Vietnamese Tone: A New Analysis (Outstanding Dissertations in Linguistics)

OUTSTANDING DlSSERTATIONS IN LINGUISTICS

Edited by Laurence HornYale UniversityA ROUTLEDGE SERIES

OUTSTANDING DlSSERTATIONS IN LINGUISTICS LAURENCE HORN, General Editor THE SYNCHRONIC AND DIACHRONIC PHONOLOGY OF EJECTIVES Paul D.Fallon GRAMMATICAL FEATURES AND THE ACQUISITION OF REFERENCE A Comparative Study of Dutch and Spanish Sergio Baauw AUDITORY REPRESENTATIONS IN PHONOLOGY Edward S.Flemming THE TYPOLOGY OF PARTS OF SPEECH SYSTEMS The Markedness of Adjectives David Beck THE EFFECTS OF PROSODY ON ARTICULATION IN ENGLISH Taehong Cho PARALLELISM AND PROSODY IN THE PROCESSING OF ELLIPSIS SENTENCES Katy Carlson PRODUCTION, PERCEPTION, AND EMERGENT PHONOTACTIC PATTERNS A Case of Contrastive Palatalization Alexei Kochetov RADDOPPIAMENTO SINTATTICO IN ITALIAN A Synchronic and Diachronic Cross-Dialectical Study Doris Borrelli PRESUPPOSITION AND DISCOURSE FUNCTIONS OF THE JAPANESE PARTICLE Mo Sachiko Shudo THE SYNTAX OF POSSESSION IN JAPANESE Takae Tsujioka COMPENSATORY LENGTHENING Phonetics, Phonology, Diachrony Darya Kavitskaya THE EFFECTS OF DURATION AND SONORITY ON CONTOUR TONE DlSTRIBUTION A Typological Survey and Formal Analysis Jie Zhang EXISTENTIAL FAITHFULNESS A Study of Reduplicative TETU, Feature Movement, and Dissimilation Caro Struijke PRONOUNS AND WORD ORDER IN OLD

iii

ENGLISH With Particular Reference to the Indefinite Pronoun Man Linda van Bergen ELLIPSIS AND WA-MARKING IN JAPANESE CONVERSATION John Fry WORKING MEMORY IN SENTENCE COMPREHENSION Processing Hindi Center Embeddings Shravan Vasishth INPUT-BASED PHONOLOGICAL ACQUISITION Tania S. Zamuner ORIGINS OF PREDICATES Evidence from Plains Cree Tomio Hirose CAUSES AND CONSEQUENCES OF WORD STRUCTURE Jennifer Hay INTERACTIONS BETWEEN MARKEDNESS AND FAITHFULNESS CONSTRAINTS IN VOWEL SYSTEMS Viola Miglio THE PHONETICS AND PHONOLOGY OF GUTTURALS A Case Study from Ju \ ’hoansi Amanda Miller-Ockhuizen MANIFESTATIONS OF GENERICITY Yael Greenberg TRUE TO FORM Rising and Falling Declaratives as Questions in English Christine Gunlogson

VIETNAMESE TONE A NEW ANALYSIS Andrea Hoa Pham

Routledge New York & London

Published in 2003 by Roudedge 29 West 35th Street New York, NY 10001 www.routledge-ny.com Published in Great Britain by Roudedge 11 New Fetter Lane London EC4P 4EE www.routledge.co.uk Roudedge is an imprint of the Taylor & Francis Group This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Copyright © 2003 by Taylor & Francis Books, Inc. All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publisher. Library of Congress Cataloging-in-Publication Data Pham, Andrea Hoa. Vietnamese tone: a new analysis / by Andrea Hoa Pham. p. cm. — (Outstanding dissertations in linguistics) Includes bibliographical references and index. ISBN 0 415 96762 7 (hb) 1.Vietnamese language—Intonation. 2.Vietnamese language—Phonetics. I. Title. II. Series. PL4373.P43 2003 495.2′216—dc21 2003046886 ISBN 0-203-50008-3 Master e-book ISBN

ISBN 0-203-57881-3 (Adobe eReader Format)

In memory of my parents

Contents

1

2

3

4

5

6

7

Foreword

ix

Acknowledgments

x

INTRODUCTION

2

Vietnamese

2

Features and Complexity

3

Overview

4

PHONOLOGICAL ISSUES

6

The Traditional Classification

6

The Mismatch Problem

7

A Structural Classification

11

Supporting Evidence

16

THE TONAL INVENTORY

23

Six Tones or Eight?

23

An Eight-tone System

27

The Additional Tones

29

THE ACOUSTICS OF TONE

34

Previous Studies

34

A New Acoustic Study

36

Summary

55

THE PHONETICS AND PHONOLOGY OF TONE

59

Resolving the Mismatch Problem

59

Tonal Shape and Phonation Types

63

The Phonetics of Tone

75

Summary

55

THE DOMAIN OF TONE

84

The Vietnamese Syllable

84

The Significance of Rhyme

86

CONCLUSION

95

Appendix

96

Bibliography

116

viii

Index

119

Foreword

This book is a ‘reader friendly’ version of my 2001 doctoral dissertation submitted to the University of Toronto (Hoa Pham, Vietnamese Tone: Tone is not Pitch). There are no substantive changes only those of an editorial nature. I have given the original dissertation a new title and also made some changes in the internal titles. There is some slight reordering of material and some pruning to eliminate various redundancies. There are also stylistic changes throughout. However, all the original data and claims remain as they were. This version was completed during the first part of a post-doctoral fellowship at York University, Toronto, and later during my first year at the University of Florida. I would like to thank the Social Sciences and Humanities Research Council of Canada for that fellowship and the University of Florida for its financial support. Andrea Hoa Pham University of Florida Gainesville

Acknowledgments

I was very lucky to work under the supervision of Keren Rice. I appreciate her devotion of time, energy, and thoughtful insights to my work. What made me deeply grateful to her was that among my early incoherent ideas and broken data, she spotted a good seed, picked it out, and gave it great care until it blossomed. Nor could the thesis have been written in time without her incredibly fast feedback. I am also grateful to all other members of my thesis committee: Hank Rogers for his great support, both academic and technical; Ron Smyth for his help with the pilot study and useful comments later, and his arrangement for a lovely French restaurant lunch after the defense; Peter Avery for helping me from the very early stage of the thesis; and Parth Bhatt for spending time to discuss my experiment with special enthusiasm. Peter and Ron also helped me considerably with the writing. I am also grateful to my external examiner, Moira Yip, for her fruitful comments. I owe Bill Idsardi for his valuable suggestions, especially with the experiment. I would also like to thank all consultants who were very generous in giving me their time. My sincere thanks also go to Diane Massam for being always so comfortable to talk to; to Jack Chambers for the firm encouragement I could always rely on, and to Nick Hostettler for helping me with the formatting. Special thanks go to a friend who did a great job of editing and wants to remain anonymous. The book would not have been in this shape without that help. The financial support for my research came from various sources: Ontario Graduate Studies Scholarship, Social Sciences and Humanities Research Council of Canada (752–2000–1577), School of Graduate Travel Grant, David Chu Travel Fellowship, General Research Grant from the linguistics department, SSHRC Grant to Professors Rice and Dresher for the Phonology Project, and a University of Toronto Foundation Graduate Award. I would not have survived the program without my friends in Toronto and Vietnam. I cannot name them all, but am grateful for all the support I have received: an encouraging smile in the computer room, a Tango in a milonga, Vietnamese materials which I did not have access to, a pointer toward a possibly related article, a comforting shoulder, and a take-a-breaktrip when desperately needed. I thank my sisters, Vinh, Hi′ n, Hi′ n, Ph′ ′c and brother Bình, for sheltering me with their love and great food during my field work in Vietnam, and my son, Andrew Nguyen, for his love, trust and companionship. Finally, my deepest gratitude to my late parents—for giving me life—and nourishing my spirit. This book is dedicated to them.

VlETNAMESE TONE

CHAPTER 1 Introduction

A very common view in the literature on tone languages is that pitch height is the primary and sometimes only phonetic correlate of tone. This view creates a problem in Vietnamese, namely a mismatch between the phonetic realization of tone and its phonological patterning in reduplication. Is pitch height really the primary phonetic correlate of tone? If ‘tone’ refers to a contrastive phonological category, that category may not in fact be pitch height. It may be that pitch height is not a basic tonal feature and that pitch register is not an adequate phonological feature for representing Vietnamese tones. The research to be reported here shows that the laryngeal features of creakiness and breathiness are primary in signaling tone and that pitch height is derived from these features and from features describing tonal shape. The complexities of Vietnamese tones offer fertile ground to study the nature of tone. First, tones are lexical in Vietnamese and every syllable must bear a tone. Second, tone is not affected by neighboring tones in any environment since there is no tone sandhi. Finally, the tonal inventory of Vietnamese is rich with either six or eight contrastive tones depending on which analysis the researcher adopts. In order to examine the various issues outlined above it is necessary to find answers to a number of questions. What are the features of tones in Vietnamese? In a language with such a rich tonal inventory are tones organized structurally? If they are, what is the internal structure of tonal features? What are the phonetic features of tones? How closely do these phonetic features resemble the phonological features? In the linguistic literature on Vietnamese the question of why a tone patterns in a particular way is seldom posed. For example, in one dialect tone A neutralizes to tone B; however, no consideration is given as to why A does not neutralize to C. Tonal features are never clearly defined according to phonological patterning; rather, they are determined by their phonetic properties. Moreover, when tonal features are discussed, the relationships among such features are never addressed. Rather, tones are represented as bundles of unorganized features such as [high], [tense], [flat], and [glottalic], e.g., Cu et al. 1977, Thompson 1965, Doan 1977, Vo 1997, and Alves 1997. Finally, so far as the phonetics of tones is concerned, the assumption in previous research is that the primary phonetic correlate of tone is pitch height. However, tones in Vietnamese also have laryngeal realizations of breathiness and creakiness. There is no detailed study of the phonetic constants associated with tones. This research proposes and justifies a hierarchical representation of the structure of Vietnamese tones that captures their markedness relations. The model accounts for the patterning of tones in reduplication and for several types of tonal neutralization found in the language. The research also provides phonetic evidence that the phonation properties of creakiness and breathiness are important phonological features of Vietnamese tones and demonstrates how tonal features are realized phonetically. This is the first occasion in the literature on Vietnamese in which both the phonetics and the phonology of tones are unified within a single model. Vietnamese Vietnamese is a Viet-Muong language in the Mon-Khmer group within the Austroasiatic family. Viet-Muong is one of nine subgroups in the Mon-Khmer group. Among the approximately 80 million speakers of Austroasiatic languages, 65 million people spoke Vietnamese in 1988 (Tran 1999:131). It is now spoken by about 78 million people in Vietnam and by about two million speakers living all over the world, with half of this number in North America. The dialects of Vietnamese can be classified into several different major groups depending on the method of classification. For example, a classification based on final consonants divides Vietnamese into two major groups: North Vietnamese and South Vietnamese (Gordina and Bystrov 1970 cited from Hoang Thi Chau 1989 and Pham 1998). Using tones, Vu 1982, Hoang Cao Cuong 1989, and Hoang Thi Chau 1989 agree that there are three major dialect groups: North Vietnamese, Central Vietnamese, and South Vietnamese. Northern dialects have the largest tonal inventory and are the major focus of this research because Hanoi, one of Northern dialects, is considered to be the standard dialect. (1) gives the overall syllable pattern of all Vietnamese dialects.

INTRODUCTION

3

(1)

(3)

Since the focus of this research is on the final segments, the initial consonant is simplified as C. C stands for any consonant element. V can be a short or long vowel or a diphthong with an on-glide. G represents an off-glide, N a nasal consonant, and T a voiceless stop. Each syllable must bear a tone. Only tone and V are obligatory in the syllable. Whereas any consonant can occur in initial position, only a limited number of consonants can occur in final position. Since the manner of final consonants is important in the work reported here (2) gives the full inventory of finals. (2)

Obstruents Nasals Glides

labial

alveolar

palatal

labio-velar

velar

p m w

t n j

c

kp ′m

k ′

The inventory in (2) consists of nasals, obstruents, and two glides. All final obstruents are voiceless stops. Stops are unaspirated or inaudibly released. Palatal, labio-velar, and velar are allophones, determined by the quality of the preceding vowel. Features and Complexity This study follows Clements 1985, McCarthy 1988, and others in assuming that features are organized in a hierarchical configuration. The hierarchical structure allows a set of features to undergo certain phonological processes without affecting features of another group. Elements that pattern together as a natural class are dominated by the same organizing node. (3) illustrates the working model. R is a root node, X and Y are organizing nodes, and z and p are content nodes. The sisterhood relationship between the two organizing nodes X and Y in (3) allows a process to affect X without affecting Y. Markedness itself is a measure of the way in which the relative complexity of the system is encoded (Archangeli 1988). The theory of Contrastive Specification (Steriade 1987, Mester and Itô 1989, and Avery and Rice 1989) in the particular version adopted here (after Avery and Rice 1989) holds that all features are monovalent. A feature must be present underlyingly if it is contrastive. Non-contrastive or unmarked features, which play no role in the phonology of a language, are absent from underlying representations. Moreover, while the unmarked features are generally absent underlyingly, organizing nodes are present in an underlying representation if they define a property inherent in the segment (following Rice 1993). A feature must be present underlyingly if it produces contrasts in the system. Markedness can be overridden by contrast within an inventory on the assumption (Rice and Avery 1993) that structure is elaborated under pressure from the phonology to contrast two sounds. As contrasts are introduced, more complex structures are added to the existing less complex ones already in the inventory. The more marked a segment the more structure it has, e.g., Avery and Rice 1989, Rice and Avery 1991, Kaye et al 1985. The complexity of structural representations, therefore, increases as the number of contrasts in the inventory increases. There are two types of complexity (see Dresher and van der Hulst 1998): local and non-local. A node C has local complexity if it branches when others do not, as in (a) in (4), or if C has an immediate dependent when others do not, as in (b). The second type of complexity is non-local and addresses issues concerning the internal structure of the daughters of a node. For example, in (5), node C has non-local complexity if C dominates D and D dominates E and F, as in (a). It is simple if C dominates D and D dominates E but does not branch, as in (b). Within non-local complexity, a structure with more specified features in a hierarchical system, either vertically or horizontally, counts as more complex than a structure with fewer specified features; consequently, (a) in (6) is vertically more complex than (b). The difference between the two is that (a) has an extra dependent feature G under F. In (7), (a) is horizontally more complex than (b). The only difference between the two is that in (a), both D and E have dependents while in (b) only D has a dependent.

