coda G a. two high vowels with a mid vowel in between b. lax vowels c. lack of round harmony d. low-high vowel sequence allowed e. mid vowels  HL, M L  The table (70) shows that the tonal and segmental contrasts between the two types of syllables in Fuzhou are similar but not identical to those in Fuqing. In Fuzhou. the tonal contrast is apparently quantity (i.e., simple tonal contour vs. complex tonal contour), while in Fuqing. it is apparently quality (i.e., the L tone occurs only in the loose syllables but not in the tight syllables). Notice that the tones in the tight syllables in both Fuzhou and Fuqing are identical, namely, they are H, M , and HM. The tonal difference of these languages lies in the loose syllables. The segmental contrasts in the two languages are of three types The first one is the quantity difference. In particular, the number of vocalic segments present in the tight syllables is one less than in the corresponding loose syllables. In Fuzhou. this contrast 63  Chapter Two  Jiang-King, 1996  appears as monophthongs vs. diphthongs (70i-a). In Fuqing, it is diphthongs vs. triphthongs (70ii-a). The second one is the difference in feature content, namely, the tenselax distinction in Fuzhou (70i-b) and Fuqing (70ii-b). The third one is the harmonic 8  restrictions in the tight syllables. Both languages require round agreement within a diphthong ((70i-c) and (70ii-c)) and prohibit low-high sequences in a diphthong ((70i-d) and (70ii-d)). The only difference between these two languages regarding the segmental contrast is the correspondence between a high nuclear vowel in the tight syllables and a mid one in the loose syllables. This correspondence exists only in Fuqing (70ii-e) but not in Fuzhou. The generalizations above raise a number of questions as to the nature of the tightloose distinction. In particular, why both languages have two types of syllables (i.e., the tight-loose distinction), even though their segmental and tonal properties differ in certain aspects. Why are there only two types of syllables in each of the languages, but not three or four? Two possible answers suggest themselves immediately. One is the quality approach and the other is the quantity approach. Both of them are evidenced in these languages since the tonal and segmental distinctions in these languages involve both quality and quantity differences. In the following sections, I will argue for the second approach, namely, the quantitative distinction between the two types of syllables, as being more promising than the first one.  8  Alternatively, the tense/lax distinction can be characterized in terms of ATR/RTR distinction. Since the  approach taken here assumes that quantity distinction is primary while quality distinction is secondary for the two types of syllables, it is not crucial which pair of features are used to characterize this distinction. 64  Chapter Two 2 5.1  Jiang-King, 1996  Tone and duration  The intrinsic correlation between fundamental frequency (F ) and duration has been found 0  in several experimental studies. First, it has been observed that there is a reverse relationship between tonal height and duration. That is, a high tone is shorter than a mid or low tone (Benedict 1948:186 for Cantonese; Pike 1974:171 for Chatino; etc.). Fuqing provides phonological evidence supporting this finding As shown in table (70), Fuqing tonal distinction in the two types of syllables involves the presence of L tone in the loose syllables and the absence of the L tone in the tight syllables. Given that syllable weight is represented by the number of mora(s), and that the mora is a tone-bearing unit, the presence of a L tone could increase syllable length by requiring another mora, assuming that L tone cannot link to the head tone-bearing unit of a syllable. Thus, the tight-loose distinction of syllable types can be viewed as a contrast between syllable weight, that is, the light vs. heavy syllables. The formal mechanism increasing syllable length by the presence of L tone will be discussed in detail in chapter 3. Second, it has been reported that pitch contours are related to duration in that a rising tone or a concave tone is longer than a high level tone or a falling tone (Zee 1978 for Taiwanese; Abramson 1962 for Standard Thai; Dreher and Lee 1966, Chuang 1972, Howie 1974 for Mandarin; Langdon 1976 for Yuman languages). The phonological correlation of this type is found in Fuzhou. Two claims have been made in previous studies regarding the tonal distinction in the two types of syllables. The first one is the "pitch height" distinction (Wang 1968, Yip 1980). Yip (1980) claims that the tones in the tight syllables are higher in pitch, whereas the ones in the loose syllables are lower in pitch. This pitch distinction is represented by the feature [±upperj\ The [+upper] tones trigger vowelraising in the tight syllables (assuming that the vowels in the loose syllables are the underlying forms). Hence, a low vowel in a loose syllable (71b) becomes a mid vowel in the corresponding tight syllables (71a). 65  Chapter Two  Jiang-King, 1996  (71)  'tight" syllables a.  tein  c.  k ai h  ML  ML  loose" syllables  "wait*  b.  tairj  'change'  d.  k^i * - * 1  LHL  1 1  Distribution  surname'  eirj- am  'approximate'  ai ~- ai  However, a low vowel behaves differently in the same tonal environment. The pair 'ai~ ai' in (71c-d) shows that a low vowel alternates along the tense/lax dimension only when a high vowel follows it without a coda consonant present. This different behavior of a low vowel with respect to the presence or absence Of the coda consonant is unexpected under Yip's proposal since tones in the tight syllables (71a) and (71c) are [+upper] in both pairs, regardless of whether or not a coda consonant is present. If the tonal distinction in question is characterized solely in terms of pitch height, it is not clear why the same tonal feature (i.e., [+upper]) would trigger different vowel alternations for low vowels in (71). The second claim regarding the nature of the tonal difference in the two types of syllables is the "shape" distinction. Chan (1985) claims that the tones in the loose syllables contain arisingcontour (MHM, HMH). while the ones (H, HM, ML) in the tight syllables do not. Therisingcontour triggers vowel-lowering by increasing syllable length, assuming that the vowels in the tight syllables are the underlying forms. Hence, a mid vowel [e] in the tight final (30c) becomes a low vowel [a] in the corresponding loose final (30d). Chan's proposal is compatible with the phoneticfindings.However, Chan claims that the syllable is the tone bearing unit and that vowels link to the CVC skeleton Since syllable length is represented by the number of moras, it is not clear how therisingcontour as a whole unit that links to a syllable node can actually increase syllable length without referring to moraic structure. Chan's insight regarding the possible correlation between syllable length and tonal contour will be formally encoded into my proposal later. The real tonal distinction between the two types of syllables in Fuzhou; I argue, is quantity rather than quality. In particular, I claim that tones in the tight syllables (H, HM, 66  Chapter Two  Jiang-King, 1996  ML) link to one TBU while those in the loose syllables (MHM, MLM) link to two TBUs, givingriseto the distinctive syllable weight (I will discuss this in detail in chapter 3). It is the distinctive syllable weight that determines the realizations of vocalic segments in the two types of syllables in both Fuzhou and Fupjng (see Chapter 4 for detailed discussion).  2.5 2  Segment and duration  As shown in table (70), the segmental differences between the tight and the loose syllables are of three kinds: (i) the number of vocalic segments differs; (ii) the feature content differs (i.e., tense/lax and high/mid distinctions for the nuclear vowel); (iii) there are harmonic restrictions on the tight syllables. The question that arises is how to encode all of these segmental distinctions in a unified manner. One approach is to characterize the differences in terms of quality, namely, the feature content This approach, however, can only be partially successful. This is because the featural differences between the two types of syllables in both Fuzhou and Fuqing are only part of the tight-loose distinction. If we characterize this distinction in terms of feature content alone, the majority of generalizations will be left unexplained. The quality approach, therefore, is untenable. The other approach is the quantitative one. Following Wright's insight, we assume that the distinction between the tight and the loose syllables lies in the syllable duration rather than feature content. In particular, I assume that the tight syllables are shorter and the loose ones are longer, hence the tight-loose distinction can be characterized in terms of syllable weight. Given that syllable weight is represented by the number of moras, the observation that there are fewer segments present in the tight syllables than in the loose syllables is explained since short syllables (i.e., the tight ones) contain only one mora, and hence incorporate fewer segments, while longer syllables (i.e., the loose ones) have two moras, hence, incorporate more segments.  67  Chapter Two  Jiang-King, 1996  One may wonder about the tense/lax distinction, which involves a difference in feature content. Phonetic studies show that tense and lax vowels have intrinsic duration differences. For example, Fischer-J0rgensen (1990) reports that the fundamental frequency of the long tense vowels [i:] and [u:] is the same as that of their short lax counterparts [i] and [u] in German. In Cantonese, the tense/lax distinction of vowels represents a length distinction rather than a quality difference (Beijing daxue JfcfClfc^ 1989): The same can be argued to be true underlyingly for English (Halle and Mohanan 1985). By the same token, it is reasonable to interpret the tense/lax distinction in Fuzhou and Fuqing as a length distinction rather than a quality difference. In general, the tense/lax distinction involves correlation between quantity and quality: either one could be the lexically distinctive property and the other one redundant. The other featural distinction observed in Fuqing is that a high nuclear vowel in the tight syllables corresponds to a mid vowel in the loose syllables. The question is, can this vowel height difference be interpreted as a length distinction as well? Possibly. The duration difference between vowels of different heights is also found in phonetic studies. For instance, it is reported that a high vowel, other factors being equal, is shorter than a low vowel (Lehiste, 1970:18): This is compatible with the phonological patterning of vowels in Fuqing. In Fuqing. there exists a correspondence between two high vowels in the tight syllables and two high vowels with a mid vowel in between in the loose ones. If the tight-loose distinction is characterized as one of length, this difference in the number of segments, as well as the difference in vowel height, can be interpreted as length difference in terms of the number of moras involved. Now we turn to the vowel harmony restrictions on the tight syllables. Recall that there are two types of harmony restrictions observed in both Fuzhou and Fuqing. One is that agreement in roundness within a diphthong is required in the tight syllables, but not in the loose ones. The other is that a low-high sequence of vowels is prohibited in the tight syllables but not in the loose ones. All these restrictions involve vowel harmony in terms of 68  Chapter Two  Jiang^King, 1996  a certain feature content. Why then do these harmonic requirements apply only to the tight syllables but not to the loose ones? If we treat the tight-loose distinction as differences in feature content alone, there is no explanation for this phenomenon. On the other hand, if we identify the tight-loose distinction as a length difference, the harmonic restrictions only on the tight syllables are expected. That is, since the tight syllable is light and contains one mora, the cooccurrence of segmental features within a single mora should be more restrictive than that of a long syllable with two moras (see chapter 4 for detailed analysis). To sum upi by identifying the tight-loose distinction as a distinction in syllable length, all tonal and segmental differences between these two types of syllables are unified in terms of syllable weight, a desirable result.  2.5 3  Further evidence from Southeast Asian languages  If the tight-loose distinction of syllable types exhibited in Fuzhou and Fuqing is identified as a difference in duration, namely, a length distinction, it should be possible to find evidence showing a cooccurrence restriction between tonal contours and vowel length This kind of evidence is indeed found in various Southeast Asian languages. For example, in Hu, one of two Mon-Khmer languages described by Svantesson (1989), a distinction of vowel length has been replaced by a tonal distinction, i.e., the former short and long vowels have acquired a high tone and a low tone, respectively. In Sre, another MonKhmer language, there is co-variation between vowel length and tone, so that long vowels always have a low tone and short vowels a high tone (Manley 1972). Gandbur (1977) cites data suggesting a similar phenomenon in Thai He points out that the loss of a phonological distinction in vowel length historically among certain Thai dialects may be seen as principally conditioned by tone. In Northern Thai and Southern Thai, historically, short vowels tend to become long under the rising tones; long vowels tend to become short under the falling tones (Gandour 1977) In the Chiang Rai dialect short non-low 69  Chapter Two  Jiang-King, 1996  vowels have become long under rising tones, long non-low vowels have become short under non-rising tones (Gandour 1977). In the Phuket dialect long non-low vowels have become short under falling tones, all short vowels have become long under non-falling tones (Gandour 1977). Also, long vowels within a checked syllable occur only with contour tones HL or L M in Siamese and Red Tai respectively, whereas their corresponding short vowels do not have this restriction (Gedney 1965, 1989). All these reported cases furnish further support for the length distinction between the tight and the loose syllables. The following table gives a summary of the correlation between tonal contour and vowel length reported in the Southeast Asian languages.  (72) Correlation between vowel length and tonal contours in Southeast Asian languages Lg's I  Light syllables  Segment  Heavy syllables  Tone  Segment  Tone  Sre  short vowels in checked a  H  long vowels in checked a L, HL  Red Tai  short vowels in a checked a  M  long vowels in checked a  LM  Siamese  short vowels in a checked a  H  long vowels in checked a  HL  Hu  short vowels  H  long vowels  L  Chiang Rai  long non-low vowels become nonshort rising T  short non-low vowels become long  rising T  Phuket  long non-low vowels become falling short T  all short vowels become long  nonfalling T  70  Chapter Two  Jiang-King, 1996  2.6 Conclusion  Three findings emerge from our investigation of Fuzhou and Fuqing First, there is a cooccurrence restriction on tonal categories and vowel distributions/alternations. This cooccurrence restriction is identified with the light-heavy distinction of syllable types. The tonal differences in the two types of syllables are primarily quantitative in Fuzhou. while Fuqing invokes the presence or absence of L. The segmental differences between the two types of syllables involve (i) differences in the number of segments; (ii) differences in feature content (i.e., the tense/lax and the high/mid distinction); (iii) harmonic restrictions on the tight syllables. Second, high vowels behave differently with respect to syllable position. They are active in alternating between monophthongs and diphthongs when they are the only vowel in a syllable, whereas they are inert when they occur with another nonhigh vowel within a syllable. Third, low vowels behave differently with respect to whether a coda consonant is present or absent. They alternate along the tense/lax dimension when they are followed by either a high vowel or a coda consonant. On the other hand, they raise to mid when they are followed by both a high vowel and a coda consonant. These two types of asymmetries, i.e., the asymmetrical behavior of high vowels with respect to syllable position and the asymmetrical behavior of low vowels with respect to presence/absence of coda, suggest that tone is not the only factor that affects vowel distributions and alternations. The prosodic structure plays an important role in triggering vowel distributions and alternations. Therefore, I conclude that the nature of tone-vowel interaction is indirect. That is, neither tone nor vowel affect each other directly. Their apparent interaction lies in the prosodic anchor that mediates between them. The phonetic correlation between intrinsic fundamental frequency and vowel height cannot furnish any explanation for these asymmetries These new findings demand an explanation in phonological theory which will be built up in the rest of this dissertation.  71  CHAPTER 3  Correlation between Tonal Contour and Syllable Weight  3.0 Introduction  In chapter two, I identified two kinds of direct relations. One is the correlation between tonal contour and syllable weight. The other is the influence of syllable structure on vowel features These new findings raise a number of questions, (i) Why is it possible for tonal contour to correlate with syllable weight? (ii) Why is it the syllable structure (but not tone) that affects vowel features directly? (iii) How are the relation between tonal contour and syllable weight, and the relation between syllable structure and vowel features regulated? The theory developed in this chapter aims to answer these questions. First, I propose a hypothesis which attributes the indirect nature of tone-vowel  interaction to a  representation in which tone and vowel are mediated by the mora Second, I examine the dual nature of the mora (i.e. being both a weight unit and tone-bearing unit), and argue that the mora is the only valid prosodic anchor capable of capturing the relations identified in chapter 2. To regulate the linking between tones and TBUs, I make use of the notions "head mora" and "nonhead mora" which are defined in term of the nuclear mora vs. nonnuclear mora distinction first introduced by Shaw (1992, 1993), and explore their asymmetric behavior with respect to their capability of bearing tones. To account for the asymmetric behavior of L tone regarding its restriction to a non-nuclear mora in Fuqing. I propose the tonal sonority hierarchy, which distinguishes L tone from H tone based on their intrinsic sonority, and the harmonic alignment hierarchy, which encodes the tonal intrinsic sonority into syllable positions. Third, I propose a set of constraints, such as Head Binarity (which requires that a nuclear mora bears two tones), and Head Prominence 72  Chapter Three  Jiang-King, 1996  (which requires a nuclear mora to be filled by the most sonorous tone on the tonal sonority hierarchy). I then demonstrate how their interaction with the faithfulness constraints (previously known as the well-formedness conditions and "automatic association", Goldsmith 1976), can successfully govern tonal distributions, giving rise to the distinctive moraic structures for both Fuzhou and Fuqing. The typological variation between syllable structures and tonal distributions observed by Hyman (1988) can be captured in terms of different rankings df constraints. Before I proceed to develop the theory, a number of assumptions must be made explicit. First, I assume the version of the prosodic hierarchy (Zee 1988, Hayes 1989, among others) in which the mora is the lowest constituent, shown as in (1):  (1) Prosodic hierarchy  Pw  Pw = prosodic word Ft = foot  Ft a  = syllable  LI =mora  The prosodic constituents in (1) that are relevant in the present context are of three types. First, the mora serves as a prosodic anchor for both tonal and non-tonal features to link to. Second, the syllable functions as a morphological domain for association of lexically specified tones and as a prosodic constituent organizing segments (i.e. syllabification). Third, the foot and prosodic word are different domains for stress assignment, which has certain effects on tone-vowel interaction. Second, I assume the representation of feature geometry (Clements 1985b, Sagey 1986, McCarthy 1988, Odden 1991, Halle and Stevens 1991, Halle 1995, etc.) in that 73  Chapter Three  Jiang-King, 1996  segmental features are organized into classes, such as supra-laryngeal, laryngeal, etc., which in turn are dominated by a segmental root, as shown in (2):  (2) Feature geometry  /  O  Root Laryngeal Supralaryngeal  Cj  Place  o  LO  In the representation above, the root node serves as an organizational device that groups segmental features together. Third, I assume the tonal features [+upper] and [-raised] proposed first by Yip 1  (1980) and later modified by Pulleyblank (1986) However, the use of these features in the present context is different from their use in Yip (1980) and Pulleyblank (1986). First, in Yip's system, both of these features are binary so that a total of four tonal features present in Yip's system (i.e. [+upper], [-upper], [+raised] and [-raised]). In contrast, I assume these features to be monovalent so that only a total of two tonal features is present in the tonal systems being discussed. Second, in Yip's system, there is a dominance relation between [+/-upper] and [+/-raised]. In particular, the feature [upper] divides the entire pitch range into two registers, which in turn are each divided into two by the feature [raised]. The relationship between these tonal features in Yip's system is represented in (3) :  1  The tonal features originally proposed by Yip (1980) are [+/-upper] and [+/-high]. The feature [+/-high]  later is renamed as [+/-raised] by Pulleyblank (1986) in order (i) to avoid confusion between a High tone •(H), which could be [+upper, -high] and a Mid tone (M) which could be [-upper, ±hjgh], and (ii) to distinguish the tonal feature H from the segmental feature [high]. 74  Chapter Three  Jiang-King, 1996  (3) Tonal features proposed by Yip (1980:45), modified by Pulleyblank (1986:125)  + upper  -upper  + raised  H  -raised  HM  + raised  M  -raised  L  As shown in (3), the feature [upper] combined with the feature [raised] gives rise to four level tones. In this chapter, I follow A & P's (1994) combinatorial specification theory and assume that features (either tonal or segmental) are combined to represent tones or segments. The combination of the two features  [+UPPER]  and [-RAISED] (i.e; the non-  default values of Pulleyblank (1986)), therefore, represents four level tones in (4):  (4) Tonal representations for four level tones a. b.  F-elements: M  H 1  [+UPR], [-RSD] L  M 2  +UPR  +UPR -RSD -RSD  (4a) gives two active F-elements [+UPR] and [-RSD]. (4b) shows that the combination of these F-elements gives rise to four level tones. For the languages investigated in this dissertation, three tone levels are sufficient . That is, either M or 2  x  2  could be redundant.  Although languages that havefivecontrastive level tones are not examined in this thesis, the tonal  theory assumed here is capable of representing five level tones by invoking an additional feature, such as the feature [EXTREME] proposed by Maddieson (1971). Thus, combinations of the three features [+UPR] and [-RSD] plus [EXTREME] giverisetofivelevel tones: H, L, M, Extra-H and Extra-L, as in (i).  75  Chapter Three  Jiang-King, 1996  As for contour tones, I assume that they are sequences of level tones (Woo 1969, Goldsmith 1976, Duanmu 1990, 1994, among others), resulting from interactions of constraints on the linking between tones and tone-bearing units (see detailed discussion later in this chapter) The contour tones are represented as in (5).  (i) Tone feature representations a.  F-elements: +UPPER, -RAISED, EXTREME  b.  M  1  H  L  *  +UPR -RSD  c.  EX-H EX-L M +UPR  -RSD EXTR EXTR EXTR  ?  »  +TJPR +UPR -RSD -RSD EXTR  Feature parasitic condition The presence of [EXTREME] depends on the presence of both [+UPR] and [-RSD].  The five tone representations proposed above differs from the three tone representations assumed in (4) in two regards. First, there is an addition of the feature (EXTREME] in (i), which reflects Maddieson's (1978) insight of "tone-space expansion" idea. That is, when tonal levels in a tone language increase; the entire tone space is expanded. Second, there is an additional condition (ic). This condition expresses the parasitic relation between the feature [EXTREME] and the features [+UPR] or [-RSD]. That is, the presence of [EXTREME] depends on the presence of both [+UPR] and [-RSD]. Thefivelevel tones defined by the combination of these three F-elements can be represented as in (ii), where the M tone could be unspecified for any features as in (iic) (in which the circle stands for a tonal root node) or could be specified for both [+UPR] and [-RSD]  (ii) Representations of five level tones a. H tone " j rtJ  b. Ltone  c. Mtone  o  e. Extra H +UPR EXTREME  R  o  o  ! /  76  f Extra L - R S D  EXTREME  {/  Chapter Three  Jiang-King, 1996  (5) Representations of contour tones a HL sequence  b H M sequence  +UPR -RSD  Q  V  c. M H sequence +UPR  +UPR  V  V  d. L H sequence -RSD +UPR  V  Notice that the representations of contour tones in (5) differ from those in Yip (1989) and Jiang-King (1994a, b, 1995a) in that a contour tone, as in Hyman and Pulleyblank (1988), involves two tonal roots in (5) but a single tonal root in Yip (1989) and Jiang-King (1994a, b, 1995a).  3.1 The prosodic anchor hypothesis  The prosodic anchor hypothesis proposed here attempts to answer the questions as to why and how tonal contours correlate with syllable weight, as well as why syllable structures have such a direct influence on vowel features. It is stated in (6):  (6) Prosodic anchor hypothesis of tone-vowel interaction  a.  Representational Requirement Both Tonal Root and Vocalic (segmental) Root must directly link to the lowest prosodic anchor on the prosodic hierarchy, that is, the mora.  b.  Constraint Satisfaction Optimal linking between the prosodic anchor and tone or vowel is determined by a set of universal output constraints.  77  Chapter Three  Jiang-King, 1996  (6) imposes two conditions on tone-vowel interaction. The condition (6a) states that tone and vowel must be represented in a particular configuration in order for them to interact. That is, they must be associated to the same and the lowest prosodic constituent on the prosodic hierarchy: the mora. This amounts to saying that the mora has a direct relationship with tone and vowel independently. This direct relationship between the mora and tone or vowel is represented in (7d) but not in (7a), (7b) and (7c) Thus, (7d) is the only representation that satisfies the condition (6a).  d. A / T V  O = segmental root T = tonal feature V = vowel feature fi = mora o = syllable  In (7a) both the tone and the vowel link to the same node. However, this node is not a prosodic constituent, it is a segmental root. In this representation, the segmental root node mediates between the mora and tone or vowel, hence it does not satisfy the condition (6a). In (7b) both tone and vowel link to a prosodic anchor, but they each link to a different anchor: the tone links to the syllable node, and the vowel to the mora. Thus, (7b) does not satisfy the condition either. In (7c) both tone and vowel link to the same prosodic anchor, but the anchor is not the lowest constituent on the prosodic hierarchy. That is, the prosodic anchor they link to is the syllable but not the mora. Only (7d) satisfies the condition in (6a) since the prosodic anchor that both tone and vowel associate to is the mora, the lowest prosodic constituent in the prosodic hierarchy. Thus, it is the only representation in which tonal contours may correlate with syllable weight (i e; the tight/loose distinction between syllable types and tonal groups), in which syllable structure  78  Chapter Three  Jiang-King,. 1996  may directly influence vowel features, and in which the indirect relation between tone and vowel may be captured. The condition in (6b) states that the linking of tone or vowel features to the prosodic anchor must be regulated by well-formedness constraints. In other words, the two kinds of relations (i.e., the correlation between tonal contour and syllable weight and the relation between the syllable structure and vowel features) are subject to certain regulations. The basic constraints governing tonal distributions are those previously known as the wellformedness conditions (WFC) proposed in autosegmental phonology (Goldsmith 1976), re-introduced here in (8) for convenience.  (8) The well-formedness conditions (WFC) (Goldsmith 1976)  a.  Each tone must be associated with at least one TBU.  (Parse)  b.  Each TBU must be associated with at least one tone.  (Fill)  c.  No association lines may cross.  (Linearity)  In the framework of Optimality Theory, the condition in (8a) can be reinterpreted as a PARSE  constraint which requires every tone to be parsed onto a prosodic anchor. The  condition (8b) can be reinterpreted as a FILL constraint which requires all prosodic anchors to be filled by a tone. The condition (8c) can be captured by LINEARITY (9c) (A & P 1994, McCarthy 1995, M & P 1995), which prevents association lines from crossing. In other words, the functions of WFC can be incorporated into Optimality Theory as a set of faithfulness constraints (M & P 1993a, b, 1994, 1995, P & S 1993, Pulleyblank 1994, 3  3  Thefaithfulnessfamily of constraints have been incorporated into correspondence theory (McCarthy  1995,  M&P  1995).  The constraints PARSE and FILL are replaced by MAX and  DEP,  respectively, in  correspondence theory. However, the two sets of terminology are not quite identical as far as the treatment  79  Chapter Three  Jiang-King, 1996  McCarthy 1995, among others). In addition, the one-to-one association between tones and TBUs proposed in autosegmental phonology (Clements and Ford 1979, A & P 1994) can be interpreted as a faithfulness constraint  UNIFORMITY  (McCarthy 1995, M & P, 1995). To  prevent the insertion of tones that are not lexically specified for a morpheme, another faithfulness constraint  LEXTONE,  which is an extension of the LEx-ct (Pulleyblank 1994,  Pulleyblank & Turkel 1995, Pulleyblank et al 1995) must be introduced. The basic faithfulness constraints governing tonal distributions are listed in (9), and their effects on tonal distributions are represented in (10) below:  (9) Faithfulness constraints on tonal distributions  a.  PARSETONE : A  tone must be incorporated into a prosodic structure.  b.  Fnx-p: A mora must be filled by a tone.  c.  LINEARITY:  d.  UNIFORMITY: N O element  e.  LEXTONE:  String, reflects the precedence structure of String , and vice versa. 2  of String has multiple correspondents in String,. 2  A tone that is present in an output must be present in an input.  The constraint  PARSETONE  (9a) rules out the representation in (lOa), in which the  tones are not parsed onto the prosodic anchors (i.e. moras). The FILL constraint (9b) rules out the representation in (10b), since one of the two moras is not filled by a tone. (10c) violates the constraint LINEARITY (9c) because two association lines cross each other. (lOd) and (lOe) violate the constraint  UNIFORMITY since there is  a multiple linking between tones  of floating tones is concerned. For example, afloatingtone satisfies MAX, but not PARSE when it is present in an output, since it is not considered to be associated to any prosodic anchor. Since the effect of floating tones are not of concern in this thesis, I leave the question of how the floating tones should be treated in an OT framework open for further research.  80  Chapter Three  Jiang-King, 1996  or TBUs in each case: two tones link to a single mora in (lOd) and two moras are linked by a single tone in (10e) . (lOf). violates (9e) since the circled tone is not specified in an 4  input.  (10)  The effects of faithfulness on tonal associations a.  b.  |i-  u  c.  u  \x  d.  u  u  e.  a  f.  u  u  One may assume that the constraints governing the linking of tones to the mora may differ from the ones governing the linking of vowels to the mora, since tone and vowel are represented on different planes. What this implies is that these two kinds of linking are independent of each other. Tones and vowels do not interact directly. They are related to each other only when the prosodic anchor that bridges them is affected. This implication will be borne out once the stress effect on tone-vowel interaction is examined (I will discuss this issue in detail in Chapter 5).  4  Both (lOd) and (lOe) may be surface-true in some languages due to higher ranked constraints, such as  PARSETN  or FILL - . In particular, ranking 1 1  PARSETN  above UNIFORMITY would give the representation in  (lOd), where tones must be parsed onto a mora even though the number of tones exceeds the number of moras, whereas ranking FILL!- above UNIFORMITY would give the representation in (lOe), where all moras 1  must befilledby a tone even though the number of moras exceeds the number of tones. 81  Chapter Three  Jiang-King, 1996  3.2 Mora: the prosodic anchor for tone-vowel interaction  The prosodic anchor hypothesis in (6) requires that both tone and vowel link directly to the mora. This allows them to interact through the mora. The question that arises is why the mora is the only eligible candidate rather than other prosodic constituents like the syllable, since both the mora and the syllable have been argued (either separately or together) to be possible tone-bearing units in the phonological literature (Hyman 1985, 1988, Hyman and Pulleyblank 1988, A & P 1989, Pulleyblank 1986, Peng 1992, among others) . To test this hypothesis, I examine various possible representations and see how 5  the correlation between tonal contour and syllable weight found in both Fuzhou and Fuqing can be captured. Three functions of mora have been proposed in the literature (Hyman 1985, Zee 1988, Hayes 1989, among others). First, it serves as a measurement for the number of segments. For example, Hyman (1985) assumes that every segment inherently comes with a weight unit (a mora or an x slot). The weight unit associated with an onset or a coda consonant will be removed by an "onset creation rule" and a "margin creation rule" respectively. Second, mora serves as a core to which segments and autosegments link (Hyman 1985). Third, it serves as a base for syllable structure to build upon (i.e. a subsyllabic constituent, the lowest prosodic constituent on the prosodic hierarchy) (Zee 1988, Hayes 1989, among others). What is relevant for our purpose here is that the mora serves as a prosodic anchor for both tone and vowel to link to. This requirement, stated in the prosodic hypothesis (6a), is rooted in Hyman's (1985) theory of phonological weight. Hyman takes the strong position of assuming that there is an identity relation between the units that contribute syllable weight (i.e. WUs) and those that bear tones (i.e. TBUs). To  5  The condition (6a) does not exclude the possibility that the syllable is a TBU. What this condition really  means is that the mora is the sole prosodic anchor for the phenomenon of tone-vowel interaction. 82  Chapter Three  Jiang-King, 1996  capture this non-arbitrary relation between WUs and TBUs, he proposes a weight tier which consists solely of weight units (i.e., moras or x's). To support Hyman's assumption that WUs are identical to TBUs, one needs to find evidence exhibiting a correlation between tones and syllable weight. Hyman (1985), however, does not provide such evidence. In this chapter, I furnish evidence to show that such a correlation between tonal contours and syllable weight indeed exists in Northern Min languages (i.e. Fuzhou and Fuqing). as well as in various Southeast Asian languages, such as in Sre. a Mon-Khmer language (Manley 1972), Hu, another Mon-Khmer language (Svantesson 1989), Thai dialects (Gandour 1977). Siamese and Red Tai IGedney 1965, 1989). The seven tones in Fuzhou and Fuqing. for instance, are divided into two groups: one group of tones occurs only in light syllables and the other group of tones in heavy syllables, as shown in (11):  (IT) Tones in Fuzhou and Fuqing  I. Tones in light syllables  II. Tones in heavy syllables  ±3* Fuzhou Fuqing  Yin Ping 44 H 53 HM  Yang Ping 53 HM 44 H  YangRu 5? H 5? H  Shang 31 ML 33 M  ra* K * YangQu 242 MHM 41 HL  YinQu 213 MLM 21 ML  mx YinRu 23? MLM 22? ML  To see how these cooccurrence restrictions on tonal contour and syllable weight can be explained by the prosodic anchor hypothesis I examine the representation requirement in (6a) and discuss its consequences in tone-vowel interaction in the two Northern Min languages. First, let's assume that both tone and vowel link to a segmental root node as in (7a), re-introduced here as (12) for convenience:  83  Chapter Three  Jiang-King, 1996  (12) Segmental root as TBU  In the representation above, tone and vowel are dominated by the same segmental root node. This implies that every vocalic segment is a potential tone-bearing unit no matter which syllable position it occupies. If this were true, we would expect a high vowel to behave identically with respect to the same tonal environment, no matter whether it appeared as a nucleus of a syllable or as a glide preceding or following a nuclear vowel. This prediction, however, is not borne out As illustrated in (13), a high vowel behaves differently with respect to its relative syllable positions. It surfaces as a monophthongal high vowel in a tight syllable (which has been identified as a light syllable) and as a diphthong containing that high vowel in a loose syllable (which has been identified as a heavy syllable) (see chapter 2 section 5 for detailed arguments).  (13)  I "tight"  Gloss  a.  tsi^  'only'  c.  tsieq"  'stick'  H "loose"  Gloss  Distribution  b  tsei ^  'will'  i ~ ei  d.  tsien^  'fight'  1  In (13a), the high vowel [i] is a nucleus in a syllable with a M L tonal contour. It appears as part of a diphthong in the corresponding loose syllable with a complex tonal contour M L M in (13b). However, the high vowel [i] in (13c-d) does not behave in this manner. Instead, the vocalic difference takes place between a tense mid vowel [e] and its lax counterpart [e]. What makes the high vowel behave differently in (13a-b) and in (13c-d) is  84  Chapter Three  Jiang-King, 1996  its different syllabic position. In particular, the high vowels are part of the nucleus in (Daft), whereas they are pre-nuclear glides in (13c-d). If the segmental root were to dominate both tone and vowel features, the asymmetric behavior of the high vowel [i] between the pair of (13a-b) and the pair of (13c-d) is left unexplained. Therefore, tone and vowel cannot be dominated by the same segmental root in the kind of tone-vowel interactions found in Fuzhou and Fuqing. and (12) is not a valid representation for tone and vowel to interact in these languages. Second, let's examine the representation in (7b), repeated as (14) for convenience. In (14), tone and vowel link to different prosodic anchors: the syllable is the anchor for tone, while the mora serves this function for vowel features.  (14) Mora and syllable serve as different anchors  T V  In this representation, syllable weight should not have any effect on tonal contour since tone links to the syllable node, and syllable weight is represented by the number of moras. Under this representation, any type of syllable (either light or heavy) should occur with any kind of tone (level, simple contour or complex contour). In Northern Min languages, however, this is not the case. We see from the data in (11) that the light syllables only have a level or a simple contour tone in both Fuzhou and Fuqing. whereas the heavy syllables only have complex contour tones in Fuzhou and the tones containing a L part in Fuqing. This kind of cooccurrence restriction between tonal contours and syllable types cannot be explained under the representation in (14). Therefore, (14) should be excluded as a possible representation in which tone and vowel interact in the languages investigated.  85  Chapter Three  Jiang-King, 1996  Third, I examine the representation in (7c) where both tone and vowel link to the syllable node, shown as in (15) below for convenience.  (15) Syllable as an anchor for both tone and vowel T o V  Although the syllable is assumed to be a possible tone-bearing unit by some phonologists (Yip 1980, 1989, Hyman 1988, Peng 1992, among others), it is not a possible prosodic anchor for vocalic segments . Treating the syllable as an anchor for 6  vocalic segments amounts to saying that there is no subsyllabic constituent within a syllable as far as features are concerned. Phonologists generally agree that syllables must have some sort of internal structure. One view is that a syllable contains one or more moras as its subsyllabic constituents (Hyman 1985, 1988, Zee 1988, among others). Another view is that a syllable has onset, nucleus, and coda as its subsyllabic constituents  6  Doug PuUeyblank (p.c.) points out that the syllable node could be a valid anchor for onsets and  morale codas in the following configuration:  RT  RT RT  The interesting point relevant to the present context is that the leftmost and the rightmost segmental roots in the configuration above could be vocalic segments, such as glides. However, as far as tone-vowel interaction in Fuzhou and Fuqing is concerned, the pre- or post-nucleus high vowels do not show any tonal effect (see chapter 2 and later in this chapter for detailed observations). 86  Chapter Three  Jiang-King, 1996  (Pike & Pike 1947, David 1988). A third view is that a syllable contains onset and rime, which in turn contains nucleus and coda (Levin 1985, Steriade 1988, Bao 1990, among others). To explain the correlation between tonal contours and syllable weight observed in Fuzhou and Fuqing under the representation in (15), one has to stipulate that a level or a simple contour tone can only occur with a syllable having a certain number of vocalic segments. However, even this is not correct, since syllable weight does not necessarily correspond to the number of segments (Hyman 1985). As shown in (13), both the tight and the loose syllables could contain from one to three vocalic segments. If we look at the segmental root alone, there is no way we can distinguish the tight syllables from the loose ones. Also, it cannot explain why the level or simple contour tones occur in light syllables and the complex contour tones in the heavy syllables in Fuzhou. Now, the only choice left is (7d), re-introduced here as (16), in which both tone and vowel link to the mora, the lowest prosodic anchor on the hierarchy.  (16) Mora as the anchor for both tone and vowel T  I  •n  I  v  In this representation, the mora serves as a core for both segment and tone. Since syllable weight is represented by the number of moras, and moras are also tone-bearing units, it is possible in this representation for tonal contours to correlate with syllable weight. This claim is plausible for Fuzhou. Recall that high vowels alternate in the fashion of single segments versus diphthongs only when they occur in nuclear position, as in (17a ~ b). If they are in non-nuclear position, they do not alternate, as in (17c ~ d), where the high vowel is an on-glide, not a nucleus.  87  Jiang-King, 1996  Chapter Three  (17)  I "tight"  Gloss  a.  tsi  'only'  c.  tsierj  'stick'  ML  H  H "loose"  Gloss  Distribution  b.  tsei^ ^  'will'  i ~ ei  d.  tsierj^M  'fight'  11  Since the light-heavy distinction is represented by the number of weight units (i.e. one mora in light syllables and two moras in heavy ones), the occurrence of the simple contour tones in the light syllables and the complex contour tones in the heavy ones is expected, given that the weight unit is identical to the tone-bearing unit. Thus, the seven tones in Fuzhou can be represented as in (18) below . I treat /53/ and /5/ as having the same 7  8  underlying tone HM, because they behave identically with respect to the tone sandhi effect. For the same reason, I treat 7213/ and 723/ as a single underlying tone MLM. Thus, the 7 citation tones in Fuzhou are reduced to 5 underlying tones.  (18) Fuzhou tones and syllable weight a. Tones in the "tight" syllables  b. Tones in the "loose" syllables  Tl:/44/  T4:/213/,723/  T2: 1531,151 T3:/31/  A [  H  7  H  f  3  °  M  . A M  L  l\ M L M  T5:/242/ [ V- 1*-]  A  B  M H M  Ignore for the moment whether both of the moras in the loose syllables have the same capacity to bear  tones. 8  The tones 151 and /23/ are called checked tones. They occur only in syllables with a glottal stop coda. No  discussion of the checked syllables is planned for this thesis. 88  Chapter Three  Jiang-King, 1996  (18) captures the cooccurrence restrictions that hold between two groups of tones and two types of syllables structurally. All tones in the tight syllables link to one mora, even though they differ in pitch level (i.e. H level in Tone 1, H M in Tone 2 and M L in Tone 3), whereas the tones in the loose syllables Unk to two moras, even though their shapes are divergent (i.e. concave M L M vs. convex MHM). Assuming for the moment that one prosodic anchor can bear a maximum of two tones, the quantitative distinction for the two groups of tones in Fuzhou will giveriseto distinct morale structures for the two types of syllables in (19):  (19) Distinctive syllable weight in Fuzhou a. light syllables  T (T)  b. heavy syllables  M  T  TI  (19) suggests that the distinctive morale structures in Fuzhou are determined by the tonal specifications for each morpheme. That is, the specification of a H level or a simple falling tone (either H M or ML) givesriseto a monomoraic structure for the tight syllables, while the specification of a concave tone M L M or a convex tone M H M results in a bimoraic structure for the loose ones. I will demonstrate in chapters 4 and 5 that it is the different moraic structures that directly trigger the vowel distributions and alternations.  89  Chapter Three  Jiang^King, 1996  3.3 Head mora vs. nonhead mora  In last section, I argued that the correlation between tonal contour and syllable weight in the Northern Min languages can be best explained if the mora is treated as a prosodic anchor for both tone and vowel. This kind of tone-vowel interaction in these languages thus furnishes empirical support for Hyman's hypothesis that the mora is an entity for both weight and tone. This argument raises two questions. First, why don't the complex contour tones (i.e. concave and convex tones) in Fuzhou give rise to three moras in the loose syllables, given the one-to-one association convention between tones and TBUs? In other words, why is the correlation between tone and syllable weight in Fuzhou a two-way contrast but not a three-way contrast as in (20)?  (20) Hypothesized three-way contrast of syllable weight in Fuzhou a. monomoraic  b. bimoraic  H  * HM vs.  |  * ML  | |  te  l^a  c. trimoraic * HMH  Ill Ml  vs.'  | |  l"o ^  u  [  * H LH  te  u uJ  a  [u u J u  c  Second, why are there no extra-complex tonal contours (such as H M H M or L M L M , as shown in (21b))- in Fuzhou (maybe in any language in general), assuming for the moment that a violation of the one-to-one association convention is allowed?  (21) Hypothesized tonal contours in Fuzhou a. monomoraic syllables H  I  tel  a  H M  M L  tel  m  V  CT  V a  b. bimoraic syllables  -  * M H M H  * M L M L  V V  [ii 90  ^ l  VV  a  In  Chapter Three  Jiang-King, 1996  There are two possibilities to rule out the unattested three-way distinctive syllable weight in (20). One possibility is to treat a simple contour tone (i.e. HM, ML) in Fuzhou as involving a single tonal roots and the complex contour tones (i.e. M L M and MHM) as involving two tonal roots, as shown in (22) proposed by Jiang-King (1994a, b, 1995)  (22) Fuzhou correlation between tones and morale structures (Jiang-King 1994a, b, 1995a)  a. Tones in the "tight" sylla>les  b. Tones in the "loose" syllables  T2: /53/, 151 T3:/31/  Tl:/44/  T4:/213/,723/  T5: /242/  U  u-  . Tonal'Rt^Ja^  ^  H Tonal Rt  ^  Tonal  H  Rt^,L^ H  Tonal  M  Rt^L^ M  L  Tonal  ^  RIJD^ M  L  M M  H  M  However, treating a complex contour tone as involving two tonal roots as in (22b) without distinguishing their two TBUs cannot explain why the extra-complex tonal contours in (21b) are unattested in Fuzhou. or more generally, in any natural language. The other possibility is to impose restrictions on linking between tones and TBUs. That is, regulations are imposed on the relation between tones and moras. This is the approach taken by Hyman (1988). Hyman proposes that there are three possibilities for representing tonal association to moras in a tritonal syllable, as in (23 a-c) and (24a-c) Hyman uses [a]'s as an abbreviation for moras in order to show that these TBUs belong to the same syllable. (23)  a. [pa  a 1  K I HLH  b. *[pa  a 1  1/1  H L H  91  c *[pa  a1  N/l HL H  Chapter Three  (24)  Jiang-King, 1996  a. [pa  a 1  K I L HL  *[pa  I  a 1 A  LHL  [pa  a L  K/l L HL  However, not all three representations in (23a-c) and (24a-c) are well-formed. Hyman assumes that only (23 a) and (24a) are well-formed and (23b-c) and (24b-c) are ill-formed. To capture the difference between the (a)'s and the (b-c)'s in (23) and (24), Hyman (1988) proposes that linking of two tones to an intrasyllabic TBU is permitted only to the first (.= head?) mora of a syllable. The insight of Hymari's proposal is that it distinguishes two types of moras, head and nonhead with respect to their toner-bearing behavior. In particular, only head moras are allowed to bear two tones. Hyman also admits certain situations where either the number of tones or the number of TBUs exceeds the other. In such cases, a nonhead mora may bear two tones. What is relevant to the present context is the distinction between a head mora and a nonhead mora. I will pursue Hyman's approach, namely, the approach in which restrictions are imposed on the linking between tones and TBUs, and propose a set of constraints governing the relation between tones and moras in the light of the nuclear/non-nuclear distinction of moras (Shaw 1992a, b, 1993, 1996). The theory of the nuclear moraic model proposed by Shaw (1992a, b, 1993, 1995) distinguishes two types of moras: nuclear moras and non-nuclear moras. These two types of mora differ in their behavior in templatic reduplications. In certain languages, only nuclear moras but not non-nuclear moras are referred to in reduplication templates. Since moras are argued to be the TBUs in the tone-vowel interactions in Fuzhou and Fuqing. it is natural to ask whether nuclear moras behave differently from non-nuclear moras regarding their tone-bearing capability. The following proposal suggests that tone-bearing moras differ with respect to whether or not they are a syllable head. In particular, a nonhead mora is more restricted than a head mora in what kind of tone and how many tones it may bear. Two possibilities regarding how to define a head mora suggest  92  Chapter Three  Jiang-King, 1996  themselves. One is to define a head mora in terms of "direction". That is, a syllable head can be designated to either the left or the right mora in that syllable. The other possibility is to define a syllable head in terms of "representation". Namely, the mora dominated by the nuclear node is the syllable head, whereas the one not dominated by the nuclear node is nonhead. Since Fuzhou and Fuqing data do not suggest the need of the two sets of notions (i.e., the nuclear vs. non-nuclear and the head vs. nonhead), thus the distinction of head mora vs. nonhead moras can be defined in terms of nuclear vs. non-nuclear mora.  3.3.1 Head Binarity: Fuzhou tonal distributions and prosodic structures  Fuzhou exhibits a cooccurrence restriction on tonal contours and syllable types. In particular, one group of tones (H, HM, and ML) occurs only in the tight syllables, while another group of tones (i.e. M L M or MHM) only with the loose syllables. The tight-loose distinction is identified as the light-heavy distinction in chapter two. The seven contrastive tones in Fuzhou and the syllable types they occur with are given in (25) below:  (25) Fuzhou cooccurrence between tones and syllable types (data are from Liang 1982) I. Tones in light syllables  IL Tones in heavy syllables  Yin Ping  !W Yang Ping  YangRu  Shang  YangQu  m^ Yin Qu  mx YinRu  44 H  53 H M  5?H  31 M L  242 M H M  213 M L M  23? M L M  tsan  tsan  tsa?  tsan^  tsarjM™  tsan^M  tsa?  'mixed'  'chop; cut'  'stand'  'raise'  'tight; bind'  H  'hairpin'  IIM  'incomplete'  H  M  (25) clearly shows a restriction on the cooccurrence between groups of tones and types of syllables. The tones in the light syllables are simpler than the ones in the heavy syllables.  93  Chapter Three  Jiang-King, 1996  To capture this quantitative distinction of tones in the two types of syllables, I propose a constraint that requires a head mora in a syllable to bear two tones.  (26) Head Binarity (HDBIN) A mora must bear two tones x and y, iff it is a syllable head (i.e. a nuclear mora).  (26) is a constraint on the relation between tones and moras. It demands that the head mora within a syllable bear two tones. Non-head moras do not have this capability. Consequently, the unattested complex tonal contours (HLHL or LHLH) are excluded. The tableaux in (27) and (28) demonstrate how the interaction of HDBIN with faithfulness in (9) gives rise to the 'tight-loose' distinction of syllable weight (i.e. [[|i]] vs. [[n]|x] ) 0  m  a  Fuzhou. The square brackets without a subscribed "o" symbol in the output candidates indicate head moras. Shading indicates that constraints are not crucial in determining an optimal output. The tableau in (27) contains three different tonal specifications in the tight syllables. The input in (27i) is a high level tone (H). Each of the three output candidates violates a constraint. (27i-a) violates HDBIN, (27i-b) violates LEXTN, and (27i-c) violates the PARSETN.  Since PARSETN and LEXTN rank above HDBIN and UNIFORMITY, (27i-a) is the  optimal output for a H-toned syllable. The cases in (27ii) and (27iii) are similar in that the input forms are all falling tones. The difference between them is that the former is a high falling tone H M while the latter is a low falling contour ML. There are four output candidates in each of the two cases Compare the (a)'s and the (b)'s in (27ii) and (27iii): the former violate UNIFORMITY once (i.e. the moras in (27ii-a) and (27iii-a) bear two tones), while the latter violate the same constraint twice (i.e. the head moras in both (27ii-b) and (27iii-b) bear two tones on the one hand and the second tone in each case links to two moras). In Optimality Theory, if competing candidates violate the same constraint, the one incurring fewer violations wins. 94  Chapter Three  Jiang-King, 1996  Therefore, the (a)'s in (27ii) and (27iii) are better than the (b)'s. Now let's compare the (c)'s and the (d)'s in (27ii) and (27iii). The (c)'s in these cases violate HDBIN since the head mora in each case bears only one tone, while the (d)'s all violate  PARSETN  because one tone  in each of these cases fails to be parsed onto the prosodic anchor. Since PARSETN and LEXTN »  dominate HDBIN, which in turn dominates  HDBIN »  UNIFORMITY,  the ranking PARSETN,  LEXTN  determines (a)*s over (b)'s, (c)'s and (d)'s for (27ii) and (27iii).  UNIFORMITY  (27) Fuzhou tone distributions in the "tight" syllables Input  Outputs a.  PARSETN  LEXTN  -  0) b.  L  A  /44/H  rm c.  •  *  *!  H  a.  *  *!  M H  M  *  M b.  (H)  H  M  \A  C  -  11  *!  nxi a.  M  c^-  /31/ML  b. -  •  c  -  '*!'  [ul -u H M  D  (iii)  * i*  Me H M  »  /53/HM  L  LA  *  nil  M  L  At Iul.U M L  * I*  11  *!  Tul u D  UNIFORMITY  *  Ful H  HDBIN  M L  M  *  *!  95  Chapter Three  Jiang-King, 1996  Notice that this ranking ensures that all optimal outputs in (27) have a single mora, hence, the light syllable: Now let's turn to the tones in the loose syllables in (28).  (28) Fuzhou tone distributions in the "loose" syllables Input  Outputs „  LEXTN  HoBtN  UNIFORMITY  MLM  V W  V 1 M u  r.  (i) A  7213/ ~  f „  *  *  MLM  \ l  1  MLM  PARSETN  *  Tul u MLM III : rm u >i MLM  *  *  rm  *  **  *  * * |*  *  **  MLM  ful  u  MLML  V  V  *!  rm u  MLM  i \ ; •rm u  *  *.!  MHM  A  *  rm u •J,  MHM 1  (ii)  -  A  /242/ MHM  *  MHM  f. C  \ l  ' 1 1 1  *  ruin ii  MHM  **  lui A  MHM  * * \* f  MHM H  rm u „  MHM  -  1 \ rm u  §  '*'!  * *  *!  96  ** -  Chapter Three  Jiang-King, 1996  The two cases in (28) are of the same type: they both involve three tones in their lexical specifications. The difference between them is the tonal shape. In particular, the tonal contour in (28i) is a concave tone ( M L M ) while the one in (28ii) is a convex one (MHM). Again, both (a) candidates in these cases violate UNIFORMITY, both (b), (c), (d) and (e) violate HDBIN, both (f) violate  LEXTN,  and all (g) violate  PARSETN.  Since UNIFORMITY  ranks below HDBIN which in turn is below LEXTN and PARSETN, the (a) are the best outputs for each case. Notice that the ranking that chooses (a) over the rest in (28) is identical to the ranking in (27). However, the syllable structures in (27) and (28) are different. The former is monomoraic since the tones are lexically specified as simple contours, while the latter is bimoraic since their tones are complex contour tones . The constraint Head Binarity and 9  its interaction with the faithfulness constraints (i.e., PARSETN, LEXTN and UNIFORMITY) plays a crucial role in deriving the distinctive moraic structures, giving rise to contrastive syllable weight. Fuzhou tonal distribution, governed by the constraint interaction (i.e. the ranking PARSETN, LEXTN »  3.3.2  HDBIN »  UNIFORMITY),  determines the syllable weight in this language.  Head prominence: Fuqing tonal distributions and prosodic structures  Another case of correlation between tonal contour and syllable weight is Fuqing. The tonal system in Fuqing differs from that in Fuzhou in that the tones in the loose syllable are  9  A question raised by Doug Pulleyblank (p.c.) is how to ensure the tonal composition of attested Fuzhou  melodies, or in general, how to constrain the basic composition of various attested tonal sequences in different languages. A tentative answer is to impose constraints on tonal combinations, such as *[+upper]/[-raised], which would rule out a HL sequence or a LH sequence, resulting in a HM or a MH sequence with L being underparsed. Even though the details need to be worked out, the general approach seems promising (see the Fuqing case in this chapter).  97  Chapter Three  Jiang-King, 1996  not complex contour tones. Instead, the tonal distinction between the tight and the loose syllables involves a quality difference, namely, the L tone Fuqing tonal distributions with respect to the syllable weight are exemplified in (29) below.  (29) Fuqing Tones I. Tones in light syllables  !W  KA  II. Tones in heavy syllables  YinPing  Yang Ping  YangRu  Shang  YangQu  Yin Qu  YinRu  53 H M  44 H  5?H  33 M  41 HL  21 M L  22? M L  sin"  sin"  sin  serj" -  serf^-  sen**  #'spirit*  £ 'solid'  #f 'ante'  ^ 'kidney'  fW letter'  H 'room*  ^•hart'  M  1  1  1  (29) shows that tones in the tight syllables are either level (H or M) or high falling (HM), whereas the ones in the heavy syllables are either low falling (ML) or high falling (HL) Crucially, the tones in the light syllables do not have L, whereas the ones in the heavy syllables all contain a L. The questions, then, are why L tones occur only in the heavy syllables and why there is no L level tone? What is the difference between L tone and nonL tones? To answer these questions, I argue that the difference between L tones and nonL tones lies in the intrinsic sonority of pitch. As with segments, tones have their intrinsic sonority. High tone has higher pitch (i.e. higher fundamental frequency) than low tones. Parallel to the segmental sonority hierarchy, this intrinsic difference of pitch can be stated as a tonal sonority hierarchy in (30) , assuming that tones are represented by the features [+UPPER] and [-RAISED], as well as their combinations.  98  Chapter Three  Jiang-King, 1996  (30) Tonal Sonority Hierarchy \+UPPER] > \-RAISED\  If tones differ in their intrinsic sonority, this property of tones must manifest itself in tonal phonology. One possibility for tones to behave differently with respect to their sonority difference is that the higher a tone is on the sonority hierarchy, the more prominent it is. This assumption can be stated in terms of the tonal harmonic alignment in (31), in light of the segmental harmonic hierarchy proposed by P & S (1993).  (31) Tonal Harmonic Hierarchy NUCW[+UPPER] > NUC /[-RAISED] U  What (31) says is that linking a H tone to a nuclear mora is more harmonic than linking a L tone to a nuclear mora. This harmonic alignment effect of tones can be encoded into a constraint ranking (32) similar to that on segmental alignment in P & S (1993)  (32) Tonal Alignment Hierarchy *NUC^/[-RSD] »  *NUC"/[+UPR]  The ranking schema in (32) means that linking a L tone to a nuclear mora within a syllable is less optimal than linking a H tone to a nuclear mora. To see how these tonal sonority alignment constraints interact with the faithfulness constraints UNIFORMITY,  PARSETN, LEXTN  and  giving the syllable weight distinction in Fuqing. look at the tableaux (33) and  (34) below. The features [+UPR] and [-RSD] represent H tone and L tone respectively. The input in (33i) is a falling contour (HM); The output set contains four candidates. The violations of PARSETN (33i-c),  *NUC /[-RSD] U  and LEXTN (33i-d) are fatal, since these  are the most highly ranked constraints in this set. Compare the first two candidates (33i-a) 99  Chapter Three  Jiang-King, 1996  and (33i-b): the former violates ranks above the  UNIFORMITY,  (33i-b) is HDBIN  »  UNIFORMITY  while the latter violates  HDBIN.  Since  HDBIN  (33i-a) wins. The particular ranking that chooses (33i-a) over  UNIFORMITY.  (33)Fuqing tone distributions in light syllables Input  Outputs a  H  PARSETN  LEXTN  HDBIN  b  753/  f. C  H M  UNIFORMITY  M  *  rm  (i)  r r  *  HM  -  1  *!  rm  HL M  A  '  V 1  M u  a.  *!  *  *!  H  *  1 Tul  (ii) /447  *NUCW[-RSD]  *! H  c.  H  d  V  a.  *!  *!  "  *  M  (iii) 733/M  *  *!  M  c.  rm  *  *•!•  M  *  *!  rm 4  M  *!  *'!  *  The cases in both (33ii) and (33iii) differ from that in (33i) in that the lexically specified tones are level tones but not a H M falling contour. The difference between (33ii)  100  Chapter Three  Jiang-King, 1996  and (33iii) is that (33ii) contains a H level, while (33iii) contains a M level. The (a) in (33ii) and (33iii) is more optimal than the (b), (c) and (d), since it satisfies the highly ranked constraints  *NUC /[-RSD], PARSETN U  and LEXTN, even though it violates HDBIN, the  lowly ranked constraint. The crucial ranking established in (33) is *NUC 7[-RSD], PARSETN, u  LEXTN »  HDBIN »  UNIFORMITY.  (34) differs from (33) in that all inputs in (34) contain a L tone and the constraint *NUC /[-RSD] U  plays an important role in choosing an optimal candidate. Parsing a  L  tone  to the nuclear mora in the (d) of (34i) and (34ii) results in violation of *NUCH7[-RSD]. The (b) and (c) violate LEXTN and satisfies all  PARSETN  *NUC /[-RSD], LEXTN U  respectively. The only candidate left is the (a). It  and PARSETN, even though it violates HDBIN.  (34) Fuqing tone distributions in the "loose" syllables Input  Outputs a  H  *NUCI»/[-RSD]  PARSETN  LEXTN  HDBIN  UNIFORMITY  L *  (i)  U  '  741/HL  R  ^  d-  i •  HM L A  rm H 1  M H  L  L  IA  *!  M L  ful  (i)  U  '  ;  I2\l ML  -  °d  *  *!  M  a  *  *!  u  U  M H L V 1  mi  *  *!  i i  M L  1 ful M  *! L  LA ful.  *  *  *!  101  Chapter Three  Jiang-King, 1996  (34) shows that ranking *NUC  U  /[-RSD], PARSETN  and LEXTN above  HDBIN  is crucial in  deriving the bimoraic structures for the loose syllables in Fuqing. The entire ranking for Fuqing is  *NUC /[-RSD], PARSETN, LEXTN » U  HDBIN »  UNIFORMITY.  This constraint  interaction givesriseto the tight/loose distinction of the syllable weight in Fuqing (35):  (3 5) Fuqing tones and syllable weight a. Tones in the "tight" syllables Tl:/53/  A  T2: /44/,/5/  T3: /33/  b. Tones in the "loose" syllables T4:/21/,/22/  T5: /4.1/  •  H  M  H  M  M  L  H  L  3.3.3 A summary  I have shown that moras behave differently with respect to how many tones they can bear and what kind of tones they can bear. These differences can be captured by the nuclear moraic model proposed by Shaw (1992, 1993). First, nuclear (head) moras can bear two tones while non-nuclear moras only bear one tone, as shown in the Fuzhou case. Second, nuclear (head) moras are restricted not to bear L tone, while non-nuclear moras do not have this restriction, as shown in the Fuqing case. By defining the nuclear mora as a syllable head,  I  propose a set of constraints (i.e. HDBIN and  *NUCM/[-RSD])  that regulates  the linking between tones and prosodic anchors. I show that the interaction of these constraints with the faithfulness constraints (i.e.  PARSETN, LEXTN  and  UNIFORMITY)  successfully derives the correlation between tonal contours and syllable weight in Fuzhou and Fuqing. I also demonstrate that the reverse ranking of these constraints accounts for the different patterns of the tone-syllable structure correlation that are observed in Fuzhou and Fuqing. The rankings are given in (36) below:  102  Chapter Three  Jiang-King, 1996  (36) Typological variations on tonal distributions in Fuzhou and Fuqing  Types  Ranking  a  PARSETN, LEXTN »  b.  *NUC /[-RSD], PARSETN, LEXTN »  HDBIN »  Languages Fuzhou  UNIFORMITY, *NUC /[-RSD] U  U  HDBIN »  Fuaine  UNIFORMITY  3.4 Hyman's typology of tonal distributions  A number of theoretical attempts to impose restrictions on tonal distributions are proposed in the phonological literature (Goldsmith 1976, Hyman 1988, Bickmore 1993, among others). Among these studies, Hyman's (1988) draws attention to typological variations of tonal distributions with respect to syllable structure. He observes that tonal languages may differ in two aspects. The first is the prosodic constituents serving as tonebearing units. Some languages choose syllable head (which is defined as the nuclear mora) as TBU, while others may choose mora as TBU. The second aspect of cross-linguistic variation lies in whether a language imposes a one-to-one constraint on linking between tones and TBUs. Thus, these two parameters define four different tone systems in (37):  (37) Hyman's typology (1988:49-50) a. a  b. a  L  a  H  o  L  L  H  103  o  a  *  H  L  H  o  * a  L  L  H  Chapter Three  Jiang-King, 1996  Languages like (37a) and (37b) choose the head mora as TBU. The difference between these two types of languages is that (37a) allows a tone-bearing unit to bear two tones but (37b) does not. Languages like (37c) and (37d) allow all moras (either head or nonhead) to serve as TBU. However, (37c) permits contour tones while (37d) does not. Furthermore, a clear generalization that can be abstracted from (37) is that whereas some languages may prohibit contour tones, no language disallows level tones. If we replace Hyman's parameters (i.e. TBU = either a or | i and whether contour tones are allowed or disallowed) with the two constraints  PARSETONENUC  u  (which requires tones to be parsed  onto a head mora in a syllable) and UNIFORMITY (which disallows multiple linking between tones and TBUs) respectively, Hyman's typological variations can be redefined in terms of whether the two constraints are enforced or optional, shown as in (38):  (3 8) Recast of Hyman's typology  PARSETNNUC  Enforced  U  Optional n  Enforced  1 T  UNIFORMITY  Optional  1 T  /\ H  L  /\ H  L  * ll  1  A  T  H L  1  A  T  H L  Notice that in the four types of tonal systems defined above, none rejects level tones. This unmarked tonal pattern is nicely captured. The cross-linguistic variation discussed in Hyman (1988) lie in whether a grammar of a particular language imposes either one or two of these constraints. In the framework of Optimality Theory, both the universal  104  Chapter Three  Jiang-King, 1996  unmarkedness and the cross-linguistic variation can be achieved by constraint ranking, shown as in (39) below:  (39) Constraint ranking in deriving Hyman's typology  PARSETNNUCM-  UNIFORMITY »  HDBIN  PARSETNNUCM' » P A R S E T N  PARSETN »PARSETNNUCM'  Type A  TvpeB  M 1 T  UNIFORMITY  * lu]  V-  /\ H  L  TvpeC HDBIN »  UNIFORMITY  In] 1 T  I  A  T  H L  TypeD  M /\ H  '* P-  1  L  (39) shows that ranking the faithfulness constraint  T  UNIFORMITY  A H L  above the structural  constraint HDBIN defines the language types A and B (i.e. contour tones are prohibited), whereas the reverse ranking gives rise to the language types C and D (i.e. contour tones are allowed). On the other hand, if PARSETNNUCM dominates PARSETN, the language types A and C (i.e. only nuclear moras are allowed to bear tones) are achieved, whereas the reverse ranking of these two constraints derives the language types B and D (ie all moras are allowed to bear tones). Thus, the universal unmarkedness and language typology are achieved by different rankings of the same set of the constraints, a desirable result.  105  Chapter Three  Jiang-King, 1996  3.5 Conclusion  The prosodic anchor hypothesis proposed in this chapter to account for the correlation between tonal contours and syllable weight and the direct relation between syllable structure and vowels contains two conditions. The first condition requires tone and vowel to link to the same mora. It is this particular representation that captures the important role the mora plays in the kind of tone-vowel interaction found in Fuzhou and Fuqing. The second condition, namely, constraint satisfaction, requires that Unking of the prosodic anchor (i.e. the mora) to either tone or vowel must be well-formed That is, constraints are imposed on the relation between the mora to both tones and vowels rather than directly imposed on the relation between tonal features and segmental features The examination of the mora with respect to its function as both a weight unit and a tone-bearing unit reveals that moras are distinguished in terms of their tone-bearing behavior. Some moras are allowed to bear certain kinds of tones and some are not. Furthermore, some moras are allowed to bear more than one tone and some are not. This asymmetric behavior of moras can be best characterized as the distinction of head mora (which is defined as nuclear mora) vs. nonhead mora (which is defined as non-nuclear mora). I further propose a set of constraints governing the linking between the mora and tones and demonstrate how their interaction with the faithfulness constraints is sufficient to derive distinctive syllable weight in both Fuzhou and Fuqing. Moreover, I show how the different ranking of the same set of constraints can successfully capture the unmarked tonal type (i.e. the level tones) and the cross-linguistic variation observed by Hyman (1988). The correlation between tonal contours and syllable weight accounted for in this chapter is compatible with Duanmu's (1990, 1993) finding that rime length affects stress and tonal patterns within a polysyllabic domain. Based on his comparative study of Mandarin and Shanghai tone sandhi, Duanmu draws attention to the relation between rime 106  Chapter Three  Jiang-King, 1996  length and tonal change, and observes that Mandarin differs from Shanghai in two respects. First, the rime length in Mandarin is more complex than in Shanghai. That is, Mandarin has more diphthongs and coda consonants. Second, in Shanghai, non-initial syllables within certain domains lose their lexical tones, whereas in Mandarin non-final syllables keep their lexical tones. These findings lead Duanmu to conclude that the tonal difference between the two types o f languages lies in their different rime structures. In particular, all M-languages (i.e., Mandarin related languages) have complex rimes which carry inherent stress, therefore retain their lexical tones. O n the other hand, all Slanguages (i.e., Shanghai-related languages) have simple rimes which do not carry inherent stress, and therefore, do not retain their lexical tones. Whatever the precise status o f Duanmu's claim that Chinese languages  fall into two  classes,  M-languages  and S-  languages, this thesis supports the claim that moraic distinctions are important. It shows that different moraic structures affecting both tonal and segmental patterning play a crucial role language-internally in Northern Min  languages.  107  CHAPTER 4  Linking between Prosodic Structure and Vowel Features  4.0 Introduction  Two direct relations emerged from the investigation of Fuzhou and Fuqing in chapter 2. One is the correlation between tonal contour and syllable weight, and the other is the direct influence of syllable positions (i.e. nucleus vs. non-nucleus) on vowel features. These relations are captured in the prosodic anchor hypothesis proposed in chapter 3, repeated here for convenience in (1):  (1) Prosodic anchor hypothesis of tone-vowel interaction  a.  Representational Requirement Both tone and vowel must directly link to the lowest prosodic anchor on the prosodic hierarchy, that is, the mora.  b.  Constraint Satisfaction Optimal linking between the prosodic anchor and tone or vowel is determined by a set of universal output constraints.  