4

VIETNAMESE TONE: A NEW ANALYSIS

(4)

(5)

(6) Vertical complexity

Default rules act as a phonetic implementation component for an underspecified feature (Rice and Avery 1989). A delinking rule serves to delink the content nodes in a neutralization environment. In Vietnamese, delinking of either the laryngeal or the contour features of tone occurs to yield unmarked features or structures. For example, a more complex, i.e., more marked, tone is reduced to a less complex, i.e., less marked, one. Overview This research study is organized as follows. Chapter Two presents a proposal for the structure of Vietnamese tones based on markedness and provides phonological evidence for that structure. Markedness of tones is based on their complexity, which is evaluated according to tonal patterning. Evidence for the markedness of tones is drawn mainly from tonal neutralization. The chapter analyzes the six tones that occur in the same environment, namely sonorant-final syllables. It also introduces the classification of tones in the traditional literature and the mismatch problem that this classification creates. Chapter Three discusses the size of the tonal inventory, either six tones or eight. Evidence is advanced to support the eight-tone hypothesis and arguments are given for the representation and markedness of the two tones that occur in obstruent-final syllables.

INTRODUCTION

5

(7)

Chapter Four deals with the phonetics of tones. It reviews the acoustic experiments in the literature and presents an experimental study of tones produced by speakers of the Northern dialect. It examines various aspects of tones: fundamental frequency, shape, length, linear portion, and phonation types such as creakiness and breathiness. One of the most important findings is the instability of the fundamental frequency, i.e., F0. The F0 of pitch varies from speaker to speaker, from token to token, and from form to form. However, phonation types are the most stable feature across speakers and tokens. Another finding is that the end point of a tone is the most crucial portion of a tone for differentiating that tone from others, except in curved tones where the middle part is crucial. Chapter Five presents the implications of the phonetic results for the phonology of tones. Phonation type is the primary phonetic correlate of tone in Vietnamese. Pitch height is not constant but is predictable on the basis of phonation type. Use of the phonation features of creakiness and breathiness as laryngeal features eliminates the mismatch between the phonetics and phonology of tone in reduplication. The chapter also includes a discussion of the phonetics of tone and shows how the phonology maps onto the phonetics. Chapter Six discusses the domain of tones and shows that in Vietnamese the domain of tone is not the syllable, nucleus, or mora. It is smaller than a syllable and larger than a nucleus; it is the rhyme. Chapter Seven provides a very brief summary of the major findings.

CHAPTER 2 Phonological Issues

This chapter is concerned with the phonology of Vietnamese tones and focuses on their structural representations and their markedness with respect to one another. The evidence for the markedness of different tones is drawn mainly from tonal neutralization in various phonological domains. The chapter provides background information on the distribution of tones and their phonetic descriptions, proposes a hierarchical structure for tones and gives evidence for the structure, reviews tonal representations in the traditional literature, and identifies some of the problems in those accounts. The Traditional Classification Vietnamese has eight tones (see also Chapter Three) ngang, huyen, sac1, nang1, hoi, nga, sac2, and nang2 and (8) shows their distribution. Ngang, huyen, sac1, nang 1, hoi, and nga occur in open or sonorant-final syllables. Only sac2 and nang2 occur in stop-closed syllables. Chapter Three discusses the phonological status of these tones and Chapter Four provides phonetic details. (8) Syllable types

e.g.

ngang

huyen

sac1

nang1

hoi

nga

CV CVG CVN CVT

la la:w l′ n la:t

+ + +

+ + +

+ + +

+ + +

+ 4 +

+ + +

sac2

nang2

+

+

(9) provides a set of minimal pairs for these tones. The Vietnamese orthography is in the third column. The orthography uses only six diacritics: sac1 and sac2 share the same diacritic. Likewise, nang1 and nang2 share the same diacritic. (9) Tone

Phonetics

Orthography

Gloss

(ngang) (huyen) (sac1) (nang1) (hoi) (nga) (sac2) (nang2)

l′ n l′ n l′ n l′ n l′ n l′ n 1′ t l′ t

lân l′ n l′ n lân l′ n l′ n l′ t lât

unicorn/neighbor/near turn/time/layer/to totter away to surpass/ to jostle to tuck/to defraud to slip away, to escape crack-brained/to confuse unstable, unreliable (l′ t l′) to overturn/to capsize/chestnut

This chapter is concerned with only the first six tones, i.e., ngang, huyen, sac1, nang1, hoi, and nga, since they occur in the same environments. Figure 1 shows the pitch tracks of the six tones in the open syllable [ta] from Nguyen and Edmondson 1997. (In this figure the transcribed vowel carries a diacritic for tone.) Tones are generally presented in pairs for both phonetic and phonological reasons. The pair in 1(a) are the level tones ngang and huyen. Ngang is high in pitch and huyen is low. The pair in 1(b) are sac1 and nang1. Sac1 is a high rising tone and nang1 is a low falling tone ending with a glottal stop. The pair in 1(c) are nga and hoi. Nga is a high falling-rising tone broken by a glottal stop in the middle of the tone. Hoi is a low falling-rising tone.

VIETNAMESE TONE: A NEW ANALYSIS

7

Figure 1. Six tones in the open syllable [ta] (from Nguyen and Edmondson 19 1997:6)

These phonetic descriptions yield two parameters that are generally used to describe Vietnamese tones, e.g., Doan 1977 and Hoang Cao Cuong 1989. The first refers to tonal shape: level (ngang, huyen); contour (sac1, nang1); or curve (hoi, nga). The second involves tonal height: high (ngang, sac, and nga) and low (huyen, nang, and hoi). These tones are thus generally classified as in (10). (10)

high low

level

contour

curve

ngang huyen

sac1 nang1

nga hoi

The Mismatch Problem The usual representation of tones in the literature on Vietnamese is one based on tonal shape and tonal height. The assumption that register is equivalent to pitch is also common. The term ‘contour’ in the traditional literature clearly refers to tonal shape. Tones are thus classified according to their pitch and shape. (11) summarizes the classification of Vietnamese tones found in Doan 1977, Vuong and Hoang 1994, and Hoang Thi Chau 1989. The first parameter is the tonal shape or contour. According

8

VIETNAMESE TONE: A NEW ANALYSIS

to this parameter, ‘level’ indicates that the tone is basically flat. The level tones are ngang and huyen. The non-level tones are sac1, nang1, hoi, and nga. Among these tones, nga and hoi are broken, i.e., the tone goes down then rises to make a curved shape, and sac1 and nang1 are nonbroken, i.e., the tone either goes down or rises (see Chapter Four for details). This view of contour is consistent with the one adopted in this study. The second parameter is tonal height, which divides tones into two registers defined by pitch: high register consists of the high tones ngang, sac1, and nga, and low register consists of the low tones huyen, nang1, and hoi. The justification for [high] and [low] in (11) is discussed in Chapter Four. (11) Traditional classification of tones according to tonal shape and height level

High Low

non-level

ngang huyen

non-broken

broken

sac1 nang1

nga hoi

Evidence for the two registers high and low is drawn from phonetics, e.g., Doan 1977 and Vuong and Hoang 1994. Evidence for tonal contour comes from tonal patterning in poetry, i.e., level tones form one group and non-level tones form the other group. These sources are briefly mentioned as evidence for such a classification but no systematic accounts are given. However, as we will see, the situation is different in reduplication. The classification in (11) is standard with the proviso that the terms there are used primarily to classify tones and not to specify the features of tones. (12) provides an alternative classification. (12) Unmarked register: Marked register:

ngang huyen

sac1 nang1

hoi nga

The major difference between the traditional classification in (11) and the classification in (12) is that the latter is based on a phonological patterning in which hoi forms a class with ngang and sac1 in terms of register and nga forms a class with huyen and nang1. In the traditional proposal, the registers of hoi and nga are reversed. (Evidence for (12) is presented later in this chapter.) Tonal features are also proposed in the literature, features such as glottalization, length, tense, and lax, but no evidence is provided. For example, (13) shows tone features from Thompson 1965 and Alves 1997. (13) also assumes that there are only six tones in Vietnamese. (13) Tone

high

tense

glottalic

short

ngang huyen sac nang hoi nga

+ − + − − +

− − + + + +

− − − + − +

+/− +/− + + − −

(13) uses tonal height to classify tonal registers. One problem with (13) is that these features do not group tones into natural classes. For example, ngang, sac, and nga are grouped under the feature [high] but they do not pattern together as a group. In addition, although nang and nga have the feature [glottalic], they do not pattern as a pair. Hoang 1986 argues that two parameters alone, pitch height and contour, cannot account for tone languages such as Vietnamese, whose tones developed from final consonants (see Chapter Three for details) and still show some relic of those segments through glottal stop or breathiness. In an attempt to incorporate phonation types as tone features, he proposes the tonal features in (14). This proposal classifies tones based on the diachronic hypothesis that a tone which developed from a final consonant may still retain some properties of that segment. For example, the glottal stop in the tone nang1 is treated as a relic of the historical final glottal stop. In (14), the phonation column indicates the presence or absence of glottalization or pharyngealization in a tone. The second column indicates whether the tone falls within its register, i.e., the pitch contour does

VIETNAMESE TONE: A NEW ANALYSIS

9

(15)

not cross two registers. The last column indicates the pitch height for register features. While Hoang classifies eight tones according to their phonetic properties, the phonological status of the sac and nang tones is left unclear. (14) Tone ngang huyen nga hoi sac1 nang1 sac2 nang2

1 2 3 4 5 6 5’ 6’

phonation

falls in one register

low

− + + − − + + +

+ + − − − − + +

− + − + − + − +

The features in (14) are classified phonetically. Although Hoang argues that this classification of tones agrees with the traditional treatment in music and in the frequency of occurrence of tones in reduplicative forms, he provides no evidence. He emphasizes that (14) is based primarily on phonation, and reflects the historical development of tones. However, like the classification in (13), these features do not account for the major patterning of tones in the language, since nga, nang1, sac2, and nang2 do not pattern as a group although, according to the classification in (14), they share the same feature for phonation. Hoang recognizes that it is problematic to use pitch height to classify Vietnamese tones because while some tones fall completely within one register (high or low), other tones fall across the two registers. Chapter Five takes up this issue again. All the featural accounts discussed in this section fail in one major way: the proposals do not capture classes that the phonology requires to be natural. Other recent works use feature geometry to represent tones as constituents but do not provide evidence for the organization of tones. For example, using the features in (14), Hoang 1986 suggests the feature geometry in (15) for Vietnamese tones. The numbers refer to the tones as follows: 1 is ngang, 2 is huyen, 3 is nga, 4 is hoi, 5 is sac1, 5’ is sac2, 6 is nang1, and 6’ is nang2. In the abbreviations ‘ng’ stands for ngang, ‘hu’ for huyen, ‘sa1’ and ‘sa2’ for sac1 and sac2, respectively, and ‘na’ and ‘na2’ for nang1 and nang2, respectively. However, these features do not group tones in the way that they pattern in the language. For example, there is no evidence that 3, 6, 5’, and 6’ (nga, nang1, sac2, and nang2) in the top row pattern as a group or that 5 and 4 (sac1 and hoi) in the last row pattern together as a pair, and so on. It is problematic to use only phonetic features such as length, glottalic, tenseness, and so on to classify tones. For example, the feature [glottalic] groups nga and nang1 into a class that does not reflect their phonological patterning. In an attempt to avoid the difficulty with the feature [glottalic], Ngo 1984 suggests another feature, one that makes use of the tonal shape, namely [concave], to group hoi with nga. In this system, tonal representation is simply a reflection of the pitch shape of tones or their contour (1984:75). (16) shows features of tones in this account. (16)

Concave Contour

ngang

huyen

sac

nang

hoi

nga

− −

− −

− +

− +

+ +

+ +

10

VIETNAMESE TONE: A NEW ANALYSIS

(17)

High

ngang

huyen

sac

nang

hoi

nga

+

−

+

−

+

−

Using the tonal features in (16), Ngo represents tones hierarchically in a binary branching model, as in (17). This model uses [concave] at the highest level to dominate tonal contour. The feature [±high] is a dependent of [contour]. The high tone hoi is [+concave, +contour, +high]. The low tone nga is [+concave, +contour, −high]. According to Ngo, this is a morphological representation, since hoi is phonetically low but morphologically high and nga is phonetically high but morphologically low. There is an unexplained gap in the model: the absence of two tones under [+concave, −contour]. This gap is also seen in Doan 1977:107, who explains that in reality there are no tones which are ‘even’ but ‘broken’ (i.e., [−contour] and [+concave] in Ngo 1984). Using binary branching with three different levels, the model seems to be appropriate for an eight-tone system because the combinations give eight possible terminal nodes, but it is not completely filled. The two tones ngang and huyen should occupy the two unspecified nodes. Authors such as Burton 1992, Avery 1983, and Ngo 1984 agree about their unmarkedness because of their frequency, their patterning in reduplication, and borrowings. However, ngang and huyen do not occupy these nodes. Furthermore, this model does not capture the distributional fact that the four tones ngang, huyen, hoi, and nga occur only in sonorant-final syllables. Although the above proposals classify tones into pairs, there are no relationships between these pairs; therefore, the tonal pairs appear to exist as autonomous components within the tonal system. Accordingly, there are no adequate representations that can account for the various patterns of tones in Vietnamese. The division of tonal registers in (11) and the assumption that register is tonal height also cause a serious mismatch between the phonetics and phonology of tones. It creates a problem in reduplication with respect to hoi and nga, one which the research reported here will resolve through proposing the alternative classification given in (12) above. In the traditional view tones pattern in two registers: ngang and sac1 pattern together and huyen and nang1 pattern together. Ngang and sac1 are high tones and huyen and nang1 are low tones. The phonetics of these tones predicts their patterning, i.e., their phonetic classification underlies their phonological classification. Given this patterning, nga, a phonetically a high tone, should pattern phonologically with ngang and sac1, and hoi, a phonetically low tone, should pattern with huyen and nang1. However, hoi and nga do not follow the predicted pattern. The patterning is as in (12). Many efforts have been made to explain the unusual behavior of hoi and nga in reduplication. Some researchers, e.g., Doan 1977:123, Hoang Thi Chau 1989:207, Vuong and Hoang 1994:100, and Nguyen and Edmondson 1997, look at historical developments for an explanation of this phenomenon. They adopt Haudricourt’ s suggestion (1954) that hoi was historically a low tone and nga a high tone, and they suggest that hoi and nga switched their registers during the evolution of tones. However, they do not clarify how this happened. Ngo 1984 and Burton 1992 propose abstract representations of tones to solve the problem with hoi and nga in reduplication. For example, in order to account for the phonetics-phonology mismatch with hoi and nga, Ngo 1984:78 posits a Concave Tone Reversal rule (18) that changes the phonological high tone hoi to a phonetic low tone and the phonological low tone nga to a phonetic high tone. This rule has no theoretical motivation: there are apparently no languages in which a phonetically back vowel patterns phonologically as if it were a front vowel or vice versa. Such a system would necessitate such a flip-flop rule; (18) is highly suspicious at the very least. Burton 1982 also examines the behavior of hoi and nga in reduplication and proposes (19) as a representation of Vietnamese tones. Adopting Yip’s 1980 model, Burton uses upper case +HIGH/ -HIGH for Register and lower case h, 1, 1h for contour features. In this model, each tonal pair shares the same feature for contour, i. e., both ngang and huyen are ‘1.’ They differ only in the register feature, i.e., +/− HIGH. (19) CONTOUR REGISTER − HIGH

+ HIGH D=1 (huyen)

A=1 (ngang) E=h (nang)

B=h (sac) F=1h (nga)

C=lh (hoi)

VIETNAMESE TONE: A NEW ANALYSIS

11

(18)

(20) Flip-Flop rule

Burton adopts the view that sac2 and nang2 are variations of sac1 and nang1 in stop-final syllables (see Chapter Three). Burton’s classification is abstract in that it groups hoi, a phonetic low tone, with phonetic high tones, and nga, a phonetic high tone, with phonetic low tones. However, at the same time, he also claims that ‘Upper vs Lower pitch-range is exactly what characterizes Register.’ Burton also classifies ‘contour features’ using both phonological and phonetic properties. (This issue is further discussed in Chapter Five.) As for tonal contour and following the theoretical approach of Yip 1980, Bao 1999b, and Duanmu 1990, Burton says that hoi and nga are rising ‘lh.’ (Burton’s theory does not allow a dipping tone, i.e., ‘hlh.’) Burton treats the initial dip in these tones as a phonetic effect. However, with respect to register, while arguing that ngang and sac are [+High] register and huyen and nang are [−High] register because all are articulated in the upper and lower parts of their respective pitch-ranges, Burton offers no evidence as to why hoi is [+High] and nga is [−High]. How does Burton account for the mismatch between hoi and nga? He proposes that hoi and nga must switch their register to achieve the appropriate phonetic representations. He argues that the surface representations of hoi and nga result from an interaction of two rules: Register Flip-Flop and Creakiness Acquisition, as in (20). Other authors have captured the fact that hoi patterns phonologically with high register tones and nga with low register tones by suggesting that, diachronically, hoi was a low tone that changed to a high tone. They must then assume that the phonology of reduplication makes no reference to natural classes. Unlike these authors, Burton argues that hoi is synchronically [+High]. He explains the phonetic surfacing of hoi as a high tone by a Flip-Flop rule that switches its register. Likewise, nga is underlyingly [−High] but surfaces as [+High] by a Creakiness Acquisition rule to acquire the phonetic feature [creaky] and then by a Flip-Flop rule to switch its register. Register Flip-Flop changes hoi from an underlying [+High] to surface [−High]. Consequently, Burton’s Flip-Flop rule is equivalent to Ngo’s Concave Tone Reversal in (18); the major difference between these analyses is simply whether the tonal switch is considered to be diachronic or synchronic. It is quite clear that tonal features have been largely based on phonetics in the traditional literature on Vietnamese. These features do not group tones into natural classes. However, the patterning in reduplication and neutralization shows that a classification of tones that relies only on a phonetic interpretation of register as high and low is quite unsatisfactory since it leads to the mismatch problem between the phonetics and phonology of the tones hoi and nga. The analyses based on this type of classification must then account for the mismatch with an unnatural rule that switches the registers of these two tones. A different approach to these issues is clearly warranted. A Structural Classification In the literature on tone, including standard works on phonetics, the terms ‘contour’ and ‘register’ are not at all well-defined. Generally, ‘contour’ refers to tonal shape and ‘register’ to tonal height. For example, Pike 1948 divides tone languages into two categories, register tone languages and contour tone languages, depending on which feature of pitch behavior is significant in the language. Laver 1994:465 proposes three types of tone languages. The first type is the register tone system. This is a system in which the relevant feature of word-identifying pitch behavior is the relative height of the syllabic pitches within the speaker’s pitch-span. The second type is the contour tone system. This is a system in which the relevant feature is less the relative height of the tone and more its shape or trajectory, together with its general placement in the speaker’s pitchspan. The third type is a mixed register/contour tone system. This is a system in which the end point cannot be identified directly with any of the level tones. Representations of tones and tonal features almost invariably reflect the assumption that register is pitch. For example, Yip 1980 proposes two features for tone, as in (21): a feature [Upper] for register, and a feature [High] for tone. The feature [Upper] is a binary feature, which, according to Snider 1999:152, is either high or low and indicates whether the tone is in the higher or lower register.

12

VIETNAMESE TONE: A NEW ANALYSIS

(23) (21)

Register: Pitch: (Yip 1980)

Upper High

Raised H

H

Mid

Low

+ +

+ −

− +

− −

Bao 1999b: 3 claims that register is the pitch level of a tone and contour is how the pitch behaves over the duration of the tone-bearing unit so that register is static and contour is dynamic. Assuming that register is tonal pitch, Bao 1999a: 487 describes a phenomenon called ‘register harmony’ in tone sandhi in Chaozhou (Chinese) as follows: ‘the pitch height of the sandhi tone—its register—is determined by the register of the following tone.’ The assumption that register is tonal pitch is also shared in work on African tone languages, e.g., Snider 1999. Clements 1981 proposes the representation of tones in (22). The register feature in the first row indicates that the tone is in the upper or lower register and the pitch feature in the second row indicates the higher or lower tone within that register. (22)

Register: Pitch:

Raised

High

High

Mid Low

h h

h 1

1 h

1 1

The concepts ‘register feature’ and ‘tone feature’ seem to include tonal height. The register feature indicates a pitch range, usually divided into two registers, high and low. The tone feature refers to the pitch of that tone within its register, e.g., a high tone in a low register is a low tone that is articulated at the highest point within the low register. It is reasonable to assume that contour indicates tonal shape. However, register does not necessarily involve pitch height: it can also be used to represent different phonation types, as in Chapter Five. Chapters Four and Five will also show that the particular register feature that is needed to describe Vietnamese tone is a laryngeal feature. Tonal features belong to a level of structure that is independent of segmental features, e.g., Leben 1973, Goldsmith 1976, and Odden 1995, and there are two major features in tonal representation, namely Register and Contour. Register is not tonal pitch but phonation type. Specifically, it refers to modal voice, creaky voice, and breathy voice. Chapters Four and Five provide the phonetic evidence for this claim. In the work reported here the term ‘contour’ refers to tonal shape: whether a tone is flat (level in tone) or whether there is some movement during the course of the tone. Movement can be of two types. It can involve a single change of direction, e.g., going up or down to make a rising or falling tone, or more than one change of direction, e.g., going down and then up to make a falling-rising tone. Tonal contour, therefore, does not refer to tonal height, so a ‘level’ contour might be a ‘high’ or ‘low’ tone, or a ‘falling’ tone might be ‘high’ in the pitch range. Tonal contour refers to the shape of the tone. In Vietnamese the structural relationship between Register and Contour is one of sisterhood not dominance. The model (similar to that in Bao 1999b) is shown in (23). In it Contour and Register are two organizing nodes dominated by a tone node. This model predicts that some phonological processes can affect the whole tone (T) while others may affect only Register but not Contour, or vice versa. The following pages provide evidence from neutralization for the independence of Contour and Register features, for constituency in the feature geometry of tones, and for a hierarchic arrangement of tones. (24) provides a structure for Vietnamese tones with the terms used here only partially defined (but see also Chapter Five where these terms are further discussed and justified). Some terms in (24) are used only for descriptive and classificatory purposes, e.g., [nonlevel] and [curve]. However, the formal representation of each tone follows in (28). The tonal root node in (23) is represented by the tone’s name in (24); later, in (28), that name is used for referential convenience. In (24), the two major components of tones are Contour (C) and Register (R), as in (23). The term [laryngeal] is used as a marked feature for register and appears in huyen, nang1, and nga. The feature [laryngeal] has [spread] as its dependent in huyen and [constricted] as its dependent in nang1 and nga. The evidence for the register features and their phonetic realization is presented in Chapter Five. With respect to the contour feature, the tones ngang and huyen are essentially flat and are not specified as [even]. The feature [non-even] (changing direction) appears for the rising and falling tones sac1 and nang1. The feature [curve] appears for the two curved tones (down and then up) hoi and nga.

VIETNAMESE TONE: A NEW ANALYSIS

13

(24)

(c) (25)

Tones can be captured through the use of points, notated by •. These points indicate whether there is movement during the tone. A single point indicates a level tone. Because contour shows the F0 value of a tone during the time of production, at least one point needs to be specified for each tone in order to generate the tone contour. If the tone is level, i.e., there is no movement during its production, one point is the beginning point (onset), marked with •, and it generates the end point (offset), as in (25). (The register feature is omitted for convenience.) The underlying contour feature is specified for one point in (a). While a level tone has only a single point phonologically, it nevertheless extends phonetically over a tone period. On the surface in (b), this point generates the second point to end the tone. If there is movement during the tone, two points are required at the phonological level to indicate the contour, i.e., movement, of the tone, e.g., whether it is rising or falling. These two points at the phonological level must be interpreted as having different values. The movement from the onset to the offset of a rising or falling tone is illustrated in (26). (Note that the height of the points makes no reference to anything that occurs in the surface forms; it is simply an illustrative device.) In (a), a contour tone (a tone with movement) has two points phonologically. If it is a rising tone, the second point is higher phonetically than the first point to indicate the rise, as in the first example in (b). If it is a falling tone, the second point is lower phonetically than the first point on the surface to indicate the fall, as in the second example in (b). If the tone is a curved tone, e.g., the tone goes down and then rises, three points are needed to indicate the complex movement phonologically, as in (27). A concave tone is phonetically indicated with the mid point lower than the first and last points to show that the tone goes down and then up, as in the first example in (b). A convex tone is indicated phonetically with the mid point higher than the

14

VIETNAMESE TONE: A NEW ANALYSIS

(26)

(27) Specification of the contour of a tone that changes direction more than once

(28)

first and last points to show that the tone goes up then down, as in the second example in (b). The phonetic interpretation of contour and register features is further discussed in Chapter Five, where pitch will be shown to be derived from the register and contour features. (28) provides a phonological representation for each tone.

VIETNAMESE TONE: A NEW ANALYSIS

15

(30)

The theory of Contrastive Underspecification, e.g., Avery and Rice 1989, Rice 1992, and Wu 1994, holds that an unmarked feature is absent underlyingly unless there is a contrast which forces it to be present, and allows for the omission of unmarked features in (28). The register feature [laryngeal] is marked; therefore, it is present underlyingly in huyen but absent in ngang in (a) in (28). The level tone ngang has no specified features. Its counterpart huyen has the register feature [laryngeal] with [spread] (breathy, see Chapter Five) as its dependent. In each level tone, one point is specified for the contour feature, marked with •, as shown in (27). This point does not need to be specified as ‘h’ or ‘I,’ because this value is predictable from the laryngeal feature. (See the discussion of the phonetic realization of tone in Chapter Five.) In (b) in (28), sac 1 does not have a register feature but does have a contour. As there is movement in this tone, two points are required to represent the contour, marked with •. Sac1’s counterpart, the falling tone nang1, has the register feature [laryngeal] with the feature [constricted] (creaky, see Chapter Five) as its dependent. On the contour side nang1 is specified for two points to show the movement. (c) in (28) shows the representation of the curved tones hoi and nga. Hoi does not have the register feature. It has a contour specified for three points to generate a curved tone. Its counterpart nga has features on both sides. On the register side nga has a laryngeal feature with [constricted] as its dependent. On the contour side, as in hoi, nga has three points specified to represent the curve. The representations in (28) show that each pair of tones shares the same contour: ngang and huyen have only one point specified phonologically; sac1 and nang1 have two points specified phonologically to show that these tones change direction; and hoi and nga have three points specified phonologically to show that these tones change direction more than once, i.e., they go down and then rise (there are no rising-falling (lhl) contours in Vietnamese). On the register side the representation in (28) shows that tones on the left (ngang, sac1, and hoi) do not have the register feature. The tones on the right share the same register feature [laryngeal] with a dependent. The markedness relations of tones are based on structural complexity. Four kinds of markedness apply to tones: markedness within register; markedness within contour; markedness between pairs; and markedness within pairs. The representations in the left column of (28) show that the unmarked register consists of tones that are unspecified for register, i.e., ngang, sac1, and hoi. In the right column the marked register consists of tones that are specified for register with the features [spread] or [constricted], i.e., huyen, nang1, and nga. Tones in the right column, therefore, are more complex than those in the left column with respect to register, and thus are more marked (see (7) in Chapter One). For example, ngang and huyen have an equal degree of complexity on the contour side, with only one specified point for each; however, on the register side huyen has the feature [laryngeal] with [spread] as its dependent. In this case huyen is horizontally more complex than ngang. In terms of contour, ngang and huyen are simpler than sac1 and nang1, and these in turn are simpler than hoi and nga. A comparison of ngang and sac1 shows that neither of these tones has features under the register node. However, on the contour side sac1 has two points specified for the contour feature while ngang has only one. Therefore, sac1 is horizontally more complex than ngang (with two points of contour). Likewise, a comparison of huyen and nang1 shows that nang1 is vertically more complex than huyen (with two points specified for the contour). Although both huyen and nang1 have the [laryngeal] feature on the register side, [constricted] is more marked than [spread]. A comparison of the pair sac1 and nang1 with the pair hoi and nga shows that on the register side sac1 and hoi are equally complex in that they lack the register feature. Nang1 and nga are equally complex: both have the feature [laryngeal] with [constricted] as its dependent. However, on the contour side sac1 and nang1 are horizontally less complex than hoi and nga. Hoi and nga have three points specified for the contour feature whereas sac1 and nang1 have two. Overall, the simplest structure is that of ngang with only one contour point specified. The most complex tone is nga with the laryngeal feature [constricted] and three points specified for the contour. (29) summarizes the markedness relationship of tones between registers. (29) Unmarked register Marked register

ngang huyen

sac1 nang1

hoi nga

Within each pair in (29), the tone on the upper row is less complex than its lower counterpart according to the complexity measure of (7). For example, in (a) in (28) ngang does not have the register feature while huyen has the feature [spread] under Register. (30) summarizes the markedness relations of tones in opposite registers. Here ‘<’ indicates that the lefthand member is less complex or less marked than the righthand member.