The condition (la) requires tone and vowel to link to the same mora in order for them to interact. The condition (lb) requires the linking between the mora and tone or vowel to be governed by a set of well-formedness constraints. In chapter 3, Imotivated the constraints governing tonal distributions and demonstrated how their interaction gives rise to 108  Chapter Four  Jiang-King, 1996  distinctive prosodic structures for both Fuzhou and Fuqing. The first direct relation, namely, the correlation between tonal contours and syllable weight, has thus been accounted for. The task of this chapter is to deal with the second direct relation, namely, the direct influence of syllable positions on vowel distributions. I will first review current syllable theory within the optimality framework. I then propose a set of constraints on linking between the prosodic structures and vowel features in Fuzhou and Fuqing. showing that their interaction with the basic syllable structural constraints proposed by M & P (1993a, b, 1994) and P & S (1993), as well as segmental sonority constraints (P & S 1991, 1993), can successfully derive attested vowel distributional pairs.  4.1 Syllable theory in OT framework  A syllable (a) is defined as a prosodic constituent, the second lowest on the prosodic hierarchy proposed by Zee (1988), and has its own internal structure. Different theories assume different subsyllabic constituents for the syllable. The syllable theory proposed by M & P (1993a, b) and P & S (1991, 1993:6 & 8) assumes that a syllable must have a peak, presumably, Nucleus . It may also have margins: Onset as leftmost and Coda as 1  rightmost. The syllable theory of P & S assumes that the relation between inputs and outputs follows from general Optimality Theory. That is, the function Gen freely builds up any number of output syllabic structures for each given input. These output candidates must be evaluated by the function Eval with respect to a set of ranked constraints. The optimal output is the one that best satisfies highly ranked constraints. There are two types of constraints proposed in the syllable theory of P & S. One type is the structural constraints, such as Nuc (syllables must have a nucleus), ONS (syllables  1  The term "Nucleus" is usedfrequentlyto formulate the constraints even though it is not explicitly  proposed to be a formal subsyllabic constraint in M & P (1993a, b) and P & S (1991, 3).  109  Chapter Four  Jiang-King, 1996  must have an onset), -COD (syllables must not have a coda), etc.. The other type is the constraints, such as *M/a (a cannot be a margin of a syllable) and *P/a (a cannot be a syllable peak). The former governs the basic syllable structures and the latter restricts association of segments to syllable positions based on universal segmental sonority. The top-down (i.e. looking for segmental material to fill certain syllable positions) and bottomup (i.e. looking for a syllable position for a segment to be parsed onto) syllabification often brings constraints into conflict. The resolution lies in constraint ranking.  4.1.1  Basic syllable structures  A number of empirical generalizations are captured by the syllable theory. The first generalization regards the universal unmarked syllable structure. It has been observed that C V syllables are invariably allowed in all languages. The structural constraints listed below in (2) ensure this unmarked syllable type:  (2) Basic syllable structure constraints (P & S 1993)  a.  N u c : syllables must have nuclei.  b.  ONS: syllables must have onsets.  c.  -COD: syllables must not have codas.  d  ""COMPLEX:  no more than one C or V may associate to any syllable position node.  The terms "nuclei", "onset" and "coda" in the constraints above are used as cover terms for different syllable positions in P & S's theory, but do not refer to any formal subsyllabic constituents. According to the constraints above, a string of syllabified as  .CVCV.  CVCV  should always be  (the dots indicate a syllable break). Any syllabification other than  C V C V . would violate one or more of the structural constraints. For instance, to parse a 110  Chapter Four string  CVCV  Jiang-King, 1996  into three syllables like  CVCV.  would violate Nuc (in the first syllable),  ONS (in the second and third syllables), and -COD (in the second syllable). The unmarked syllable structure thus is defined by satisfying all the structural constraints listed In (2). The second generalization captured by the syllable theory in the OTframeworkis referred to as "Jakobson's typology" of syllable structures. It has been observed that the three elements of a syllable, namely, the onset, nucleus and coda, do not have equal status in terms of obligation. The nucleus is the only required element in all types of syllables, whereas the onset and coda may be optional in some languages. Furthermore, although not all languages require the onset in every syllable, no language requires the coda in every syllable. Syllable structure, therefore, varies from language to language depending on whether the onset is required or the coda is forbidden. A chart (3) showing "Jakobsonian typology" (1962) is presented below.  (3) Jakobsonian typology of syllable structures (from P & S 1993:85)  Onsets Required  Optional  £cv  I(C)V  £CV(C)  £(C)V(C)  Forbidden Codas Optional  This cross-linguistic variation can be captured quite nicely in the syllable theory in the OT framework by the possible rankings of the structural constraints with respect to faithfulness. In particular, if ONS dominates optional. Similarly, if -COD ranks above  PARSE,  PARSE,  the onset is required; otherwise, it is  the coda is forbidden; otherwise, it is  optional. The typological variation derived by the interaction of the structural constraints 111  Chapter Four  Jiang-King, 1996  with faithfulness is given in the following chart. The capital letter F denotes the faithfulness set, while the capital letter F with a subscribed "i" stands for a member in this set, namely, either PARSE or FILL.  (4)  Jakobson's typology derived by constraint ranking (P &  S 1993:86)  Onsets ONS  -CoD»F  i  »F  F»Oxs  {  lev  £(C)V  £CV(C)  £(C)V(C)  Codas F»-COD  Chart (4) shows that four types of syllable structure can be derived by the four pairs of constraint-ranking. The ranking of both ONS and -COD above faithfulness (i.e. ONS, -COD » F) gives only the most unmarked syllable structure (i.e. £ ) , which has onset but no cv  coda. Conversely, the ranking of faithfulness constraints above both ONS and -COD (i.e. F » ONS, -COD) allows the most marked syllable structure (i e. £( ) ( )), optionally allowing c  v  c  both onset and coda. In between there are two types of syllable structures, namely, S and  Z < ), CV  C  (C)V  which result from the ranking of ONS and -COD between the two faithfulness  constraints. Notice that the difference between F and F is not trivial. F denoting the entire {  set of faithfulness constraints must be respected, whereas F- standing for a member of the t  faithfulness set needs not be. In terms of ranking, F can be ranked below other {  constraints, while F cannot. Ranking F at the bottom amounts to saying that there are {  cases where lexical properties are not respected at all. However, such cases have not been found in any language so far. Moreover, what this chart demonstrates is that ranking of constraints plays a crucial role in determining different types of syllable structure.  112  Chapter Four  4.1.2  Jiang-King, 1996  Segmental sonority constraints  Another generalization encoded in the syllable theory vvithin OT is universal segmental sonority. It has been observed that the suitability of each segment to syllable positions is largely determined by the intrinsic prominence of segments, which is represented by the sonority hierarchy in (5):  (5) Segmental sonority hierarchy (Zee 1988) 2  z  a,  i,  d, t  greater sonority \  (5) indicates that low vowels are more sonorous than high vowels, which in turn are more sonorous than voiced stops, and so on. Universally, the more sonorous a segment, the better it serves as a syllable peak. Conversely, the less sonorous a segment, the better it serves as a syllable margin . These generalizations can be captured by the two sets of 3  2  Segmental sonority could also be based on features like the tonal sonority hierarchy proposed in chapter  three (see section 3 in chapter 3 for detailed arguments). 3  The term "margin" in P & S's (1993) theory refers to both onset and coda which are not distinguished.  This problem is noted but not addressed seriously in P & S (1993:162). Also, it is observed that more sonorous codas are better than less sonorous ones (Prince 1983, Zee in prep, Clements 1990). This is also true for Fuzhou. Since this study focuses on tone-vowel interaction, I leave the question of how onset and coda should be distinguished open for future research.  113  Chapter Four  Jiang-King, 1996  association constraints: the peak hierarchy and the margin hierarchy , as shown in (6) and 4  (7) respectively.  (6) Peak Hierarchy (P & S 1993:129) * P / t » * P / d ... * P / i » * P / a  (7) Margin Hierarchy ( P & S 1993:129) * M / a » *M/i ... * M / d » *M/t  (6) and (7) require that more sonorous segments make more harmonic peaks and less harmonic margins. Within the OT framework, it is possible for other constraints to be interspersed into the two hierarchies, giving rise to various syllable structures. For example, problematic inputs like /CCW7 bring the segmental association constraints and the structural constraints into conflict. An input C ideally parsed as a margin may actually be parsed as a peak, or vice versa, in response to top-down constraints on syllable shape. These two conflicting sources of constraints must be harmonized in syllabification by constraint-ranking. This will be demonstrated in the following two Northern Min languages, namely Fuzhou and Fuqing. The syllable structure assumed in this chapter is as follows. A syllable node (a) may contain one or two moras (u). A syllable head is a nuclear mora which is indicated by a Nuc node. Segments directly linking to the syllable node are referred to by the term Onset (leftmost of a Nuc). The term Coda refers to a segment rightmost of a Nuc, which may  4  The peak-margin dichotomy is argued to be insufficient to account for pre-nucleus glides in Fuzhou  "cutting foot words" (i.e. disyllabic words formed by the principle of Fanqie, see the section 5.4 to follow) (Jiang-King 1994a, b, 1995a, b). 114  Chapter Four  Jiang-King, 1996  link to the syllable node directly or indirectly (i.e. via a mora). (8) illustrates the syllable structure I assume.  (8) Syllable structure assumed in general  H (u) V (V/C)  (C)  A  A  A  coda nucleus  onset  The general syllable structure assumed in (8) may vary in a particular language along the lines of whether a coda is moraic, or whether a language has contrastive syllable weight.  4.2 Fuzhou syllabification  Fuzhou distinguishes between two types of syllables: the light syllable (i.e. monomoraic) and the heavy one (i.e. bimoraic). The different number of moras in different types of syllables is determined by tonal properties. That is, a level tone or a simple contour tone (i.e. H, H M or ML) gives a single mora to the tight syllables, whereas a complex contour tone (i.e. M L M or MHM) gives two moras to the loose syllables (see detailed arguments in chapter 3). What is relevant to the present context is that these distinctive syllable structures have a direct influence on vowel features. In other words, the vowel distributions and alternations observed in chapter 2 are highly restricted by the established syllable structures. This direct relationship between the syllable structures and vowel  115  Chapter Four  Jiang-King, 1996  features is captured by the proposed theory since it allows the distinctive syllable structures to play an important role in linking segments to different syllable positions. Before I discuss Fuzhou syllabification, I will first lay out the basic syllable structures in Fuzhou and the relevant constraints assumed.  4.2.1  Fuzhou syllable structure  Most of Fuzhou syllable structure falls comfortably within the purview of the basic theory of syllable structures (M & P 1993a, P & S 1993). Fuzhou admits a wide range of segmental sequences within a syllable: V, CV, W , C W , VC, CVC, V W , C V W , W C , C W C (Liang 1982). Onsets and codas are optional, underpaying of segmental root nodes is disallowed. Nuclei are the only elements required. Fuzhou thus exemplifies the typological family  S( M ), C  C  in the terminology of the basic C V syllable structure theory (P  & S 1993, chapter 6). This means that the faithfulness constraints dominate both ONS and COD, allowing the V syllable when there is no segmental material for an onset and a coda. Various syllables without a consonant are exemplified in (9). Data are from Liang (1982).  monosyl. Gloss a.  i  b.  u  c.  o"  d:  H  H  monosyl.  Gloss 'aslant'  •cloth'  e.  uai  •black'  f.  uo«  'put together'  'stick'  g-  ygML  'shrivel up'  'Asian'  h.  au  'sunken'  H  H  The syllables in (9a-d) contain a single vowel, hence, the nucleus only. There is no margin at all. The ones in (9e-h) show that a syllable in Fuzhou may have more than one vowel without any consonant. This seems to suggest that O N S and - C O D do not play any role in  116  Chapter Four  Jiang-King, 1996  Fuzhou syllabification. However, the following data in (10) show that ONS is respected where possible in disyllabic words.  (10)  monosyl.  Gloss  +  monosvl.  Gloss  - * disyl words Gloss  a.  ts'uonMHM  'similar'  yorjMHM  'shape'  ts'uonnyon  'presentable'  b.  xuon ^  'powder'  0yn  'red'  xuogg0yn  'pink'  1  HL  The examples in (10) show that when a coda consonant is followed by an onsetless syllable, gemination (which is indicated by an underscore in the above examples) occurs. The occurrence of consonant gemination can be attributed to the effect of the constraint ONS. Since ONS is active, it puts pressure on the syllabification so that the second onsetless syllable gets an onset from its preceding coda. The contrast between (9) and (10) lies in the interaction between the structural constraint ONS (M & P 1993a, b, P & S 1993) and the faithfulness constraint LEX-O: (Pulleyblank 1994, P & T 1995, Pulleyblank etal. 1995): the former requires a syllable to have an onset, while the latter prohibits insertion of any Felement which is not present in the input. By ranking LEX-OC above ONS, the lack of onset in (9) is accounted for since there is no consonant present in the input. On the other hand, the inputs in (10) contain a consonant in an appropriate position for gemination (i.e., from the previous morpheme). The occurrence of consonant gemination is expected since the output with the onset satisfies both LEX-O: and ONS, while the one without an onset would violate ONS. This shows that the structural constraint ONS does play a role even though it ranks below the faithfulness LEx-a. Fuzhou has two types of syllables: the light (i.e. monomoraic) syllables and the heavy (i.e. bimoraic) ones, given in (1 la) and (1 lb) respectively:  117  Chapter Four  Jiang-King, 1996  (11) Fuzhou syllable structures  a.  The light syllable in tight  (C)  finals  b.  The heavy syllable in loose finals  (C)  (V/C)  V  V V (V/C)  The contrastive syllable structures above are motivated by the constraints on tonal distributions. In particular, the monomoraic structure in (11a) results from the simple tonal contours (i.e. H, HM, M) while the bimoraic one in (lib) from the complex tonal contours (i.e. M H M and MLM) (see chapter 3 for detailed discussions). Before I proceed to demonstrate Fuzhou syllabification, a number of assumptions must be made explicit. First, following A & P (1994), I assume the theory of Combinatorial Specification in that featural elements combine to represent segments. The seven vowels in Fuzhou i . y. u, E (e/e), 0, O (O/D), A (a/a), thus can be represented as in (12):  (12) Fuzhou vowel representation  a. b  '  F-elements: +Lo, +Hi, E  i  A  \  + L O  + H I  c.  °  E  u 7  +RD, +FRONT  0  *  *  •  y  *  +LO +LO +LO +LO +HI +HI +H1 +HI +m +RD +RD +RD +RD +RD +RD +FRT +FRT+FRT +FRT +FRT  Conditions: Hi/Lo:  if+Hi, then not+Lo.  LO/RD:  if +Lo, then not  LO/FRT:  if +Lo, then not +FRT. 118  +RD.  *  *  *  +LO +LO +LO +HI +H1 +RD +RD +FRT +FRT +FRT  Chapter Four  Jiang-King, 1996  I propose four active vowel features in (12a), whose combinations are represented in (12b). Columns headed by an * indicate the absence of featural combinations ruled out by the three grounding conditions in (12c) which are undominated in Fuzhou. Second, I assume the faithfulness families of constraints (M & P 1993a, b, P & S 5  1993, Pulleyblank 1994, P & T 1995, Pulleyblank et al. 1995, among others) in (13).  (13) Faithfulness constraints  a.  PARSE-CC:  an F-element (feature or node) a must be parsed onto an appropriate  prosodic constituent. b.  FILL: A prosodic constituent  c.  LEX-CC:  must be filled by an F-element.  an F-element (feature or node) a that is present in an output form is also  present in the input: [...a ...Joutput -> [- a .Jinput  Third, I assume the basic structural constraints in (14) (M & P 1993a, P & S 1993), in which the constraint constraints:  5  *COMPLEX  proposed by M & P (1993a) is decomposed into three  *COMPLEX-COD, *COMPLEX-ONS  and *COMPLEX-NUC:  The faithfulness set of constraints are later incorporated into the correspondence theory developed by  McCarthy (1995) and M & P (1995). The terminology has been changed such that PARSE and FILL correspond in large measure to MAX and DEP, respectively. For the purposes of this thesis, either approach to faithfulness would be sufficient for the present context.  119  Chapter Four  Jiang-King, 1996  (14) The basic structural constraints  a.  Nuc: syllables must have nuclei.  b.  ONS: syllables must have onsets.  c.  -COD: syllables must not have codas.  d.  *COMPLEX-COD:  No more than one post-nucleus segment may link to o directly  e.  *COMPLEX-ONS:  No more than one pre-nucleus segment may link to a directly.  f.  *COMPLEX-NUC:  No more than one segment may link to nuclear mora directly.  Fourth, I reformulate the harmonic nucleus constraint *P/a (P & S 1993) as *Nuc/a, since Nucleus is proposed to be a syllable head, hence, a formal constituent within a syllable (Shaw 1992, 1993). *Nuc/a is a set of constraints restricting certain segments to a nuclear position. It contains a number of members with the ranking given in (15).  (15) The set of harmonic nucleus constraints ( H A R M N U C ) *Nuc/C » *Nuc/[+Hi]» *Nuc/[+Lo]  With the assumptions above made clear, I now demonstrate how the interaction of these constraints can automatically derive the attested vowel distribution pairs in Fuzhou.  4.2 2  The contrast between monophthongs and diphthongs  As observed in chapter 2, the vowel distributions in Fuzhou exhibit a correspondence between monophthongs in the tight syllables and diphthongs in the loose ones. The distinctive syllable types are argued to reflect a difference in syllable weight (i.e. monomoraic vs. bimoraic), resulting from their tonal properties (i.e. H, HM, M L vs. M L M , MHM) (see chapter 3 in detail). Given these contrastive syllable structures, linking 120  Chapter Four  Jiang-King, 1996  an underlying high vowel to different syllable structures would give a single high vowel in the tight syllables on the one hand and a possible long high vowel in the loose syllables on the other hand. However, the expected vowel length distinction does not show up There is no surface long high vowel in Fuzhou phonology. Instead, the segmental contrast shows up as a difference between a monophthongal high vowel and a diphthong containing that high vowel. The corresponding vowel pairs are illustrated in (16) below.  I "tight"  Gloss  a.  tsi  •f. word'  g  t s e  b.  piq  'guest'  h.  pein  c.  kuML  'ancient'  i.  kouMLM  'old; reason'  d.  tsun  'permit'  j  tsoun  'handsome'  e.  sy  Tjeard'  k.  «j0yMLM  *(cotton)wadding'  f.  tynHL  'repeat'  1.  t0yrjMHM  'middle'  H  H  ML  H  H "loose" jMLM MLM  MLM  Gloss  Distribution  'will'  / i / —> i ~ ei  'combine' In! -» u ~ ou  /y/ - * y ~ 0y  Moreover, the examples in (16) show that the component vowels of a diphthong share all vocalic features except [+Hi]. The lack of the length contrast for high vowels can be captured by ranking PARSEHI above *Nuc/Hi . As a result, a high vowel links to a nuclear 6  mora only when there is no other vowel present. On the other hand, if there is another non-high vowel occurring with a high vowel, the high vowel can not link to a nuclear mora It can only be parsed onto a non-nuclear mora, giving rise to a diphthong in the loose syllables. To ensure (i) that the outputs in the loose syllables consist of a non-high vowel followed by a high vowel and (ii) that the two component vowels within a diphthong share all vocalic features but [+Hi], as exemplified in (16g-l), the faithfulness constraint Fiix-p, in (13b) and its interaction with *Nuc/Hi play an important role in  I owe this insight to Doug Pulleyblank (p.c.). 121  Chapter Four  Jiang-King, 1996  deriving the alternating pairs. Mapping a high vowel N to the monomoraic structure is relatively straightforward, and is demonstrated in the tableau (17) below Following standard notational conventions, an exclamation mark ("!") signals a fatal violation; an asterisk ("*") represents a single violation; and shading indicates that constraints are not crucial in determining an optimal output. Technically, the mora need not be in the input, but it is required by tonal distributions (see discussions in chapter 3). This will be generally true and will not be repeated for every tableau.  (17) Output candidates [pin ] 'guest' H  Input  P  Cand,  N  c  |Q  /\  /' • \ p  f|  Cand,  Cand,  Cand  a A  a  a  A  / \I)  P  O  / /'  / P  / /•  J X  Cand,  J  / p  3 rj  4  / p  /\ 1  y  n  i+Hr  [pirj ] 'guest' H  [phj]  [pon]  [Pi]  LEXT^X  [pin] *!  PARSERT  *!  *Nuc/C  *!  PARSEHI  *'! *  *NUC/HI  *  The last two candidates in ( 1 7 ) show that violation of  LEXU  (which disallows adding  moras that are not present in an input) or PARSERT (which demands a segmental root to be parsed onto a prosodic structure) is fatal, so they are out. Compare the first three candidates: Cand violates *Nuc/C because the nuclear mora is filled by both the high 3  vowel [i] and the velar consonant [rj]. Cand violates 2  *NTJC/HI.  PARSEHI,  while Cand, violates  Since PARSEHI ranks above *Nuc/Hi, Cand, wins. The two faithfulness constraints 122  Chapter Four LEX-W  Jiang-King, 1996  and PARSERT are highly ranked in all cases. Any output candidates violating them will  be out, thus will not be included in the following tableaux. Notice that the crucial ranking in this case is PARSEHI »  *Nuc/Hi.  (18) Output candidates for [peirj  Input  MLM  Cand,  ] 'combine'  Cand,  Cand,  Cand  4  Cand,,  uu  [+FRT1 [+H0  [peirj  MLM  ] 'combine'  [poq]  PARSEHI  *!  *!  *Nuc/Hi  *!  *!  FlLL-U  The ranking PARSEHI  »  *Nuc/Hi established in (17) rules out the last candidate in  (18), since it violates PARSEHI (i.e. the feature +Hi in Cand is left unparsed). Both Cand 5  3  and Cand in (18) violate *Nuc/Hi in different ways. Cand does so by parsing a high 4  4  vowel to the nuclear mora only, while Cand violates it by linking the high vowel to both 3  the nuclear and non-nuclear moras. Compare Candj and Cand,: the former violates Fnx-p; because the left mora is unfilled, while the latter satisfies it by linking the feature +FRONT to the nuclear mora. Thus, Cand, is the best one. Since the optimal output satisfies all constraints, there is no crucial ranking in this case. The alternating pair u ~ ou can be derived in the same way as the pair i ~ ei. The difference is that the F-element [+RD] is involved in the u ~ ou pair, but not in the i ~ ei pair. The sharing of all features other than [+Hi] within a diphthong can be captured by the  123  Chapter Four  Jiang-King, 1996  interaction between *Nuc/Hi and FILL-U: the former prevents the nuclear mora in a syllable from being filled by [+Hi], while the latter demands that all moras must be filled. This conflict between the two constraints triggers parsing any feature other than [+Hi] to the nuclear mora. Consequently, the roundness or frontness harmony exhibited in (16g-l) has been ensured. The evaluation for the u ~ ou pair is illustrated in (19) and (20) respectively.  (19) Output candidates for [tsun ^] 'permit' 1  Input  Cand,  k  N  u  [+RD]  A  Cand,  [+HJ  b  ts  5^ q  ts  0  [tsorj]  [tsun^] 'permit'  Cand, a  ll [+RD]  1  '  [tsurj]  *Nuc/C  *!  PARSEHI  *  *  *NUC/HI  In (19), linking the velar consonant [rj] to a nuclear mora, as shown in Cand , results 3  in a violation of *Nuc/C. Given the ranking PARSEHI »  *Nuc/Hi established in (17), the  first candidate wins since it violates *Nuc/Hi, while the second candidate violates PARSEHI. Again, we see that the ranking *Nuc/C, PARSEHI »  *Nuc/Hi plays a crucial role in  choosing Candj over Cand . 2  Tableau (20) shows the [+Rr>] spreading that is triggered by the interaction between *Nuc/Hi and FDLL-H in deriving the correct output form [ou]. Cand violates PARSE-HI since 5  the feature [+Hi] is left unparsed. Cand and Cand are ruled out by *Nuc/Hi since the high 3  4  vowel [u] links to a nuclear mora in both cases, the difference between them is that the  124  '. Chapter Four  Jiang-King, 1996  i high vowel links to both moras in Cand but only the nuclear mora in Cand . Candj 3  4  F  violates FILL-JI (i.e. the nuclear mora is left unfilled), while Cand, violates nothing. | Therefore, Cand, wins. Comparing Cand, with Cand , FILL-U crucially chooses Cand, over 2  j Candj since the nuclear mora in the first candidate isfilledby [+RD], whereas the nuclear  imora in the second one is left empty. i I  I (20) Output candidates for [tsoun  ] 'handsome'  MLM  !  Input  Cand,  Cand,  k  N  " /V  A te  [+RD] [+H0  A  A  5  [+M>] J-fHTJ  [tsouq ]'handsome' MLM  Cand,  [tseuq]  A [+RD] I+HI]  Cand  4  A ts  Cand,  A AT  ts  oq  [-HiDi [+m\  [tsuq]  [tsoq]  [tsu:q] PARSEHI  *! *!  *NUC/HI FILL-p.  *!  *}  *!  The corresponding pair y ~ 0y exemplified in (16e, f, k, 1) is similar to the pair u ~ ou. The difference is that the y ~ 0y pair involves feature agreement in both roundness and frontness: [y] occurs with [0], not *ey, *oy. This featural agreement for roundness and frontness within a diphthong can be seen as an instance of the parasitic harmony proposed by Cole and Kisseberfh (1994). The essential idea of Cole and Kisseberth's proposal is that  i the harmonic domain of a feature x co-exists with the harmonic domain of a feature .y. This captures the empirical observation that prosodic anchors within a certain domain tend to share more features if they already share a certain feature. To derive an output where the  125  • Chapter Four  Jiang-King, 1996  two component vowels within a diphthong share all vocalic features but [+Hi], I postulate | a parasitic constraint in (21), along the lines of Cole and Kisseberth (1994:10). I  \ (21)  Parasitic Constraint (PARAS(RD,  FRNT))  Two anchors within a syllable agree in their values for [RD], iff they agree for [FRONT]. I  The following tableaux (22) and (23) demonstrate how the parasitic constraint 1  proposed in  (21)  and its interaction with  PARSEHI, *NUC/HI  and  FILLH  can successfully  j derive the y ~ 0y pair. Linking a front round high vowel lyl to the monomoraic structure in (22) is relatively • simple. The high vowel y must be parsed onto the nuclear mora as in Cand, even though it I violates *Nuc/Hi. Otherwise, it would violate a highly ranked constraint *Nuc/C, as in  j Cand , or PARSEHI, as in Candj. Notice that the parasitic constraint 3  (21)  does not play any  i  ! role in determining an optimal output since there is only one mora in this case. i  'j (22) Output candidates for [tyn ] 'repeat* HL  Input  Cand,  Cand,  Cand,  A A A 0  N  [+RD] \ r+HTJ [+FRNT)  ts  f+FRNT]  [tyqHL] 'repeat'  o  [+RD]  *Nuc/C  1  1  [tyn] *!  *!  PARSEHI *NUC/HI  q.  <f+HII>  *  *  126  Chapter Four (23)  Jiang-King, 1996  Output candidates for  Input  Cand,  o n.  [FRNT]  1  [+RD]  \  :  "  .  [+HI]  'middle'  Cand,  Cand,  k k /r\  N  ts  [WyqMHMJ  [tK6i  ©  A  r 1*101  |*raMT)  [t0y MHM] RJ  t+w>j  A A  [«nj  [+FBNT]  [toyrj]  *  A -  Cand  Cand,  A  A  4  :  r t [ + S D ] \t+mi  [+HD]  [WKNT]  [teyrj]  [tyq]  [t0q]  'middle'  *!  PARSEHI  *!  *NUC/HI FlLL-U,  *•!  PARAS(RD, FRNT)  Tableau  «t+HJ)>  [+FKNT1  (23)  *j  shows that violation of PARSEHI in Cand , 5  *Nuc/Hi  in Cand and Fnxu in 4  Candj is out. Compare the first two candidates. Cand satisfies Fnxu, but violates 2  PARAS(RD, FRNT),  since only the feature Rr> links to two moras. The feature  FRONT  only  links to one mora. Thefirstcandidate satisfies all the constraints, hence, is the optimal one. The crucial ranking suggested in this subsection is given in ( 2 4 ) below:  ( 2 4 ) The ranking for the contrast between monophthongs and diphthongs  L E X - U , PARSE, FILL-IJ,, * N U C / C , PARSEHI »  127  * N u c / H i , PARAS(RD, FRNT)  Chapter Four 4.2.3  Jiang-King, 1996  The tense/lax distinction  As observed in chapter 2, there is a tense/lax distinction between the two types of syllables when a nucleus contains a non-high vowel. It is argued that the tense/lax distinction represents a primary quantity difference with derivative differences in featural content (see chapter 2 section 5 for details). This tense/lax distinction between the two types of syllables is illustrated in (25) below.  I "tight"  Gloss  a.  tsieq™-  'felt'  b.  ko  'song'  e  c.  kua  'few, scant'  H  ML  H "loose"  Gloss  Distributions  d.  tsieqM™  'fight*  / E / - » e ~e  _  foMLM  'individual'  /O/ -> o ~ o  f.  kuaML  'hung up'  IAI -> a ~ a  M  The data in (25) show that mid vowels surface as [e] and [o] in tight syllables, while they appear as [e] and [o] respectively in the corresponding loose ones. Similarly, a low vowel also varies along the tense/lax dimension. It appears as [a] in the tight syllables and [a] in the corresponding loose ones. If the length distinction for the e ~ e, o ~ 3 and a ~ a pairs is primary (which is derived by linking an underlying non-high vowel to a monomoraic structure on the one hand, and a bimoraic structure on the other hand), and the featural distinction (i.e. tense vs. lax) is secondary, we need a constraint assigning the feature [lax] (or whatever other feature appropriately characterizes the distinction between e, o, a and e, D, a) to a long non-high vowel. Following Cole and Kisseberth's (1995) proposal for Yawelmani vowel lowering (i.e. VUJI -> [Low]), I propose a length dependence constraint in (26), which derives a lax vowel from a non-high vowel linked to a bimoraic structure.  (26) Length Dependent Constraint (LAXTNG or Vu|i - » [Lax]) Iff a is parsed onto two moras, then a is [LAX], (where a * [Hi]) 128  Chapter Four  Jiang-King, 1996  What the constraint (26) requires is that the presence of the feature [LAX] depends on the configuration in which a non-high vowel links to two moras This means that the vowel quality difference rests on their quantity difference. The tableaux (27) and (28) below illustrate that given these constraints, linking of an underlying non-high vowel to the different moraic structures can automatically derive the tense/lax pairs.  (27) Output candidates for [ko ] 'song' H  Input  Cand,  Cand,  //r  Cand,  0  /  N  /  |  /  u  k> k  ko  / 1  o  [ko ] "song" H  k  o  [ko]  Cand  0  i i k  1  o  [ke]  4  0  /N\  I VI/ 1 1 / /  M  o  k  [ko:]  LEX-JI  *!•  PARSERT  *!.  LAXTNG  In (27), each of the last two candidates incurs a fatal violation mark since the constraints  LEX-U  and PARSERT are undominated. The second candidate violates the length  dependent constraint  LAXTNG  because the appearance of the feature [LAX] does not depend  on the length. The first candidate does not violate anything, hence is optimal. Notice that the length dependent constraint  LAXTNG  does play a role in ruling out an inappropriate  insertion of LAX in (27), even though there is only one mora in the input.  129  Chapter Four  Jiang-King, 1996  (28) Output candidates for [ k o ^ ] 'individual'  Input  Cand,  Cand,  Cand,  a  o  0  N u  AA  ko  / n  &  k  [koMLMj  o  'individual'  "  k a o  o  [kao]  [ko:]  FlLL-U LEXF LAXTNG  Cand,  A  A  / ^ j»  / IA  k  4  o  A  /N\  u  Cand  k  o  k  o  [ko]  [ko]  *!  *!  *! *!  (28) differs from (27) in that the input contains two moras (which are determined by tonal specifications). Linking the mid vowel to either mora alone, as in Cand and Cand , would 4  5  violate FILL-U. Inserting a feature [Low] as in Cand would violate LEXF which prohibits 3  insertion of any feature that is not present in an input. Comparing Cancl, with Cand,, the former violates LAXING (i.e. a non-high vowel links to two moras without becoming lax), while the latter violates nothing (i.e. a long mid vowel becomes lax). Cand,, therefore, is optimal. The important constraint that chooses Cand, over Cand is 2  LAXTNG.  Since the  optimal outputs in both (27) and (28) satisfy all the constraints, there is no crucial ranking in these cases.  4.2.4  The harmonic restriction on tight syllables  The harmonic restriction on the tight syllables observed in chapter 2 is that when a high vowel and a coda consonant (either a glottal stop or a nasal velar) are both present after a low vowel, the low vowel becomes mid, as shown in (29).  130  Chapter Four (29)  Jiang-King, 1996  I "tight" Gloss a.  mei?  b.  H "loose"  Gloss  Distributions  'strange(lit)'  e.  mai?  teirjML  'wait'  f.  tain™  c.  tsourj  'stolen goods'  g. tsaurj ^  'to bury'  d.  pou? ^  'thin (lit.)'  h. pau? ^ ^  'explode'  HL  H  1  MLM  1  1  1  'strange (collq.)'  eiG~aiC  'chair' ouC ~ auC  r  The data in (29) raise a question as to why the low vowels in the loose syllables become mid in the corresponding tight ones. To answer this question, I propose that the presence of a coda consonant may force the high vowel to link to the mora, resulting in both a high vowel and a low vowel linked to the same mora in the "tight" syllables, as in (30a), whereas they link to two different moras in the "loose" syllables, as in (30b). The square brackets indicate the nuclear mora in a syllable. (30)  a. Short diphthongs in "tight" finals  M  *  e  i  a  [+LO]  b. Long diphthongs in "loose" finals [ul  M i [+HI]  '  a  [+LO]  .u i [+HI]  To account for the corresponding pairs ein. ~ ain and oun. ~ ainj, I propose that the condition Hi/Lo (i.e. if +Hi, then not +Lo) proposed by A & P (1994) applies to the domain of a mora , thus forcing either [+Lo] or [+Hi] to be unparsed in the "tight" 7  syllables. This condition can be formulated as in (31) below:  7  The grounding condition originally proposed by Archangeli and Pulleyblank (1994) is the path  condition that prohibits two phonetically incompatible F-elements from cxxniring on a single path. For instance, the HI/LO condition prevents the features [+HTJ and [+LO] from linking to a single segmental root. The HI/LO condition used here differs from that proposed by Archangeli and Pulleyblank in that it  131  Chapter Four  Jiang-King, 1996  (31)Hi/Lo Condition extended into the domain of mora •Hi/Lo^: a mora cannot be filled by both [+Hi] and [+Lo] F-elements.  The function of (31) is to prevent the features [+Hi] and [+Lo] from linking to the same mora. If that happens, one of these two features must be underparsed. In terms of Optimality Theory, both *Hi/LoH and PARSEHI must rank above PARSELO . In other words, 8  it is better to underparse [+Lo] than to violate *¥h/Lo^ and PARSEHI. The following tableaux (32) and (33) illustrate how syllabification governed by the interaction of *Hi/Lo , PARSEHI and PARSELO derives the alternating pair eij} ~ airj u  The last two candidates in (32) each incur a fatal violation mark since *COMPLEXCOD and PARSERT are undominated. Cand violates *Hi/Lo'- -, which ranks above 1  3  the rest of the constraints, and so is out. Compare thefirsttwo candidates, both of which satisfy *Hi/Loi in different ways: underpaying [+Hi] in Cand results in the diphthong A  2  [ae], whereas underpaying [+Lo] in Cand, gives rise to the diphthong [ei]. The former violates PARSEHL while the latter violates PARSELO. Since PARSEHI ranks higher than PARSELO, thefirstcandidate wins, even though it violates PARSELO and *Nuc/Hi, the lowest ranked constraints. The ranking iri this case is: *COMPLEXCOD, PARSERT, *HI/LO, PARSEHI »  PARSELO, *NUC/HI.  involves two segmental root nodes. In other words, it prevents a high vowel and a low vowel from linking to the same mora rather than prevents them from linking to the same segmental root node. 8  1 am grateful to Pat Shaw and Doug Pulleyblank (p.c.) for suggesting the use of the ranking PARSE-HI »  PARSE-LO to account for this case. 132  Chapter Four  Jiang-King, 1996  (32) Candidate outputs [tein ] "wait" ML  Input  Cand,  n  [+LOI [+HI]  Cand,  Cand  Cand,  4  A A /AA A Ax  N  t a i rj 1 1  Cand,  t .e iq •  x  t a iq  t a  1  1  [+L0]  <[+LO]> [+K]  [tein ^] « j i 1  wa  t  t  <[+H3>  [taen]  a i II  [+LO]  q  [+mi  [tain]  iq  t  a iq 1 1  1 1 [+LO] t+ra)  t+LO] [+H]  [tai]  [taiij]  *COMPLEXCoD  *!