16

VIETNAMESE TONE: A NEW ANALYSIS

(32)

In terms of contour, the markedness of tones is also evaluated among tonal pairs across registers. Ngang-huyen is the least marked of the three pairs. These tones have only one point specified for the Contour node. The second least complex pair is sac1-nang1 Both tones have two points specified for the contour node and sac 1 is unspecified for register. This pair is also horizontally more complex than the pair ngang-huyen according to (7). Finally, the most complex pair is hoi-nga: it is horizontally more complex than sac1-nang1 with three points specified for the contour node. (31) summarizes the markedness relations among tone pairs. (31) ngang-huyen<sac1-nang1
tan (huyen) lanh (nang1) mo (nga)

‘worn out’ ‘cold’ ‘grease’

> > >

tan ta (huyen—nang1) lanh leo (nang1—nga) mo mang (nga—huyen)

‘worn out’ ‘very cold’ ‘very greasy’

Register preservation in tone harmony is strictly preserved even when there is no distinction between certain tones in some dialects. For example, in Southern dialects nga is neutralized to hoi. Because the distinction between hoi and nga is still expressed

VIETNAMESE TONE: A NEW ANALYSIS

17

in the orthography and in productive reduplication, Southerners must memorize the rule of ‘ngang-sac-hoi’ and ‘huyen-nangnga’ in order to decide whether a morpheme has the tone hoi or nga (Hoang Thi Chau 1989: 201). For example, in ‘nong nay’ (temper), if ‘nong’ has sac1, ‘nay’ must be hoi, which is in the same register as sac1. In reduplication register must be retained in tone harmony; however, contour can vary. Tones in the unmarked register pattern together in one group (ngang, sac1, and hoi), and tones in the marked register comprise another group (huyen, nang1, and nga). This patterning provides evidence for two registers in Vietnamese. These registers are independent of contour features since tonal patterning in reduplication affects only Contour but not Register, e.g., an unmarked register tone can pattern with another unmarked tone that has a different contour feature. Researchers such as Bao 1999a, b, Duanmu 1990, and Snider 1999 also argue for Register and Contour being sisters. Yip 1989, 1995 proposes a model of tonal representation in which a register feature dominates a pitch feature and argues that tones spread as a whole. (34) shows a high rising tone in Yip’s model with ‘H’ standing for register and ‘I, h’ standing for pitch. The fact that Register is preserved in Vietnamese reduplication is a challenge for such a system, regardless of the formal details of how tones are determined in reduplication, by either spreading or copying of register. If Register dominates Pitch as in (34), the reduplicant tone would have the same pitch feature, i.e.,‘Contour’ in this study, as its base, but such is not the case in Vietnamese, where reduplicant tones do not share contour features with their base tones. Yip 1995:487 also notes that spreading in African tone languages usually shows that voicing and Register do not interact: high and low tones spread freely across syllables and across any kind of consonant, as Hyman et al, 1987 show for Luganda. This type of spreading shows that only Pitch is involved not Register. Yip suggests that there are different types of dependency relations between Pitch and Register in Chinese and African languages. The model in (23), in which Contour and Register are independent, suggests a straightforward account for both types of spreading: in African languages only Contour spreads; in Vietnamese reduplication only Register spreads. Reduplication, therefore, provides evidence for the independence of contour and register features. In addition, it provides evidence that the tonal pair ngang-huyen is unmarked compared to the other tones. It is widely recognized that in certain positions only a limited number of segments or features is allowed, e.g., Rice 1996 and Paradis and Prunet 1989. For example, almost all segments in the inventory of a language are allowed in syllable-initial position but fewer occur in syllable-final position. In Korean, regardless of their particular coronal place and manner features, all coronal obstruents are neutralized to [t] in coda position (Cho 1991:171). In the Saigon dialect of Vietnamese, while all consonants (except /p/ (Pham 1998)) occur initially only coronals or velars can surface syllable-finally. With respect to laryngeal features, German allows only voiceless obstruents in final position and voiced obstruents must devoice in this position (Brockhaus 1995 and Jessen 1998). Features that are required in certain environments are said to be unmarked. Therefore, in the examples involving coronal consonants in which [coronal] is the only place of articulation allowed in the coda, [coronal] is unmarked. (See Greenberg 1966 for a discussion of neutralization as a characteristic of lack of markedness.) If unmarked features occur in neutralization environments, the structure in (28) predicts that if there are tones occurring in such an environment, they should be ngang or huyen, the least marked ones. This is indeed the case: ngang and huyen are found in neutralizing environments, as in productive reduplication. This process neutralizes a base tone to ngang or huyen. The particular form depends on the register of the base tone. If the base tone is an unmarked register tone, either sac1 or hoi, the reduplicant tone must be ngang, another unmarked register tone, as in (a) in (35). If the base tone is a marked register tone, either nang1 or nga, the reduplicant tone must be huyen, another marked register tone, as in (b) in (35).

18

VIETNAMESE TONE: A NEW ANALYSIS

(36)

(38) 1 E E

2 E E

3 N N

4 N N

5 E E

6 E E

7

8

N

E

The model used here accounts for this process, as shown in (36). The reduplicant tone is pre-specified as a level tone, i.e., its contour is specified. Its register is predictable, being the same as the register of the base. (Note that there is no register feature in (a), but there is one in (b)). Further details of this process are discussed in Chapter Five; the important issue here is that less complex contours occur in neutralization environments. A level contour is a less complex contour than a rising and falling one. This process shows that ngang and huyen are two unmarked tones, here occurring in a neutralizing environment. Frequency of occurrence can be a further diagnostic for the markedness of tones. A less marked feature occurs more frequently than a more marked one, e.g., Greenberg 1966, Maddieson 1984, and Hamilton 1996. The ranking of markedness of tones in (29), (30), and (31) predicts that among tones the least marked tone ngang should have the highest frequency and the most marked tone nga should have the lowest frequency. Among pairs the least marked pair ngang-huyen should occur the most frequently and the most marked pair hoi-nga the least frequently. Vo 1997:20 examined 4243 monosyllabic words in a Vietnamese dictionary. Omitting syllables ending in the stops p, t, and k, he provides the raw numbers of occurrence for each tone in sonorant-final syllables, as in (37). (37) Total 4243

ngang 1029

huyen 840

sac1 845

nang1 606

hoi 570

nga 353

(37) shows a striking match between frequency and markedness. The unmarked tone ngang occurs more than its more marked counterpart huyen (1029 and 840 instances respectively). Likewise, sac1 occurs more than its more marked counterpart nang1 (845 and 606 instances respectively) and hoi occurs more than its more marked counterpart nga (570 and 353 instances respectively). Finally, among pairs, the least marked pair ngang-huyen occurs the most often and the most marked pair hoi-nga occurs the least often. Poetry provides additional evidence that is compatible with the claim that ngang and huyen form a tonal pair. In a traditional verse called six-eight verse, there is a pattern of alternation between even tones, i.e., Vietnamese ‘bằ ng’ (ngang and huyen) and Vietnamese ‘trằ c,’ the so-called sharp tones, (all the other tones). This verse has a six-syllable line followed by an eight-syllable line. The two lines form a pair. A poem can have any number of pairs of lines. (38) shows the patterns of tones in the six-eight verse. ‘E’ stands for even tones underlined. Tones in the first, third, and fifth positions are flexible with respect to whether they are even or sharp tones. Tones in the second, fourth, sixth, and (ngang and huyen), and ‘N’ stands for non-even tones. The rhymed position is eighth positions must strictly follow the pattern. For example, a tone in the first position can be either E or N, but a tone in the second position must be E (ngang or huyen). Thus only even tones can occur in the rhymed position, underlined in (38).

VIETNAMESE TONE: A NEW ANALYSIS

19

(39)

(40)

The fact that tones in six-eight verse pattern in two groups, one group consisting of ngang and huyen and the other group consisting of the remaining tones, shows that ngang and huyen form a pair. This pair shares the same contour feature; both are level tones. Furthermore, in poetry, rhyme is found in positions in which only even tones can occur: the last syllable of the six-syllable line rhymes with the sixth syllable of the eight-syllable line; the eighth syllable of this line then rhymes with the last syllable of the next six-syllable line; and so on. With the dominant occurrence of even tones, which do not have complex contours, sixeight verse is regarded as the smoothest verse in Vietnamese poetry in terms of sound harmony. Since only even tones can occur in rhyme position, this position is heavy enough, i.e., it is sufficiently metrically prominent, that an unmarked tone is the best candidate to maintain the general smoothness of the poem in order to compensate for the heavy position. This neutralization is not active since it does not change a lexical tone (an active neutralization changes a lexical tone in some processes of reduplication) but rather requires a choice of words with tones ngang and huyen. The fact that only ngang and huyen can occur in such a pattern supports the claim that they are unmarked and form a contour pair. The model in (28) also predicts that if neutralization takes place, the more marked member of a pair will be neutralized to the less marked member. Moreover, because Contour and Register are separate from each other, the model also predicts the specific types of neutralization: in one type Contour contrasts will be retained with Register neutralized; in the other type Contour contrasts will be lost but Register distinctions will remain. Such patterns of neutralization are found in Southern and many Central dialects of Vietnamese (see Doan 1977 and Ngo 1984). In terms of contour the markedness relation between the pairs in (29) predicts that neutralization should occur with the most marked pair hoi and nga. In terms of register the markedness relation between the tones of each pair in (30) predicts that, because hoi is less marked than nga, hoi should be retained in neutralization and nga lost, and merge with the unmarked hoi. In these dialects hoi and nga neutralize and the result is hoi (Doan 1977, and Ngo 1984). The hierarchical analysis of tones previously proposed offers a straightforward and simple account of such a neutralization of hoi and nga to hoi: the contour contrast remains but the register contrast is lost (delinking of the register feature [laryngeal]). (39) shows the process. There are also markedness relations between tones in the same register. As (29) showed, the markedness between the pairs of tones increases from left to right. In the unmarked register ngang is the least marked tone and hoi the most marked. In the marked register huyen is the least marked tone and nga the most marked. (29) predicts that in neutralization within register the lost tone in the unmarked register will be hoi and that in the marked register it will be nga. These are the two most marked tones in their registers. These predictions prove to be correct, although here the evidence is mainly diachronic. In all Nghe Tinh dialects spoken in North Central Vietnam, neutralization occurred in the marked register: nga neutralized to nang1 (Hoang Thi Chau 1989 and Vuong and Hoang 1994). The model offers a simple account of this fact. In this process Register remained intact. Nang1 resulted from delinking of one point of the contour feature in nga. A complex contour (curve in nga) was neutralized to a less complex one (nang1). Athough (40) shows delinking of the last point, the deleted point could be any of the points because the result would be the same since, according to (28), a tone with two points specified and a register feature is nang1. The register feature was preserved in this process. The fact that neutralization results in nang1 shows that nga is more marked than nang1. There is also a neutralization process in the unmarked register. In Mai Ban, another dialect of Nghe Tinh, nga merged with nang1 in the marked register. Moreover, in the unmarked register hoi and sac1 also neutralized, resulting in sac1 (Hoang Thi Chau 1989 and Vuong and Hoang 1994), as shown in (41).

20

VIETNAMESE TONE: A NEW ANALYSIS

(42)

(43)

(41) ngang huyen

sac1 nang1

<---hoi <---nga

This evidence supports the claim that hoi is more marked than sac1 and that nga is more marked than nang1. Register remained intact (no register) and the contour contrast was lost by the delinking of one point of the contour feature in hoi. (42) shows this neutralization of a tone from a more complex contour feature (curve) to a less complex one (non-curve). Once again, it is not necessary that it must be the beginning point in the contour feature of hoi that is delinked because the result would be the same no matter which point is lost: according to (28) a tone without the register feature and with two specified points for the contour is sac 1. The fact that within Register neutralization occurs with the two tones hoi and nga provides further evidence for hoi and nga being the most marked pair. Moreover, in all dialects, the two tones ngang and huyen always remain regardless of different patterns of neutralization. This evidence strongly suggests that ngang and huyen are the least marked tones. Neutralization of tones in dialects occurs in two dimensions: across registers and within register. In all cases the surface tone in neutralization is the less marked member of a pair. Across registers the unmarked register tone remains, e.g., hoi in Southern dialects. Within register the less marked tone remains, e.g., sac1 or nang1 in Nghe Tinh dialects. All patterns of neutralization in both reduplication and poetry are alike in this regard. These patterns strongly support the claim that markedness relations between tones do exist and that the features that make up tones are structurally organized. Vietnamese clitics provide still other evidence for the markedness of tones. In the construction of phrases with clitics a word is cliticized onto an adjacent word either to the left or to the right. After attaching to the adjacent word, all segments of the clitic are deleted but its tone remains. If the clitic associates with an obstruent the clitic tone is realized on a homorganic nasal. If it associates with a vowel or glide the clitic tone is realized on a geminated vowel or glide (Pham 1997). (43) provides data from Pham 1997 transcribed here phonetically but with only the tones of the relevant syllables indicated. Clitics and their tones are underlined. For example, in (a) the clitic tone is huyen. The word [la:m] ‘how’ with tone huyen cliticizes onto the host [x′ n] ‘to pray’ (sac1), which ends in a coronal nasal. The clitic also surfaces as a coronal nasal bearing its original tone huyen. In (b) the clitic [n ] ‘he’ with tone sac1 cliticizes onto the preceding host [c ] ‘to let’ (ngang), which

VIETNAMESE TONE: A NEW ANALYSIS

21

ends in a vowel. The vowel is lengthened and the tone of the clitic is realized on the lengthened part of the vowel / / In Vietnamese, the underlying velar final consonant is phonetically doubly-articulated as a labio-velar, i.e., /k/ is [kp] and /′ / is [′ m] after the back round vowels /u, o, / a process shown in (c). In (c) the host [x kp] ‘to cry’ bearing tone sac2 ends with a labio-velar obstruent [kp]. The clitic [t ′ (ngang) ‘inside’ cliticizes onto the final consonant [kp] of the host and assimilates to the place of the segment [p] of the labial consonant. The clitic surfaces as a labial nasal that bears tone ngang. In (d) the clitic [ku′ m] ‘also’ with tone nga cliticizes onto the host [an] ‘eat’ with tone ngang. The clitic surfaces as a coronal nasal bearing its original tone nga. (Although sac2 and nang2 are not discussed until Chapter Three, these tones are included here in (e). There is a single clitic bearing tone nang2, the determiner [mot] ‘a.’ In (e) it cliticizes onto the host [k ] ‘have’ bearing sac1, which ends in a vowel. The vowel in [k ] is lengthened and the tone nang2 of the clitic is realized on the lengthened part of that vowel. Because sac2 and nang2 occur only in stop-final syllables, the nang2 of the clitic [mot] realized on a vowel must surface as huyen.) Since cliticization restricts the range of tones, the most frequent tones in clitics should be the less marked tones. This is indeed the case because in terms of frequency of occurrence, ngang, huyen, and sac1 occur most frequently in clitics. Of the 24 clitics found in the recorded conversations and interviews in Pham 1997, there was only one clitic with the tone nga, (d) above, and only one clitic, [mot] ‘determiner,’ with the tone nang2, which surfaced as huyen in (e). The rest were ngang, huyen, and sac1 Of the 22 clitics, seven had ngang, six had huyen, and nine had sac1 Clitics do not provide a large quantity of data because only function (grammatical) words can be clitics. Nonetheless, the similar distribution of ngang, huyen, and sac1 in clitics can be explained if we assume the structure of tones in (28), i.e., that ngang, huyen, and sac1 are the least marked tones. This finding is consistent with the claim that the smaller the inventory, the simpler the segments it has, e.g., Maddieson 1984 and Rice and Avery 1993. Borrowings from non-tonal languages such as French or English also produce evidence compatible with the claim about the markedness of tones. The proposed feature structure predicts that the tones in borrowed words should be ngang or huyen, the least marked pair. This prediction is borne out. Furthermore, most borrowings bear the tone ngang, the least marked tone. For example, among the examples given in (44), taken from dictionaries (Nguyen 1998 and Le 1988), very few forms occur with huyen, e.g., ‘cà’ in cà phê ‘coffee,’ ‘mùi’ in mùi-xoa ‘handkerchief,’ ‘xì’ in xì-nách-ba ‘snack-bar’, and xì-c′ ng-′an ‘scandal.’ The majority of examples have ngang. Every syllable in Vietnamese must have a tone and in (44) these tones are given beside the orthography. Sac2, the less complex tone of the pair, occurs in stop-final syllables. Ngang also predominates in borrowings, e.g., hai-phai from English ‘hi-fi,’ (with the meaning ‘gay’) instead of ái nam ái nằ , and em-xi from English ‘emcee,’ and in some abbreviations, e.g., xi-′i from English ‘CD,’ and ti-vi from English ‘TV.’ Because of the dominance of ngang in borrowed forms in sonorant-final syllables, Avery 1983 states that ngang is the only tone found in borrowings. However, huyen also occurs. Moreover, (44) shows that borrowings that are monosyllabic words occur only with ngang. The fact that only ngang and huyen occur in borrowed forms is used in many studies as evidence that these tones are unmarked. For example, Avery 1983 argues that ngang and huyen have no tonal tier and Burton 1992 argues that ngang and huyen are unmarked contour tones. (Hoi does occurs in a very few items, e.g., o′ n (hoi) tù (huyen) tì (huyen) from English ‘one two three,’ and m′ (hoi) lét (sac2) from French ‘molette.’ However, there seems to be no evidence for the use of nga. Such a distribution of hoi and nga is not surprising: they are the most marked tones in the group that does not end with a stop consonant.) The foregoing provides strong evidence for a structural representation of tones that incorporates markedness relations. The model predicts the various types of neutralization that occur in reduplication, dialects, and poetry. Tones can be neutralized in the same register, e.g., nga to nang1 in Nghe Tinh dialects, or across register, e.g., nga to hoi in Southern dialects. The frequency of occurrence of tones in clitics provides evidence for markedness and the treatment of borrowings is compatible with the claim that ngang and huyen are the least marked tones.

22

(44)

VIETNAMESE TONE: A NEW ANALYSIS

CHAPTER 3 The Tonal Inventory

The size of the tonal inventory is much debated in the literature on Vietnamese: the number of tones and the number of features required to classify them. The discussion so far has focused mainly on the tones that occur in open and sonorant-final syllables and ignored the two tones that occur in stop-final syllables. Two different positions may be taken with respect to these latter tones. The six-tone hypothesis holds that these tones belong with sac and nang; the eight-tone hypothesis treats them as independent tones. It is necessary to resolve this issue before proceeding further. Six Tones or Eight? In the six-tone system both nasals and stops are phonemic in final position, e.g., Doan 1977 and Vuong and Hoang 1994. In the eight-tone system there is no manner distinction in final consonants, e.g., Cao 1998 and Hoang Thi Chau 1989. It should also be noted that smaller inventories have also been proposed (see Doan 1977 and Vu 1988). Only a limited number of segments can occur in final position in Vietnamese: glides, nasals, and voiceless stops. In nonstop-final syllables six tones are found, but in stop-final syllables only sac2 and nang2 occur. The tonal distribution was given in (8) above. Vietnamese orthography distinguishes only six tones: ngang, huyen, hoi, nga, sac, and nang. Therefore, the orthography assumes that sac and nang occur in both syllable types whereas other tones occur only in open or sonorant-final syllables. The discrepancy in distribution between sac and nang and the other tones has the consequence that only sac and nang appear in all syllable types, whereas the other tones are restricted to syllable types that are not closed by obstruents. The standard position in the literature, reflected in the orthography, is that there are six tones in the inventory with final consonants distinctive in manner, e.g., Doan 1977 and Vuong and Hoang 1994. In this view sac1 and sac2 and nang1 and nang2 are allophones in complementary distribution, i.e., sac1 and nang1 occur in non-stop-final syllables and sac2 and nang2 occur in stop-final syllables, and there is no phonological tonal distinction between sac1 and sac2 or between nang1 and nang2. The distinction lies rather with the final stop or sonorant, because tone is predicted from the manner of that consonant. For example, sac is realized as sac1 in an open syllable or before a sonorant but as sac2 before an obstruent. The other tones have only a single form because no manner contrast is present in a following consonant. Thus, consonantal manner is basic to predicting tonal shape. (45) shows the phonological tonal inventory of the six-tone system. The sacs and nangs are categorized together, i.e., as noneven, non-broken. The two registers are unmarked and marked. (45) even non-broken

broken

Unmarked Marked

ngang huyen

non-even

sac nang

hoi nga

Only sonorants follow ngang, huyen, hoi, and nga. However, this predictability of manner disappears when sac and nang are added since both sonorants and obstruents follow sac and nang. Therefore, manner of articulation must be distinctive with these tones and listed as part of the lexical entry. Why are final stops so restricted in their distribution? This question has been of considerable concern in the traditional literature on Vietnamese tone. For example, Doan 1977 and Hoang 1986 offer an explanation based on phonetics: a rhyme with a final voiceless stop is too short for other tones to be realized, i.e., ngang and huyen must occur on a long enough span to show their level contour and hoi and nga require a certain length to show their complex contour. Sac and nang are simply rising and falling, their contour is neutralized, and they need only be distinguished by their registers (Hoang Cao Cuong 1989). However, such an explanation based on timing is not satisfactory. For example, in the Hue and Saigon dialects a phonetically complex contour tone can occur with the final stop (Hoang Thi Chau 1989, Vu 1982, and Hoang Cao Cuong 1989). Figure 2 from Hoang Cao Cuong 1989 shows the tonal system in the Hue dialect. This figure shows the tonal contours of only seven tones because in the Hue dialect nga is neutralized with hoi. Only sac1 and

24

VIETNAMESE TONE: A NEW ANALYSIS

Figure 2. The phonetic contours of sac1 and sac2 in the Hue dialect (from Hoang Cao Cuong 1989)

sac2, the two tones under consideration here, are labeled. Speaker 1 is male and speakers 2 and 3 are female. Hoang does not label the duration of tones. The clearest pitch graph, that of speaker 2, shows that the two sacs, which are rising tones in the Hanoi dialect, are falling-rising tones in Hue. They are in fact the only curved tones in Hue. Therefore, in the Hue dialect the rising tones sac1 and sac2 of the Hanoi dialect surface as the curved tone hoi, i.e., a falling-rising tone, in the Hanoi dialect. Figure 3 is also from Hoang Cao Cuong 1989. This figure shows the pitch graphs of four informants from the Saigon dialect. Speakers 1 and 2 are male and speakers 3 and 4 are female. This figure labels only nang1 and nang2, the tones under consideration here. These tones, which are falling tones in the Hanoi dialect, are falling-rising tones in the Saigon dialect. The third curved tone in this dialect is hoi. The absence of hoi and nga in stop-final syllables in the Hanoi dialect has been attributed to phonetics, the rhyme being too short to carry the complex contour tones. Figures 2 and 3 demonstrate that this proposal does not explain why a tone with a phonetically complex contour can surface in a stop-final syllable in the Hue and Saigon dialects. In other tone languages the presence of a rising or a falling tone in the system implies the existence of a level tone in that system (see Anderson 1978: 151). Because of this implication researchers have wondered whether contour tones should be treated as single units or as sequences of two level tones (Anderson 1978 and Yip 1995). According to Anderson, Woo 1969 claims that the universal system of tonal features includes only a level-tone element. If stopfinal syllables provide an environment that neutralizes tonal contours, the two simplest tones ngang and huyen should occur in that environment. There

VIETNAMESE TONE: A NEW ANALYSIS

25

Figure 3. The phonetic contours of nang1 and nang2 in the Saigon dialect (from Hoang Cao Cuong 1989)

are indeed languages in which stop-final syllables allow only level tones, e.g., Be (Hashimoto 1985), a Be-Tai language in the Tai-Kadai family, and Sgaw and Pwo (Benedict 1972), two Karen languages in the Sino-Tibetan family. These languages have both level and contour tones in non-stop-final syllables. Yip 1995:487 reports that Cantonese, a Chinese language, has seven distinctive tones in non-stop-final syllables but only three level tones in stop-final syllables. Burton 1992 addresses the issue of why only sac2 and nang2 occur in stopfinal syllables. (19) showed this classification. Burton argues that only sac and nang can occur in stop-final syllables because they have simple contours, represented as ‘h’ in (19). Hoi and nga cannot occur in stop-final syllables because they have complex contours, represented as ‘lh’ in (19). However, why is it that ngang and huyen, which also have simple contours, represented as ‘l’ in (19), do not occur in stopfinal syllables? Burton answers by proposing a constraint that ‘l-pitch tones cannot occur before a voiceless consonant.’ He claims phonetic motivation: there is a relationship between voicelessness and high pitch, i.e., initial voiced consonants raise the pitch of the vowel and voiceless consonants lower the pitch (Ohala 1973). All final stops in Vietnamese are voiceless. Ngang and huyen are ‘l’ therefore, they cannot occur with final stops, which are voiceless. Sac and nang are ‘h’ tones; it follows that they can occur before voiceless consonants. However, Burton provides no further explanation for this constraint. Burton classifies tonal contours phonetically. Figure 4 shows a pitch graph of the Hanoi dialect from Vu 1982. Burton argues that ngang and huyen are classified as ‘1’ because they are articulated in the lower part of their respective registers. Burton perhaps bases his claims on Vu’s placement of the middle line in Figure 4, which divides the tones into two registers. This line shows that ngang and huyen are articulated in the lower parts of their registers, ngang in the upper register and huyen in the lower register; therefore, they are both ‘l.’ However, this claim fails with nang, which is classified as ‘h’ although it is in the lower part of its register. Figure 4 shows that the sacs (sac1 and sac2) occur in the higher part of their registers and are therefore ‘h’ tones. However, a comparison of nang1 and nang2 shows that except for their beginnings, both are low in their registers, even lower than huyen. Chapter Four will show that it is the end points not the beginning points of nang1 and nang2 that are distinctive. Consequently, phonetic evidence does not support the claim that the nangs are ‘h.’ Burton also says that sac and nang can occur with final stops because they are high in pitch. Since final stops are always voiceless in Vietnamese, it is impossible to test such a claim.