•  *!  PARSERT  *! *!'  PARSEHI PARSELO  *  *NUC/HI  *  *  Notice that decomposing the P A R S E family plays a crucial role in this case. Its internal ranking, namely,  PARSERT »  PARSEHI »  PARSELO,  is important in deriving the correct  output. Compare (18) and (32): the former contains the diphthong [ei] in a loose syllable (j-peiqMLMj  'combine') while the latter has the diphthong [ei] in a tight syllable ([teirj^]).  The difference between them, however, is that the diphthong [ei] in a loose syllable is long since it links to two moras required by the complex contour tone, whereas the one in a tight syllable is short because it links to a single mora with only a simple tonal contour. Tableau (33) below shows that once an input contains two moras and a string with a low vowel followed by both a high vowel and a consonant,  PARSE-HI  and  PARSE-LO  play no  role in determining the optimal output. Since the syllable contains two moras, both a high vowel and a low vowel are able to be parsed to different moras. Therefore no feature  133  Chapter Four  Jiang-King, 1996  conflict occurs, even though the presence of a coda consonant forces a high vowel to be linked to a mora.  (33)  Output candidates for [tain^HM] "chair"  Input  Cand,  Cand,  Cand^  CT  |  A  a  N  i  rj  hi  A  /r\ •  / v\  uu  t  CT  o  A  N  Cand,  / u u / N1  a i q  t  [tain] 'chair'  :  ail)  A  /K\  [tab]]  / M- M/ 1. 1  t  a i r)  [taq]  t  a  i rj.  [tain]  "COMPLEXCOD PARSERT  *! *!  ••HI/LOH  The last two candidates in (33) are ruled out by *COMPLEXCOD and PARSERT, respectively. The second candidate violates *HJ/LO^, while Cand, does not. The first candidate, therefore, is better than the second. There is no crucial ranking in this case. The harmonic restriction demonstrated in this section suggests that PARSERT  must rank above *Hi/Lol ,  4.2 5  The asymmetric behavior of high vowels  A  PARSEHI,  *COMPLEXCOD  which in turn rank above PARSELO,  and  *NUC/HI.  The investigation of Fuzhou tone-vowel interaction in chapter 2 also reveals that a high vowel behaves differently with respect to its relevant syllable positions. It manifests a correspondence between a monophthong and a diphthong when it appears as a nucleus of a syllable (see section 4.2.2 in this chapter), whereas it does not alternate at all when it precedes or follows another non-high vowel. In the latter case, the vowel distribution 134  Chapter Four  Jiang-King, 1996  effect takes place in the non-high vowels along the tense-lax dimension. The asymmetric behavior of the high vowels is illustrated in (34)  (34)  I "tight" Gloss ka  ML  c.  kua  e.  kau  ML  H  H "loose"  Gloss  Distribution  'ancient'  b.  JJQUMLM  'old- reason'  u ~ ou  'few, scant'  d.  kua  MLM  'hung up'  *u ~ ou  'suburbs'  f.  kau  MLM  'enough'  *u ~ ou  (34a-b) show that when a high vowel occurs as a nuclear vowel, it surfaces as [u] in the tight syllables and as [ou] in the loose ones. This corresponding pair u ~ ou does not show up when the high vowel u precedes (34c-d) or follows (34e-f) a non-high vowel [a]; The asymmetry can be accounted for by the constraint interaction. The tableaux (35) and (36) demonstrate how the inert behavior of a high vowel following a non-high vowel can be accounted for by the interaction between PARSEHI,  PARSELO  and  *Nuc/Hi.  (3 5) Output candidates for [kau ] 'suburbs' H  Input  Cand, a  N  / n &> k  kau  :  /N\  a  \ u  [kau ] "suburbs" H  Cand,  Cand,  Cand  0  a  0  A /i\  k  <a>  u  [keu]  k  k  a  PARSELO  u  [kau]  L  k  a<u>  [kae]  PARSEHI  *.!  *NUC/HI  *!•  *!  135  d  Chapter Four  Jiang-King, 1996  In (35), the last three candidates violate PARSEHI,  *NUC/HI  and PARSELO respectively.  In particular, underpaying the feature [+Hi] in Cand violates PARSEHI, while linking both 4  [a] and [u] to the same mora violates  *Nuc/Hi  in Cand . Cand; violates PARSELO because 3  the feature [+Lo] is left unparsed. Only the first one incurs no violation, and is therefore the optimal one. There is no crucial ranking in this case. When an input is bimoraic it should be possible for a and u each to link to a separate mora. In such cases, the low vowel a should surface as short and tense in a loose syllable. However, the example (34f) shows that a becomes a when it precedes a high vowel u in a bimoraic syllable even though there is no coda consonant present. To prevent the  high  vowel u from being parsed onto a non-head mora, a constraint (36) is needed:  (36) Harmonic Mora  (HARM ) U  * | i / C » * u / H i » *u/Lo.  The effect of (36) is to express the observation that the more sonorous a segment is, the more harmonic a mora it makes. The following tableau illustrates how the constraint HARM! interacts with other constraints, giving rise to an optimal output. 1  In (37), Cand is out since the non-nuclear mora is left unfilled. Cand violates HARMI-  1  4  3  because the non-head mora is filled by the high vowel u. The first two candidates both satisfy  FILL-H  and  HARM! . 1  However, Candj violates  LAXTNG  because the low vowel is  parsed onto two moras without becoming lax. Cand satisfies all of them except t  the lowest ranked constraint, and is therefore optimal  136  *COD/HI,  Chapter Four  Jiang-King, 1996  (3 7) Output candidates for [kau  Input  MLM  ] 'enough'  Cand,  Cand;,  a N  Cand,  o  N  H k  \i  a  u  /^ A / V \ k a u  ..  [ k a& u ^ ] 'enough'  AX  k  4  o  ;  / \\  Cand  ':  a u [kau]  A A ? k  a u  [kau]  M\ /^ ^X k a u  [kaeu]  FlLL-H HARM  *!  U  *!  LAXTNG *COD/HI  *  *  *  Having explored the possibilities for a high vowel following a low vowel to link directly to either o or u, I now turn to the cases where a high vowel precedes a non-high vowel. In traditional Chinese phonology, the high vowel in front of a non-high vowel is called the "on-glide". The data in (34c-d) suggest that the "on-glide" cannot be part of the nuclear mora. If it were linked to the nuclear mora, the low vowel in (34c) would become a high vowel + a mid vowel [uo], since the nuclei in these cases are monomoraic. My analysis for the off-glide predicts that linking a high vowel and a low vowel to the same mora would violate the condition  *Hi/LoH',  thus resulting in underpaying of  [+Lo].  However, the data in (34c-d) show that the presence of an on-glide has no effect on the low vowel: Namely, a low vowel in these cases does not become a mid vowel; therefore the on-glide must not link to the nuclear mora. If the on-glide is not part of the nuclear mora, where should it be? One possibility is to treat it as an onset, which directly links to the syllable node. However, the evidence from so-called "cutting-foot" words shows that this is not the case. The "cutting-foot" words in Fuzhou are disyllabic words formed from monosyllabic words by a process 137  Chapter Four  Jiang-King, 1996  resembling partial reduplication. In particular, the first syllable in the output disyllabic words is a monomoraic syllable, which shares the most sonorous vowel and any segmental material before the nuclear vowel with the original syllable. The second syllable of the output retains everything of the original syllable except the onset. The data in (38) exemplify this type of word.  (38)  Original  Derived  a.  pQJjjMLM  pa lairj  b.  tia" 1  c.  uai  H  d.  uo  Output pattern 'turnaround'  CV.1VGC  tiaMLliaHL  'too tired to keep eyes open'  CGV.1GV  ua luai  •aslant'  GV.1GV *GV.|V  •put together'  GV.1GV *GV.1V  L  MLM  ML  H uo  Mq H uo  H  The examples in (38) show that when an input contains a string CVGC in (38a) and CGV in (38b), the first syllable of the output contains the most sonorous vowel of the input, and any segmental material preceding that vowel, with a new tone; while the second syllable of the output retains everything of the input (including the tone of the input), except that the initial consonant is replaced by a liquid [1]. If the on-glide were treated as an onset, then the input string GV(G) in (38c-d) would yield the output [GV.IV(G).], where the on-glide in the second syllable is replaced by the liquid [1]. The actual output, however, is [GV.1GVC.]. The on-glide G is not replaced by the liquid [1], but remains in the second syllable of the output with the addition of a preceding [1]. Therefore, the evidence (38c-d) from the "cutting-foot" words suggests that the on-glide cannot be the onset or part of the onset. Based on the same type of data (i.e. the "cutting-foot" words), Qu (1995) also argues that an on-glide cannot be part of the onset. He proposes for similar reasons that it links to the leftmost nuclear mora together with another non-high vowel This option has been rejected due to the evidence from the alternating behavior of low vowels In particular, it is observed that an on-glide behaves differently from an off-glide in triggering 138  Chapter Four  Jiang-King, 1996  the vowel-raising effect. When an off-glide (i.e. a high vowel following a low vowel) is forced to link to a mora with a low vowel, it triggers vowel-raising. However, an on-glide (i.e. a high vowel preceding a low vowel) never triggers a vowel-raising effect. This suggests that an on-glide cannot be parsed onto the same mora as a low vowel. Otherwise, the lack of vowel-raising is left unexplained. If the on-glide cannot be either an onset (or part of an onset) or the nucleus (or part of the nucleus), where should it link to? I propose that it links to the Nuc node directly, in the same way an initial consonant links to a syllable node directly. If the nucleus is a formal constituent, as argued by Shaw (1992, 1993), in principle nothing can prevent an on-glide from linking to a Nuc node directly. (39) and (40) show that it is optimal to link an on-glide to the Nuc node rather than to link it to a nuclear mora.  (3 9)  Output candidates [tsieqHL] 'felt'  Input  Cand,  Cand,  N  u  ts p  CJ  ts  jb  [«m|  b  q  (+H!T1  [tsien™-] 'felt' q  A |5  ts 3 q hflUl+FRTT  [tsien]  *COMPLEX-ONS [+HI] [+FRT1  *NUC/HI  Cand,  A tS CD jO  I)  r+Hl]t+fRTl  [tsien] *!  * l  In (39) above, the last candidate incurs a violation mark for  *COMPLEX-ONS  since both the  initial consonant [ts] and the on-glide [i] link to the syllable node directly. Cand violates 2  139  Chapter Four  Jiang-King, 1996  *Nuc/Hi, while Cand, violates no constraints . Therefore, Cand, is the optimal one. 9  (40) Candidate Outputs for [tsierj* ^] 'fight' 11  Input  Cand,  Cand,  J DJ DA k  Cand,  N  u u  ts  /  [+MI+FRT]  i+mi |+™T] [LAX]  [tsienM-^] 'fight* 1  ts jb |c/ IJ [+HTj [+FRT)  c  Cand, 0  A • J D D c  \  .ts  3  [tsierj]  [tsie:rj]  4  0  .ts  n  £>  Cand  0  b b  n  Q  :tS  O  [4HTJ l+FRTl  q[ •+BIJI+FRT]  [tsien]  [tsen]  *COMPLEX-ONS  *!  HARM^  '*•!  PARSERT *NUC/HI  *!  LAXTNG  *.!  In tableau (40), the last candidate incurs a fatal violation mark for *COMPLEX ONS. It also violates H A R M since the non-nuclear mora is filled by a consonant rather than a vowel u  Parsing a velar nasal to the right mora, as in Cand , violates HARM* , since the right mora is 1  4  not filled with the most sonorous vowel [e]. It also violates PARSERT, since the high vowel fails to be parsed onto any prosodic constituent. Candj violates *Nuc/Hi, while Candj violates  9  LAXTNG.  Only Cand, satisfies all constraints, and hence is optimal. The optimal  Notice that linking an on-glide to the Nuc node directly, as in Cand^ does not incur a violation for  •Nuc/Hr, since the nuclear mora is defined as a syllable head (see chapter 3), and the Nuc node indicates a syllable head with a mora as its content.  140  Chapter Four  Jiang-King, 1996  outputs in both (39) and (40) satisfy all constraints, and there is no crucial ranking for these cases.  4.2.6 A summary  I have demonstrated in this section that Fuzhou vowel distributions can be derived by a set of constraints on the linking of vowel features to the existing contrastive syllable structures. First, the correspondence between monophthongs and diphthongs can be achieved by linking the same set of high vowels to the monomoraic syllables on the one hand and bimoraic syllables on the other hand The lack of a vowel length contrast for the high vowels can be attributed to ranking parsed onto a prosodic anchor) above  PARSEHI  *Nuc/Hi  (which requires a high vowel to be  (which prohibits a high vowel from being  parsed onto a nuclear mora). Second, the tense/lax distinction between the two types of syllables can be captured by the length dependent constraint  LAXTNG,  which requires a  doubly linked non-high vowel to become lax. Third, the harmonic restriction (i.e. the lack of low-high vowel sequence) on the tight syllables can be explained by the harmonic constraint:  *Hi/LoH,  which extends the grounding condition "if + H i , then not +Lo" to the  domain of the mora and prevents both [+Hi] and [+Lo] from linking to the same mora. Last, the asymmetric behavior of a high vowel with respect to its relevant syllable positions (i.e nucleus vs. non-nucleus) can be accounted for by the interaction of PARSEHI and  *Nuc/Hi  The entire set of constraints and their ranking for Fuzhou syllabification is  given in (41):  (41) Constraint Ranking for Fuzhou Syllabification  L E X - U , LEX-F, FILL-U,, PARSERT, *COMPLEXONS, *COMPLEXCOD, *HI/LO, * N u c / C , PARSEHI »  PARSELO, *NUC/HI, PARAS(RD, FRNT), LAXTNG »  141  *COD/HI  Chapter Four  Jiang-King, 1996  4.3 Fuqing syllabification  The investigation of Fuqing tone-vowel interaction in chapter 2 shows that as in Fuzhou. there is a correlation between tonal contours and syllable types. This correlation is captured by the prosodic anchor hypothesis proposed in chapter 3. It is also shown that the vowel distributions in Fuqing are similar to those in Fuzhou in four respects. First, there is a tense/lax distinction between the two types of syllables (i.e. the tight syllables and the loose ones). Second, like Fuzhou. the cooccurrence of vocalic features is more restricted in the tight syllables than in the loose ones In other words, the harmonic restrictions are also active in Fuqing. Third, as in Fuzhou. there is an asymmetry where high vowels behave differently with respect to different syllable positions. In particular, they surface as high in the tight syllables and as mid ih the corresponding loose ones when they occur as nuclei of syllables However, they do not alternate at all when they occur with another non-high vowel Last, there is a correspondence between monophthongs and diphthongs when there is a on-glide present (or equivalently, a correspondence between diphthongs and triphthongs). Despite these similarities, Fuqing vowel distributions also exhibit some differences from those of Fuzhou. For instance, an underlying high vowel surfaces as high in the tight syllables, whereas it appears as mid in the corresponding loose syllables. This high/mid correspondence does not exist in Fuzhou. The goal of this section is to demonstrate how the set of constraints motivated for Fuzhou can also apply to Fuqing cases and the different vowel distribution effects can be captured by different rankings of the same set of constraints. Before I go on to discuss Fuqing syllabification, a number of assumptions must be made explicit. First, I assume that the basic structural constraints and the association constraints on linking segments to syllable structures (M & P 1993a, b, P & S 1993) are also suitable for Fuqing. Second, I follow the combinatorial specification proposed by A & P (1994) and assume that features are combined to represent segments. Thus, Fuqing 7 142  Chapter Four  Jiang-King, 1996  vowels i, y, u, E (e/e), a (0/os), O (oh), A (a/ate) can be represented as in (42). Fuqing data in this work are all from Feng's (1990, 1993) exhaustive descriptive works.  (42) Fuqing vowel representation  a. b.  F-elements: +Lo, +Hi, E  1  A  i  o  E  ?  +RD, +FRNT  u  i  +HI  +HI  ?  (E  +LO  »  »  y  +LO +LO +LO +HI +RD  +RD +FRT  c.  »  Conditions: Ffi/Lo  +HI +RD  +RD  +FRT+FRT  »  »  *  *  +LO  +LO +LO +LO  +HI  +HI  +HI  +RD  +RD  +FRT +FRT  +HI +RD +RD  +FRT+FRT+FRT  if+Hi, then not+Lo.  LO/RD  if +Lo, then not  LO/FRT  if +Lo, then not +FRT.  +RD.  Four active F-elements are postulated in (42a) and their combinations are represented in (42b). The *'s indicate the impossible combinations ruled out by the feature cooccurrence conditions in (42c).  4.3.1  The high/mid correspondence  I begin with Fuqing syllabification by examining the correspondence between high vowels in tight syllables and mid ones in the corresponding loose syllables. That is, an underlying high vowel appears as a single high vowel in the tight syllables, while it occurs as a mid vowel in the corresponding loose syllables. This correspondence does not exist in Fuzhou. and is illustrated in (43) below.  143  Chapter Four  (43)  Jiang-King, 1996  I "tight"  Gloss  II "loose"  Gloss  Distributions  a.  tsi™^  £  'a pronoun'  b.  tse™-  ^  'character"  / i / -> i ~ e  c.  ts'uHM  jjfi  'rough'  d.  t'b  ^  'rabbit*  /u/-»u~o  e.  tsy  =f  'boil'  f.  ts'^  &  'place'  /y/-»y~0  M  HL  The data in (43) show that the high vowels [i], [u] and [y] in the tight syllables correspond to the mid vowels [e], [o] and [0] in the loose syllables, respectively. This high/mid correspondence raises a question as to how the vowel differences in height between these two types of syllables can be achieved, given that the tight/loose distinction is argued to be a light-heavy distinction (i.e. monomoraic vs. bimoraic) of syllable weight (see chapter 2 section 5 for details). Comparing the high/mid correspondence with the similar cases in Fuzhou, where a monophthongal high vowel corresponds to a diphthong containing that high vowel, it is clear that the constraint ranking  PARSEHI  »  *Nuc/Hi  established from the monophthong/diphthong distinction in Fuzhou is not sufficient to account for the high/mid correspondence in Fuqing. We need a constraint that prevents a high vowel from linking to two moras within a syllable and a constraint that demands that two moras within a syllable agree in the feature value for [Hi]. These two constraints can be formulated as in (44) and (45), respectively:  (44) *Hi-j iu r  A feature [+Hi] cannot link to two moras within a syllable.  (45) Height Agreement within a syllable (HTAORJXU) Two moras within a syllable must agree in feature value for height.  144  Chapter Four  Jiang-King, 1996  Combination of these two constraints will rule out cases where a high vowel links to two moras in a bimoraic syllable or either one of the two moras. The effect of *Hi-p.p. and on deriving the high/mid distinction in Fuqing will be demonstrated in tableaux  HTAGRJXH  ( 4 6 ) and ( 4 7 ) . Linking an underlying high vowel to a monomoraic syllable is simple, as is shown in ( 4 6 ) .  ( 4 6 ) Output candidates for [ts'u ^ 'rough' 1  Input  Cand,  Cand,  CT  CT  I  /r  N  H  /=j  * k  '/\  a  [+RD]  k  [+H0  ts'  [ts'u™-] 'rough'  [tso]  PARSERT  Cand, cr  k 1  ts' p  [tse] *!'  PARSEHI *  *NUC/HI  The last candidate in ( 4 6 ) incurs a fatal violation for PARSERT, since the high vowel [u] fails to be parsed onto a prosodic anchor. There are two candidates left to compete with one another. Cand violates 2  PARSEHI,  while Candj violates *Nuc/Hi. The ranking between  these two constraints established in Fuzhou is  PARSEHI »  *Nuc/Hi. Thus, the first  candidate wins since it only violates *Nuc/Hi, the lowest ranked constraint. Notice that the constraints  *HI-HJX  and  HTAGRJXU,  proposed above are not included in tableau  (46)  since  these constraints only affect a bimoraic syllable and there is only one mora in this case. They will become important in determining an optimal output in a bimoraic syllable ( 4 7 ) . It is important to point out that the ranking  145  PARSEHI »  *Nuc/Hi motivated by the  Chapter Four  Jiang-King, 1996  correspondence between monophthongs and diphthongs in Fuzhou works equally well in determining the optimal output for the tight syllables for the high/mid distinction in Fuqing.  (47) Output candidates for [t'o^] 'rabbit*  Input  Cand,  :  N uu  A  Cand,  Cand,  A A ts DJD  [+KDH+HI1  [tmj <|»ffll>  [t'o ]'rabbit' ML  A a  ' 1 1  *  "  V\  \  • A  [•JSD) [+HI]  l+RD) l+HI]  [tsou]  [tsu:]  FlLL-U *Hi-HM,  4  Cand,  A •A K  ts  L  A  l+RDl |+an  t+»D) (+HI1  [tseu]  [tsue]  *'!  *!  *|  *»  *•!  *!  HTAGRUU PARSEHI  Cand  * *  *  *NUC/HI  In (47), the last two candidates all violate FILL-U, since one of the two moras in each of them is left unfilled. The middle two candidates violate  *HI-LIU  and  HTAGRUU,  respectively. In particular, linking a high vowel to both moras, as in Cand , violates *Hi3  \x,]x, whereas parsing it to the right mora only, allowing the feature [RD] to fill the nuclear mora (as in Cand^), violates HTAGRUU because the two moras within a syllable do not agree in the feature value for height (i.e. the left mora is [-Hi], while the right mora is [+Hi]). The first candidate satisfies all constraints except  PARSEHI.  Notice that  underpaying [+Hi] is the only way to satisfy the three constraints Fnxu, *Hi-uu and HTAGRUU.  In other words, not ranking these three constraints forces [+Hi] to be unparsed,  146  Chapter Four  Jiang-King, 1996  resulting in a long mid vowel. The crucial ranking for the high/mid correspondence is: *Hi-p.u, HTAGRU^ »  PARSERT, FILLU,  4.3.2  PARSEHI »  *Nuc/Hi.  The tense/lax distinction  As in Fuzhou, Fuqing vowel distribution also exhibits a tense/lax distinction between the two types of syllables. This kind of vowel quality difference is argued to represent primarily a length distinction which has certain featural content (see chapter 2 section 5 for details). That is, when a non-high vowel links to a monomoraic structure, it becomes a tense vowel (i.e. [e], [o], [0] or [a]) in the tight syllables. When a non-high vowel links to a bimoraic structure, it appears as a lax vowel (i.e. [e], [D], [ce] or [a]) in the loose syllables. This is illustrated in (48) below.  (48)  I "tight" a.  nieq  c  ts'yorj"  e.  t0n  g.  pa ^ 1  M  Gloss  II "loose"  Gloss 'art'  2fe 'dye'  b.  qie™-  i§  'wall'  d.  ts'yoq  'winter'  f.  tceq™-  g!j 'to move'  'white'  h.  pa^  ~g 'hundred'  HM  £  n|§ 'sing'  ML  Distributions /e/ —» e ~ e lol -» o ~ /CE/ ->  D  0 ~ ce  /a/ -> a ~ a  To account for the tense/lax distinction shown in (48), the Length Dependent Constraint  (LAXING) in (26)  (i.e. If a is parsed onto two moras, then a is [LAX]) proposed  for the similar case in Fuzhou is suitable here. I demonstrate in the following tableaux (49) and (50) that it is the constraint  LAXTNG,  along with the faithfulness Fux-u., that gives the  tense/lax distinction for the non-high vowels. The two candidates in (49) show that linking the velar consonant [q] to a mora in Candj violates *Nuc/C, while underpaying it in Cand violates 2  PARSERT.  The violation is  avoided in Cand] by linking the consonant to the syllable node. Since Cand, satisfies both 147  Chapter Four  Jiang-King, 1996  constraints, it is the optimal one. Notice that the constraint  LAXTNG plays  no role here since  the input for this case is a monomoraic structure.  (49) Output candidates for ^q™" ] 'winter' 1  Input  Cand,  Cand,  A  N M-  1  [+RD][+FRT| b  0  A  /N  /1  • A  A '  1  'A'  [t0]  [«u>| [+FRT]  [t0qHM]  Cand,  0  -winter'  1-WD} [+FRT|  [t0q]  *Nuc/C  *!  *!  PARSERT  (50) Output candidates for [tceq ] 'to move HL  Input  Cand,  N  1  1  Cand,  Cand,  A A AV A A /A AA A •.  1 [+RDH+FRT]  ^  Cand^  [+RD] I+FKT)  [tceq™-] 'to move'  • A  "  I+RDJ [+FRTI  [t0:q]  '  A  [+RDJ (+FRT]  [te0q]  '  t  ^  1+RD] [+fKTj  q  Cand, o  /\ N  .  ,  / /  ^ H |  A  1 ,  t+KDJ [+FRT]  [t0eq]  *u/C  *'!  FlLL-U  *!  LAXTNG  •*1  *!  In (50), the violations of *u/C and Fiix-p. in the last three candidates are fatal. Compare the first two candidates. Cand violates LAXTNG (i.e. a front round vowel links to 2  148  Chapter Four  Jiang-King, 1996  two moras without becoming lax) while Cand j does not. Thus, the first candidate wins. Notice that in these cases, there is no crucial ranking among these constraints since the optimal outputs satisfy all constraints.  4 3.3  The harmonic restrictions on the tight syllables  The vowel distributions in Fuqing also exhibit the harmonic restrictions on the tight syllables, as found in Fuzhou. The difference between these two languages in this regard is that the prohibition of the low-high sequence within a monomoraic syllable and the agreement in roundness and frontness are required at the same time in Fuqing. whereas these two restrictions apply to different cases separately in Fuzhou. The data in (51) illustrate these restrictions.  (51)  I "tight" a.  pieu  c.  puor™  M  Gloss  II "loose"  Gloss  Distribution  z% 'watch'  b.  kieu™-  jjft 'sedan'  ieu ~ ieu  $fC 'cup'  d.  puBi*^  Jjj  uoi ~ UBi  'shell'  Two generalizations that can be abstracted from the data in (51) are (i) a mid vowel in the tight syllables corresponds to a low vowel in the loose syllables when it is flanked by two high vowels; (ii) the mid vowels in the tight syllables (51a, c) agree in frontness and roundness with the preceding high vowel but not with the following high vowel. The first generalization can be explained by the *Hi/Lo condition proposed for Fuzhou. That is, u  linking both a low and a high vowel to the same mora is disallowed. Since the tight syllables are monomoraic, the only way to satisfy the *Hi/Lo condition is to underparse u  either the feature [+Hi] or the feature [+Lo]. To ensure that the feature that is underparsed is [+Lo] but not [+Hi], P A R S E H I must rank above  PARSELO,  as in Fuzhou Moreover, to  account for the second generalization, i.e., the feature agreement between the mid vowel 149  Chapter Four  Jiang-King, 1996  and its preceding high vowel in the tight syllables, I propose that the roundness and frontness agreement between an on-glide and its following non-low vowel can be achieved by a feature agreement constraint, as stated in (52):  (52) Feature Agreement Constraint (F-AGR[RD] or [FRNT])  A non-low nuclear vowel must agree with its preceding on-glide in feature value for [RD] and [FRONT].  The function of (52) is to demand a harmonic feature cooccurrence between a nuclear vowel and its preceding on-glide. Thus, the roundness and frontness agreement are accounted for. Tableaux (53) and (54) demonstrate how F-AGR[RD] interacts with other constraints, such as *Hi/Lo , PARSEHI and PARSELO in deriving the optimal outputs. u  (53) Output candidates for [puoi ] cup' HM  Input  Cand,  Cand,  Cand,  Cand  a  a  o  o N  // /N  p u e i  /  1  /  A  /[+M>H+io]/[+Hri (•HD) [+FRT|  p U  /N  /f\ O  1  /  A  e  [+HI1  i  <q+LOJ>/\  /I+»D] S3  /N  /f\  pu  j  [+FRTJ  / f+BTJ  [puoi™] 'cup'  [puei]  B  1  pu  e  '/  f-fFRTJ  [purje]  /  / K 6  i  1  h  /[+8D][+LO]/[+HI]  LPUBi]  p U  /  8  1  i  h ••  /[+RD][+LO]^I+HIl I+MJ  1+FM1  [puBi]  *COD/HI  *!  *HI/LO"  *!  PARSEHI  *!  F-AGR[RD] PARSELO  A  /N  / / r\ p u  Cand,  4  *!  •*»  *  150  *t  Chapter Four  Jiang-King, 1996  In (53), the last two candidates violate * C O D / H I and  *Hi/Lo'r  respectively, since the  a  high vowel [i] is parsed onto a syllable node directly in Cand , and the nuclear mora is 5  filled by both the high and low vowels in Cand . The first three candidates violate P A R S E H I , 4  F-AGR[RD]  and  respectively; In Cand,, the feature  PARSELO  Cand,, the feature  [+Lo]  is unparsed, while in  is unparsed. Candj violates F - A G R [ R D ] , because the nuclear non-  [+Hi]  low vowel does not agree with its preceding on-glide in feature value [+RD] Since PARSEHI  and F - A G R [ R D ] rank above P A R S E L O , Cand, wins. The crucial ranking in this case  is that * C O D / H I ,  * H I / L O H , PARSEHI,  and F - A G R [ R D ] must rank above P A R S E L O .  (54) Output candidates for [pum ^] 'shell' 1  Input  Cand,  1  N  pus i 1 11 [+EE|tLOI+FRT|  Cand,  PUB  *,  1  1  Cand,  /w 1  A 1  pus  i  i  1 1 1  1  [»m|tLon«ni  [pum^] 'shell'  / p  Cand  IK UB  / 1  i  p  1  UB  /  4  i  1  1  [•aqi+Loiitaq  [•Bll [•KJH+H11  [pum]  [pum]  *COD/HI  *!  *NUC/HI  *!•  *Hi/LoH  *!  •  *!  *!  *!  Now let's turn to the case in (54) where the input contains two moras All candidates except the first one incur a violation mark That is, Cand violates both 4  *COD/HI  and  *Nuc/Hi, whereas Cand and Cand violate both *Nuc/Hi and *Hi/LoM\ Only the first 2  3  candidate does not violate any of these constraints, thus is the optimal one. Notice that the constraints (i.e. *YhlLo\ , x  PARSEHI, F - A G R [ R D ] , PARSELO)  151  and their ranking that are crucial  Chapter Four  Jiang-King, 1996  in (53) no longer matter in (54), since the input is bimoraic, allowing both the high and low vowels to link to a separate mora.  4.3.4  The asymmetry of the high vowels  The asymmetric behavior of high vowels with respect to their syllable positions in Fuqing is similar to that found in Fuzhou. The data in (55a-d) show that a high vowel in the tight syllables corresponds to a mid vowel in the loose syllables. However, this kind of vowel distribution does not show up in (55e-h). The difference between the high vowels in (55ad) and the ones in (55e-h) is the syllable positions they occur in. In particular, the high vowels in the former appear as syllable nuclei, whereas the ones in the latter case occur with other non-high vowels (i.e. as on-glides).  (55)  I "tight"  Gloss  ITloose"  a.  tsi™*  £  'a pronoun'  b.  c.  ts'u™  &  'rough'  e.  siarj™  g. puaq"  Distributions  ^  'character'  i ~e  d. t ' o ^  %  'rabbit'  u ~o  jSr 'sound'  f.  j£  'line'  * i~e  &  h. puan^  ^  'half  *u~o  'dish'  tse™-  Gloss  siarjML  The different syllable positions affecting the behavior of the high vowels can be accounted for in the prosodic anchor hypothesis proposed since the linking between syllable structures and vowel features is direct and is governed by the interaction of the constraints, such as the faithfulness FILL-U, PARSEHI, and the association constraints like *NUC/HT.  The tableaux (56) and (57) below show how the non-alternation of a high vowel  in a pre-nuclear position is achieved.  152  Chapter Four  Jiang-King, 1996  (56) Output candidates for [sian ^ 'sound* 1  Input  Cand,  A A AA AA  Cand  :  N  s i a rj  Cand,  fa  s i a q  [siarj ^] 'sound' 1  4  /N\  Cand,  Cand,  A • AV. 1 /N  / /I  /  H \  s<i>a q  s i<a>rj  [sag]  [sirj]  s i a q  s i a q  [siag]  [siag]  *Nuc/C *Hi/Lx>u  *!  *!  PARSEHI  *!  PARSELO  In (56), linking both a low vowel and a velar consonant to the same mora, as in Cand , 5  violates *Nuc/C, whereas linking a high vowel and a low vowel to the same mora, as in Cand , violates the harmonic condition *Hi/Lo . Both the last two candidates are out. u  4  Compare the middle two candidates. Underpaying either [+Hi] or [+Lo] violates or  PARSELO,  PARSEHI  as in Cand and CancL, respectively. Cand, violates no constraints, thus is 3  optimal. When an input is bimoraic, it is possible for a low vowel and its following consonant to link to separate moras. In such a case the Harmonic Mora constraint  (HARMP ) 1  proposed  in (36) plays an important role in determining the optimal output. What H A R M requires is u  that the most sonorous segment makes the most harmonic mora (i.e. *u/C »  *u/Hi  »  *u7Lo). Tableau (57) illustrates how H A R M and its interaction with other constraints such u  as *Nuc/Hi and L A X T N G , give rise to an optimal output in a loose syllable.  153  Chapter Four  Jiang-King, 1996  (57) Output candidates for [siarj ^] 'line' 1  Input  Cand,  N  sian  h  Cand,  Cand,  Cand  k k /^\ J i  s  a  i]  i a  s  [siarj ^] 'line'  i]  s  [sia:rj]  i aq  4  s  i aq  [sian]  [sian]  HARM!"  *!'  *!  *NUC/HI  *!.  1  1  *!  LAXTNG  In (57), the last two candidates violate HARM*- in different ways. In Cand , the right mora 1  4  is filled by a consonant, while In Cand , the left mora is filled by a high vowel. The first 3  two candidates satisfy both HARM! and *Nuc/Hi by linking the low vowel to both moras. 1  However, the second candidate violates the length dependent constraint LAXTNG, because a doubly linked non-high vowel does not become [lax] in Cand . The first candidate satisfies 2  all constraints, and therefore is the best output. Since the optimal outputs in this pair satisfy all constraints, there is no crucial ranking in these cases.  4.3.5  The diphthongs vs. triphthongs  The correspondence between diphthongs and triphthongs in Fuqing differs from that between monophthongs and diphthongs in Fuzhou in that it involves two high vowels. In particular, the triphthongs occur ih the loose syllables only when there are two high vowels present. This is shown in (58) below.  154  Chapter Four (58)  Jiang-King, 1996  I "tight" a. tui  H  c. ts'iu™  Gloss  TI "loose"  Gloss  Distributions  g  'thump'  b. tuoi™-  ^  'team'  ui ~ uoi  ^  'autumn'  d. kieu™-  % 'uncle'  iu ~ ieu  Two properties emerge from the examination of the data in (58). One is that when an input contains two high vowels in the tight syllables, the corresponding loose syllables have the form of a mid vowelflankedby the two high vowels. The other one is that the mid vowel must agree in roundness and frontness with its preceding high vowel but not with its following high vowel Under the current theory, the first property is expected, since the tight/loose distinction is identified as a weight distinction between the monomoraic and bimoraic syllables. It is natural for a bimoraic syllable to contain one more segment than a monomoraic syllable. The second property, namely, the roundness and frontness harmony, is by no mean a unique property for this type of distinction. It is also found in the corresponding pairs uoi ~ uui and ieu ~ hsu in which uoi and ieu occur in the tight syllables. Comparing the data in (58) with the data in (51), it becomes clear that the harmonic restrictions for the feature [RD] and  [FRONT]  between a non-low nuclear  vowel and its preceding on-glide is a general property in Fuqing. It applies to both types of syllables. Therefore, the Feature Agreement Constraint ( F - A O R [ R D ] or [FRNT]) proposed in (52) for the harmonic restriction on the tight syllables also applies to the loose syllables here. Moreover, it is important to note that in a tight syllable, it should be possible for either i or u in (58a) and (58c) to link to the nuclear mora. Our account proposed for the data in (51) shows that a high vowel linked to a syllable node directly is disallowed by the constraint  *COD/HI,  constraint  *COD/HI  how the constraints FILL-U.  and that a mid vowel is analyzed as absence of +Lo and +Hi. The  is also applicable in the current case. The tableaux (59) and (60) show F-AGR[RD]  and  *COD/HI,  as well as their interaction with *Nuc/Hi and  are sufficient to derive the ui ~ uoi pair.  155  Chapter Four  Jiang-King, 1996  (59) Output candidates for [tuFJ 'thump'  Input  Cand,  /,  N  1  A  u t  u  Cand,  I 1 \ t u i | | |*M>1 [+FRT]  i  1 1  fa  [+HD)[+FRT1  [tui ] 'thump' 11  Cand,  /  A  /\  /\  t  u  Cand  i  | | [+BD1 [+FRT]  I  /N / 1  / 1  /  /  u  l i  [tu]  /t i i  [+RDJ+FRT1 .  *!  PARSERT  1  t u i 1 1  i  I+RD] l+FRT]  [tm]  I  / 1  / 1  t  4  *!  *!  *COD/HI  In (59), the last two candidates violate P A R S E R T , since one of the two high vowels is left unparsed in each of the candidates. Compare the first two candidates. Cand violates 2  *COD/HI  (that is, a high vowel is not allowed to link to a syllable node directly), while  Cand! does not. Therefore, thefirstone is optimal since it does not violate anything.  (60) Output candidates for [tuoi™-] 'team'  Input  Cand, 7  t  u  Cand,  Ai •A n  N  i  //i  A/'  'TV . fa  |*RD| |*F»T]  [tuoi -] 'team' 151  I  t a  1  i  1 [+RDJ f+FETl  [tuoi]  Cand,  1  A? A  t  u  1  i  1  [tui]  *!  *NUC/HI  *•!  156  4  I A :  / A 11  i  t u o i 1 1 I+EDJ l+FRT]  [t0i] *!  *COD/HI  FlLL-U  Cand  *!  Chapter Four  Jiang-King, 1996  In (60), the last candidate violates  *COD/HI  and *Nuc/Hi since the two high vowels u and i  are parsed onto the nuclear mora and the syllable node respectively: Candj violates  FILL-JA  while Cand, violates nothing. Therefore, thefirstcandidate wins. Since all constraints are respected in the optimal outputs, there is no crucial ranking for this pair;  4.3.6  A summary  Fuqing syllabification demonstrated in this section shows that the different vowel distribution pairs can be derived by linking the same sets of vocalic features to the distinctive syllable structures which are determined by the tonal specifications. Five types of vowel distribution between the tight and the loose syllables are captured in this section: (i) the high/mid contrast; (ii) the tense/lax distinction; (iii) the harmonic restrictions on the tight syllables; (iv) the asymmetric behavior of high vowels with respect to their different syllable positions; and (v) the correspondence between diphthongs and triphthongs in the case where the input contains two high vowels. Constraints invoked in Fuqing syllabification are summarized in (61) and their ranking is given in (62):  (61) Constraints for Fuqing syllabification a.  