26

VIETNAMESE TONE: A NEW ANALYSIS

Figure 4. Normalized mean F0 in the Hanoi dialect (from Vu 1982)

The alternate hypothesis is that there are eight tones in the phonological inventory. For example, traditional poetry classifies tones according to their register and contour. There are two registers, ‘phu’ (high) and ‘tram’ (low). Contour tones are divided into ‘binh’ (even) tones and ‘trac’ (non-even) tones. Among non-even tones there are ‘thuong’ (falling and rising) tones, ‘khu’ (rise or fall) tones, and ‘nhap’ (checked) tones, i.e., the two tones that occur in stopfinal syllables. These terms were used as the names of the tones in the literature before the current terms were introduced. However, in the system based on tonal patterning in poetry that uses the traditional names, e.g., phu, tram, and so on, researchers show inconsistency concerning the register of hoi and nga. Some, e.g., Hoang Thi Chau 1989:202 and Phan 1997:23, classify hoi as ‘phu’ (high) and nga as ‘tram’ (low), but others, e.g., Doan 1977:120, place hoi in the ‘tram’ register and nga in the ‘phu’ register. The classification in (46) shows the former view. The traditional eight-tone classification describes hoi as phonetically low and nga as phonetically high. In this respect it resembles the six-tone classification. (46) binh (even)

trac

thuong (fall & rise)

khu (rise/fall)

nhap (checked)

PHU (high) TRAM (low)

ngang huyen

hoi nga

(non-even)

sac1 nang1

sac2 nang2

VIETNAMESE TONE: A NEW ANALYSIS

27

(48)

However, rather than being predictable variants of sac and nang in non-stopfinal syllables, sac1 and nang1 are phonologically distinct from the tones in stop-final syllables, i.e., sac2 and nang2. The eight-tone hypothesis claims that the contrast between sac and nang in stop-final syllables and sac and nang in sonorant-final syllables lies with the tones and that the manner of the final consonant is conditioned by tone. For example, [mat] with the tone sac2 and [man] with the tone sac1 contrast in the two different sacs not in the two different final consonants. Thus, there is no underlying manner contrast in final consonants. Because voiceless stops can occur only with sac2 and nang2, voiceless stops are conditioned by these tones: sac2 and nang2 predict voiceless stops. It is thus manner that is allophonic rather than realizations of sac and nang. (47) shows an underlying eight-tone system, together with an example, based on Cao 1998. The example in (47), aC, shows the Vietnamese orthography for tones in which the final consonant is unspecified for manner, here represented by ‘C’ (Cao 1998 uses only the coronal in the example.) In the surface forms ‘N’ represents a final sonorant and ‘T’ a final stop. Two registers, ‘phu’ (high) and ‘tram’ (low), have four high tones (ngang, nga, sac1, and sac2) and four low tones (huyen, hoi, nang1, and nang2). In terms of contour the even tones are ngang and huyen. The six non-even tones are divided into three groups: curved tones (hoi and nga); rising/falling tones (sac1 and nang1); and two additional tones (sac2 and nang2), which are non-curve. As in (45) and (46), hoi is placed in the high register and nga in the low register. (47) EVEN

NON-EVEN

rising/falling

curve

non-curve

Phu Tram

aC (ngang) => [aN] àC(huyen) =>[àN]

aC (sac 1) => [áN] aC(nang1) =>[aN]

′ C (h o i) =>[′ N] ãC (nga) =>[ãN]

−aC (sac2) => [áT] −aC (nang2) => [′ T]

The eight-tone system allows the manner of the final coronal to be predicted from tone. The difference between [áN] (sac1) and [áT] (sac2) lies with tones. Sac1 and nang1 predict a final sonorant. Sac2 and nang2 predict a final stop. The different tones are thus represented as /aT, sac1/ and /aT, sac2/ rather than as /aT, sac1 and /aN, sac1, as in the six-tone view. Cao 1998 captures this distribution by assuming that [±nasal] is a tonal feature that is realized on the final consonant. Sac2 and nang2, are [-nasal] so the final consonant surfaces as a stop. Sac1, nang1, and the other remaining tones are [+nasal] so the final consonant surfaces as a sonorant. In (48), C is a consonant that has no manner specification. Cao notes that because the writing system uses the same diacritics for these two tones in all syllable types, it misleads phonologists to propose a six-tone inventory. If the founders of the current writing system had used distinct diacritics for sac and nang in stop-final syllables, this confusion of tones would not be a problem. For example, in (47) all consonants would share manner but use different diacritics for sac and nang in non-stop-final and stop-final syllables. An Eight-Tone System While the eight-tone analysis is little discussed in the current literature on Vietnamese, it was assumed in traditional poetry before the current Vietnamese writing system was created (Cao 1998). Adding two tonemes also reduces the phonemic final consonant inventory from six to three, i.e., from /p, t, k, m, n, ′/ to /M, N, ′/, where /M, N, ′/ represent places of articulation without specification for manner. It also provides a simpler account of tone harmony in reduplication (Cao 1998). For example, the reduplication process discussed in Chapter Two shows that the reduplicant tone must be either ngang or huyen depending on the register of the base tone. It is ngang if the base tone is an unmarked register tone and huyen if the base tone is a marked register tone. Since the reduplicant carries either ngang or huyen, it must therefore end in a sonorant consonant. If the base tone is other than sac2 and nang2, as in (49), tone harmony is the same in both systems. (The base tone is once again underlined.) (49) /toj/ (sac1)

‘dark’

>

toj (ngang)+toj (sac1)

‘rather dark’

28

VIETNAMESE TONE: A NEW ANALYSIS

However, if the base ends in a stop, the reduplicant must still have either huyen or ngang as its tone. Since these tones are incompatible with final stops, a final nasal surfaces instead. (50) shows this: the base tone sac2 in (a) has the reduplicant tone ngang, which surfaces with a final velar nasal, and the base tone nang2 (b) has the reduplicant tone huyen, which surfaces with a final labial nasal. (50) (a) (b)

sac (sac2) dep (nang2)

‘sharp’ ‘beautiful’

> >

sang sac (ngang—sac2) dem dep (huyen—nang2)

‘rather sharp’ ‘rather beautiful’

In this case tone harmony is represented differently in the two systems, as (50) shows. The six-tone analysis requires a rule to change a stop to a homorganic nasal, i.e., a final stop consonant must change to a homorgarnic nasal if it occurs with tones other than sac2 and nang2, as in (a). On the other hand, the eight-tone system in (b) does not need such a rule because the manner of the consonant is predictable from the tone, i.e., ngang predicts a final sonorant. Here the final consonant in the eighttone system, unmarked for manner, is represented by a nasal. (51)

The eight-tone system provides a more elegant account of reduplication in Vietnamese in that the manner becomes simply a consequence of default. Hoang Thi Chau 1989:202 supports the eight-tone system using not only arguments from traditional poetry, i.e., (46), but also because it accords with the hypothesis that tones developed from segments. According to Haudricourt 1954, Vietnamese tones developed from voicing of initial and final segments, as (52) shows. Haudricourt uses p and b to represent initials since only the voicing is important here. s and x represent the voiceless fricatives, alveolar and velar, respectively. There are no final stops in (52) except ?. (52)

In this view early Vietnamese was non-tonal. By the 6th century three tones were established: ngang in open syllables; huyen from the final fricatives [s] and [h]; and sac from the final [x] and glottal stop. By the 12th century the voicing contrast was lost in the initial consonants and six tones resulted in two registers. The original voiced initial gave rise to low tones (huyen, nga, and nang). However, it is not clear how hoi and nga developed from huyen. Finally, in modern Vietnamese, a voicing contrast is reestablished to give six tones. Hoang Thi Chau 1989 and Vu 1988 claim that when final consonants disappeared they created tones with their traces: the final stops disappeared and left glottalization in sac and nang; and the final fricatives disappeared and left creakiness in hoi and nga. Therefore, an understanding of how tones developed helps in understanding the origin of phonation types in the current tonal system. However, what the final *-x from *pax and *bax actually leaves in a tone remains undisclosed. This historical hypothesis is very influential in the literature on Vietnamese since it supports the eight-tone view and offers an account of the intimate relationship that exists between the final stops and tones.

VIETNAMESE TONE: A NEW ANALYSIS

29

(53)

There are still further problems with Haudricourt’s hypothesis. For example, Chapter Four will show that it is reasonable to claim that the historical final glottal stop left a trace in nang1 where the tone has either glottal stop or creakiness. However, that claim cannot be made for sac1, where glottal stop is not found in the tone (see Alves 1997). Hoi and nga also developed from final *-s and *-h and it is claimed that the breathy voice found in these tones has its origin in these fricatives. However, Chapter Four will show that hoi has breathiness while nga is characterized by creakiness or glottal stop, a characteristic which comes from the historical final glottal stop. The eight-tone view gains support even from authors who use the six-tone one in their work. For example, in a standard book on Vietnamese phonology, Doan 1977 presents the six-tone system as the formal, official one. However, in his discussion he says that he makes this choice because it is very popular and familiar in the Vietnamese literature, and is reflected in the orthography. After showing several different tonal inventories including the six-tone and eight-tone systems, he comments that the eight-tone system is probably best, because it reflects the traditional classification in poetry before the orthography was created. The Additional Tones Only the two tones sac2 and nang2 can occur in stop-final syllables. The manner of articulation of the final consonants is also predictable from the tones. Ngang, huyen, hoi, nga, sac1, and nang1 predict that a final consonant will be a sonorant, and sac2 and nang2 predict that a final consonant will be an obstruent. The structures of sac2 and nang2 contain the feature [obstruent] and tones with this feature are more marked than those without it. It is not unusual to find that obstruents are neutralized in the coda position in various languages. For example, in Spanish and Italian, except for s, only sonorants can occur in final position (Clements 1990a: 312). Such evidence supports the claim that obstruents are more marked than sonorants in this position. Support for this claim is also found in the final inventory of Vietnamese and in clitics. The phonetic final inventory was given in (2). Under the eight-tone hypothesis, final stops and nasals are in complementary distribution and are thus allophones. According to the feature organization of segments, as developed in Clements 1990b and Clements and Hume 1995, a consonant has a C-place node in (a) in (53) under the root node, and a vowel has a V-place node and an aperture node as its sister in (b) in (53) under the vocalic node. An aperture node specifies vowel height. The difference between a vowel and a glide is that the former has an aperture node, which the latter lacks, as in (c). X is a place feature. If nasal and stop in final position are allophones, there are two types of underlying segments in final position, one for consonants and the other for glides. The representation of a final consonant is illustrated in (54) below with only relevant features given. In (a) the surface final voiceless stops and nasals share the same underlying representation, a Root node with C-place as a dependent. On the surface there are two types of consonants: one with the [obstruent] feature and the other without. In the type with the [obstruent] feature the consonant is realized as a voiceless stop. It acquires the feature [obstruent] from the tone. The eight-tone hypothesis assumes that this feature is a tone feature so [obstruent] is shown in the representations of sac2 and nang2 only. In the second type the feature [nasal] is inserted by a default rule (Rice 1993) and yields a surface nasal.

30

VIETNAMESE TONE: A NEW ANALYSIS

(54)

(b) (56)

(b) assumes that the glides [w] and [j] have the underlying V-place node under the vocalic node. The Nasal Default rule does not apply to glides and the result is a plain sonorant, the consequence of a constraint that prohibits a node from being specified for nasal if the vocalic node is present in the segment (there are no nasalized vowels in Vietnamese). The conclusion is that there is no underlying manner feature for syllablefinal obstruents or sonorants; the surface feature for sonorants is a default feature inserted at the phonetics. Sac2 and nang2 force the insertion of obstruent. Clitics provide further evidence that in Vietnamese sonorants are unmarked in the syllable-final environment. Clitics in the coda position are unspecified for manner and [nasal] is a default feature. (55) gives examples of clitics and cliticization. Only the forms involved in cliticization are shown, and the clitics are underlined. (55)

As Chapter Two showed and from (55) it is apparent that after cliticizing onto the host all segments of the clitic disappear and only the tone remains. In (a) the clitic surfaces as a nasal if the host ends in an obstruent. It so happens that this clitic [la:m] ends in a nasal. In this case the nasal has the place of articulation of the final obstruent, i.e., it is [n] in (a) because [ha:t] ends in [t]. The tone of the clitic is realized on the nasal. If the host ends in a vowel, as in (b), or a glide, as in (c), the vowel or glide is lengthened and the tone is realized on the lengthened part. (The clitic [mot] in (b) occurs with huyen not the original nang2 tone.) No matter what tone the base has the clitic must have an unmarked register tone that predicts a sonorant. The fact that only sonorants (nasal, vowel, and glide) can surface in the clitics supports the claim that in that environment sonorants are less marked than obstruents. In order to represent sac2 and nang2 structurally it is necessary to assume that the feature [obstruent] is part of these tones and that the predictability of a final stop with these tones follows from its presence. Since this feature is a marked feature in this position, it must be present underlyingly in the representation of sac2 and nang2, as in (56). Sac2 and nang2 have the feature [obstruent] under the contour node in (56) because these are the only tones that occur with stop-final syllables. These structures can be compared with those of sac1 and nang1 in (57).

VIETNAMESE TONE: A NEW ANALYSIS

31

(57)

The only difference under the contour node is the presence of [obstruent] in sac2 and nang2. As with sac1 and nang1, sac2 and nang2 have two points specified under the feature [obstruent] to show the movement. Neither sac1 nor sac2 has a register feature because it is unmarked for register. On the register side nang2 is marked with the register feature [laryngeal] just as is nang1. However, unlike nang1, [laryngeal] in nang2 has [spread] as its dependent. Chapter Five provides justification for the register feature. While sac2 and nang2 are equally complex in (57) on the contour side, sac2 is less complex than nang2 on the register side with the feature [spread]. The restricted distribution of sac2 and nang2 is an indicator of the markedness of these tones. A comparison of the structures of sac2 and nang2 in (57) with the structures of the other six tones in (28) from Chapter Two shows that sac2 and nang2 are the most complex tones vertically with the feature [obstruent] under the contour node. In the process of reduplication discussed in Chapter Two tones in the unmarked register (ngang, sac1, and hoi) pattern together in one group and tones in the marked register (huyen, nang1, nga) form a second group. The register of the base is always replicated in the reduplicant. The data in (58) show this process with sac2 and nang2. (The base tone is underlined.) (58)

(The fact that a final stop in the base changes to a homorganic nasal in the reduplicant was dealt with earlier and is not relevant to the argument here.) In (a), when the base is sac2, the reduplicant is always ngang, an unmarked register tone. In (b), when the base is nang2, the reduplicant is always huyen, a marked register tone. Sac2 patterns as if it were in the same group as sac1 and hoi, i.e., all have ngang in the reduplicant. Nang2 patterns as if it were in the same group as nang1 and nga, i.e., all have huyen in the reduplicant. This pattern shows that sac2 and nang2 share the same register feature with other unmarked and marked register tones. Sac2 does not have the register feature and nang2 has the marked feature [laryngeal] with [spread] as its dependent, e.g., in huyen. The use of the feature [spread] will be justified in Chapter Five. The behavior of sac2 and nang2 in reduplication shows that sac2 is an unmarked register tone and nang2 is a marked register tone. (59) summarizes the markedness of tonal registers in the eight-tone system. (59) Unmarked register Marked register

ngang huyen

sac1 nang1

hoi nga

sac2 nang2

Borrowings provide evidence that sac2 is less complex and, therefore, less marked than nang2. Borrowed forms that end in non-sonorant segments can occur only with final stops. Consequently, the two possible tones in this position are sac2 and nang2. (60) gives some examples of borrowings (from Nguyen 1975 and Nguyen 1998). The tone sac2 is represented by the diacritic ‘′’ above the vowel, and nang2 is represented by the diacritic ‘.’ underneath the vowel. Some of the borrowed forms in (60) can occur with either sac2 or nang2, e.g., (h), (i), and (j). The data in (60) show that sac2 is more common than nang2 in borrowings.

32

VIETNAMESE TONE: A NEW ANALYSIS

(60) Borrowings with sac2 or nang2 (a) x′ p (b) dip (c) xà lách (d) cà r′ t (e) gác (f) séc (g) mù t′ c (h) gác do bu or g′ c do bu (i) xi′ c (sac2) (j) soó (sac2) (k) các vi dít or các vi dít or các vi dít

(Fr. chef) (Fr. jupe) (Fr. salade) (Fr. carotte) (Fr. garde) (Fr. cheque) (Fr. moutarde) (Fr. garde-boue)

‘chief’ ‘short skirt’ ‘salad’ ‘carrot’ ‘guard’ ‘cheque’ ‘mustard’ ‘mudguard’

(sac2) (sac2) (huyen-sac2) (h huyen-sac2) (sac2) (sac2) (huyen-nang2) (sac2-huyen-ngang) (nang2-huyen-ngang) (nang2) (nang2) (sac2-ngang-sac2) (nang2-ngang-sac2)

(Fr. cirque)or xi′ c (Eng.) or so′ c (Fr. carte de visit)

‘circus’ ‘short’ ‘business card’

(k) in the above does not have nang2 in both syllables of the word, i.e., *cac vi d′t (nang2-ngang-nang2) is impossible. Moreover, (except for (g), which is rare), if only a single form is available, it occurs with sac2 not nang2. Examples are lúp/* l′ p (Fr. loupe), típ/*t′p (Fr. type), xúp/*x′ p (Fr. soupe), ráp/*r′ p (Eng. rap music), x′ p/*x′p (Fr. chef), r′ c/*r′c (Eng. rock music), a-xít/ *a-xit (Fr. acid), tu′ c-no-vít / *tu′ c-n′-v′t (Fr. tournevis), xà lách/*xà l′ ch (Fr. salade), and phát phút/ *phat ph′ t (Eng. fast food). The dominance of sac2 over nang2 in this type of borrowing has led researchers, e.g., Burton 1992, to claim that only sac2 occurs in this syllable type. However, nang2 also occurs. The more frequent distribution of sac2 over nang2 in this type is explained if sac2 is less marked than nang2. Frequency effects also lend some support to the claim that sac2 is less marked than nang2 and to the eight-tone hypothesis. (37) in Chapter Two showed that the pair ngang and huyen occur more frequently than the pair sac1 and nang1. However, if sac and nang in both syllable types are regarded as the same, within a six-tone system the frequencies are quite different. (61) from Vo 1997 shows the actual numbers of tones when sac and nang include both types of syllables: stop and non-stop final. (61) Total 4243

sac 1426

nang 1045

ngang 1029

huyen 840

hoi 570

nga 353

In (61), ngang and huyen, the two unmarked tones, have a lower frequency than sac and nang. If the less marked the tone the more frequently it occurs, the markedness relationship between the pairs is strange. If ngang and huyen are unmarked because of their patterns in reduplication and in borrowings, they should occur more frequently than sac and nang. However, they occur less frequently. The six-tone hypothesis is unable to explain this fact. (37) and (61) provide the information needed to calculate the frequency of sac2 and nang2 in stop-final syllables. (62) shows that frequency. (62) 581

sac2:

439

nang2:

Nang2 occurs less frequently than sac2, a confirmation of the claim that nang2 is more complex than sac2. Moreover, sac2 and nang2 occur less frequently than ngang and huyen, the two unmarked tones in all analyses. (63) summarizes the frequency of occurrence of the eight tones. The bold border separates the two tones in stop-final syllables from the rest. (63) huyen 840

sac 1 845

nang 1 606

ngang 1029

hoi 570

nga 353

sac2 581

nang2 439

(63) shows that among tones in sonorant-final syllables the unmarked tones ngang and huyen occur most frequently (1029 and 840 times respectively) with the ranking of markedness among tones shown in (64). The ranking in (64) summarizes the

VIETNAMESE TONE: A NEW ANALYSIS

33

ranking of six tones in sonorant-final syllables in (30) and (31) in Chapter Two with the addition of the two tones in stop-final syllables. ‘<’ means ‘less marked than.’ (64)

If frequency of occurrence can be used as a diagnostic for the markedness of tones, the ranking of the markedness of tones in (64) predicts that among tones in sonorant-final syllables, the least marked tone ngang has the highest frequency and the most marked tone nga has the lowest frequency. Among pairs, the least marked pair ngang-huyen occurs most frequently and the most marked pair hoi-nga occurs least frequently. In the two tones in stop-final syllables, nang2 occurs less frequently than sac2. These predictions about frequency in Vietnamese are borne out; consequently, there is good evidence forthe markedness relations proposed here.

CHAPTER 4 The Acoustics of Tone

This chapter is devoted to a review of previous experimental studies of Vietnamese tones and the presentation of the findings of an acoustic study of Northern dialects. The particular concerns addressed are the fundamental frequency (F0), phonation types, length, and linear portion of tones. Previous Studies There have been a number of acoustic studies of Vietnamese tones, e.g., Han 1969, Vu 1982, Hoang 1986, Nguyen and Edmondson 1997, and Vu 1999. However, those of Nguyen and Edmondson, Vu, and Hoang merit special consideration. Nguyen and Edmondson 1997 is the only study using modern instrumental techniques that focuses on phonation types in Northern Vietnamese tones. They argue that in addition to fundamental frequency, voice qualities are used distinctively in Northern dialects. They examined [ta] with 6 tones in an open syllable; they did not investigate tones in stop-final syllables. They recorded six speakers three times each and analyzed the responses using CECIL 2.1. They calculated the F0 value of each syllable. Figure 1 in Chapter Two shows the normalized F0 values of the three pairs of tones. (It is not clear whether the values given in Figure 1 are averaged across speakers or are from one speaker.) Nguyen and Edmondson provide the following description of the tones and assigned each tone a relative pitch value in numbers using the system in Chao 1930 (cited in Bao 1999b), with 1 representing the lowest point and 5 the highest point in the pitch range. The first pair, in Figure 1(a), is ngang and huyen. Ngang (high level) is a high tone with a flat contour gradually going down just a little; it is a 33 tone. Huyen (low level) is a low tone with a flat contour and starts lower than ngang and then goes down slightly at the end; it is a 21 tone. The second pair, in Figure 1(b), is sac1 and nang1. Sac1 (rising) starts from a lower point than ngang and rises to the highest point in the pitch range; it is a 35 tone. Nang1 (falling) is shorter than the other tones; it is a 32 tone. The third pair, in Figure 1(c), is nga and hoi. Nga starts as high as ngang and then rises to the highest point in the pitch range. It is interrupted by a glottal stop at approximately 225 ms (milliseconds). Hoi starts between the beginning points of ngang and huyen, goes down, and then rises to a point that is close to the onset. They do not assign tonal numbers to hoi and nga. Nguyen and Edmondson observe that there is variation in tones from speaker to speaker in terms of length, contour, and F0 level. Figure 5 displays the six tones using the syllable [ta] from two speakers. In 5(a) the sac tone for speaker 6 starts higher than ngang, huyen, and hoi, at approximately 44 semitones. However, for speaker 2 in 5(b) sac has the lowest starting point. The nang1 tone is very short for speaker 6. It is also higher than the same speaker’s ngang. Figure 5 shows that for both speakers nga is broken in the middle by a glottal stop. Speaker 6 in 5(a) starts nga at 46 semitones and breaks it with a glottal stop from 100 ms to approximately 320 ms. After the glottal stop the rising portion emerges at between 48 and 50 semitones. For speaker 2 in 5(b), the broken portion in the nga tone is shorter than that of speaker 6, from approximately 90 to 180 ms. After the glottal stop the rising portion in 5(b) emerges at between 50 and 54 semitones, a point higher than that for speaker 6. The hoi tone does not show any rise in either speaker. Nguyen and Edmondson also investigated the phonation types and their results are discussed in Chapter Five. Vu was concerned with the general F0 of tones. He recorded eleven speakers from Northern dialects reading word lists. One speaker was from Thanh Hoa province and one was from between Nam Dinh and Thanh Hoa. He also included the two tones that occur in stop-final syllables, sac2 and nang2. The diagrams in Figure 6, from Vu, show the mean F0 of eight tones plotted against mean duration (in centiseconds) from eleven speakers. Two tones appear in Figure 6 that do not appear in Nguyen and Edmondson: sac2 and nang2. Sac2 is the highest tone, short and rising steadily to the highest point; nang2 is the second lowest tone, short and falling steadily. The six remaining tones in Vu are similar to those in Nguyen and Edmondson. For example, sac1 and nga have the highest end points and nga is broken by a glottal stop, represented by a dotted portion. However, Vu shows nga as going down steadily and then rising very quickly to form a V shape. The difference in F0 between tones is larger in Vu than in Nguyen and Edmondson because the scales differ. Vu’s nang1 is much lower than ngang. The curved tone hoi, the lowest tone in Vu, clearly has a rising part which is not present in Nguyen and Edmondson. Therefore, although the two studies agree on many points, they do show some variation, e.g., the lengths of the glottal stop and of nang1,

VIETNAMESE TONE: A NEW ANALYSIS

35

Figure 5. Six tones from speakers 6 and 2 (from Nguyen and Edmondson 1997:8)

and the shape of the curved tone hoi. Hoang 1986 shows tonal shapes in the Hanoi dialect that are similar to those in Vu. However, there are differences in F0 between the two studies. Hoang recorded eleven speakers reading a word list of 150 syllables. Figure 7 shows the pitch graphs of eight tones from four speakers. Speakers 1 and 2 are male and speakers 3 and 4 are female. All the tones are labeled for speaker 4, but Hoang does not show duration. Hoang like Vu shows that the three tones with the highest end points are sac 1, sac2, and nga. Nga also has the clear V shape, but there is no break for the glottal. Hoi and nga show variation between speakers. Hoi is represented by a line broken by dots. It is the lowest tone for speaker 1; it has a rising part for speaker 2; and it falls and levels off for speakers 3 and 4. Nga is represented by a broken line with small circles. It also falls and rises. The most noticeable difference between the two curved tones hoi and nga in this study and in the studies of Nguyen and Edmondson and Vu is that for the latter, the lowest point of nga is higher than that of hoi. Hoang also shows that hoi bisects nga, as with speakers 3 and 4. This point will be discussed further in Chapter Five. These studies show variation in both the pitch height and contour of tones, especially in the non-level tones. However, it is important to note that all these researchers describe nga as having a glottal stop in the middle and all share the view that tones are classified according to their contours and registers. (65) shows the classification of tones that arises from these phonetic studies. In these studies (and in almost all traditional studies of Vietnamese tones) ‘register’ must be interpreted as ‘tonal height.’ High register tones include ngang, sac1, nga, and sac2. Low register tones include huyen, nang1, hoi, and nang2. (65) High Low

ngang huyen

sac1 nang1

nga hoi

sac2 nang2

Once again there is a serious problem involving a mismatch between the phonetics and phonology of tones.