Faithfulness: F I L L - U ,  b.  Segmental sonority: *Nuc/Hi, * N u c / C ,  c.  Harmonic: H T A G R U U ,  d.  Parasitic: *Hr-uu,  PARSERT, PARSEHI, PARSELO *COD/HI, H A R M  U  F-AGR [RD], * H I / L O , U  LAXTNG  (62) Constraint ranking in Fuqing  L E X - U , F I L L - U , PARSERT, * N U C / C , * C O D / H I , * U / C , H T A G R U U , F - A G R [RD], * H I / L O H , H A R M , * H I - U U , LAXTNG, PARSEHI » U  PARSELO, * N U C / H I  157  Chapter Four  Jiang-King, 1996  4.4 Conclusion  I have motivated a set of constraints governing linking of vowels to the distinctive moraic structures in both Fuzhou and Fuqing. and demonstrated that the interaction of the constraints is successful in deriving the vowel distribution pairs in these two languages. The constraints governing syllabification in Fuzhou and Fuqing are of four types. The first type are the faithfulness constraints, such as the PARSE family (i.e. PARSERT, PARSEHI, PARSELO), the FILL family (i.e. Fnx-u), and the LEX family (i.e. LEX-U, LEX-F). The second  type of constraints are the ones encoding the intrinsic segmental sonority to certain syllable positions, such as *Nuc/Hi, * N u c / C , *COD/HI. The third type are the harmonic ones like HTAGRHH, F-AGR [RD], *Hi/Lol . This set of constraints requires certain featural l  agreement within a certain prosodic domain or prohibits conflicting features from occurring the same domain. The last type of constraints is the parasitic group, such as *HiHji and LAXTNG. This set of constraints is related to syllable length. They either prevent a high vowel from linking to two moras or assigning a certain feature to a segmental root doubly linked to two moras. The vowel distribution patterns in both Fuzhou and Fuqing are derived by the different rankings of the crucial constraints, as shown in (63) and (64), respectively.  (63) Fuzhou ranking *Nuc/C »  *Nuc/Hi »  *COD/HI  (64) Fuqing ranking * N u c / C , *COD/HI »  *Nuc/Hi  Notice that the majority of the constraints proposed are common in both languages. Both Fuzhou and Fuqing. for instance, invoke the same types of faithfulness, such as LEX 158  Chapter Four and  FILL  Jiang-King, 1996  which are undominated in these languages. The internal ranking of the  family is that both  PARSERT  and  PARSEHI  must dominate  PARSELO.  PARSE  Moreover, there is a  difference between these two languages in terms of ranking schema of the constraint restricting certain segments to certain syllable positions. Ih Fuzhou.  *Nuc/C  above *Nuc/Hi, which in turn ranks above * C O D / H I , whereas in Fuqing. both *COD/HI  must rank  *Nuc/C  and  dominate * N U C / H I . The third difference between these two languages in terms of  ranking is the relation between *Hi/Lol and L A X T N G . l  *HI/LO^  must rank above L A X T N G in  Fuzhou. while they are not crucially ranked with respect to each other in Fuqing.  159  CHAPTER 5  Stress effects on tone-vowel interaction  5.0 Introduction  Two asymmetries regarding tone sandhi and vowel alternation in Fuzhou and Fuqing emerge from the investigation of disyllabic words in chapter 2. One is the asymmetric behavior of syllables in different positions within a disyllabic domain. In particular, a syllable in a non-final position (i.e., a of a sequence Oja ) usually undergoes either tonal r  2  or both tonal and vocalic changes In contrast, the syllables in a final position (i.e., o of a 2  sequence o ^ ) do not change their lexical properties (either tone or vowel quality). The other asymmetry is the different behavior of the tight syllables and the loose ones in the very same non-final position. The loose syllables in a non-final position always undergo both tonal and vocalic changes, whereas the tight syllables in the same position never change their vowels, even though some of them do have tonal changes. Furthermore, there is an identical patterning between vowel alternations and distributions. That is, the vowel alternation patterns involved in the disyllabic words are the same as the vowel distribution patterns exhibited in monosyllabic words. Moreover, the tonal changes in the loose syllables are predictable. That is, the outputs of the changed tones must fall into the tonal categories in the tight syllables. Four questions arise from the observations described above, (i) Why do tone sandhi and vocalic changes only happen to a non-final syllable within a disyllabic domain? In other words, what is the difference between a final syllable and a non-final syllable in that domain? (ii) Why do only the loose syllables undergo the vocalic change but not the tight ones? (iii) Why do the vocalic changes in disyllabic words follow the vowel distributional 160  Chapter Five  Jiang-King, 1996  patterns in the monosyllabic words, namely, the changes are from the loose syllables to the tight ones, but not vice versa? (iv) Are tone sandhi and vocalic changes related to each other? If not, what triggers their changes? This chapter aims at providing an answer for these questions. First, I address the question of how stress affects tone-vowel interactions and identify what are the exact factors that govern the asymmetry of the syllables in different positions within the same domain and the asymmetry between the loose syllables and the tight ones in the same position ( i e , the non-final position). Second, I extend the analysis proposed for the vowel distributions in monosyllabic words to the vowel alternations in disyllabic compounds, and investigate whether the prosodic anchor hypothesis proposed in chapter 3 and the constraints on syllabification motivated in chapter 4 can account for the identical patterns between vowel distributions and alternations. Third, I explore the vowel alternations ih the reduplication forms, and provide an account for the vowel changes between a base and a reduplicant based on the prosodic anchor hypothesis and prosodic morphology within the OT framework (M & P 1993a, b, 1994, 1995). Last, I examine the similarities and differences between the reduplication forms and Fanqie words (i.e., the "cutting foot words"), showing that the apparent different outputs between these two types of forms (i.e., the full copy in the reduplications and the partial copy in the Fanqie words) lies in the interaction of alignment constraints and structural constraints. Thus, the vowel variations in various forms (i.e., disyllabic compounds, reduplications and Fanqie words) in Fuzhou and Fuqing can be uniformly explained.  5.1 How does stress affect tone-vowel interaction?  This section focuses on the asymmetric behavior of syllables with respect to their different positions within a disyllabic domain, and examines how stress actually affects tone-vowel interaction. By a close examination of the spectrographic evidence provided by Wright 161  Chapter Five  Jiang-King, 1996  (1983), I attempt to identify the exact factors that govern these asymmetries. I will then propose a set of constraints which incorporate Wright's insight and account for the asymmetries observed. I will further demonstrate how featural stability on the one hand and the lack of featural stability on the other hand can be achieved by the interaction of these constraints.  5.1.1  Identifying stress effects  The first asymmetry observed in chapter 2 is that the loose syllables in both Fuzhou and Fuqing behave differently with respect to the different positions in a disyllabic compound. They undergo both tonal and vocalic changes when they do not occur domain-finally. The tonal and vowel changes of the loose syllables in a non-final position are illustrated in (1). The data of Fuzhou disyllabic compound nouns are from Liang (1983a). The morphemes that undergo changes are emphasized by shading. The underscore indicates the tone and vowels that will undergo changes in the shaded column, and their corresponding outputs in the disyllabic column. The dots in the compounds signal syllable/morpheme breaks.  (1)  Morph.  Gloss +  Morph.  Gloss  ->  Disyllabic N  Gloss  a.  kign^^  'mirror'  suon  *box'  -»  kiarjH.nuon  'jewellery box'  b.  hourjMU&  -neck'  lien* ™  'chain'  ->  hurpUienMHM  'necklace'  H  1 1  11  The first morpheme (ie., the ones in the shaded column) of a compound in both (la) and (lb) contains a complex contour tone: that is, a M L M contour in (la) and a M H M contour in (lb). They are loose syllables. When they combine with a following morpheme to form a compound, the vowel [a] in (la) becomes its tense counterpart [a], and the diphthong [ou] in (lb) becomes its tight counterpart monophthongal [u] Furthermore, the complex tonal contours in both (la) and (lb) are simplified in the compounds. The M L M 162  Chapter Five  Jiang-King, 1996  contour in (la) changes into a H level tone, while the M H M contour in (lb) changes into a simple contour HM. Of interest is that not only the vowel changes are from the vowels in the loose syllables to the forms in the tight syllables, the output tones of the changed syllables also belong to the tonal categories in the tight syllables. The similarities between the syllables in a non-final position within a disyllabic domain and the tight syllables standing alone as monosyllabic words suggest that these two forms must have something in common. This kind of co-variation between tone and vowels in (1), however, does not show up in (2), where the loose syllables occur in a final position within a disyllabic compound. That is, the morphemes in the shaded cells become the second morphemes in the disyllabic compounds.  Morph  Gloss •+  a.  mi^  •rice"  b.  kieH  'platform'  tSO»HM  c.  tiar}M=  'pot'  p ierj -  *be bossy'  ts'uoi  d  Morph.  ,  Gloss  —> Disyllabic N  Gloss  •bag'  '-»  'rice bag'  miH.lDyMHM  'step'  ikU  M  MLM  'steps'  'thin piece' -»  tiarjM.mierj -  Kfl M  'lid, cover  •beak*  saRsuoi ^  1  'servant'  ->  1  1  Like the morphemes in the shaded cells in (1), the ones in the shaded column in (2) contain complex tonal contours (i e., either M L M or MHM), hence are loose syllables. When they occur domain-finally ih a disyllabic compound (i.e., the column after the arrows), their tonal contours and vocalic properties remain unchanged . 1  Comparing the lack of feature changes in the domain-final syllables in (2) with the changes of the tone and vowels in the non-final position in (1), it becomes clear that  1  Notice that the consonants [ts] and [ts'J in (2b, d) become [3] Jntervocalicalry, and [p'] in (2c) becomes  [ro] when it follows a nasal of its preceding morpheme. Since this thesis focuses its attention on tonevowel interaction, this kind of consonant change will not be discussed! 163  Chapter Five  Jiang-King, 1996  different positions within a disyllabic domain play a crucial role for the contrast between the featural stability on the one hand and the featural change on the other. The question that arises from these observations is what makes these positions different from each other. In other words, how can we explain the contrast between the featural stability in final position and the lack of featural stability in non-final position within the same domain? The other asymmetry observed in chapter 2 is that tight syllables in a disyllabic domain behave differently from the loose ones in the same domain. They do not change their vowels no matter whether they occur domain-finally or not. That is, the contrast between a non-final and a final position exhibited in (1) and (2) does not show up in (3), where both morphemes within a disyllabic compound are tight syllables. The Fuzhou data in (3) are from Liang (1983a).  (3)  Morph.  Gloss  a.  haP^  b.  a  c.  tseirj^  H  +  Morph.  Gloss  —> Disyllabic N  Gloss  'earthenware'  kuo  'pot'  -> hai^.kuo"  'earthenware pot'  'girl'  k'uaq™  'circle'  -> a^k'uaq ^  'maid'  'cut'  yrj  'flannel'  -» tseirjknyn™  'cotton flannel'  H  104  1  The morphemes in the compounds in (3) contain either a H level tone or a simple contour tone (I.e., H M or ML), hence they are tight syllables. This type of syllable does not exhibit vowel changes at all, no matter where the vowels occur. For instance, the compound noun [hai^ kuoH] 'earthenware pot' in (3 a) is comprised by the two morphemes [hai^s] 'earthenware' and [kuo* ] 'pot', whose vowels [ai] in the non-final position and [uo] in the 1  final position remain unchanged The same kind of vowel stability is also observed in (3b) and (3c). The question raised from the comparison of the lack of contrast in (3) with the contrast in (1) and (2) is that if the vowel changes were due to the different positions within a disyllabic domain, why don't the different positions In the same domain trigger 164  Chapter Five  Jiang-King, 1996  any vowel alternations for the tight syllables in (3)? The two asymmetries observed in (1), (2) and (3) can be summarized in (4).  (4)  Two asymmetries of syllables in a disyllabic domain  DOMAIN:  tight a  looseCT  tighter  looseo  VOWEL CHANGE  NO  YES  NO  NO  TONAL CHANGE  SOME  ALL  NO  NO  The chart (4) shows that the contrast between a and a in terms of their feature stability x  2  is quite straightforward. That is, the second syllable (i e., o ) within a disyllabic domain 2  never changes its lexical properties, no matter whether it is tight or loose. On the other hand, the change of features in the first syllable within a disyllabic domain is more complex. Different syllables in this position behave differently. In particular, the tight syllables do not have any vowel changes, even though some of them might have tonal changes Conversely, the loose syllables in this position always have both tonal and vowel changes. Our task then is to find an explanation for these asymmetries. Attempting to account for the tone sandhi and vowel alternations in Fuzhou. Wright (1983) conducted a series of experimental studies, and found that the length of the first syllable in a disyllabic word reduces nearly two thirds in duration from its citation form (i.e., the form which stands alone as a monosyllabic word). Some of the examples provided by Wright (1983) are given in (5) below. The duration in milliseconds of each monosyllabic morpheme is listed in the left shaded column The right shaded column indicates the duration of a disyllabic word with the numbers before 7" for thefirstsyllable and the ones after "/" for the second syllable. The dots in the disyllabic words indicate syllable and/or morpheme boundaries. The underscore indicates tonal and vocalic changes.  165  Chapter Five  Jiang-King, 1996  (5) Duration change for loose syllables in Fuzhou (Wright 1983:36-38)  morph  2  a.  kep  b.  koul  c.  top  ft 2  £ ft  Gloss  msec,  disyl. word  'record'  320  ki52.au  'rent'  400  ku^.tswo  facing'  368  t^.pi  12  2 2  12  Gloss  msec ar,/a  ftft  'mark, sign'  136/432  $  'rent house'  152/432  xttfc  'to contrast'  144/280  2  The morphemes in (5) contain a tonal category and vocalic forms that belong to the loose syllables. When they combine with a following morpheme to form a disyllabic compound, they change both their tonal contour and vowel features. Comparing the numbers in the two shaded columns, the monosyllabic morpheme [kejU-] 'record' in (5a) has a duration of 2  320 msec. When it co-occurs with a following morpheme in a disyllabic compound [kj52.au ] 'mark, sign', its duration reduces to 136 msec. The same pattern of duration 12  reduction is also observed in (5b) and (5c), where the morphemes [koul -] 'rent' and [toi* -] 2  2  'facing' reduce their duration from 400 msec, and 368 msec, to 152 msec, and 144 msec, respectively. Moreover, comparing the numbers before the slash "/" with the ones after the slash "/" in the last column, shows clearly that the duration of the first syllable is significantly shorter than that of the second syllable within the same domain. The data in (5) seems to suggest a correlation between duration reduction and the change of vowel features for the non-final syllables. A closer examination of the same type of data in (6), however, shows that the evidence for the apparent correlation between the shortening of syllable and the change of features in (5) is inconclusive. Cases where syllables in a non-final position reduce their duration without involving any vocalic change  2  The 12 tone in this column belongs to the category of Yin Qu, which is a MLM contour tone in our  representation.  166  Chapter Five  Jiang-King, 1996  are also found in Wright's spectrographic studies. The examples in (6) are drawn from Wright (1983).  (6) Duration change for tight syllables in Fuzhou (Wright 1983:36-38)  morph  Gloss  msec  disyl. word  %  'fly'  304  hi^.ki  3  Gloss  msec. o,/o  ^  'airplane'  112/280  a.  hi  b.  ku^2  $  'paste'  400  ku^.tsai  22  l/flfc  'paste paper'  144/366  c.  ts0y52  $  'to cut'  456  t0y22.pun  %M  'tailor*  128/336  44  44  52  2  All morphemes in (6) have tones that belong to the tight syllables. Comparing the two shaded columns, the duration of the morpheme [hi ] 'fly' in (6a) is 304 msec. When it 44  combines with another morpheme to form a disyllabic compound [hi^.ki ] 'airplane', its 44  duration reduces to 112 msec, nearly one third of its duration as a citation form. The same shortening effect is also observed in [ku^ -] 'paste' (6b) and [ts0y52] 'to cut' (6c), 2  where the duration for these monosyllabic morphemes reduces from 400 msec, and 456 msec, to 144 msec! and 128 msec., respectively. Note that, unlike the data in (5) where the shortening of duration is accompanied by the vocalic changes, the vowels in the non-final syllables in (6) do not undergo any change, even though their duration has been shortened and some of them have tonal changes, such as the examples in (6b) and (6c). Comparing (5) with (6), reveals that the vowel alternations in a disyllabic compound relate to, but do not necessarily result from, the duration reduction. Based on these findings, Wright claims that the tone sandhi and the vowel alternations in Fuzhou are independently triggered by stress, and that stress in Fuzhou is assigned to an  3  The 44 and 53 tones in this column belong to the categories of Yin Ping and Yang Ping respectively.  They are H level and HM contour tones in our representation.  167  Chapter Five  Jiang-King, 1996  iambic foot with the final syllable being a metrical head The stress effect on tone sandhi is more direct. That is, the shortening of an unstressed syllable (or "weakly stressed syllable" in Wright's term) in a disyllabic domain gives rise to the loss of a mora, hence triggering tone sandhi. However, the relation between the shortening of the non-final syllables and the vowel alternations is not quite obvious. Wright proposes a constraint that prohibits an unstressed syllable from having a branching nucleus As a result, only syllables violating this constraint undergo the vocalic changes, whereas the ones which do not violate this constraint do not undergo any vocalic change. Wright's proposal focuses on the change of the tones and vowels in the non-final syllables (i.e., the unstressed syllables), hence is only partially successful in accounting for the contrast between the change of tones and vowels in a non-final position and the lack of changes in a final position. As for the featural stability exhibited in the final syllables, one could infer from Wright's proposal that since stress is assigned to an iambic foot, the final syllable of a disyllabic domain always bears a stress, hence neither loss of mora nor violation of the constraint (that disallows a branching nucleus) takes place in a stressed syllable. Therefore, the tonal and vowel features must be retained in a domain-final syllable. Thus the first asymmetry, that is, the contrast between the different positions within a disyllabic domain is accounted for explicitly and implicitly by Wright's proposal. Although Wright's proposal says nothing about the contrast between the lack of the vowel changes in the tight syllables and the change of vowel in the loose ones in the same non-final position, as shown in (4), and the similarities between the vowel alternation in the disyllabic compounds and the vowel distributions in the monosyllabic words, she does provide strong evidence showing that the difference between the two positions in a disyllabic domain is their duration. Phonetically, duration is diagnosed as one of the strongest correlates of stress (Fry 1958, 1976). Phonologically, vowel quality or other segmental features often interact with stress, such that schwa in some languages is stressless (Hayes 1991, 1995). The Fuzhou data examined above provide another type of 168  Chapter Five  Jiang-King, 1996  phonological diagnostic for stress. That is, the featural content in the stressed syllable (ie., the domain-final syllable) is more stable than that of the unstressed syllables within a disyllabic domain. Based on the observations from the Fuzhou data in (1), (2) and (3), as well as the experimental evidence provided by Wright (1983) in (5) and (6), I build on Wright's insight and propose that the stress effect on tone-vowel interaction is twofold On the one hand, it preserves all lexical properties in the stressed syllable; on the other hand, it reduces the weight of an unstressed syllable, hence triggering various tonal and vocalic changes. To assign prominence to the final syllable within a disyllabic domain, I propose Prominence Alignment (7) which assigns a metrical grid mark to the rightmost syllable of a disyllabic domain, and Prominence Reduction (8) which prevents an unstressed syllable from being bimoraic.  (7) Prominence Alignment ( P R O M A L I G N ) Given a domain x, a metrical grid mark must be aligned to the right edge of x. x = morphonological word.  (8) Prominence Reduction ( P R O M R E D U C ) If a is not assigned a metrical grid mark, a cannot be bimoraic.  The function of  PROMREDUC  is to require a non-prominent syllable to, be monomoraic,  triggering both tonal and vowel changes in the non-final syllable, if that syllable is bimoraic in the first place. This constraint reflects the spirit of Prince's (1990:358) "Weight-toStress Principle" in that it makes the weak weaker. It must be pointed out that the constraints proposed above aim to account for one of the asymmetries summarized in (4). That is, the contrast between the feature stability in a  2  and the tonal and vowel changes in o,. The other asymmetry in (4), namely, the contrast 169  Chapter Five  Jiang-King, 1996  between the tonal and vowel changes in the loose syllables and the lack of vowel changes in the tight ones follows from our proposal accounting for the tonal and vowel distributions in the monosyllabic words. It will be discussed in detail in the latter part of this chapter.  5.1.2  Asymmetric behavior of syllables in Fuzhou  Having proposed the relevant constraints for the contrast between different positions within the same domain, I now proceed to demonstrate how these constraints and their interaction with the constraints on tonal distribution can successfully derive (i) the featural stability in final position, (ii) the possible tonal and vowel changes in non-final position of the same domain, and (iii) the identical tonal partem between the output of the tone sandhi in the loose syllables and the output of the tonal distributions in the tight syllables. Fuzhou tonal distributions demonstrated in chapter 3 show that there are two types of syllables: tight and loose. The former are monomoraic, and the latter bimoraic. Their combinations in a disyllabic domain give four possibilities: (i) loose-loose, (ii) loose-tight, (iii) tight-loose, (iv) tight-tight. In the following, I show how the surface forms for each of the four pairs of syllables in a disyllabic domain can be achieved by the constraints proposed in the last section and the constraints on tonal distributions proposed in chapter 3. I assume that the  PROMAUGN  (which assigns prominence to the rightmost syllable  within a domain) is highly ranked; there are no constraints higher than  PROMALIGN  that  could cause it to be violated. Outputs violating this constraint will therefore not be included in tableaux below. The square brackets without a subscribed symbol "a" indicate the head of a syllable, namely, the nuclear mora.  170  Chapter Five  Jiang-King, 1996  (9) Loose-loose compounds in Fuzhou  "Input"  Cand,  4  TTT  Cand,  T T T  T TTTTT  h  V I VI II  RTRT  V 1  V  II RTRT  •. V  [M-y-MuL.  •'  1  K  RTRT  RTRT  IM [ C  .  V1 MuL.  K 1 1  RTRT RTRT  Cand, TTT  VI  T T T  VI  RTRT  RTRT  1  *!  PROMREDUC  *!  HDBIN  *  PARSETN  The input in tableau (9) contains two bimoraic syllables. It is shown that violation of PROMREDUC  or  HDBIN  (which Is motivated by the tonal distribution In Fuzhou. see the  detailed arguments in section 3.3.1) is fatal since these constraints are highly ranked. Cand violates P R O M R E D U C because the unstressed syllable has a bimoraic structure. The 3  first two candidates satisfy  PROMREDUC  in the same way. That is, the non-final syllable in  both of them lost one of their two moras, hence is monomoraic. The difference between them is that the tone left by the loss of the mora in Cand is parsed onto the remaining 2  mora, resulting in a H D B I N violation, whereas the tone left by the loss of mora in Cand, stays unparsed, a violation of P A R S E T N , the lowly ranked constraint. Cand,, therefore, is optimal. The crucial ranking for the loose-loose compound in Fuzhou is HDBIN »  4  PROMREDUC,  PARSETN.  By "Input**, I mean the form that would be optimal on its own. It is not necessarily an input in the sense  of "underlying representation".  171  Chapter Five  Jiang-King, 1996  It is important to notice that by satisfying P R O M R E D U C , there is a concomitant violation of Faithfulness: namely, the bimoraic structure of the first syllable in a disyllabic form becomes monomoraic in both (9) and (10). (10) differs from (9) in that its input is a loose-tight compound rather than loose-loose one. The last candidate violates the faithfulness constraint Lroqi (which prohibits insertion of any F-element that is not present in the input) because the domain-final syllable with a single mora in the input becomes bimoraic. Cand violates  because the  PROMREDUC  3  unstressed syllable keeps its two moras. Comparing the first two candidates, Cand violates  HDBIN  by linking three tones to one mora, while Candi violates  leaving a tone unparsed. Since  PARSETN  2  by  is the lowest ranked constraint, Candj wins.  PARSETN  Again, the ranking established in (10) also applies to this case.  (10) Loose-tight compounds in Fuzhou  "Input" TTT  Cand,  T(T)  V 1 V •II RTRT  N RT(RT)  T(T)  V  V  K  b  Cand,  TTT  K  R T R T RT(RT)  TTT  T(T)  v/ V  TMLIML  K  N  RTRT RT(RT)  Cand, T(T)  VI  V  I 1 RTRT  K  T T  V  4  TTT  V 1  [My[u]n] RT  a  RT RT  RTTRT)  *!  LEXU  *!  PROMREDUC  *!  HDBIN PARSETN  Cand  TTT  *  Both (11) and (12) have one property in common, that is, the non-final syllable in both cases is monomoraic. Their difference lies in the final syllable. It is a loose syllable in (11) and a tight one in (12). The last two candidates in each of them incur two violation marks  172  Chapter Five for  PARSE.  Jiang-King, 1996  This time, it is the tone and the segmental root in the non-final syllable that get  unparsed Only the first candidate in each case incurs no violations, hence is optimal.  (11) Tight-loose compounds in Fuzhou  "Input" T(T)  V  Cand,  TTT  T(T)  V !  Cand,  TTT  V  V 1  fM] [MKl 0  RTCRT) RT RT  <b  K M  T(T)  TTT  \  V 1  0  RTCRT) RT RT  Cand, T(T)  TTT  V  V  [Ml [Mn]  K  0  1  RT(RT) RT RT  RTCRT) R T R T  *|*  *|*  0  PROMREDUC PARSETN  (12) Tight-tight compounds in Fuzhou  "Input" T(T)  V K  Cand,  T(T)  T(T)  V  V K  RT(RT) RT(RT)  Cand,  T(T)  V  [M1JML  K  b  N  RT(RT) RT(RT)  T(T)  \  T(T)  V  IMUML RT(RT) RTCRT)  Cand, T(T)  TCV)  V  \  RT(RT) RT(RT)  PROMREDUC PARSETN  *|*  Notice that the constraints PROMREDUC and H D B I N db not play any role in determining the optimal output in (11) and (12) since the inputs in these cases contain a tight syllable in the non-final position. I have shown in this section that regardless of what combinations of the two types of syllables are in a disyllabic domain, the rightmost syllable within this domain always keeps its lexical properties. This is because the alignment constraint  PROMALIGN  and  LEXJI  are  highly ranked. Any loss of lexical properties (either prosodic or featural) would be 173  Chapter Five  Jiang-King, 1996  prevented. The asymmetrical behavior between the loose syllables and the tight ones in the non-final position can be accounted for by the constraint  PROMREDUC  in the following way.  Since the input for a loose syllable contains two moras while a tight one only contains a single mora, the former violates  PROMREDUC,  hence triggering tonal and vowel changes,  whereas the latter does not violate P R O M R E D U C , therefore, no vowel changes take place. I also demonstrate that the identical output tone sandhi between the non-final loose syllables and the tonal distributions in the tight syllables can be achieved by the interaction of the constraints on stress effects, namely, tonal distributions  HDBIN  P R O MA L I G N  and  PROMREDUC  with the constraint on  proposed in chapter 3, as well as the faithfulness constraint  PARSETN.  5.1.3  Asymmetric behavior of syllables in Fuqing  Fuqing disyllabic compounds also exhibit the same types of asymmetries as those observed in Fuzhou. First, the loose syllables, i.e., the syllables with a falling tonal contour containing a L tone, and undergo both tonal and vocalic changes when they combine with a following morpheme to form a disyllabic compound. In other words, when a loose syllable occurs in a non-final position, its vowel changes into the corresponding tight form, and its tone also becomes a tone belonging to the tight categories. These kinds of tonal and vocalic changes in the loose syllables are illustrated in (13).  (13)  Morph  Gloss  ->  Disyl. word  Gloss  'ear'  pa  'pick'  ->  qiHpa"  'earpick'  b. pon^  'excrement'  t'0n  c. t0?^  'bamboo'  ts'yo™*  a.  Morph  Gloss  ne_a  +  H  M  'bucket' ->  purpit'0n  'mat'  typats'yoHM  174  ->  M  |$j§  'manure bucket'  ftjft  "bamboo mat'  Chapter Five  Jiang-King, 1996  The morphemes in (13) all have a contour tone that is comprised of a non-L tone (i.e., either H or M) and a L tone. Therefore, they belong to the loose type of syllables, and hence are bimoraic. When they occur in a non-final position, their vowel [e], [o] and [0] become the corresponding tight forms [i], [u] and [y], respectively. Meanwhile, their tones also get changed, and the tonal changes are precisely the loss of the L part of the entire tonal contour. For instance, the morpheme [ne™^] 'ear' in (1-3a) has a contour tone HL. When it occurs in the non-final position within a disyllabic compound [rjiHpa ] 'earpick', 11  its original tonal contour gets simplified and becomes a H level tone; Its vowel [e] changes into the corresponding tight form [i]: The same patterns of tonal and vocalic change are also observed in (13b) and (13c).  (14)  Morph  Gloss +  Morph  a. tieqa  'electricity' pieu  M  b. norj^  'tender'  muoi  ML  c. pugrj^  'half  ni?  H  Gloss  ->  Disyl. word  'watch' —> tierj pieu H  Gloss 'meter'  M  'sister'  -» norj ^mu3i  'young sister'  'day'  ->  'half a day'  ffi  ML  puan ni? H  H  As in (13), the morphemes in the left column in (14) contain a tone with a L tone as the second part of the contour, hence they are loose syllables with a bimoraic structure. When they occur in a non-final position in a disyllabic compound, their main vowels [e], [o] and [a] become the corresponding tight forms [e], [o] and [a], respectively. Meanwhile, the L tone in their original tonal contour gets lost. For example, the monosyllabic morpheme [tien^] 'electricity' in (14a) changes into [tierjH] when it cooccurs with another morpheme in a disyllabic compound ftienH pieuM] 'meter'. The vowel change in this case is from [e] to [e], and the tonal change is from H L to H, that is, the loss of the L part of the entire tonal contour. Compare the two morphemes [norj^] 'tender' and [muoi^] 'sister' in (14b), they both have the same M L contour tone, and their vowels include [o], hence they are loose syllables. When they occur together as a 175  Chapter Five  Jiang-King, 1996  disyllabic compound fnon^ muoi^] young sister , the first morpheme changes its vowel 1  from [o] to [o], and its tone from M L to H M The tonal and vowel changes in the first morpheme, however, do not happen to the second morpheme, the one which occurs domain-finally. This contrast shows clearly that different positions within a disyllabic domain play a crucial role in determining the featural change on the one hand, and the featural stability on the other. However, the tonal and vowel changes of the non-final morphemes exhibited in (13) and (14) are not found in (15), where the non-final syllables in the disyllabic compounds are originally tight. In particular, the vowels [e], [o] and [a] in the tight syllables in (15a), (15b) and (15c) do not change into their corresponding forms [i], [u] and [a]. Their tonal change is not the same type of change observed in (13) and (14).  (15)  Morph Gloss + Morph  Gloss  -»  Pisyl word  a.  se  -»  se^ts'iu ^  b. c.  'wash'  ts'iu ™  'hand'  ton™  "become'  rja™"  'family' ->  torjHna™  ha?  'combine'  pa?  'arm'  haP^pa?  M  H  1  1  2  ->  1  2  1  Gloss ^  'wash hands'  ^jgj  'decision-maker'  ^  'purse'  Comparing the lack of vowel change in (15) with the vowel alternations in (13) and (14), it becomes clear that different types of syllables behave differently with respect to their moraic structures. The vowel alternations take place only in the bimoraic syllables occurring in a non-final position, whereas no vowel alternations happen to the monomoraic syllables, whether they occur domain-finally or not. This finding is compatible with the constraints proposed for Fuzhou disyllabic compounds. In particular, the non-final syllable in a disyllabic domain is unstressed, hence subject to the constraint Prominence Reduction, which requires an unstressed syllable to be monomoraic. A bimoraic syllable in that position must be made monomoraic syllable, triggering the vowel change. A tight syllable, however, is monomoraic; and does not violate this constraint; 176  Chapter Five  Jiang-King, 1996  hence it need not change its vowel As for the contrast between a loose syllable in different positions within a disyllabic domain, it can be accounted for by the constraint Prominence Preservation, which ensures that all lexical properties of a stressed syllable, i.e., the final syllable, must remain unchanged. Now we have clearly identified (i) the contrast between the loose syllables in different positions in a disyllabic domain, (ii) the contrast between the loose syllables and the tight ones in the same position of a disyllabic domain, and (iii) the identical patterning between tone sandhi and vowel alternations in the loose syllables, and the tonal and vowel distribution in the tight ones. In the following, I will demonstrate how the constraints proposed for Fuzhou disyllabic compounds can be extended to the Fuqing cases, accounting for the same types of asymmetries and identities between tonal and vowel distributions in the tight syllables and their alternations in the loose ones.  (16) Loose-loose compound in Fuqing  "Input" T L  II  Cand,  T L  T L  II  II  II  RTRT  RTRT  1  , ; &  Cand,  T L  1 1  TL  V  Cand,  T L  T L  [[UJJMHL, K  1  1  RT RT RT RT  N  T L  II  1 1  II  [MuL[[u]nL  II  II  RTRT RTRT  II  RTRT  PROMREDUC  RTRT  *!  *NUC/[-RSD1  *'!  *  PARSETN  The input in tableau (16) contains two bimoraic syllables. It is shown that violation of PROMREDUC  or  *NUC/[-RSD]  is fatal since these constraints are highly ranked. Leaving  everything in the input unchanged, as in Cand , violates PROMREDUC since an unstressed 3  syllable (i.e., the non-final syllable) has two moras. The first two candidates satisfy  177  Chapter Five PROMREDUC  Jiang-King, 1996  in the same way. That is, all lexical properties in the rightmost syllable are left  unchanged, and the non-final syllable in both of them has lost one of their two moras, hence is monomoraic. The difference between them is that the L tone left by the loss of mora in Candj is parsed onto the remaining mora, resulting in a  *NUC/[-RSD]  violation,  whereas the L tone left by the loss of mora in Candj stays unparsed, a violation of PARSETN.  Since * N U C / [ - R S D ] ranks above P A R S E T N , Cand therefore, is the optimal one. As l5  for the loose-loose compound in Fuzhou, the crucial ranking for the loose-loose compound in Fuqing is P R O M R E D U C ,  *NUC/[-RSD]  »  PARSETN.  (17) Loose-tight compounds in Fuqing  "Input" T L  I  !  II  RTRT  Cand,  TCT)  T L  Cand,  T(T)  1/  V  V [M-yiMiL  N  RT(RT)  &  Cand,  T I. TCT)  R T R T RT(RT)  K  N  RTRT RT(RT)  :  T L  TCT)  i i  v  11 RTRT  K  RTfRT)  *!  PROMREDUC  *!  *NUC/F-RSD1  *  PARSETN  (17) differs from (16) in that its input is a loose-tight compound rather than a loose-loose one. As in (16), Cand violates 3  PROMREDUC  because the unstressed syllable (i.e., the left  syllable) is bimoraic, and so is out. Comparing the first two candidates, Candj violates *NUC/[-RSD]  by linking the L tone to the nuclear mora, while Cand, violates P A R S E T N by  leaving the L tone unparsed. Since P A R S E T N is the lowest ranked constraint, Candj wins. Again, the ranking established in (16) also applies to this case.  178  Chapter Five  Jiang-King, 1996  (18) Tight-loose compounds in Fuqing  "Input" T(T)  V  Cand,  T L  T(T)  V  1 1  [•MyMuL,  K  1  1  [ M y MuL  K  1 1  RT(RT) RT RT  Cand,  T L  h  1  I-  RT(RT) RT RT  T(T)  \  Cand,  T L  IM y  T(T)  1 I  M u y  1  [ M y M u ]  II  1  T L '  V  N  c  1  RT(RT) RT RT  RTTRT) RT RT  *|*  *|*  PROMREDUC PARSE  Both (18) and (19) have one property in common, that is, the non-final syllable in both cases is monomoraic. Their difference lies in the final syllable. It is a loose syllable in (18) and a tight one in (19). The last two candidates in each of them incur two violation marks for  PARSE.  This time, it is the tone and the segmental root in the non-final syllable that get  unparsed. Only the first candidate in each case violates nothing, and hence is optimal Notice that the constraint  PROMREDUC  plays no role in determining the optimal output in  (18) and (19) since the "Input"s in these cases contain a tight syllable, hence are monomoraic in the non-final position.  (19) Tight-tight compounds in Fuqing  "Input" T(T)  V  T(T)  V  I M U M L  K  K  RT(RT) RT(RT)  Cand, T(T)  V  Cand,  T(T)  V  [ M y M L  K  K  RT(RT) RT(RT)  T(T)  \  T(T)  V  [M1JML 1 K  RT(RT) RTTRT)  Cand,  •  T(T)  T(T)  V  \  [[u]] [ML 0  RT(RT) RT(RT)  PROMREDUC  *|*  PARSE  179  Chapter Five «  ma  t  Jiang-King, 1996  I have shown in this section that no matter what combinations of the two types of syllables are in a disyllabic domain, the rightmost syllable within this domain always keeps its lexical properties. This is because the alignment constraint P R O M A L I G N is highly ranked. The asymmetrical behavior between the loose syllables and the tight ones in the non-final position can be explained by the constraint  PROMREDUC,  since the input for a loose syllable  contains two moras while a tight one only contains a single mora. The former violates PROMREDUC,  hence triggering tonal and vowel changes, while the latter does not violate  PROMREDUC,  hence no vowel changes take place.  I  also demonstrate that the identical  output tone sandhi between the non-final loose syllables and the tonal distributions in the tight syllables can be achieved by the interaction of the constraints on stress effects, namely, P R O M A L I G N and P R O M R E D U C with the constraint on tonal distributions proposed in chapter  5.1.4  3,  as well as the faithfulness constraint  *NUC/[-RSD]  PARSE.  A summary  I have identified two asymmetries regarding syllables in different positions within a disyllabic compound. The first one is the contrast between the featural stability in a final syllable and the lack of featural stability in a non-final syllable. The spectrographic data provided by Wright (1983) show that the duration of a non-final syllable is significantly shorter than one in a final position. This shortening effect in the non-final position is characterized as a stress effect which causes tone sandhi and vowel change in an unstressed syllable (Wright  1983).  Following Wright's insight, I propose two constraints,  Prominence Alignment and Prominence Reduction, to account for the contrast between the featural stability on the one hand and the featural change on the other. The second asymmetry is that different types of syllables in the same non-final position of a disyllabic compound behave differently. The loose syllables in this position always change their tone and vowels, and the tone sandhi and vowel alternations for these 180  Chapter Five  Jiang-King, 1996  types of syllable are identical to the tonal and vowel distributions observed in the tight syllables. On the other hand, the tight syllables in the very same position never change their vowels, even though some of them do have tonal changes. This contrast between the loose and tight syllables in the same position is identified as being due to their different moraic structures. That is, the loose syllables are bimoraic, hence subject to the constraint Prominence Reduction which requires an unstressed syllable to be monomoraic. On the other hand, the tight syllables are monomoraic; hence not subject to Prominence Reduction; therefore, do not have any vowel change. My proposal differs from Wright's account in that it encodes the different moraic structures between the two types of syllables into the constraint on an unstressed syllable, namely, Prominence Reduction, correctly predicting that only the bimoraic syllables (i.e., the loose syllables) undergo vowel changes but not the monomoraic syllables (i.e., the tight ones). Since the different moraic structures for the different types of syllables have already been motivated by their tonal distributions in chapter 3, and this structural difference affects their vowel distributions in monosyllabic words, it is expected that the vowel alternation in the bimoraic syllables of the disyllabic compounds would be identical to the vowel distributions in the tight syllables of the monosyllabic words; Thus, we achieve a unified account for the vowel distributions and alternations in both monomoraic and bimoraic forms. Moreover, I have shown that the striking similarity between the tone sandhi in the loose syllables and the tonal categories in the tight syllables lies in the interaction of the constraint Prominence Reduction with the constraints on the tonal distributions H D B I N and  *NUC/[-RSD],  as well as the faithfulness constraint P A R S E T N .  5.2 Vowel alternations in disyllabic compounds  In the last section I proposed a set of constraints on stress, successfully predicting the contrast between the featural stability in the final syllables on one hand, and the featural 181  Chapter Five  Jiang-King, 1996  change in the non-final syllables on the other hand. In particular, the feature stability is guaranteed by Prominence Alignment (which assigns a final syllable) and  PARSE  (which  prevents a final syllable from losing its lexical properties), whereas the featural change is triggered by Prominence Reduction which requires an unstressed syllable to be monomoraic. In this section, I focus on the identical patterning between the vowel alternations in the loose syllables that do not occur domain-finally and the vowel distributions in the tight syllables, and answer the question of why the vocalic changes involved in the non-final syllables always fall into the vowel distributional patterns between the loose syllables and the tight ones in monosyllabic words. In other words, the outputs of the vocalic changes in the non-final position are always the forms occurring in the tight type of monosyllabic words, and not vice versa. Under the prosodic anchor hypothesis proposed in chapter 3, the tight-loose distinction is argued to be a structural difference (see chapter 3). That is, the loose syllables are bimoraic, while the tight ones are monomoraic If the effect of Prominence Reduction is to require the non-final position to be monomoraic, then the identical patterning between the vowel alternations and the vowel distributions is expected. Since the tight syllables and the syllables that undergo changes in the non-final position have identical moraic structure (i.e., monomoraic), their featural similarities are no longer a mystery. The task of this section is to demonstrate how the interaction of the sets of constraints, that is, the constraints on stress proposed in the last section and the constraints on syllabification proposed in chapter 4, can successfully account for the identical patterning between the vowel alternations in the disyllabic compounds and distributions in the monosyllabic words.  182  Chapter Five  Jiang-King, 1996  5.2.1 Fuzhou disyllabic words  As shown in section 5.1.2, Fuzhou disyllabic words exhibit three characteristics. First, syllables in a non-final position in a disyllabic domain undergo certain changes, whereas syllables in a final position of the same domain never change anything. Second, syllables in the same non-final position behave differently with respect to their moraic structure. In particular, loose syllables (i.e., bimoraic syllables) always change both tone and vowels, while tight syllables (i.e., monomoraic ones) never have any vocalic change, even though they may or may not change their tone depending on the particular tonal configuration. Third, the vowel alternations in non-final loose syllables are identical to their corresponding tight forms in the monosyllabic words. These properties are illustrated again in (20).  (20)  Morph.  Gloss  +  Morph.  Gloss  a.  hourj^  'neck*  lien^  b.  hai*^  'earthenware' kuo  H  -»  Disyllabic N  Gloss  'chain* ->  hurja^lien^  'necklace'  'pot'  hai^.kuo  'earthenware pot'  -»  11  The disyllabic word [hurj^.lienM * ] 'necklace' in (20a) is composed of two monosyllabic 11  1  morphemes [hourj!^] 'neck' and [lierj ^ ] 'chain'. They both contain a complex tonal 1  1  contour that belongs to the loose type of syllables, and hence are bimoraic. Interestingly, the output of the disyllabic word shows that the tonal and vocalic features in the first syllable of this disyllabic word differ from the tonal contour and the vowels in the morpheme [hounM^j 'neck'. The diphthong [ou] becomes a monophthongal [u], and the M H M tonal contour becomes HM. On the other hand, the tonal contour and the vowels in the second syllable of this disyllabic word remain unchanged. Recall that an underlying high vowel lui surfaces as a monophthongal [u] in a tight syllable and as diphthong containing the high vowel [ou] in a corresponding loose syllable; it becomes clear that the 183  Chapter Fixe  Jiang-King, 1996  change from [ou] to [u] in (20a) is exactly the same as the corresponding pair u ~ ou in monosyllabic words. This identical relation between the vowel alternations in disyllabic words and the vowel distributions in monosyllabic words can be explained easily under the theory proposed in this work. That is, since the tight syllable is monomoraic and the loose syllables bimoraic, linking of an underlying high vowel IvJ to the distinctive moraic structures would give rise to the monophthongal [u] in a tight syllable on the one hand, and the diphthong [ou] in a loose syllable on the other hand. Now the question is why does the diphthong [ou] in a non-final loose syllable become [u], but not *[o]? The answer is that the constraint Prominence Reduction only allows a monomoraic syllable in that non-final position, forcing the loose syllable in that position to lose a mora. As a result, the [ou] which underlyingly is a high vowel /u/ becomes a monophthongal [u] but not [o]. In other words, the stress effect forces a non-final loose syllable to lose a mora, giving rise to a monomoraic structure which is the same as the moraic structure for the tight syllables, therefore, the vowel alternation pattern u ~ ou follows. The following tableau (21) shows how the change from ou to u can be achieved by the interaction of the two sets of constraints, that is, the constraints related to stress, i.e., P R O M A L I G N and the ones on syllabification, such as  PARSEHI.  PROMREDUC,  and  Following the conventional notation for  representing stress, the x's on top of a's stand for stressed syllables, while the dots on top of cr's indicate unstressed syllables The input in (21) contains two bimoraic morphemes (i.e., two loose syllables). The last candidate violates  (which requires a metrical grid to be aligned to the  PROMALIGN  rightmost syllable within a word) since the prominence indicated by the "x" on top of the syllable node is aligned to the left, resulting in a misplacement of stress. Leaving everything unchanged, as in Cand , violates P R O M R E D U C , 3  SO  Cand is out. Comparing the 3  first two candidates, both satisfy the stress-related constraints. The difference between them is that the high vowel [u] which stands for the root node containing a feature [+Hi] gets parsed onto a prosodic anchor in Cand^ while in Cand it is left unparsed, resulting in 2  184  Chapter Five  Jiang-King, 1996  violation of P A R S E H I . Since Candj satisfies all constraints, it is the optimal output. There is no crucial ranking in this case.  (21) Output candidates for [hurj^.lieq ^] 'necklace' in Fuzhou 1  "Input"  Cand,  i i  AA h ou q 1 i e  Cand,  ( •  q  hourjkfflM 'neck' + lienM™ 'chain'  x)  :  ( •  I I  Cand,  x)  ( •  I I  Cand (x  x)  AA AA AA  h u  q 1i e  q  [hurja^.lienMH^ ] 1  h eftpq l i e  q  [hon.lien]  h ou q1le  q  [hourj.lien]  'necklace'  4  .)  kk I  I  h ou q 1 ie  q  [hourj.lien] *!  PROMALIGN PROMREDUC  '*!'  *!  PARSEHI  Notice that once the constraints related to stress have been satisfied, the output vocalic form is determined by the constraints on syllabification, which are the same as that used for the vowel distributions in monosyllabic words. This explains why the output of vowel changes in a loose syllable is always identical to the output of the vowel distributions in a tight syllable. An examination of the data in (20b) shows that when a disyllabic word [hai^.kuo ] 11  'earthenware pot' is comprised of two morphemes [haP^J 'earthenware' and [kuo"] 'pot' belonging to the tight type of syllables, neither the first morpheme nor the second one has any vocalic change. The lack of vowel alternation in either the final or the non-final position for the tight type of syllables is expected under the current theory, since there is no structural change involved. This is demonstrated in tableau (22) below.  185  Chapter Five  Jiang-King, 1996  (22) Output candidates for [hai^.kuo ] 'earthernware pot' in Fuzhou 11  "Input"  Ai  ha  i k uo  'earthernware' + kuo 'pof  hajHM  H  Cand,  Cand, (.  7  X)  Al  ha  f  i k uo  [hai^.kuo *] 'earthernware pot'  Cand,  M il (•  x)  7  7  h a ik <ite  [hai.ko]  1  PROMALIGN  (x  .)  1  ha  1  i k uo  [hai.kuo] *'!  *!  PARSE  The last candidate in (22) violates P R O M A L I G N because the prominence is aligned to the left rather than to therightmostsyllable of the word. The middle candidate satisfies  PROMALIGN  by aligning the stress to therightmostsyllable of the word. However, it violates PARSE since the high vowel [u] in the final syllable is left unparsed. Only the first candidate violates no constraint, and is optimal. Notice that the constraint P R O M R E D U C is not violated in any member of the candidate set, since the morphemes involved are all monomoraic.  5.2 2 Fuqing disyllabic words  As shown in Fuzhou. Fuqing disyllabic words also exhibit three characteristics. First, syllables in a non-final position in a disyllabic domain undergo certain changes, whereas syllables in afinalposition of the same domain never change anything. Second, syllables in the same non-final position behave differently with respect to their moraic structure. In particular, loose syllables (i.e., bimoraic syllables) always change both tone and vowels, while tight syllables (i.e., monomoraic ones) never have any vocalic change, even though they may or may not change their tone. Third, the vowel alternations in non-final loose  186  Chapter Five  Jiang-King, 1996  syllables are identical to their corresponding tight forms in the monosyllabic words. These properties are illustrated here in (23) for convenience.  (23)  Morph  Gloss +  a.  nor$& 'tender'  b.  toq ^ 12  'become'  Morph  Gloss  ->  Disvl. word  Gloss  muoi^  'sister'  -»  noq^muai^ -  qa™  'family' —»• toqHna™^  11  jjfy^  "young sister' 'decision-maker'  The disyllabic word [noq^ muoi ^] 'young sister' in (23 a) contains two monosyllabic 1  morphemes [noq^] 'tender' and [muoi ^] 'sister'. They both have a tonal contour that 1  includes a L tone, hence are loose syllables (i.e., bimoraic). The output of the disyllabic word shows that the tonal and vocalic features in the first syllable of this disyllabic word differ from the tone and the vowels in the morpheme [naq^J 'tender'. The lax mid [o] becomes its tense counterpart [o], and the M L tone becomes HM. On the other hand, the tone and the vowel in the second syllable of this disyllabic word remain unchanged. Recall that an underlying mid vowel IOI surfaces as tense [o] in a tight syllable and as lax [o] in a corresponding loose syllable; it becomes clear that the change from [o] to [o] in (23a) is ;  exactly the same as the distributing pair o ~ D in monosyllabic words. As in Fuzhou. this identical relation between the vowel alternations in disyllabic words and the vowel distributions in monosyllabic words can be easily explained in the current theory. In particular, the constraint on stress forces a bimoraic syllable in a non-final position to become monomoraic, the moraic structure that is identical to the tight syllables. Thus, the vowel alternation pattern o ~ o follows. The following tableau (24) shows how the change from a to o can be achieved by the interaction of the constraints related to stress and those constraining syllabification. The input in (24) contains two bimoraic morphemes (i.e., two loose syllables). The last candidate violates  PROMALIGN  (which requires a metrical head to be aligned to the  rightmost syllable within a word) since the prominence indicated by the "x" on top of the 187  Chapter Five  Jiang-King, 1996  syllable node is aligned to the left, resulting in a misplacement of stress. It also violates PROMREDUC  because of the bimoraic syllable in the unstressed position. The middle  candidate violates P R O M R E D U C because of the bimoraic structure in the unstressed syllable. Thefirstcandidate wins since it violates no constraints.  (24) Output candidates for fnon^muoi ^] 'young sister' in Fuqing 1  "Input"  Cand,  k k II ( •  l v\ Uv\ no  qm uo  i  norj^ 'tender' + muoi^ 'sister'  Cand,  Cand,  I I1 I  x)  x)  ( •  (x  .)  kk kk kk  n o n muo i  [norjf^muoi ] 'young sister* ML  n o q muo i  [hon:muoi]  q muo i  no  [non.muoi]  PROMALIGN PROMREDUC  .  *•!  *!  *!  The examination of the data in (23b) shows that when a disyllabic word [torjH na ™] 1  "decision-maker" is comprised of two morphemes [ton ™] "become' and [rja™] 'family' 1  1  belonging to the tight type of syllables, neither thefirstmorpheme nor the second one has any vocalic change. The lack of vowel alternation in either the final or the non-final position for the tight type of syllables is expected under the current theory, since there is no structural change involved. This is demonstrated in tableau (25) below. The last candidate in (25) violates P R O M A L I G N because the prominence is aligned to the left rather than the rightmost of the word. The middle candidate satisfies P R O M A L I G N by aligning stress to the rightmost syllable of the word. However, it violates P A R S E since the onset consonant [n] in thefinalsyllable gets underparsed. Only thefirstcandidate incurs  188  Chapter Five  Jiang-King, 1996  no violations, and thus is optimal. Notice that the constraint  PROMREDUC  is not violated in  any member of the candidate set, since the morphemes involved are all monomoraic.  (25) Output candidates for [toq rja ] 'decision-maker' in Fuqing H  "Input"  HM  Cand,  M  /  n  \\\ \  /H /H t/1o \rj rj/1a  torj "become' + na ™ 'family' H  1  Cand,  I\ I  (.  X)  toqq a [tonHqa ™] 1  (•  Cand, x) : (x  \  /1 \ i t o q <q\,a  [torj.a]  'decision-maker'  A A  .)  I  /1 \ / i to qq a  [torj.rja] *!  PROMAUGN PROMREDUC  *!  PARSE  5.2.3 A summary  The demonstration above shows that the identical patterning between vowel distributions in monosyllabic words and vowel alternations in disyllabic words is fully predictable under the proposed theory. This is because a moraic contrast (i.e., monomoraic vs. bimoraic) exists in monosyllabic morphemes. Once the stress effect takes one mora away from a loose syllable (i.e., bimoraic syllable) in a non-final position, that syllable becomes monomoraic, hence its vocalic output is the same as that in the tight monosyllabic words. Our theory also correctly predicts that a tight morpheme does not undergo any vocalic change, whether it occurs domain-finally or not, since it is monomoraic, and does not violates any stress-related constraint.  189  Chapter Five  Jiang-King, 1996  5.3 Vowel alternations in disyllabic reduplications  Vowel alternation in disyllabic reduplications exhibits the same characteristics as that in disyllabic compound nouns: First, the segmental properties in a reduplicant are identical to those in its base, if the base is the tight type of morphemes. That is, the copy of segments from the base is total for tight morpheme. In contrast, the segmental identity between a reduplicant and its base exhibited in the tight type of morphemes does not show up in the loose type of morphemes. In particular, if the base is the loose type of morpheme, the tone and vowels in the reduplicant are not totally the same as those in the base. Some modification occurs in the reduplicant. Second, non-identical segmental properties between the reduplicant and the base (only for the loose type of syllables), rather, vocalic segmental properties in the reduplicant of the loose syllables are identical to the corresponding forms in the tight ones. In a disyllabic reduplication, the reduplicant and the base are identical in their segmental properties The questions that arise are why are the vowel alternations in disyllabic reduplications the same as in disyllabic compounds? In other words, why is the segmental change from a base to its reduplicant the same as the vocalic changes from a loose morpheme to its corresponding tight one? Is there anything in common among the three types of morphemes (i.e., reduplicants, morphemes in a non-final position, and tight morphemes)? In this section, I start by investigating the segmental relation between the reduplicants and their base in Fuzhou. and explore its deeper relation with other types of morphemes (such as the tight morphemes in monosyllabic words and the loose ones undergoing vocalic changes in a non-final position) by examining their similarities in terms of their prosodic structures. I will argue that the similarity among these types of morphemes lies in their identical moraic structure, that is, they all have a monomoraic structure. Once this structural property has been identified, I then introduce a set of constraints on reduplication that are relevant to the present context, and show that the vowel alternation 190  Chapter Five  Jiang-King, 1996  between a reduplicant and its base follows from the analysis proposed for the vowel distributions in chajpter 4.  5.3.1  Fuzhou verb reduplications  Fuzhou morphology exhibits a rich reduplication system. It has nominal reduplication (Liang 1983a), adjective reduplication (Chen & Zheng 1990, Zheng 1988) and verbal reduplication (Zheng 1983, L i 1984). In this section, I focus on verb reduplication. In particular, I examine vowel alternations between a reduplicant and its base, and answer the question of why the alternations in this kind of reduplication are identical to the vowel distributions in monosyllabic words. Reduplication of monosyllabic verbs in Fuzhou is very common. Zheng (1983) reports that 921 out of 1019 monosyllabic verbs he investigated can be reduplicated in different ways. The Fuzhou verb reduplication data are from Zheng (1983). The underscore indicates tonal and vowel changes from the base to the reduplicant. The dots in the reduplicated verbs signal syllable and/morpheme boundaries.  (26)  Verbs  Gloss -> Redupl. Vs  a.  pei^  'to comb' p p 4 . p e i  b.  SOU^M  'to count'  su^  c.  ts'0yMLM  'to look'  ts^yia4.ts'0y  Gloss  Alternation  -ffig 'just comb hair'  M L M  .SOU^M MLM  ei -> i  ffcffc  'just count'  ou -> u  ••  'just take a look'  0y -> y  The data in (26) show that when a monosyllabic verb becomes disyllabic by reduplication, the tone and vowels of the second syllable are identical to those of the monosyllabic verb, whereas the tone and vowels of the first syllable differ from those of the monosyllabic verb. For example, the monosyllabic verb in (26a) is [pei" - ^] 'to comb', 1 1  when it becomes a reduplicated disyllabic verb, its first syllable has the form [pp^], that is 191  Chapter Five  Jiang-King, 1996  different from the original monosyllabic form, while its second syllable [pei  MLM  ] is the  same as the original form. The same is true for (26b) and (26c). The tonal change involved can be characterized as tonal simplification, since the original tonal contour M L M is complex, while the changed tonal contour is simple H M . Of interest are the vocalic changes. In particular, the diphthongs [ei], [ou] and [0y] of the original verbs in (26a), (26b) and (26c) become monophthongal high vowel [i], [u] and [y], respectively. This kind of vowel alternation, i.e., i ~ ei, o ~ ou and y ~ 0y, is exactly the same as the vowel alternation in the disyllabic compounds, as well as the contrast between monophthongs and diphthongs as exhibited in the monosyllabic words. Furthermore, the data in (27) show that the lax non-high vowels [e], [o] and [a] in the monosyllabic verbs become their tense counterparts [e], [o] and [a], respectively, in the first syllable of the disyllabic reduplicated verbs. Again, these alternating pairs e ~ e, o ~ o, a ~ a are identical to the tense/lax distinction exhibited in the vowel distributions in the monosyllabic words.  (27)  Verbs  Gloss  a.  k'ieM*™  'stand against'  k^ieHM.k'ie* ™  b.  muDn  'ask'  muor)^. muorj  c.  k'grjMLM  'look'  k'aq^ k'arj^  MLM  -»  Redupl. Vs  Gloss 41  MLM  Alternation  $f$f  'stand everywhere'  e -> e  |EJ[BJ  'just ask'  o -» o  ^f^f  'just take a look'  a -> a  Moreover, the examples in (28) reveal that the diphthongs [ai] and [oy] in the original verbs become [ei] and [0y] respectively in the first syllable of the disyllabic reduplicates. Recall that vowel distributions exhibit some feature co-occurrence and feature agreement restrictions in the tight syllables. In particular, a low-high sequence of vowels is disallowed in the tight syllables so that a diphthong ai in the loose syllables corresponds to a diphthong ei in the tight ones. Also, frontness agreement between the two elements of a diphthong is more restrictive in the tight syllables than in the loose ones. That is, a 192  Chapter Five  Jiang-King, 1996  diphthong like [ay] with frontness disagreement is only allowed in the loose syllables, and becomes [0y] in its corresponding tight ones. These two pairs ei ~ ai and 0y ~ ay are exactly the vowel alternations observed in the verb reduplications in (28).  (28)  Verbs  Gloss  ->  Redupl Vs  Gloss  a.  k'air)^ ^  'to cover*  b.  s2yr|  'to see sb. off s^yrp* .s3yq  1  MLM  k^ejrj^k'airj  MLM  MLM  Alternation  'just cover it'  ai-> ei  'just see sb. off  oy -» 0y  However, unlike the vowel alternations in the verb reduplication observed in (26), (27) and (28), (29) does not exhibit any vowel alternation: the original monosyllabic verbs all contain a simple contour tone (i.e., H, HM or ML) that belongs to the tight syllables.  Verbs  Gloss  a.  lin  'carry sth.'  lirj .lirj  m  'carry everywhere'  b.  ny?  •  'step on'  nyML. ly?H  a a  'step on everywhere'  w  *hold with hands' p'un^.-p'ur|  m  'rob with hands'  ts'o" ts'oH  nien .nierj  H  H  c.  —>  ts'o  e.  nierj  &  'weight in hand'  j ML  n  'write'  f.  H  s  a  H  Gloss  H  HM  d.  11  Redupl. Vs  H  s  j Mjj a  WW 'hold with hands' 'keep robbing with hands'  H  siaML  m  'pick up sth'  MM  'write everywhere'  The question then is why don't the tight type of verbs have any vowel alternation when they reduplicate? What is the difference between verbs with a complex contour tone and those with a simple contour or a level tone? The answer provided by the prosodic anchor hypothesis and the constraints on tonal distributions in chapter 3 is that the difference between morphemes with a complex contour tone and ones with a simple contour or a level tone lies in their moraic structure. That is, the former are bimoraic and the latter are monomoraic. Then a further question is what does this distinctive moraic 193  Chapter Five  Jiang-King, 1996  structure have to do with vowel alternations in verb reduplication? In other words, why should the distinctive moraic structure determine whether a reduplicated disyllabic verb should change its vowel or not? The answer offered by the analysis of the vowel alternations in disyllabic compounds in the earlier part of this chapter is that the different moraic structures are closely related to stress. In particular, a bimoraic morpheme in an unstressed position is disallowed, hence has to become monomoraic. Consequently, the tonal and vocalic features must change accordingly Now applying this analysis to the verb reduplication case, it is obvious that it is the first syllable but not the second one in a bimoraic verb reduplication that has to change its lexical properties, since that particular position in a disyllabic domain is unstressed, hence subject to  PROMINENCE REDUCTION.  On  the other hand, a monomoraic verb is not subject to this constraint, so there is a lack of vowel alternations for this type of verb reduplication. Having argued that the identical patterning between vowel alternations and distributions among the three types of morphemes (i.e., reduplications, disyllabic compounds, and monosyllabic words with the tight/loose distinction) is due to their moraic structure, it remains to show how these vowel alternations can be achieved. First, I treat the reduplicant R E D as prefix, a morpheme having solely a prosodic category without featural content, that is, R E D = O\ Second, I show that this morpheme R E D gets its featural content from its base (i.e., the second syllable of the reduplicate). The theory of prosodic morphology developed by M & P (1986, 1993a, b) provides us with a relevant constraint ANCHORING,  (30)  which is given in (30) below:  ANCHORING  (M & P  1993a:63)  In R + B, the initial element in R is identical to the initial element in B In B + R, the final element in R is identical to the final element in B.  194  Chapter Five What  Jiang-King, 1996  ANCHORING  requires is that the reduplicant  R  and the Base B must share an edge  element, initial in prefixing reduplication, final in suffixing reduplication (M & P 1986:94, 1993:63). In the following, I first demonstrate how the different sets of constraints interact with each other in deriving the alternating pair ei ~ i in (31).  (31) Output candidates for [piHMpei^q  "Input"  Cand,  Cand,  1 A A r hA (•  x)  (•  p e i  pi  Al  RED  FRT  peiMLM'to  H  comb'  Cand, (•  Cand  x)  (•  f  f\  Cand,  4  x)  %  / A A l\ r /N\  /II  /l  p e i  HI  p e  N • 1/1  F R T HI +  x)  t  J\  /l  'just comb hair' in Fuzhou (ei •-> i)  p e i  1  F R T HI  [pjHMpeiMLM]  FRT<M>  'just comb hair'  [pe.pei]  .)  f  f ™\  / N \  /l\ / l \ /l\ / l . \ /II / l 1 / 1 1 / 1 1  p e i p e i  1/1 . 1/1 1/1  FRT HI  / N \  (x  F R T IB  F R T HI  [peLpei]  pei  l/l  p e i  n  F R T HI  F R T HI  [pei.pei]  PROMALIGN  pei  p e i  1/1 1/1  F R T HI  F R T HI  [pei.pei]  *'!  PROMREDUC  *!  *COD/HI  *!  *!  *!  PARSEHI  The last candidate in (31) incurs two fatal violation marks for PROMREDUC  PROMALIGN  and  since the metrical grid is aligned at the left edge rather than the right one and  the unstressed syllable is bimoraic. Cand violates PROMREDUC in the same way as the last 4  candidate, and so is out. Cand satisfies 3  monomoraic. However, it violates  *COD/HI  PROMREDUC  by making the first syllable  because the high vowel [i] is parsed onto the  syllable node directly. Both of the first two candidates satisfy *COD/HI.  PROMALIGN, PROMREDUC  and  The difference between them is that both features [Hi] and [FRONT] in Candj are  parsed onto the prosodic anchor, giving rise to a monophthongal high vowel [i], whereas  195  Chapter Five only the feature  Jiang-King, 1996 [FRONT]  but not [Hi] is parsed in Candj, resulting in a monophthongal [e].  Since Candj violates PARSEHI, while Candj violates nothing, Candj wins. Second, I demonstrate the interaction of constraints on deriving the vowel alternation from lax to tense in (32). Since the constraint PROMALIGN is highly ranked, and any output violating it must be out, I will not include it in the following tableaux. All of the candidates in (32) incur a fatal violation mark. The last candidate violates since the stressed syllable reduces its mora. Cand violates PROMREDUC because of  PARSEU  3  the bimoraic structure in the unstressed position. It is interesting to compare the first two candidates. Cand violates the parasitic constraint LAXTNG (which requires a long non-high 2  vowel to be lax), because a low vowelbecomes lax without being doubly parsed onto two moras. The first candidate satisfies  LAXTNG  by simply not incorporating the feature [lax]  into the segmental root. Since the first candidate violates nothing, it is optimal  (32) Output candidates for fk'an ^ k ' a n ^ ] j f i f 'just take a look' in Fuzhou (a 1  "Input"  A  k' a  q  N  Cand, (.  x)  LO LAX  'to look'  x)  Cand, (•  x)  A  1  LO  K LOLAX  'just take a look'  I  4  x)  I  I  k'o i | k ' a q K l\ LOLAX  LOLAX  [k'an.k'an]  k' a qk' a q N K LOLAX  LOLAX  [k'an.k'an]  PARSER  I  k' a qk'a q  1  •I LO  LO  [k'ank'aq] *!  PROMREDUC LAXTNG  Cand (•  kk kk kk kk A  k' a q k' a q RED +  Cand, • (•'  *! *!  196  a)  Chapter Five  Jiang-King, 1996  It is important to notice that in this case, the vowel change from the base to RED is governed by the constraint proposed for the vowel distribution in chapter 4; we haven't introduced any new constraints on vowel alternations. Lastly, I illustrate in (33) the case where the vowel change from the base to the R E D involves the constraint on feature co-occurrence proposed for the monosyllabic words. Tableau (33) shows that the last candidate is the worst output since it violates (because the stressed syllable loses a mora), has two moras) and  PARSEHI  unparsed). Cand violates 3  PROMREDUC  PARSEU  (because the unstressed syllable  (i.e., the high vowel [i] in the second syllable is left  *HI/LO^  because both the high vowel and the low vowel are  linked to the same mora in unstressed position. It is interesting to compare the first two candidates. Candj violates  PARSEHI,  while Candj violates  PARSELO.  Since P A R S E H I ranks  above P A R S E L O , Candj wins.  (33) Output candidates for fk'em^k'ainMLM] mm 'just cover it'in Fuzhou (ai -> ei)  "Input"  A / RED + k a i tj  k'ain^ 'to cover'  Cand, :  (•  x)  Cand, (•  x)  Cand, (•  Cand  x)  x)  I  t  Ai kk kk kk A  A  k<a>iij k a i IJ  rk'ein^k'ainMLM]  k a<i> g k a i ij  fkan.kain]  J!  1  k a i  q k ai  IJ  [kain.kain]  k a i q k a=i> rj  [kian.nuorj]  'just cover if* PARSEU  *!  PROMREDUC  *!  *HI/LOU  *!•  *!  PARSEHI PARSELO  d  (•  *  197  *!  Chapter Five  Jiang-King, 1996  The interesting point in this case is that the constraints *Hi/Lo , u  the ranking *Hi/Lo , u  PARSEHI »  PARSELO  PARSEHI  and P A R S E L O and  established for the vowel distributions in  chapter 4, play a crucial role in determining the optimal vowel alternating form. We have not proposed any particular constraint for reduplication alone. The full range of vowel alternation effects in the reduplication cases follows from the constraints on vowel distributions and the constraints on stress assignment in the disyllabic compounds, which is a desirable result.  5.3.2  Fuqing reduplications  Fuqing morphology also exhibits various types of reduplications: nominal reduplication, adjective reduplication, adverb reduplication, and verbal reduplication (Feng 1993). In this section, I examine vowel alternations between a reduplicant and its base, and explore the similarities among different types of morphemes regarding their output vocalic forms. Reduplicated adjectives, adverbs and verbs in Fuqing denote intensity, momentarity, etc. The data in (34) show that when a monosyllabic word becomes disyllabic by reduplication, the tone and vowels of the second syllable are identical to those of the original form, whereas the tone and vowels of the first syllable undergo some changes. For example, the monosyllabic word in (34a) is [se ^] 'thin and long', but when it becomes a 1  reduplicated disyllabic word, the vowel in the first syllable changes from [e] to [i], while the tone and vowel in the second syllable [se^] are the same as those in the original form. Note that all of the morphemes that undergo reduplication in (34) contain a L tone in their tonal contour, hence belong to the loose type of syllables Interestingly, the vocalic changes involved in these cases mirror the vowel distributing pairs between tight/loose morphemes. For instance, the lax non-high vowels [e], [o] and [a] in (34b), (34c) and (34d) become their tense counterparts [e], [o] and [a], respectively, in the first syllable of the disyllabic reduplicated forms. These alternating pairs e ~ e, o ~ o, a ~ a are identical 198  Chapter Five  Jiang-King, 1996  to the tense/lax distinction exhibited in the vowel distributions in the monosyllabic words. Further, the example in (34e) reveals that the triphthong [ieu] becomes a diphthong [iu] in the first syllable of the disyllabic reduplicant Recall that vowel distributions exhibit a contrast between diphthongs in tight syllables and triphthongs in the corresponding loose syllables. That is, diphthongs [ui] and [iu] in tight syllables correspond to the triphthong [uoi] and [ieu] in loose syllables, respectively. They are the same as the alternating pair ieu ~ iu in (34e). The tonal change that occurs in these cases is exactly the loss of the L part from the tonal contours in the original words.  monosyl  Redupl. words  ->  a  s e  b.  men™-  merjH.men™-  c.  huorj™-  huorjH.huor) ^  d  tsgrj ^  tsarjH .tsorjM-  ML  Gloss  m  ^j^c  1  1  1  e.  t s  'iUHM 'ieuML t s  Alternation  Very thin and long'  e -»i  'slowly'  e -» e  'far away' 'how come'  a ->a  'just smile'  ieu -> iu  However, the vowel alternations in the reduplicated forms observed in (34) do not show up in (35), where the original monosyllabic words do not contain a L tone in their tonal contours. The tones in this set of data are either level tones (i.e., H or M) or a H M contour tone.  (35)  monosyl  Gloss —> Redupl. words  a.  p'ierjBM  ^  ' ntime'  b.  hoM  £p 'good'  ho^.ho™  c. <t>rf  &  0n .0n  d. t a ^  % 'dry'  0  'red'  p'ienH.mierj™  H  H  taHta™  199  Gloss fl^g  'just inrighttime' 'properly'  £riH 'very red' Very dry'  Chapter Five  Jiang-King, 1996  What is the difference between the forms in (34) and the ones in (35)? Their tonal patterns in the original forms show that the difference lies in their prosodic structures. That is, the forms in (34) are bimoraic since they have either a H L or a M L tone, whereas the ones in (35) are monomoraic, since they do not have a L tone in their tonal patterns. Extending the analysis proposed for the disyllabic compounds in the earlier part of this chapter to the disyllabic reduplications, the vowel alternations in these cases can be viewed as a stress effect. In particular, since stress is assigned to the rightmost syllable within a disyllabic domain, P R O M I N E N C E  REDUCTION  requires the first syllable of that domain to be  monomoraic. Thus, a bimoraic syllable in that position must lose one of its moras, triggering the vowel alternations. On the other hand, a monomoraic syllable in that position satisfies  PROMINENCE REDUCTION,  hence such a syllable does not undergo any  vocalic changes. The structural distinction (i.e., bimoraic vs. monomoraic), thus explains why the vowel alternations in the reduplicated forms are identical to the vowel alternating pairs in the disyllabic compounds and the vowel corresponding pairs in the tight/loose monosyllabic words. As in Fuzhou reduplication, I assume that the morpheme R E D in Fuqing is solely a prosodic category with no phonetic content. It gets featural content by copying from its base. I treat R E D as a prefix; subject to the constraint  ANCHOR-L.  In the following, I will  demonstrate first how the vowel alternation from mid to high is achieved by constraint interaction. The last candidate in (36) is the worst candidate. It violates PROMREDUC  PROMALIGN  and  since the metrical grid is aligned at the left edge rather the right one and the  unstressed syllable is bimoraic. It also incurs two counts of P A R S E H I violation because the feature [Hi] in both syllables is left unparsed. Cand is the same as Cand except that it 4  does not violate  PROMALIGN.  Cand satisfies 3  PROMREDUC  5  by making the first syllable  monomoraic. However, it violates *Hiu|i because the high vowel [i] is parsed onto two moras. Both of the first two candidates satisfy 200  PROMALIGN, PROMREDUC  and *Hiuu, but  Chapter Five  Jiang-King, 1996  violate P A R S E H I . The difference between them is that Candj violates P A R S E H I once, while Cand violates it twice. Therefore, Candj wins. 2  The interesting point in this case is the interaction of the constraints on stress and those on syllabification. As shown in the tableau, the set of constraints on stress ranks higher than the set of constraints on syllabification. More interestingly, the internal ranking of the constraints on syllabification in reduplication is identical to the ranking for the vowel distribution. This explains why the output vocalic forms in R E D are identical to those in the tight type of morphemes.  (36) Output candidates for [si™ se^] Very thin and long' in Fuqing (e -» i)  "Input"  Cand,  /  s +  e  A  FRT S6  x)  7  1  RED  Cand,  A IA 7i A  (•  <H>  ML  'thin and long'  I  /ft  / ft  \v •  si  s e  K  I FRT<Ht>  [si^.se ^] 'very thin and long' 1  x)  J • /i  X A  1*  /i  FRT. HI  (•  Iv  s i s e  1  1  FRT<H> FRT<H>  [si.se]  Cand, (•  Cand  x)  f  t  /N  / N \  / fl  / f t ft  /l  / V  si  si  K FRT  HI  K FRT  [si-si]  A/i\ A MA/i\ A x)  ?  /l / l \ - .  HI  f  /i\  / ji-ft  s e  (X  s e  s e  1  1 FRT<H>  [se.se]  7  / ft^  / V  FRT<H>  .)  7  / ft ft  / IX  / 1/  PROMALIGN  / ft ft  1  1/  s e  1  1  FRT<H>  FRT<Ht>  [se.se]  *'!  *!  PROMREDUC  *Hip:p, PARSEHI  Cand,  4  (•  *!  *! *  The next case I am going to demonstrate is the vowel alternation involving the tense/lax distinction. Since P R O M A L I G N is highly ranked, any candidate violating it will be out, and hence will not be included in the following tableaux.  201  Chapter Five  Jiang-King, 1996  All of the candidates except the first in (37) incur a fatal violation mark. The last candidate violates PROMREDUC  PARSER  since the stressed syllable loses a mora. Cand violates 3  because the unstressed position has bimoraic structure. It is interesting to  compare the first two candidates. Candj violates the parasitic constraint  LAXTNG  (which  requires a long non-high vowel to be lax), because the mid vowel in the first syllable becomes lax without being doubly parsed onto two moras. The first candidate does not violate L A X T N G simply because there is only one mora, hence no need to copy the feature [lax]. Since thefirstcandidate violates nothing, it is optimal.  (37) Output candidates for [me^H meqML] 'slowly' in Fuqing (e -» e)  "Input"  Cand, (.  I  k  Cand, (.  X )  II  Cand,  x)  (•  /I kk kk  me RED  +  ™  K T  rj  rj m e  K  1 L A X  men" - 'slow' 1  me  I  FRNT  rj  FRNTLAX  fmeq .merj ] 'slowly' H  ML  me rime  K  K  rj  FRNTLAXFRNTLAX  Cand  x)  i i II rkrh  me  rjme  FRNXAX  (•  rj  FRNXAX  [men.merj]  [men.merj]  d  x)  kk  me rjme  1 FRNT  rj  1•  FRONT  [men.men]  PARSER  *!  PROMREDUC  LAXTNG  *!  It is important to note that in this case, the vowel change from the base to R E D is governed by the constraint proposed for vowel distribution in chapter 4; we haven't introduced any new constraints solely for vowel alternations. The last case I illustrate is (38) where the vowel change from the base to the R E D involves the constraints on feature alignment in specific syllabic positions, proposed for  202  Chapter Five  Jiang-King, 1996  monosyllabic words in chapter 4. Tableau (38) shows that the last candidate violates since the unstressed position has two moras. The middle two candidates  PROMREDUC  violate  *COD/HI,  because the high vowel in both cases links to the syllable node directly.  The first candidate violates *Nuc/Hi. Since * C O D / H I ranks above *Nuc/Hi, Candj wins.  (38) Output candidates for [ts'iua^.ts'ieu ^] 'just smile' in Fuqing (ieu -» iu) 1  "Input"  I  A  Cand,  Ih  ts' i e u  A l 1 R E D + HI FRT HI  ts'ieu^'to smile'  Cand,  A k Ih (•  x)  r,  ts'i  (•  x)  7  ts'i e u  A II FRTM  A H FRTM  A ffi FRTM  rts'iuHMts'ieu ] ML  'just smile'  7  7  1  ts'i e u  1  A M FRTM  [ts'ieu.ts'ieu]  ts'i  ts'i e u  u  A 1 H FRT HI  A 1 M FRTM  [ts'iu.ts'ieu]  PROMREDUC  *!  *  *COD/HI »  4  x)  7  7  / /M- n  ll\  ts'i e u  ts'i e u  A K A N M FRTM RD HI FRTM RD  [ts'ieu.ts'ieu]  *! *  The interesting point in this case is that the constraints ranking  (-  *!•  *COD/HI *NUC/HI  Cand x)  n  ts'i e u  1  (•  i A A A /A A k Ih Ih IA Ih i / / i i 7  u  1  Cand,  *COD/HI  and *Nuc/Hi and the  *Nuc/Hi, established for the vowel distributions in chapter 4, play a  crucial role in determining the optimal vowel alternating form. We have not proposed any special constraint for the vowel alternations in reduplication. The full range of vowel alternation effects in the reduplication cases follows from the constraints on vowel distributions and the constraints on stress in the disyllabic compounds, which is a desirable result.  203  Chapter Five 5.3.3  Jiang-King, 1996  A summary  Fuzhou and Fuqing exhibit the same contrast between the first syllable and the second in a disyllabic reduplication. First, the first syllable always undergoes tonal and vocalic changes, while the second has neither tonal nor vocalic change. Second, the vowel alternations in disyllabic reduplication in both languages are identical to the ones observed in disyllabic compounds In particular, the vowel changes are alwaysfromthe forms in the loose syllables to the ones in the tight syllables. The analysis proposed for the reduplication cases is based on the account of the disyllabic compounds. That is, the set of constraints on stress and its interaction with the set governing vowel distributions I have show that the interaction of these two sets of constraints is sufficient to derive the full range of vowel alternation pairs. No extra constraint is needed.  5.4 Vowel alternations in Fuzhou Fanqie words  The "cutting-foot" words (data are from Liang 1982) are disyllabic words formed from monosyllabic words by a process resembling partial reduplication. In particular, the first syllable in the output shares the most sonorous vowel and any segmental material before that vowel with the original word, while the second syllable of the output retains the tone and all segmental material of the original word except the onset which is replaced by a new onset [1]. This is illustrated in (39), (40) below.  Originals a.  kuHM  b.  saq  c.  mirj  d.  ny?  H  "cuttine foot"  Gloss 'tie up'  sa^.larp  1  H  'arrest' 'hide'  H n  yMLly?H  'fill in; squeeze 204  Chapter Five  Jiang-King, 1996  The data in (39) show that when the original word is a CV syllable, like [ku ™] in (39a), 1  the disyllabic output is [ku^.lu ™], where the segmental material [ku] in the first syllable 1  is identical to that in the original morpheme, but the tone is different. On the other hand, the second syllable [lu™] in the output retains the tone of the original syllable, as well as the segmental material except the onset [1], which is not present in the original morpheme. However, when the original morpheme is a CVC syllable, such as [min ] in (39c) and 11  [ny? ] in (39d), the first syllable in each output does not contain a coda C. Comparing the 11  data in (39) with those in (40), where the original morphemes have more segmental material than those in (39), the first syllable of the output keeps the most sonorous segment and any segmental material before that nucleus.  (40)  Originals  "cutting foot"  Gloss  a  tien  tie^.lien"  "be given to'  b.  kuon ^  H  1  kuo .luorj ML  ML  'roll up'  The original morphemes in (40a) and (40b) all have a segmental sequence C W C , the first syllable of the output in each case is C W , while the second syllable of the output has the form 1WC, where the tonal and segment properties are retained except the onset consonant [Tj. The questions that arise are: Why does the first syllable in the output contain the segmental material before but not after the nuclear vowel? Why does the second syllable acquire a new onset? Prosodic Morphology developed by M & P (1986, 1993a, b, 1994, 1995), provides a framework within which a possible solution can be found. Assume that the Fanqie words are formed by reduplication, and the reduplicant R E D is prosodicalfy defined. The question is then what kind of prosodic category might the R E D be? Two possible candidates suggest themselves immediately The first possibility is R E D = a. This is apparently surface-true, 205  Chapter Five  Jiang-King, 1996  because the tonal patterns show that the output of the Fanqie word is disyllabic with one syllable being the base and the other being the reduplicant. Suppose that the relevant constraints that govern the Fanqie type of reduplication are those in (41), and their interaction with other constraints, such as the structural constraints ONS, - C O D , as well as the faithfulness L E X , is given in (42) below. 5  (41) Constraints on Fanqie reduplication  (42)  a.  R E D = oVNuc: the reduplicant template is the syllable O R the nucleus.  b.  MAX:  c.  LEFTMOST:  every element of B A S E has a correspondent in R E D . align R E D to the left edge of a syllable.  Output candidates for the Fanqie word [tieML.lierjH] T?e given to' in Fuzhou  Input  Cand,  Cand,  Cand,  Cand  4  Cand,  Cand  /tier)/  ti.tierj  tie.tierj  tierj.tierj  ie.tien  tie.ierj  &tie.lier|  *  *  **  *  *  *  -COD RED  =a  LEFTMOST  *  ONS  *  * *  LEX-F MAX  5  **  **  *  I thank Pat Shaw for working through the details in tableaux (42) and (43) with me. 206  **  fi  Chapter Five  Jiang-King, 1996  Cand in tableau (42) is the actual output. However, there is no possible ranking of these 6  constraints that can select this output. For example, i f - C O D were ranked at the bottom, the output with a total copy of the base (in Cand ) would be the optimal output, since it 3  satisfies all other constraints except - C O D . On the other hand, if - C O D were ranked on the top, ruling out Cand , then Candj (in which R E D copies a continuous string from the base 3  except the coda) would be optimal. The problem here is that there is no way to select a surface true output, no matter how one ranks this set of constraints. Now we are forced to reconsider what other possible prosodic category the reduplicant R E D might be. The other possibility is R E D = Nuc. If the reduplicant template is a Nucleus, it should contain the segmental material that occurs within Nuc. In Fuzhou, this material includes both the nuclear vowel which is dominated by the mora and the one that links to the Nuc node directly, that is, the on-glide. Now that the prosodic category for R E D has been redefined, I examine in (43) whether the same set of the constraints as in (42) is successful in selecting the actual output.  (43) Output candidates for the Fanqie word [tie^.lien ] *be given to' in Fuzhou 11  item548  RED  Input  Cand,  Cand,  Cand,  Cand  4  Cand,  Cand  /tien/  ti.tien  tie .tien  tierj.tierj  ie.tien  tie.icq  tie.lien  *!•'  *!  *! *!  *!  = Nuc  ONS  MAX  *|*  *  *\*  LEFTMOST  fi  * *  *  *  *  LEX-F -COD  *  *  **  207  *  Chapter Five  Jiang-King, 1996  The set of output candidates in (43) is identical to that in (42). It is shown that the first three candidates all violate the template R E D = N u c , because both the reduplicant and the base share an identical onset. The last three candidates satisfy the constraint R E D = N u c in different ways. In Cand R E D is onsetless, while the base has onset that is the same as the 4  original word. Conversely, in Cand it is the R E D but not the base that has the onset. By 5  violating LEFTMOST, Cand shows that R E D in this case is actually a N u c with two elements 5  inside, namely, an on-glide [i] and a nuclear vowel [e]. The same is true for Cand since 6  the base does not retain the original onset [t], it has a new onset [1], which is assumed to be the default consonant in this language. Comparing the first three candidates which violate the template constraint R E D = N u c with the last three candidates which satisfy it, it becomes clear that it is the different onsets in the reduplicant and the base, as well as the lack of coda in the reduplicant, that determines that the reduplicant is a N u c rather than a syllable. Looking down further, we see that the structural constraint O N S rules out Cand  4  and Cand because both of them contain an onsetless syllable. The last candidate satisfies 5  ONS  by inserting a default consonant in the base, even though it violates L E X - F as a cost.  Tableau (43) shows that the templatic constraint R E D = N u c and its interaction with other constraints are successful in selecting the optimal output for the Fanqie words. The constraint ranking for this case is  RED  = Nuc,  ONS »  MAX, LEX-F,  -COD.  Now examine the interesting cases in (44), where vowel alternations take place between the original morpheme and the first syllable of the outputs. The underscore in the outputs indicates the changed vowels, and the dots signal syllable boundaries.  (44)  Originals  "cutting foot"  Gloss  a.  hien" - ^  hie> l i e q ^  'throughout'  e -> e  .  sopMt-M  spL.b?MLM  'tie tightly'  o->o  lairj  laklain  'stand on tiptoe'  D  c.  1 1  MLM  1  MLM  208  Alternations  a -» a  Chapter Five  Jiang-King, 1996  The vowels in the original morphemes are [e] in (44a), [o] in (44b) and [a] in (44c). They become [e], [o] and [a], respectively in the first syllable of the outputs. The vowel change involved is from lax to tense, exactly the same as the tense/lax distinction exhibited in monosyllabic words. That is, the tense non-high vowels only occur in the tight syllables, while their lax counterparts appear only in the corresponding loose syllables. Moreover, this type of vowel alternation also shows up in disyllabic compounds and disyllabic reduplications. The question that arises from this observation is: Why are the vowel alternations in Fanqie words identical to the distribution patterns in monosyllabic words and the distribution patterns in other types of disyllabic forms. From the structural point of view, I found that the Fanqie words that exhibit vowel alternations in (44) are originally bimoraic, since their original tones are complex contour tones (i.e., MLM). Their original vocalic forms also indicate that they are bimoraic, because the set of lax vowels occur only in loose syllables By examining their outputs, we found that the first syllable in the output differs from the second syllable in two regards. First, the output tone in the first syllable cannot be a complex contour tone like that in the original morpheme, whereas the second syllables in (44) always keep the complex contour tones. Second, the segmental material in the first syllable is always less than that in the second syllable. It never has a coda C or an off-glide high vowel. Furthermore, the first syllable within a disyllabic domain is always subject to the stress constraint  PROMREDUCTION,  which prohibits a bimoraic syllable from  occurring in that position. Once the monomoraic status of the first syllable in the Fanqie words has been identified, the vowel change that occurs in that position is expected. Having identified the monomoraic structure for the reduplicant in a Fanqie word, I now demonstrate how the vowel alternations between the reduplicant and the base in the Fanqie words in (44) can be achieved. Following the analysis proposed for the disyllabic compounds and the disyllabic reduplications, I assume here that the constraints on stress,  209  Chapter Five namely,  PROMALIGN  Jiang-King, 1996 and  PROMREDUC,  are highly ranked Any output candidate violating  them will be out, and will not be included in the following tableaux.  (45) Output candidates for [hie .lierj ] throughout' in Fuzhou L  Cand,  "Input"  (.  I k hi  1 RED +  MLM  in  e rj K  HI FRNTLAX  Cand,  x) / N \  / ii //K\ /M / //MA 1 l\ II v\ hi e I i e q N  r\r\.  HI F R N T HI  FRNILAX  [hieL.lien ^ ] 1  1  (•  x)  f  1  Cand, (•  t  i k hi eqIie  q  N N \ HIFRNTHIFRN1LAX  [hien.lierj]  hie  (•  4  x)  I  / N  Ih  Cand  x)  N  \  lv\ i e  q  N NJ\ [hie.ierj]  HI F R N T HI  FRNTLAX  / N  / N \  / (  / h. \  1 k 1 /MM\ / / l II v\  hi e hi e N  q  NT\  HI F R N T HI  FRNTLAX  [hie.hierj]  'throughout' RED =  *!  Nuc  *!  ONS -COD  *  *l*  *  *  The last candidate in (45) violates the constraint R E D = N u c because the reduplicant and the base have identical onsets (which shows R E D is a syllable rather than a N u c ) . Cand violates ONS because the base is onsetless. Comparing the first two candidates, they 3  both violate - C O D . The difference between them is that Candj violates it once, while Cand  2  does so twice. Thus, the first candidate wins. The ranking for (45) and (43) is the same. The most interesting cases are those in (46) below, where the first syllable in an output has two alternative forms. The original morphemes in (46) are the ones belonging to the loose type of syllables, since they contain complex contour tones (i.e., MLM), and hence, are bimoraic. The first syllable of the output in each case has two alternative forms. These two alternative forms are respectively from each of the vowels of the diphthongs in the original morpheme. For instance, the original morpheme in (46a) is [tsei? ^ * ], which 1  210  1  1  Chapter Five  Jiang-King, 1996  contains a diphthong [ei]. The first syllable of the output can be either [tsi ] or [tse ] with L  L  the former as the more preferred output. The original morphemes in (46b) and (46c) are [ts'our™ ™] and [t'0y? 1  MLM  ] respectively. Again, each of them has two optimal outputs. The  brackets indicate the less preferred output.  (46)  Original  "cutting foot" word  Gloss  a.  tsei?  tsi (tse ).lei?  'squeeze'  i~e  b  ts'ourj ^™  'wring out wet clothes'  o- u  '  0-  MLM  L  L  MLM  ts'u (ts'o ^.loun ^  1  L  C. I'QylMLM  1  1  t'yL(t'0t) l0y?MLM  Q  r  a  w  Alternation  D  a  c  k«  y  To account for the alternative outputs in (46), I propose in (47) that the two constraints PARSEHI  and *Nuc/Hi are not ranked with respect to each other: These two constraints  play crucial roles in determining the optimal outputs for vowel distributions between the tight syllables and the loose ones (see chapter 4 tableaux (17) and (18) for details).  (47) Output candidates for [ts'u .loim^^ or [ts'oL.lOuqM™] 'wring out wet clothes' L  "Input"  A ts' o u q  RED  1/1 +  RCW  Cand, (.  x)  ; i  1 OU 1)  ts' U  K  14  MRD  son  [ts'uUounM™] RED  Cand, (•  Cand,  x)  /1  (•  I I  li /f\ 1 ou q  ts' o  \  .  1. 1 \  / 1  K  REM  [ts'oMoun  / \  ts' u  l/l  <HPRD  x)  ou n  M  MRD  MHM  ]  ROD  [hie.ierj]  = Nuc *!.  *  PARSEHI  *  211  4  x)  /A  / 1 l\  /l  ts u ;  ts' o n q  K M M R D  ROT  [hie.hierj] *!  ONS  *Nue/Hi  Cand (•  Chapter Five  Jiang-King, 1996  (47) shows that violation of R E D = N u c and O N S is fatal in the last two candidates. Cand  4  violates R E D = N u c because the reduplicant has identical onset and base. Cand violates 3  O N S because the base is onsetless. Comparing the first two candidates, Candj violates PARSEHI,  while Candj violates *Nuc/Hi. Since these two constraints are not crucially  ranked, both candidates are optimal. Thus, the cases where the output has two alternative forms are successfully accounted for by not ranking these two constraints. Apparent problematic cases for the constraint R E D = N u c are those in (48), in which the original words are monomoraic and contain a complex nucleus . If the constraint R E D 6  = N u c were fully satisfied, that is, the entire nucleus [ei] in the base gets copied to the reduplicant, the first syllable of the output should have a nucleus [ei] rather than [e].  (48)  original  "cutting foot"  doss  a.  teirj  te^leirj  'poke (sand in the shoes)'  b.  rjou?  c.  m0yrj  H  H  rjo^.lou? *  H  look upward'  1  m0 .l0yq  H  ML  H  'puffy'  As can be seen (48), the first syllable of the outputs in (48a), (48b) and (48c) contains a single mid vowel [e], [o] and [0], respectively This suggests that the constraint *COMPLEXNUC  (which disallows a nuclear mora from having more than one vowel) must  rank above R E D  = Nuc . 7  The question that arises is: why is a complex nucleus allowed in  nonreduplicated forms but prohibited in the reduplicant. An explanation comes from the proposals concerning the "emergence of the unmarked" (M & P 1994). A complex nucleus is more marked than a simple nucleus, and it is disallowed unless a higher ranked  6  1 thank E. G. Pulleyblank for bring this case to my attention (pc).  7  1 owe this insight to Doug Pulleyblank (p.c). 212  Chapter Five  Jiang-King, 1996  constraint forces it to be present. In the case of a nonreduplicated base, the constraint PARSERT  ranks above  *COMPLEXNUC  In other words, all segments must be parsed onto a  prosodic anchor, even though it results in a complex nucleus. On the other hand, a reduplicative morpheme contains a solely prosodic constituent without any featural content. It gets its featural content from the base. Even if a reduplicant copies only one vowel from the nucleus of the base, it does not violate  PARSERT  since segmental roots in  the base have already been parsed. In the following tableau, I demonstrate how the constraints  *COMPLEXNUC  and P A R S E R T interact with each other in deriving the unmarked  type of nucleus in the reduplicant of the "cutting foot words" in (48).  (49) Output candidates for [te^.lein"] 'poke'  "Input"  Cand,  A  IA (•  /f  t e i rj  v\  t e  RED + FRTHi  1 FRT  [ten]"]  x)  k  1  e i rj  1/1  FRTHI  Cand,  if A (•  t ei  x)  1  e i rj  1/1 1/1  FRTHI FRTHI  [te^.teirp]  Cand, (•  x)  N  A  Cand.* (•  x)  \ k  I I bib  e  t e  1 FRT  t ei q M FRTHI  [teMLlein ] 11  1 FRT  t ei q  1/1  FRTHI  [teML.leinH]  'poke*  *!  RED—Nuc ONS  *l  PARSERT  *!  *COMPLEXNUC  The last two candidates each incur a fatal violation mark. By copying the onset [t] in the reduplicant, Cand violates R E D = N u c because both the reduplicant and the base share the 4  same onset. On the other hand, having an onsetless reduplicant, as in the Cand , violates 3  213  Chapter Five the constraint  ONS.  Jiang-King, 1996 Comparing the first two candidates, Cand violates 2  *COMPLEXNUC  since  the reduplicant has a complex nucleus [ei]. Candj does not violate any of the constraints, therefore is optimal. Notice that. Cand] does not violate P A R S E R T either because all segmental roots in the base are properly parsed and the reduplicant does not contain any segmental root. Therefore, the constraint  PARSERT  (which plays a crucial role in deriving the monosyllabic  form [tein ] in the base) is inert in determining the segmental property of the reduplicant. 11  Importantly, the constraint  *COMPLEXNUC  seems not to be respected in nonreduplicated  forms. However, it plays a crucial role in selecting the optimal reduplicant in the "cutting foot words". This kind of discrepancy between lexical forms and reduplicated forms is exactly the type of case illustrating the emergence of the unmarked, and furnishes support for Optimality Theory.  214  Chapter Five  Jiang-King, 1996  5.5 Conclusion  I have shown in this chapter that a parallel relation exists among three types of phenomena: (i) vowel alternations between the reduplicant and the base, (ii) vowel alternations between the non-final syllable and the loose syllable in isolation, (iii) vowel distributions between a tight morpheme and a loose one. The factor that governs this relation is the distinctive moraic structure. In particular, the common ground for the three kinds of elements (i.e., the tight morpheme, the non-final syllable in a disyllabic compound, and the reduplicant) is the prosodic anchor, that is, the monomoraic structure, even if this monomoraic structure arises from the effect of different constraints. The constraint deriving the monomoraic structure for the reduplicants and the loose morphemes in the non-final position is  PROMREDUCTION,  which restricts an unstressed syllable to being  monomoraic, while the constraints determining the monomoraic structure for the tight morphemes in isolation are those that govern tonal distributions. It is this particular prosodic structure that unifies the different types of elements. The constraints on stress have two effects. First, P R O M A L I G N assigns a metrical grid to the rightmost syllable within a disyllabic domain, giving rise to featural stability in final position. Second,  PROMREDUC  forces a loose syllable in the non-final position to lose a  mora, triggering various vowel changes. The optimal vocalic forms are derived by the same set of constraints on syllabification. This furnishes further support for the prosodic anchor hypothesis in that tone and vowel do not interact directly. Rather, they interact only when the mora gets affected. Moreover, it supports the analysis proposed in chapter 4 for the linking between syllable structures and vowel features in that the vowel distributions exhibited in monosyllabic words are derived by linking the same set of underlying features to different syllable structures.  215  LIST OF REFERENCES  Abramson, A. S. 1962. The vowels and tones of standard Thai: acoustical measurements and experiments. International Journal ofAmerican Linguistics 28.2 (part II). (Also publication 20.) Bloomington: Indiana University Research Center in Anthropology, Folklore and Linguistics. Abramson, A. S. 1972. Tonal experiments with whispered Thai. In A. Valdman (ed), Papers in Linguistics and Phonetics to the memory Pierre Delattre. The Hague: Mouton. Pp. 31-44.  Abramson, A. S. 1975. The tones of Central Thai: some perceptual experiments. In J. G. Harris and J. Chamberlain (eds), Studies in Tai Linguistics. Bangkok: Central Institute of English Language, Pp 1-16. Abramson, A. S. 1976. Static and dynamic acoustic cues in distinctive tones. Haskins Laboratories Status Report on Speech Research, New Haven, Conn., SR-47, 121127. Adam, T. B. 1891. An English-Chinese Dictionary of the Foochow Dialect. Fuzhou: Methodist Episcopal Mission Press.  Adam, T. B , and L. P. Peet. 1923 An English-Chinese Dictionary of the Foochow Dialect. Fuzhou: Methodist Episcopal Mission Press. Anderson, Stephen R. 1982. The Analysis of French shwa. Language 58:534-573. Anderson, Stephen R. 1984. A Metrical Interpretation of Some Traditional Claims about Quantity and Stress. In M . Aronoff and R. Oehrle (eds.) Language Sound Structure 83-106, MIT Press.  Archangeli, Diana. 1988. Aspects of underspecification theory. Phonology 5:183-207. Archangeli, Diana. 1992. Underspecification. An Encyclopedic Dictionary of Linguistic Terminology, B.V. Prakasem, Chief Investigator. 216  List ofReferences  Jiang-King, 1996  Archangeli, Diana, and Douglas Pulleyblank. 1986. The content and structure of phonological representations, ms., University of Arizona and University of Southern California.  Archangeli, Diana, and Douglas Pulleyblank. 1989. Yoruba vowel harmony. Linguistic Inquiry 20:173-217.  Archangeli, Diana, and Douglas Pulleyblank. 1994. Grounded Phonology, MIT Press, Cambridge, MA.  Baldwin, C. C. 1871. A Manual oftheFoochow Dialect. Fuzhou: Methodist Episcopal Press. (Revised and enlarged edition, Fuzhou College Press, 1909). Ballard, W. L. 1988. The History and Development of Tonal Systems and Tone Alternations in South China, Study of Languages and Cultures of Asia and Africa Monograph Series No: 22, Institute for the Study of Languages and Cultures of Asia and Africa, Tokyo.  Bao, Zhiming. 1990a Fanqie languages and reduplication. Linguistic Inquiry 21.3 317350.  Bao, Zhiming. 1990b. On The Nature of Tone, Doctoral dissertation, MIT. Bat-El, Guti. 1988. Remarks on tier conflation. Linguistic Inquiry 19.3:  Beijing daxue zhongguo yuyan wenxuexi yuyanxue jiaoyanshi IkMJ&'PMTfenX^Jik irfW¥lfc9fS (eds.): 1959. Hanyu fangyan zihui ^ ^ T O i C [Phonetic Dictionary of Chinese Dialects (1st edition)]. Wenzi Gaige Chubanshe X%$r$.&Wdk, Beijing.  Beijing daxue zhongguo yuyan wenxuexi yuyanxue jiaoyanshi i£MJ& ¥MT&nX¥ %i mm¥$m%(eds). 1989 HanyufangyinzihuiUWTt^C[PhoneticDictionaryoj Chinese Dialects(2nd edition)] Wenzi Gaige Chubanshe X%%.$-&Mk, Beijing. t  217  :  List ofReferences  Jiang-King, 1996  Benedict, Paul K. 1948. Tonal systems in Southeast Asia. Journal of the American Oriental Society 68 4:184-191. Benedict, Paul K 1973. Austro-Thai Studies. HRAF Press, New Haven. Benedict, Paul K. 1975. Austro-Thai Language and Culture with A Glossary ofRoots, HRAF Press. Bickmore, Lee S. 1994. Tone distribution in Optimality Theory, ms. University At Albany. Bickner, Robert J. (al ed.) Selected Papers on Comparative Tai Studies, Michigan Papers on South and Southeast Asia Center for South and Southeast Asian Studies No. 29. The University of Michigan. Black, J. W. 1949. Natural frequency, duration, and intensity of vowels in readings. Journal of Speech and Hearing Disorders 14:216-221. Blicher, D; L., R. L. Diehl, and L. B. Cohen. 1990. Effects of syllable duration on the perception of the Mandarin tone2/tone 3 distinction: evidence of auditory enhancement. Journal of Phonetics 18:37'-49.  Blight, R. C , and Pike, E. V. 1976. The phonology of Tenango Otomi. International Journal ofAmerican Linguistics 42.1:51-57.  Bogers, K., H. van der Hulst and M . Mous. 1986. (eds). ne phonological representation of suprasegmentals. Dordrecht: Foris. Broselow, E., and John J. McCarthy. 1983. A Theory of Internal Reduplication. The Linguistic Review 3:25-88.  Brown, J. Marvin. 1965. From Ancient Thai to Modern Dialects. Bangkok.  218  List of References  Jiang-King, 1996  Brown, M . 1975. The great tone split: did it work in two opposite ways? In Harris and Chamberlain (eds.) Studies in Tai linguistics in honor of William J. Gedney, pp. 3348, Central Institute of English Language, Bangkok. Cao, Congsun WW- 1992. Hanyu yingyu lueshuo R M M i f e [On Chinese secret languages]. Zhongguo Yuwenrf IlBifjt 1: 45-49. 3  Cairns, Charles E., and Mark H. Feinstein. 1982 Markedness and the theory of syllable structure. Linguistic Inquiry. 13.2:193-225. Chan, Majorie Kit Man. 1985. Fuzhou Phonology: A Non-Linear Analysis of Tone and Stress, Doctoral dissertation, University of Washington. Chang, Mei-chih Laura. 1993 The representation of tone and the parametric variations of tonal systems. The Proceedings of BLS. Chao, Yuen-Ren. 1930. A system of tone letters. La MaitrePhonetique 30:24-27 (Reprinted in English orthography in Fangyan 1980.2:81-83.) Chao, Yuen-Ren. 1931. Fanqie yu ba zhong K ^ A f t [Eight Varieties of Languages Based on the Principle of Fanqie]. Bulletin of the Institute of History and Philology, Academic Sinica 2.3:312-354.  Chao, Yuen-Ren 1933. Tone and intonation in Chinese. Academic Sinica, Bulletin of the Institute ofHistory and Philology IV2:121-134.  Chao, Yuen-Ren. 1934. The non-uniqueness of phonemic solutions of phonetic system. In Joos, ed. (1957). Readings in Linguistics, pp. 38-54. (Originally published in Academic Sinica, Institute of History and Philology. Bulletin (1934) part 4:363-397. Chao, Yuen-Ren. 1943. Languages and dialects in China. Geographical Journal 102 6366. (Reprinted in Chao 1976.)  Chao, Yuen-Ren. 1968. A grammar of spoken Chinese. University of California Press, Berkeley.  219  List of References  Jiang-King, 1996  Chao, Yuen-Ren. 1976. Aspects of Chinese Sociolinguistics: Essays by Yuen Ren Chao  (selected and introduced by Anwar S. Dil), Stanford University Press. Chen, Leo. 1967. Foochow reduplication. TsingHua Journal of Chinese Studies 6:1-20, 200-213. Chen, Leo 1969. English-Fuzhou: Fuzhou-English Dictionary. San Francisco: Asia Language Publishing Co. Chen, Leo., and Jerry Norman. 1965a An introduction to the Foochow dialect. San Francisco: San Francisco State College. Chen, Leo., and Jerry Norman. 1965b. Foochow-English Glossary. San Francisco: San Francisco State College. Chen, M . 1984. Abstract Symmetry in Chinese. Linguistic Inquiry 15:167-170. Chen, Yacuan BM.f\\, and Yide Zheng 1990. Fuzhou hua xingrongci chongdesi de yinbian fangsi jiqi leixing » W # i ^ M ^ W W ^ r ^ ^ H S l [The reduplication patterns and classes of adjectives in Fuzhou]. Zhongguo Yuwen t ®if X 5:362-370. Cheng, Chi-chuan 1973. A quantitative study of Chinese tones. Journal of Chinese Linguistics 1.1:93-110.  Cheung, S. H. N . 1973. Tonal redistribution in the Omei dialect. Paper presented at the 6th International Sino-Tibetan Conference, San Diego.  Chomsky, Noam and Morris Halle. 1968. The Sound Pattern ofEnglish, New York: Harper and Row.  Chuang, C. K. 1972. An acoustical study on the Chinese four tones. Doctoral dissertation  [in Japanese], Tohoku University, Sendai, Japan [abstract in English].  220  List ofReferences  Jiang-King, 1996  Chuang, C. K., and William S. Y. Wang. 1976. Influence of vowel height, intensity, and temporal order on pitch perception. Journal of the Acoustical Society of America 601 (A), LL13.  Chuang, C. K., and William S. Y. Wang. 1978. Psychophysical pitch biases related to vowel quality, intensity difference and sequential order. Journal of the Acoustical Society ofAmerica 64:1004-1014. Clements, George N. 1984. Principles of tone assignment in Kikuyu. In G. N. Clements and J. Goldsmith eds. Autosegmental Studies in Bantu Tone. Foris, Dordrecht, Holland, 281-339. Clements, George N. 1985a. The Problem of Transfer in Nonlinear Morphology. Cornell Working Papers in Linguistics 7, Cornell University, Ithaca, New York. Clements, George N . 1985b. The geometry of phonological features. Phonology Yearbook 2:225-252.  Clements, George N. 1990. The role of the sonority cycle in core syllabification. In Papers in Laboratory Phonology 1, ed. John Kingston and Mary Beckman, 283-333. Cambridge University Press. Clements, George N , and S. J. Keyser. 1983. CV Phonology, M T Press, Cambridge, Massachusetts.  Clements, George N , and Kevin Ford 1979. Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry 10:179-210. Cole, Jennifer, and Charles W. Kisseberth. 1994. An optimal domains theory of harmony. Cognitive Science Technical Report UIUC-BI-CS-94-02, University of Illinois. Cole, Jennifer, and Charles W. Kisseberth 1995. Levels of representation in Nonderivational Phonology. Opaque rule interaction in Yawelmani vowel harmony. Handout, University of Illinois.  221  List of References  Jiang-King, 1996  Davis, Stuart. 1988. Topics in Syllable Geometry, Garland Publishing Inc., New York & London. De Chene, Brent Eugene. 1979. The Historical Phonology of Vowel Length, Doctoral dissertation, UCLA. Delattre, P. 1962. Some factors of vowel duration and their cross-linguistic validity. Journal of the Acoustical Society ofAmerica 34:1141-1143. Di Cristo, A., and Chafcouloff, M . 1976. An acoustic investigation of microprosodic effects in French vowels. Paper presented at the 14th Conference on Acoustics, High Tatra, Czechoslovakia. Diffloth, Gerard 1985 The registers of Mon vs. the spectrographist's tone, UCLA Working Papers in Phonetics 60:55-58. Downer, Gordon. 1963. Chinese, Thai, andMiao-Yao. In Shorto, H. L. (ed.) Linguistic comparison in Southeast Asia and the Pacific, School of Oriental and African Studies, London.  Dreher, J. J., and P. C. Lee. 1966. Instrumental investigation of simple and paired Mandarin tonemes. DARL Research Communication, Advanced Research Laboratory, Douglas Aircraft Company. Huntington Beach, California, 13. Duanmu, San. 1990. A Formal Study of Syllable, Tone, Stress and Domain in Chinese Languages, Doctoral dissertation, MIT.  Duanmu, San. 1993. Rime length, stress, and association domains. Journal of East Asian Linguistics 2, 1-44  Duanmu, San. 1994. Against contour tone units. Linguistic Inquiry 25:4, 555-608. Egerod, S0ren. 1961. Studies in Thai Dialectology. Acta Orientalia 26:43-91.  222  List ofReferences  Jiang-King, 1996  Elimelech, D. 1974. On the reality of underlying contour tones. UCLA Working Papers in Phonetics 27:74-83.  Erickson, D. 1976. An electromyographic study of Thai tones. Thesis Storrs. Esling, John H. 1983. A Laryngographic investigation of phonation type and laryngeal configurations, Working Papers of the Linguistics Circle of the University of Victoria 3 1:14-36.  Esling, John H. 1984. Laryngographic study of phonation type and laryngeal configurations, Journal of the International Phonetic Association 14:56-73. Ewan, W. E., and Krones, R. 1974 Measuring larynx movement using the thyroumbrometer. Journal of Phonetics 2:327-335. Ewan, W. G. 1975. Explaining the intrinsic pitch of vowels. 5th Calif. Lin. Ass. Conf., San Jose.  Fant, Gunnar and C. Scully (ed.) 1977. The larynx and language, Phonetica 34 (special issue).  Feng, Aizhen J ^ ^ . 1988. Fujian sheng Fuqing Fangyan de Yuyin Xitong MM^W^^m # f ^ # 3 ^ [Fuqing phonological system in Fujian Province]. Fangyan jf f 4 2873G0.  Feng, Aizhen 1990. Fuqing Fangyan Shengmu Yu Guangyun Shengmu Bijiao W$1fBP-% ?r®p-®&Mt. [A comparison of Fuqing finals with Guangyun finals] Fangyan 2:109-116. J  Feng, Aizhen &g&- 1993a. Fuqing Fangyan Yanjiu-ffltiTWmt [A study on Fuqing dialect]. Shehui Kexue Wenxian Chubanshe &4tft$XtiL&1fcSL Beijing&tt  Feng, Aizhen 1993b; Fuzhou Fangyan de RushengM'M3rm$lXF [The tones of the Ru category in Fuzhou dialect]. Fangyan 2:101-118.  223  List ofReferences  Jiang-King, 1996  Fischer-J0rgensen, Eli. 1990. Intrinsic F in tense and lax vowels with special reference to German. Phonetica 47:99-140. 0  Fromkin, Victoria. A. 1972. Tone features and tone rules. Studies in African Linguistics 3:47-76.  Fromkin, Victoria. A. (ed.) 1978. Tone: a linguistic survey, Academic Press, New York.  Fry, Dennis. B. 1958. Experiments in the perception of stress. Language and Speech, vol. 1:126-152  Fry, Dennis. B. (ed.) 1976. Acoustic Phonetics: A course of basic readings. Cambridge University Press, Cambridge, London.  Gai, Xingzhi £MZ- 1994. Zhang Mian yu de shong jin yuanyin W^SM%TCU [The tight/loose vowels in Tibetan and Burmese]. Minzu Yuwen X 5:49-53. Gandour, Jackson T. 1974 Consonant types and tone in Siamese. Journal ofPhonetics 2:337-350.  Gandour, Jackson T. 1975. Evidence from Lue for contour tone features. Pasaa 5.23952.  Gandour, Jackson T., and Maddieson, Ivan. 1976. Measuring larynx movement in standard Thai using the cricothyrometer. Phonetica 33:41-67'.  Gandour, Jackson T. 1977. On the interaction between tone and vowel length: Evidence from Thai dialects. Phonetica 34:54-65.  Gandour, Jackson T. 1978 The perception of tone, in V. Fromkin (ed.) Tone: a linguistic survey, Academic Press, New York.  224  List of References  Jiang-King, 1996  Gandour, Jackson T. 1981. Perceptual dimensions of tone: evidence from Cantonese. Journal of Chinese Linguistics 9.1. Gandour, Jackson T., and B. Weinberg. 1980. On the relationship between vowel height and fundamental frequency, evidence from esophageal speech Phonetica 37:344354. Gedney, William J. 1964. A comparative sketch of White, Black and Red Tai. The Social Science Review (Bangkok), Special number 14:1-47. Reprinted in Bickner, Robert J. (ed.) Selected Papers on Comparative Tai Studies, pp. 415-462. Gedney, William J. 1965. Yay, A Northern Tai Language in North Vietnam, Lingua 14:180-93.  Gedney, William J. 1973. A checklist for determining tones in Tai dialects. In Smith (ed.) Studies in Linguistics in honor of George Trager, pp. 423-437, Teuton, The Hague. Gedney, William J. 1989. Selected Papers on Comparative Tai Studies, ed. by Robert J. Bickner, John Hartmann, Thomas John Hudak, and Patcharin Peyasantiwong, Michigan Papers on South and Southeast Asia, Center for South and Southeast Asian Studies, The University of Michigan.  Glover, Warren W. 1971. Register in Tibeto-Burman languages of Nepal: a comparison with Mon-Khmer. In Glover (et al), 1-22. Glover, Warren W., Maria Hari andE. R. Hope (ed.)-1971. Papers in South East Asian Linguistics 2, Pacific Linguistics Series A, No. 29, Australian National Universitv Canberra.  Goldsmith, John 1976a. Autosegmental Phonology, Doctoral dissertation, MIT, Cambridge, Massachusetts.  Goldsmith, John. 1976b. An Overview of Autosegmental Phonology. Linguistic Analysis 2:68.  225  List ofReferences  Jiang-King, 1996  Goldsmith, John. 1993. (ed) The Last Phonological Rule: Reflections on Constraints and Derivations. Studies in Contenporary Linguistics, The University of Chicago Press, Chicago.  Greenberg, Joseph H. 1978. (ed ) Universals of human language: phonology. Stanford University Press. Haas, Mary R. 1958. The tones of four Tai dialects. Academic Sinic^ Bulletin of the Institute ofHistory and Philology 29:817-826.  Halle, Morris., and Kenneth N. Stevens 1971. A note on laryngeal features. Quarterly Progress Report 101, Research laboratory of electronics, MIT, Cambridge, Massachusetts. Halle, Morris., and Kenneth N. Stevens 1991. Knowledge of language and the sounds of speech. In Music, language, speech and brain, ed. Johan Sundberg, Lennart Nord, and Rolf Carlson, 1-19. London: Macmillan Academic and Professional. Halle, Morris and Jean-Roger Vergnaud. 1980. Three-dimensional Phonology. Journal oj Linguistic Research 1:83-105 Halle, Morris. 1995. Feature geometry and feature spreading. Linguistic Inquiry 26.1:146.  Halle, Morris., and K. P. Mohanan 1985. Segmental Phonology of Modern English. Linguistic Inquiry 16.1:57-116.  Hanson, R. J. 1975. Fundamental frequency dynamics in V C V sequences. Proceedings of the 8th International Congress ofPhonetic Sciences, Leeds, Aug. 1975. Hartmann, J. F. 1980. A Model for the Alignment of Dialects in Southwestern Tai, Journal of the Siam Society 68.1:72-86.  Hayes, Bruce. 1981. A Metrical Theory of Stress Rules, Doctoral dissertation, MIT.  226  List of References  Jiang-King, 1996  Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20:253- 306.  Hayes, Bruce 1990. Diphfhongisation and Coindexing. Phonology 7.1:31-71. Hayes, Bruce. 1995. Metrical Stress Theory: Principles and Case Studies, University of Chicago Press, Chicago. Henderson, Eugenie, J. A. 1965. The Topography of Certain Phonetic and Morphological Characteristics of South East Asian Languages, Lingua 15:400-434. Henderson, Eugenie, J. A. 1967. Grammar and Tone in South East Asian Languages, in Treseal Anthology. Hoekett, C. 1947. Peiping Phonology. Journal of the American Oriental Society 67:253Hombert, Jean-Marie. 1976. Development of tones from vowel height. UCLA. Working Papers in Linguistics 33:55-66.  Hombert, Jean-Marie, Ohala, John J., and William G. Ewan. 1976: Tonogenesis: theories and queries. Report of the phonology laboratory, University of California, Berkeley. Hombert, Jean-Marie. 1977. Consonant types, vowel height and tone in Yoruba. Studies in African Linguistics 8.2.173-190.  Hombert, Jean-Marie. 1978. Consonant types, vowel quality, and tone: In Victoria A. Fromkin (ed.) Tone: A Linguistic Survey, pp. 77-112. Academic Press, New York. House, A. S., and G. Fairbanks 1953. The influence of consonant environment upon the secondary acoustical characteristics of vowels. Journal of the Acoustical Society of America 25:105-113.  Howie, John Marshall. 1974. On the domain of tone in Mandarin. Phonetica 30:129-148.  227  List ofReferences  Jiang-King, 1996  Howie, John Marshall. 1976. Acoustical Studies ofMandarin Vowels and Tones, Cambridge University Press, New York. Huang, Diancheng H&M. 1984. Minyu de Tezheng WfeWlWt [The characteristics of Min languages]. Fangyan 3:161-164. Hyman, Larry M . 1973. (ed.) Consonant Types and Tones, Southern California Occasional Papers in Linguistics No. 1, Published by the Linguistics Program, University of Southern California, Los Angeles. Hyman, Larry M . 1977. On the Nature of Linguistic Stress. In Larry M . Hyman (ed.) Studies in Stress and Accent, 37-82. Southern California Occasional Papers in Linguistics No. 4.  Hyman, Larry M . 1978. Tone and/or Accent. In Donna Jo Napoli (ed.) Elements of Tone, Stress and Intonation 1-20 Georgetown University. Hyman, Larry M 1984. On the weightlessness of syllable onsets Proceedings of the 10th Annual Berkeley Linguistic Society Meeting 10:1-14. Hyman, Larry M . 1985. A Theory of Phonological Weight, Dordrecht, Foris. Hyman, Larry M 1986. The representation of multiple tone heights, in Bogers et al (eds) 1986, 109-152.  Hyman, Larry M . 1988 Syllable Structure Constraints on Tonal Contours Linguistique AfricaineNo. 1:49-60.  Hyman, Larry M 1992. Moraic mismatches in Bantu, ms., Winter LSA Meeting, Philadelphia.  228  List of References  Jiang-King, 1996  Hyman, Larry. M , and D. Pulleyblank 1988. On Feature Copying: Parameters of Tone Rules, in L. M . Hyman and C. N Li, eds. Language, Speech and Mind, Studies in Honor of Victoria A. Fromkin, Routledge, London. Hyman, Larry. M , and C. N . Li, (ed.) 1988. Language, Speech and Mind, Studies in Honor of Victoria A. Fromkin, Routledge, London. Iivonen, A. K. 1987. Regional differences in the relation of Standard German vowels, in Proc. 11th Int. Congr. Phon. Sci., Tallinn, Academy of Sciences of the Estonian S SR. Jakobson, R. 1931. Prinzipien der historischen phonologic Trav. Cercle ling. Prague 4:247-267. Jiang-King, Ping 1994a. An Optimality Account of Fuzhou Diphthongisation. Proceedings ofCLA.  Jiang-King, Ping 1994b. An Optimality Account of Fuzhou Tone-Vowel Interaction ms. University of British Columbia. Jiang-King, Ping 1995a. Fuzhou Tone-Vowel Interaction. Proceedings of Phonology Conference on Feature Distributions and Optimality Theory, University of Arizona. Jiang-King, Ping 1995b The structural status of on-glides and the peak-margin dichotomy. Paper presented at Annual Conference of Canadian Linguistic Association, University of Montreal, June 3rd, 1995  Kaye, Jonathan D. 1981 Les diphthongues cachees du vata. Studies in African Linguistics 12:225-244.  Kaye, Jonathan D , and Jean Lowenstamm. 1981 Syllable Structure and Markedness Theory. In A. Belletti et al. (eds.) Theory ofMarkedness in Generative Grammar 287-316, Pisa.  Kenstowicz, Michael. 1970. On the Notation of Vowel Length in Lithuanian. Papers in Linguistics 3:73-113.  229  List ofReferences  Jiang-King, 1996  Kim, C. W. 1968. Review of Lieberman 1967. Language 44:830-842. Kiparsky, Paul. 1979. Metrical Structure Assignment is Cyclic. Linguistic Inquiry 10:421441. Ladefoged, Peter. 1964. A Phonetic Study of West African Languages: An AuditoryInstrumental Study, West African language Monographs 1, Cambridge University Press, Cambridge. Lan, Ya-shiu. MMM- 1953. Phonetic system of the Foochow College ofArts, National Taiwan University 5, 241-331.  ffi^^.  Bulletin of the  Langdon,M. 1976 Stress, length and pitch in Yuman languages. Proc. Stressfest, University of Texas Press, Austin. Lea, Wayne A. 1972. Intonational cues to the constituent structure andphonemics of spoken English, Doctoral dissertation, Purdue University, Indiana. Lea, Wayne A. 1973. Segmental and Suprasegmental Influences on Fundamental Frequency Contours. In Larry M . Hyman (ed.) Consonant Types and Tones, Southern California Occasional Papers in Linguistics No. 1:15-70, Published by the Linguistics Program, University of Southern California, Los Angeles. Leben, William R. 1973. The role of tone in segmental phonology. In Larry M . Hyman (ed.) Consonant Types and Tones, Southern California Occasional Papers in Linguistics No. 1:117-149, Published by the Linguistics Program, University of Southern California, Los Angeles.  Leben, William R. 1973 Suprasegmental Phonology, Cambridge thesis. Leben, William R. 1978. The representation of tone. In Victoria A. Fromkin (ed.) Tone: A Linguistic Survey, pp. 177-220. Academic Press, New York.  230  List of References  Jiang-King, 1996  Leben, William R. 1980. A Metrical Analysis of Length. Linguistic Inquiry 11:497-509. Lehiste, I. 1970. Suprasegmentals. MIT Press, Cambridge. Lehiste, I., and Peterson, G. E. 1961. Some basic considerations in the analysis of intonation. Journal ofAcoustic Society ofAmerica 33:419-425. Levin, Juliette. 1985. A Metrical Theory ofSyllabicity, Doctoral dissertation, MIT. Li, Fang-Kuei. 1939. Languages and Dialects of China. 77*? Chinese Year Book 19381939, Shanghai Commercial Press, 44-46. (Reprinted in Journal of Chinese Linguistics 1973. 1.1:1-13]. Li, Fang-Kuei. 1947. The hypothesis of pre-glottalized series of consonants in primitive Tai. Academic Sinica, Bulletin of the Institute of History and Philology 11:177-188. Li, Fang-Kuei. 1959. Classification by vocabulary: Tai dialects. Anthropological Linguistics 1.2:15-21. Li, Fang-Kuei. 1960. A tentative classification of Tai dialects. In Stanley Diamond, (ed.) Culture in History: Essays in Honor of Paul Rodin, pp. 951-959. Columbia University Press, New York. Li, Fang-Kuei. 1992. The Tai Languages. In Compton, Carol J., and John F. Hartmann (ed.) Papers on Tai Languages, Linguistics, and Literatures, In Honor of William J. Gedney on his 77th Birthday, Monograph Series on Southeast Asia, Occasional Paper No. 16, Center for Southeast Asian Studies, Northern Dlinois University. Li, Rulong ^ S f t f c . et al. 1979. Fuzhou Fangyan Yuyin Bianhua Gaishuo ffiWlfn^^fa [An outline of phonetic change in the Fuzhou dialect] Zhongguo Yuwen + ® if ± 4:287-293.  Li, Rulong afsfttfe. 1984. Min fangyan he Miao, Zhuang, Tai, Zhang zhu yuyan de dongci tesi chongdie WrMmfctk, im^mmn^M^ [The special verb  231  List of References  Jiang-King, 1996  reduplication patterns in Min, Miao, Tai, Tibetan languages]. Minzu Yuwen Rjfc&X 1:17-25.  Liang, Yuzhang %3sM- 1982. Fuzhou fangyan de 'qiejiaoci' MtWlTm&l'tyfflM- [The "cutting foot words" in Fuzhou]. Fangyan #-#1:37-46.  Liang, Yuzhang 1983a. Fuzhou fangyan chongdiesni mingci M'M-^M^^M [Thenominal reduplications in Fuzhou]. Zhonggou Yuwen t ® i f ;£3:177-184. Liang, Yuzhang &W£$£^%XftM jft 3:166-169.  1983b. Fuzhou Fangyan Liandu Yinbian Yu Yuyi Fenbie [The semantic distinctions of tone sandhi in Fuzhou]. Fangyan  Liang, Yuzhang Hffifc. 1984 Fuzhou hua de wen bai yiduffiJHffififcfcS& [The literary and colloquial distinctions in Fuzhou]. Zhongguo Yuwen >f» ® i f X 6:434-440. Liang, Yuzhang ^ S t . 1990. Fuzhou hua de 'GEI' ffi^frgftl Fuzhou]. Zhongguo Yuwen t MT§X 4:28-283.  [The verb "to give" in  Li, P. J.-K. 1985. A Secret Language in Taiwanese. Journal of Chinese Linguistics 13:121.  Liberman, Mark and Alan Prince: 1977. On Stress and Linguistic Rhythm. Linguistic Inquiry 8:249-336.  Lowenstamm, Jean. 1981. On the Maximal Cluster Approach to Syllable Structure Linguistic Inquiry 12:575-604.  Luo, Meizhen W^^. 1984. Daiyu changduan yuanyin he fuyin yunwei de bianhua &MjG^ffi$$^Mtfy$t'ik [The change of long/short vowels and coda consonants in Tai]. Minzu Yuwen ¥kM$X 6:20-25.  Maddieson, Ian 1971. The inventory of features. Research Notes 3, University of Ibadan.  232  List of References  Jiang-King, 1996  Maddieson, Ian. 1976. The intrinsic pitch of vowels and tones in Foochow. UCLA Working Papers in Linguistics 33:191-202. Maddieson, Ian. 1978. Universals of tone, in Greenberg, Joseph H. (ed.) Universals oj human language: phonology. Stanford University Press. Maddieson, Ian., and Peter Ladefoged. 1985. 'Tense' and Lax' in four minority languages of China, UCLA Working Papers in Phonetics60:59-83. Manley, T. N . 1972. Outline of Sre Structure, Oceanic Linguistic Special Publication No. 12, The University Press of Hawaii. Manley, T. N. 1976. Pharyngeal expansion: its use in Sre vowels and its place in phonological theory, et al. by P. Jenners etc., Austroasiatic Studies 2, Oceanic Linguistic Special Publication No. 13, the University Press of Hawaii, 833-841. Marantz, Alec. 1982 Re-Reduplication. Linguistic Inquiry 13:435-482. Matisoff, James A. 1970. Glottal dissimilation and the Lahu high-rising tone: a tonogenetic case-study, Journal of the American Oriental Society 90.1:13-44. Matisoff, James A. 1973a. The grammar of Lahu. University of California Publications in Linguistics No. 75, Berkeley and Los Angeles: University of California Press. Matisoff, James A. 1973b. Tonogenesis in Southeast Asia In L. Hyman (ed.) Consonant Types and Tones, Southern California Occasional Papers in Linguistics No. 1, Published by the Linguistics Program, University of Southern California, Los Angeles. McCarthy, John. 1979. Formal Problems in Semitic Phonology and Morphology, Doctoral dissertation, MIT, Cambridge, Massachusetts.  McCarthy, John. 1981. A Prosodic Theory of Nonconcatenative Morphology Linguistic Inquiry 12, 373-419.  233  List ofReferences  Jiang-King, 1996  McCarthy, John. 1988. Feature geometry and dependency: A review. Phonetica 45:84108. McCarthy, John 1993 The parallel advantage: containment, consistency, and alignment, handout from the Rutgers Optimality Workshop I, Rutgers University, New Brunswick, NJ. McCarthy, John. 1995. Extensions of faithfulness: Rotuman revisited, ms, University of Massachusetts, Amherst. McCarthy, John and Alan Prince 1993a Prosodic Morphology I. Constraint interaction and satisfaction, ms., University of Massachusetts, Amherst, and Rutgers University. RuCCS-TR-3. [To appear Cambridge, MA: MIT Press] McCarthy, John, and Alan Prince 1993b. Generalized alignment In Geert Booij & Jaap van Marie, eds., Yearbook of Morphology 1993. Dordrecht: Kluwer. Pp. 79-153. McCarthy, John, and Alan Prince 1994. The emergence of the unmarked: Optimality in Prosodic Morphology. In Merce Gonzalez, ed., Proceedings of the North East Linguistic Society 24. Amherst, MA: Graduate Linguistic Student Association Pp 333-379.  McCarthy, John, and Alan Prince 1995. Faitlifulness and reduplicative identity, ms., University of Massachusetts, Amherst, Rutgers University. McCawley, James D. 1977. Accent in Japanese. In Larry M . Hyman (ed.) Studies in Stress and Accent 261-302. Southern California Occasional Papers in Linguistics No. 4. Mohanan, K.P 1979. On Syllabicity. In Safir 1979b, 182-190 Mohanan, K. P. 1993 Fields of attraction in Phonology. In John Goldsmith (ed.) The Last Phonological Rule: Reflections on Constraints and Derivations. Studies in Contemporary Linguistics, The University of Chicago Press, Chicago. Mohr, B. 1971. Intrinsic variations in the speech signal. Phonetica 23:65-93.  234  List ofReferences  Jiang-King, 1996  Newman, Paul. 1972. Syllable weight as a phonological variable. Studies in African Linguistics 301-3233.  Newman, Paul. 1973. Grades, vowel-tone classes and extension in the Hausa verbal system. Studies in African Linguistics 4:297-346.  Newman, Paul. 1975. The non-correlation of tone and vowel height in Hausa. Studies in African Linguistics 6:207-213.  Newman, Paul 1986. Contour tones as phonemic primes in Grebo. In Bogers, et al. (eds.) Norman, Jerry. 1973. Tonal development in Min. Journal of Chinese Linguistics 1 22736.  Norman, Jerry. 1977. A preliminary report on the dialects of Mingtung. Monumenta Serica Vol. XXXIII:326-348.  Odden, David 1991. Vowel geometry. Phonology 8:261-289.  Ohala, John J. 1973. The physiology of tone. In Larry M . Hyman (ed ) Consonant Types and Tones, Southern California Occasional Papers in Linguistics No. 1:1 -14, Published by the Linguistics Program, University of Southern California, Los Angeles.  Ohsiek, Deborah. 1973 Heavy Syllables and Stress. In Alan Bell and Joan B. Hooper (eds.) Syllables and Segments 35-43. Amsterdam, North-Holland. Oyang, Jueya ffl®&. 1979. Shengdiao yu yuanyin de xianghu zhiyue guanxi pm^7&^ #}ffi~SM%!%M [Tone-vowel interplay]. Zhongguo Yuwen tSiiS?: 5:359-362. Pan, Maoding-ffigUft. etc. 1963. Fujian hanyu fangyan fenqu lueshou M&BlWTmftffifcM [A brief outline of the classification of Chinese dialects in Fujian province]. Zhongguo Yuwen fynx 6:475-495.  235  List ofReferences  Jiang-King, 1996  Parker, Edward H. 1879a. New Foochow colloquial words. China Review 7:415-418. Parker, Edward Ff. 1879a. Tonic and vocal modification in Foochow dialect. China Review 7:182-187. Parker, Edward Ff. 1881. Foochow syllabary. China Review 9:63-82 Peng, Long. 1992. A Unified Theory of Tone and Voice, Doctoral dissertation, University of Arizona Petersen, N R. 1976. Intrinsic fundamental frequency of Danish vowels. Annual Report of the Institute of Phonetics, University of Copenhagen, 1-27. Peterson, G E., and H. L. Barney 1952. Control methods used in a study of the vowels. Journal of the Acoustical Society ofAmerica 24:175-184. Peterson, G. E., and Use Lehiste 1960. Duration of syllable nuclei in English. Journal of the Acoustical Society ofAmerica 23:693-703. Pike, Eunice V. 1948. Problems in Zapotec tone analysis. International Journal oJ American Linguistics 14.3:161-170.  Pike, Eunice V. 1956. Tonally differentiated allomorphs in Soyaltepec Mazatec. International Journal ofAmerican Linguistics 21.1:51-71. Pike, Eunice V. 1974. A multiple stress system versus a tone system. International Q Journal ofAmerican LmguisticsA<X.3:\69-l7 5.. Pike, Kenneth L., and Eunice V. Pike. 1947. Immediate Constituents of Mazateco Syllables. International Journal ofAmerican Linguistics 13:78-91. Pike, Kenneth L. 1948. Tone languages. Ann Arbor: University of Michigan Press.  236  List of References  Jiang-King, 1996  Pikczkowa-Chodak, Nina. 1972. Tone-vowel height correlation and tone assignment in the patterns of verb and noun plurals in Hausa. Studies in African Linguistics 3.3:399-422.  Pilzczkowa-Chodak, Nina. 1975. On the correlation of tone and vowel height in Hausa: a reply to Newman. Studies in African Linguistics 6.3:315-321. Prince, Alan 1983. Relating to the Grid Linguistic Inquiry 14:19-100. Prince, Alan. 1990. Quantitative Consequences of Rhythmic Organization. Proceedings of CLS, vol. 2, The parasession on the syllable in phonetics andphonology. Prince, Alan, and Paul Smolensky 1993. Optimality Theory: Constraint Interaction in Generative Grammar, ms, Rutgers University, and University of Colorado, Boulder Pulleyblank, Douglas. 1986. Tone in Lexical Phonology.Dordrecht: D. Reidel. Pulleyblank, Douglas. 1988 Vocalic Underspecification in Yoruba Linguistic Inquiry 19:233-270.  Pulleyblank, Douglas. 1993. Underlying Mora Structure. Linguistic Inquiry Pulleyblank, Douglas. 1994. Vowel Harmony and Optimality Theory. Proceedings of the workshop on phonology, University of Coimbra, Portugal, September 1993. Pulleyblank, Douglas, and William J. Turkel. 1995 Optimality theory and learning algorithms: the representation of recurrent featural asymmetries, to appear in J. Durand and B. Laks (eds.) Current Trends in Phonology: Models and Methods. Cms, Paris-X and University of Salford. University of Salford Publications, Pulleyblank, Douglas, Ping Jiang-King, Myles Leitch, and Nike Olanike. 1995. Low vowel in tough root harmony: typological variations through constraint ranking. Proceedings ofPhonology Conference on Feature Distributions and Optimality Theory, University of Arizona.  237  List of References  Jiang-King, 1996  Pulleyblank, Edwin G. 1983. Vowelless Chinese: An Application of the Three Tiered Theory of Syllable Structure to Pekingese. Paper presented at the XVI International Conference on Sino-Tibetan Languages and Linguistics. Pulleyblank, Edwin G. 1986. Some issues in CV Phonology with reference to the history of Chinese. Canadian Journal of Linguistics 31.3:225-266. Qu, Aitang - S U I T and Kerang Tan Wffik- 1983. AH Zhang yu HJMKgf' [Ali-Tibetan]. Zhongguo shehui kexue chubanshe + ® <f Beijing Qu, Yanfeng. 1995. Adjective reduplications in Fuzhou: a morpho-phonological analysis. Journal of East Asian Linguistics 4.1:1-28 Reinholt Petersen, N. 1978. Intrinsic fundamental frequency of Danish vowels. Journal of Phonetics 6:177-189 Reinholt Petersen, N. 1986. Perceptual compensation for segmental conditioned fundamental frequency perturbation, Phonetica 43:31-42. Rose, Phil 1990. Acoustics and phonology of complex tone sandhi. Phonetica 47:1-35 Safir, Ken. 1979. Metrical Structure in Capanahua In Ken Safir. Papers on Syllable Structure, Metrical Structure and Harmony Processes 95-114, MIT Working Papers in Linguists Vol. 1  Sagey, Elizabeth. 1986. The representation of features and relations in non-linear phonology. Doctoral dissertation, MIT, Cambridge, Mass. Sawusch, James R., and Jane S. Tsay. 1994. The interaction of FO and F l in tonal and nontonal languages. Poster at the Acoustical'Society ofAmerica annual meeting June 1994.  238  List of References  Jiang-King, 1996  Selkirk, Elizabeth O. 1980. The Role of Prosodic Categories in English Word Stress. Linguistic Inquiry 11:563-605. Schuh, Russell G. 1978. Tone rules. In Victoria A. Fromkin (ed.) Tone: A Linguistic Survey, pp. 221-256. Academic Press, New York. Shaw, Patricia A. 1992a. Templatic Evidence for the Syllable Nucleus. Proceedings of NELS 23. Shaw, Patricia A. 1992b. Fanqie Languages: At the Interface of Prosodic Phonology and Templatic Morphology. Paper presented at the Annual Conference ofCLA, University of PEL May 1992. Shaw, Patricia A. 1993 The Prosodic Constituency of Minor Syllables. Proceedings of WCCFL 12. Shaw, Patricia A. 1995. Minimality and Markedness. paper presented at the Workshop in Prosodic Morphology, Rijks Universiteit Utrecht, Nederland, June 1994. Shaw, Patricia A. 1996 The non-nuclear status of syllabic obstruents in Berber. Paper presented at Annual Conference of Linguistic Society ofAmerica, January 1995. Shih, Chi-Lin. 1986 The prosodic domain of tone sandhi in Chinese. Doctoral dissertation, University of California at San Diego. Smalley, William A. 1954 Sre phonemes and syllables. Journal of the American Oriental Society 74:217-222.  Steriade, Donca. 1982. Greek Prosodies and the Nature of Syllabification, Doctoral dissertation, MIT.  Steriade, Donca. 1988. Reduplication and syllable structure. Phonology'5:73-155.  239  List ofReferences  Jiang-King, 1996  Steriade, Donca. 1993. Underspecification and markedness. To appear in Handbook of Phonology, ed. by John Goldsmith. Basil Blackwell. Sundberg, J. 1969. Articulatory differences between spoken and sung vowels in singers. Quarterly Progress Status Report, Speech Transm. Lab., R. Inst. Technol. Stockh, No. 1:33-42. Svantesson, Jan-Olof. 1987; Tonogenesis in Northern Mon-Khmer. Proceedings Xlth International Congress of Phonetic Sciences, Talinn, Estonia, U.S.S.R. Volume 2:269-272 (se 31.1.1) Talinn: Academy of Sciences of the Estonian S. SR. Svantesson, Jan.-Olof. 1989. Tonogenetic mechanisms in Northern Mon-Kmer. Phonetica 46:60-79. Svantesson, Jan-Olof. 1991. Hu - A Language with Unorthodox Tonogenesis, In J. H. C. S. Davidson (ed.. Austroasiatic Languages: Essays in honor of H. L. Shorto, School of Oriental and African Studies, University of London Tao, Yu-min. 1930. Minyin yanjiu [Phonetics of the Foochow dialect]. Bulletin of the National Research Institute of History and Philology 1.4:445-470. Trigo,Loren. 1991. On pharynx-Larynx mteractions, PAowofogv 8, 113-136. Valdman, A. (ed.) 1972. Papers in Linguistics and Phonetics to the Memory ofPierre Delattre. The Hague: Mouton.  Vance, Timothy J. 1977. Tonal distinction in Cantonese. Phonetica 34:93-107. Wang, Tien-ch'ang 3E^g. 1965. Fuzhouyu ziyin huayin de fenxi [An analysis of the spoken and literary pronunciations in the Foochow dialect]. Tunghai Hsueh Pao 7.1:101-121 E. S.  Wang, Tien-ch'ang 1968. Fuzhou hua li jieyin de hunyao xianxiang [The phenomenon of the confusion of medial vowels in the Fuzhou dialect]. Tunghai Hsueh Pao 9.1.45-51, 51:E. S.  240  List ofReferences  Jiang-King, 1996  Wang, Tien-ch'ang 1969. Phonetic Research on The Foochow Dialect, Tunghai University, published with the assistance of the Sun Yat-Sen Cultural Foundation, Taiwan. Wang, William S-Y. 1967. Phonological features of tone. International Journal of American Linguistics 33:93-105. Wang, William S-Y. 1968. The many uses of F . Project on Linguistic Analysis Reports, 2nd series 8: wl-w35. (Reprinted in: Valdman (ed.) 1972, 487-503) 0  Wang, William S-Y £ ± T G . 1988. Shengdiao fazhan fangshi yishuo PM^Mff^r^ development of tone]. Yuwen Yanjiu if X*ft%. 1:38-42.  [The  Wright, Martha Susan. 1983. A Metrical Approach to Tone Sandhi in Chinese Dialects, Doctoral dissertation, University of Massachusetts. Yip, Moira. 1980. The Tonal Phonology of Chinese, Doctoral dissertation, MTT. Yip, Moira. 1982. Reduplication and C-V Skeleta in Chinese Secret Languages. Linguistic Inquiry 13:637-661.  Yip, Moira. 1989. Contour Tones. Phonology 6:149-174. Yip, Moira. 1992. Reduplication with Fixed Melodic Material. NELS 22. Yip, Moira 1993. The Interaction of ALIGN, PARSE-PLACE and *ECHO in Reduplication, handout in the Optimality Workshop, Rutgers University, October, 1993. Yip, Moira. 1995. Echo-avoidance in phonology and morphology. Paper presented at the Phonology Workshop: Feature Distributions and Optimality Theory, Tucson, April  241  List of References  Jiang-King, 1996  Yuan, Jiahua. et al. 1960. Hanyu fangyan gaiyao WWTmWig: [Outline of Chinese Dialects}, Beijing (second edition 1983)  Yue, O.-K. 1976. Interplay of Vocalic Segments and Tones in The Yueh Dialects. In Genetic Relationship, Diffusion and Typological Similarities of East and Southeast Asian Languages, Papers for the 1st Japan-US Joint Seminar on East and Southeast Asian Linguistics, Japanese Society for The Promotion of Science.  Zec,Draga. 1988. Sonority Constraints on Prosodic Structure, Doctoral dissertation, Stanford University.  Zee, Eric, and Jean-Marie Hombert 1976. Intensity and duration as correlates of F . Journal of the Acoustic Society ofAmerica 60, S44. 0  Zee, Eric. 1978. Duration and intensity as correlates of F . Journal ojPhonetics6213220. 0  Zee, Eric. 1980. Tone and vowel quality. Journal of Phonetics 8:247-258.  Zhan, Bohui MfaM- 1981. Xiandai Hanyu Fangyan MftMWTg [Modern Chinese dialects]. Hubei Renmin Chubanshe $MhARifcJfcjt, Wuhan Zhang, Zhenxing 1989. Min yu de fenbu he renkou ®^ft##fHAn [The distribution and population of Min language of Chinese], Fangyan ^ "ff 1:54-59.  Zheng, Yide %X$. 1983. Fuzhou fangyan danyin dongci chongdieshiMWTm^^Jffi Tt$£ [The reduplication of the monosyllabic verbs in Fuzhou]. Zhonggou Yuwen *B**l-:30-39.  Zheng, Yide fcStffc 1988. Fuzhou fangyan xingrongci chongdieshi WHJSW&tfft M$£& [The adjective reduplications in Fuzhou]. Fangyan %t 4:301-311.  242  APPENDIX  LIST O F CONSTRAINTS  I. Faithfulness constraints  (1)  PARSE  family: P A R S E - C I : an F-element (feature or node) a must be parsed onto an appropriate prosodic constituent.  a.  PARSETONE:  A tone must be incorporated into prosodic structure.  b.  PARSERT:  A segmental root must be incorporated into prosodic structure.  c.  PARSEHI:  An F-element [ H i ] must be incorporated into prosodic structure.  d.  PARSELO:  An F-element [Lo] must be incorporated into prosodic structure.  (2) F I L L family:  A prosodic constituent must be filled by an F-element.  (3) L E X family:  LEX-OC:  an F-element (feature or node) a that is present in an output  form is also present in the input: [... a .Joutput ~* [••'•  a.  LEXTONE:  a.  LEX|X:  a  • linput  A tone that is present in an output must be present in an input,  A mora that is present in an output must be present in an input.  (4) L I N E A R I T Y :  String, reflects the precedence structure of String;,, and vice versa.  (5) U N I F O R M I T Y :  N O element  of String has multiple correspondents in String,. 2  243  Appendix: List of Constraints  Jiang-King, 1996  II. Syllable structure constraints  (6) N U C L E U S  (7)  ONSET  (ONS):  (8) N O C O D A  (9)  (Nuc):  Syllables must have nuclei.  Syllables must have onsets.  (-COD):  *COMPLEX  Syllables cannot have codas.  (*COMPX):  No more than one C or V links to one syllable position.  a.  *COMPX-COD: N O  b.  *COMPX-ONS:  c.  *COMPX-NUC: N O  more than one post-nucleus segment may link to cr directly.  N O more than one  pre-nucleus segment may link to a directly.  more than one segment may link to nuclear mora directly.  (10) Prominence Reduction ( P R O M R E D U C ) If a is not assigned a metrical grid mark, a cannot be bimoraic.  III. Alignment and association constraints  (11) Generalized alignment AuoNiCm,  Edge 1, Cat2, Edge 2)•= def  V Cat! 3 Cat2 such that Edge 1 of Cat 1 and Edge 2 of Cat 2 coincide, Where  Catl, Cat2 e PCat u GCat Edgel, Edge2 e {Right, Left}  (12) Peak Hierarchy * P / t » *P/d ... * P / i » *P/a 244  Appendix: List of Constraints  Jiang-King, 1996  (13) Margin Hierarchy *M/a »  * M / i ... * M / d »  *M/t  (14) Tonal Alignment Hierarchy *NUCW[-RSD] »  *NUCW[+UPR]  (15) Harmonic Mora Hierarchy ( H A R M  1  ^)  *u/C » * u / H i » *u/Lo.  (16) Harmonic Nucleus Hierarchy ( H A R M N U C ) *Nuc/C » *Nuc/[+Hi] » *Nuc/[+Lo]  (17) Prominence Alignment  (PROMALIGN)  Given a domain x, a metrical grid mark must be aligned to the right edge of x. x - a morphological word.  IV. Constraints on reduplication  (18) R E D = a/Nuc: The reduplicant template is the syllable O R the nucleus.  (19) M A X : Every element of Stringj has a correspondent in Stringy  (20)  LEFTMOST:  Align R E D to the left edge of a syllable.  (21)  ANCHORING  In R + B , the initial element in R is identical to the initial element in B. In B + R, the final element in R is identical to the final element in B. 245  Appendix: List of Constraints  Jiang-King, 1996  V. Other constraints  (22) *Hi/LoK : a mora cannot be filled by both [+Hi] and [+Lo] F-elements. i  (23) Feature Agreement Constraint ( F - A G R [ R D ] or [ F R N T ] ) A non-low nuclear vowel must agree with its preceding on-glide in feature value for [RD] and  [FRONT].  (24) Parasitic Constraint ( P A R A S ( R D ,  FRNT))  Two anchors within a syllable agree in their values for [RD], iff they agree for  [FRONT].  (25) Length Dependent Constraint (LAXTNG or N\i\x —> [Lax]) Iff a is parsed onto two moras, then a is [LAX], (where a ^ [Hi])  (26) *Hi-|j.fi: A feature [+Hi] cannot link to two moras within a syllable.  (27) Height agreement: Two moras within a syllable must agree in feature value for height.  (28) Head Binarity (HDBIN) A mora must bear two tones x andy, iff it is a syllable head (i.e., a nuclear mora).  246  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0087288/manifest

Comment

Related Items