36

VIETNAMESE TONE: A NEW ANALYSIS

Figure 6. Mean F0 of eight tones in Northern Vietnamese (from Vu 1982)

A New Acoustic Study This study investigated the fundamental frequency of all eight Vietnamese tones and the spectral characteristics of phonation types such as creakiness and breathiness. Speakers of Vietnamese were asked to read controlled lists of words. Each word consisted of a CV or CVC syllable. Only the vowel [a] was used. It was chosen because other vowels are very often diphthongized. The initial consonants in open syllables include a nasal /m/, two stops /t, k/, and two fricatives /s, z/. The final consonants include stops and velar nasals. The two final glides were tested only with hoi and nga. Some other forms were mixed in with the tested forms. These included the long vowel /a:/ with the final nasal /m/, (t′ m tãm, t′ m t′ m), the short vowel /a/ with the final nasal /n/ (t′ n tãn, t′n t′n), and closed syllables with the diphthong /i /. The set of tested syllables thus included the segmental shapes [ma], [ta], [sa], [za], [ka], final consonant /′ /, and stops plus appropriate tones. There is no tonal sandhi in Vietnamese and all tested forms were put in a fixed elicitation frame. The frame sentence was Ch′ …ph′i ngay [c ……fa:j ′ aj] ‘The word…has to be straight.’ The fricative /f/ was chosen to follow the target form because there is always a clear boundary between the target form and /f/ in the spectrogram, indicated by noise at higher frequencies. The st′ dy involved nine speakers, three males and six females, of the Northern dialects. They were between 17 and 49 years of age and lived in Toronto. Six speakers were born and grew up in the North (Hung, Hoang, Phuong, An, Binh, and Dung). The other speakers were born in the North and spent their childhood there. Later they left for the South but lived with family of the same background and within Northern communities (VanKhanh, Son, and Van). These three also spoke Northern dialects and regarded themselves as Northerners. The nine speakers had been in Toronto from one month to about eighteen years. All were completely unaware of what being tested in their speech. 112 items were randomized 3 times and presented to each speaker individually. Each speaker read all items. With nine speakers the total number of tokens was 3,024. The spoken responses of the subjects were recorded in a quiet room. They were then digitized on an iMac computer. Each data file contained the target form with its surrounding environment, i.e., after an open syllable [c ] and before [f], the onset of the following syllable [fa:j]. SignalyzeTM3.12, created by Eric Keller 1994, was used for the analysis. Fundamental

VIETNAMESE TONE: A NEW ANALYSIS

37

Figure 7. Eight tones from four speakers (from Hoang 1986:24 and 1989:2)

frequency was not measured by pitch extraction from the program but was measured manually by determining the length of the cycle at 30 ms intervals. The signal was digitized. The raw signal was then enlarged in order to measure individual cycles. Signalyze was also used to produce spectrograms and Excel was used to graph pitch. In addition to F0, both creakiness and breathiness were clearly present in the waveforms and they were also examined. F0 was measured every 30 ms, starting immediately after the burst of an initial stop or after an initial nasal segment. Measurements were taken at 30, 60, 90, 120, 150, 180, 210, 240, and 270 ms. For the tones with creakiness or glottal stop, i.e., nang1 and nga, F0 was measurable only when there was a periodic glottal pulse evident in the spectrogram. With breathy tones, F0 was measured when individual cycles could be clearly identified. When the tone was heavily breathy, i.e., it contained glottal pulses with reduced amplitude making it difficult to identify individual cycles, the recorded signal was filtered using a low pass (0–500Hz) Butterworth filter to suppress the high frequency components. After filtering, individual cycles were more easily determined when only low frequency components were preserved. If there was no voicing in the breathy portion, F0 was unmeasurable, a rare occurrence. With a nasal-final syllable, the second part of the tone was realized during the nasal. Although voicing was not strong in the nasal portion of the spectrogram, F0 was still measurable. If this part was ignored, the tone lost its second part and became unusually short. In such cases there was not enough information about the contour of the tone and the tone became unrecognizable. Usually the first two tokens of a form were measured. The third token was examined only as required, e.g., when the speaker made an unusual pronunciation of a form due to a slip of the tongue or had an unusually long pause before or after the form. A spectrogram of the signal was produced using Signalyze’s extra-wide (300Hz) setting for a female voice and the very wide (200Hz) setting for a male voice. Fundamental frequency was the first concern, particularly the F0 contours of six tones with the same syllable [ta] and of sac2 and nang2 with [ta:k] from eight of the speakers. Figure 8 is a set of graphs that show the F0 contours of all eight tones from three males and five females. The values of tokens of the same form from each speaker as well as across speakers are not averaged. The legend of tones in the graph is as follows: the speaker’s name, followed by the repetition of the token, then the tone: ‘ng’ is ngang (high-level); ‘hu’ is huyen (low-level); ‘sac’ is sac1 (high-rising); ‘na’ is ‘nang1’ (low-falling); ‘saccac’ is sac2 (high-rising in stop-final syllables [ka:k]); ‘na-cac’ is nang2 (low-falling in stop-final syllables); ‘hoi’ is hoi (falling-rising); and ‘nga’ is nga (falling-rising with a glottal stop). For example, in (a) VKhI.ng-ta means speaker VanKhanh, first token, tone ngang, syllable [ta]. The form in the legend box represents the writing system, e.g., ta is the written form of

38

VIETNAMESE TONE: A NEW ANALYSIS

[ta] and da is the written form of [za], The phonetic form is used in the title of the graph, e.g., 8 speakers.ngang-ta means the pitch graphs (F0 graphs) of eight speakers using tone ngang in [ta]. In Figure 8, the three males Hung, Hoang, and Son always have the lowest frequencies. Hung has the widest frequency range and Hoang is in the middle of the range except in sac1 (c) and sac2 (g). Ngang in 8(a) is level and usually falls slightly, during the last 30 ms. Ngang starts at between 210 to 270Hz for females and 120 to 175Hz for males. Huyen in 8(b) is also level and falls slightly. Huyen starts at between 150 to 250Hz for females and 110 to 160Hz for males. Sac1 in 8(c) starts slightly lower than huyen, at between 180 to 230Hz for females and 100 to 150Hz for males. It is mainly level and rises during the last 60 ms. The length of the level part varies from speaker to speaker. In An and Hoang the level part is quite long until approximately 150ms. However, generally after about 120ms this tone rises and ends between 220 to 320Hz for females and 120 to 190Hz for males. Depending on the speaker, the F0 difference between the beginning and end points of this tone is between 40 to 90Hz for females and 20 to 40Hz for males. The height of the ending of sac1 varies from speaker to speaker, e.g., around 250Hz for VanKhanh, but up to 320Hz for Binh. It also varies from form to form for the same speaker. Nang1 in 8(d) starts between about 180 to 280Hz for females and between 100 to 155Hz in males. It falls steadily after approximately 90 ms for most speakers. It has either a glottal stop (a blank in the line showing pitch) or a very creaky portion from between 150 ms to 180 ms, where it looks very chaotic in the pitch graph. Then the tone remains level or rises slightly

VIETNAMESE TONE: A NEW ANALYSIS

39

before the end. Whether the tone ends with a glottal stop or continues for a while varies from speaker to speaker and from token to token. Hoi in 8(e) starts the same as the onset of ngang or nang1 and gradually falls to between 140 to 180Hz for females and 90 to 120Hz for males. Then it rises slightly again. In some cases, the rising part is absent and the tone continues to fall lower than huyen. Nga in 8(f) usually starts higher than hoi, around the onset of huyen, and falls steadily during the first 90 ms. At this point it is either broken by a glottal stop or becomes very creaky between 90 and 120 ms. If the tone is broken by a glottal stop, the pitch curve is interrupted in the middle of the tone. If it is creaky, the lowest point in nga is around 70Hz in Phuong. After the creaky part, the tone rises. The height of the ending varies from speaker to speaker and from token to token. In order to examine sac2 and nang2, it is necessary to use obstruent-final syllables. Sac2 in 8(g) has the highest starting point and is short. It rises and ends after about 150ms for females and 100 ms for males. This difference between female and male speech could also be a sociolinguistic marker because a long sac2 sounds more feminine. Nang2 in 8(h) is also very short. This tone is very similar to huyen from beginning to end, but it is much shorter. Figure 9 shows six tones for [ka] along with sac2 and nang2 for [ka:k] from a female. The range of F0 is between 50 and 350Hz with the highest point in sac2 and the lowest point in nang1. There is a variety of tonal shapes: tones are flat (level), go

40

VIETNAMESE TONE: A NEW ANALYSIS

up (rising), go down (falling), and go down and then up (curved). However, it is difficult to say whether the three lowest curved tones hoi, nang1, and nga are always curved and only different in F0. The results show that it is possible to classify tones by their contour: rising tones, falling tones, and falling-rising (curved) tones. (66) summarizes the findings about contour. (66) is similar to (65), except that ‘unmarked’ and ‘marked’ have replaced ‘high’ and ‘low’ (see Chapter Two). Hoi is classified as an unmarked tone and placed in the same category as ngang, sac1, and sac2. Nga is a marked tone belonging to the same category as huyen, nang1, and nang2. (66) divides tones into two categories: even (level) tones and non-even tones, i.e., tones that have a change in fundamental frequency. The latter are then divided into three types by tonal shape. (66) EVEN

NON-EVEN

rise/ fall

rise/ fall

curve

UNMARKED MARKED

ngang huyen

sac 1 (rising) nang 1 (falling)

sac2 (rising) nang2 (falling)

hoi nga

VIETNAMESE TONE: A NEW ANALYSIS

41

Figure 8. Tones from eight speakers

There are variations in the tonal shapes of the two curved tones hoi and nga, and of the two falling tones nang1 and nang2. Hoi usually goes down and then slightly up. However, each previous study has differed in describing how curved this tone is. For example, Vu 1982 says that it is curved but Hoang 1986 says that it goes down and then up slightly. Vu 1999 observes that the rising part is clearer and stronger in males than in females. This investigation shows variation in the rising part of hoi from speaker to speaker. For example, in Figure 10 hoi is curved in 10(a), but not in 10(b). (Nga is included in this graph to show the glottal stop in the middle of the tone and the low ending.) The curve in hoi does not appear consistently even between tokens from the same speaker using the same form. In Figure 10(a) from Binh hoi is curved in [ta]. Figure 11 shows a second token of hoi in the sane form [ta]. There hoi goes down and flattens out. Nga is not necessarily broken by a glottal stop. It can have a creaky portion instead. Figure 10(a) and Figure 11 show nga broken by a glottal stop; however, in Figure 10(b) there is no glottal stop. Vu 1999 also notes this variation. If a glottal stop breaks the tone, a discontinuous portion in the pitch curve shows the stop. If the tone is creaky instead, the fundamental frequency can be measured only when it is clear enough to observe the individual cycles. In all studies nga always falls and then rises to make a clear curve. However, unlike Vu 1982 and Hoang 1986, this study did not find that nga always had a very

42

VIETNAMESE TONE: A NEW ANALYSIS

Figure 9. Eight tones in [ka] and [ka:k] from Binh

Figure 11. Hoi in a second token of [ta] from Binh

high ending. Figures 10(a) and 10(b) and Figure 11 show that nga ends at a very high point for An, but Figures 10(a) and Figure 11 show that it does not for Binh. The rising part of nga also varies from form to form for the same speaker. In Figure 12 Phuong has nga with a high ending in [ma] but a low ending in [sa]. Nga also varies from token to token for the same speaker, as Figure 13 shows. It has a high ending in 13(a) and a low ending in 13(b). Nang1 is a falling tone with a glottal stop. Vu 1988 says that the tone ends with a glottal stop. However, as Figure 8(a) showed, after the glottal stop or a very creaky portion, the tone goes either flat or slightly down sometimes and even rises. Contrary to the descriptions in Vu 1999:40 and Vuong and Hoang 1994, nang2 does not fall steadily. It is just slightly higher than huyen, except that it is much shorter because of the final stop consonant, as in Figure 8(h). This result agrees with Hoang 1986. The study also investigated the phonation types of creakiness and breathiness. Nguyen and Edmondson 1987 used the term ‘voice quality’ to refer to phenomena such as pharyngealization, laryngealization, glottal stop, and breathy voice. This study (see also Ladefoged 1988) avoided the term ‘voice quality’ because in some work (e.g., Laver 1994), it also refers to the presence or absence of nasalization. Ladefoged acknowledges that ‘phonation type’ is hard to define precisely if the intent is to include not only those states in which the glottis vibrates in some way but also uses of the abducted vocal folds in voiceless sounds (1988:374). Nevertheless, this term is useful and is used here. Laver 1984:184 says that phonation is the action of the larynx that transforms the airflow into the audible acoustic energy necessary to provide voice quality. The term covers modal, breathy, and creaky voice.

VIETNAMESE TONE: A NEW ANALYSIS

43

Figure 10. Hoi from two speakers

Figure 14 shows the waveforms of three laryngeal settings (modal, creaky, and breathy) from Ladefoged and Maddieson 1996. Modal voice is shown in the middle row. It is a laryngeal setting in which the arytenoid cartilages are in a neutral position for voicing, neither pulled apart nor pushed together (Stevens 1988, from Ladefoged and Maddieson 1996:50). This setting produces periodic pulses with a moderate level of frequency and intensity, characterized by regular cycles in the waveform. Creaky voice is shown in the top row. In creaky voice the vocal folds are strongly abducted and longitudinal tension is weak (Marasek 1997). The vocal folds thicken and the result is more irregular glottal pulses (more jitter) at a very low frequency (25 to 50 Hz). The amplitude is reduced a little more than in modal voice. Breathy voice is shown in the bottom row. In breathy voice the vocal folds are weakly abducted, i.e., the contact between the vocal folds is poor. Vocal fold vibration is not efficient and is at a lower frequency than in modal voice. Closure of the glottis is incomplete. The result is glottal leakage, causing audible friction noise (Marasek 1997). In some cases it is very difficult to identify individual pulses in the waveforms. The glottal pulses are regular but the amplitude is noticeably reduced. This study measured phonation types through the use of digital sound spectrography (discussed in Ladefoged et al. 1988). Kirk et al. 1984 also used this technique to measure phonation types in Jalapa Mazatec, as reported in Ladefoged et al. 1988:

44

VIETNAMESE TONE: A NEW ANALYSIS

Figure 12. Nga from Phuong

Figure 14. Waveforms of modal, creaky, and breathy vowels from one speaker of Jalapa Mazatec (from Ladefoged and Maddieson 1996)

298. They say that ‘this technique provides an excellent general view of many aspects of differences in phonation types’ (1988:300). They also investigated phonation types in Jalapa Mazatec by measuring the variation in the interval between glottal pulses in waveforms (1988:302). (See Dilley 1996 for more on this technique.) Technical problems prevented a complete replication of the spectral sections in Ladefoged and Maddieson 1996 so a visual inspection for specific cues of phonation in the waveforms and wide-band spectrograms was substituted in order to examine the characteristics of these three phonation types in Vietnamese vowels. One speaker in the research, Son, was chosen for close examination. In the data that follow the filtered waveforms help identify individual pulses more clearly in breathy and creaky tones. All syllables have the same form [ta:] except for sac2 and nang2 with CVC syllables, exemplified here with the final consonant [k], i.e., [ta:k]. Figure 15 shows the waveform and filtered form of the two level tones ngang and huyen for [ta:] in (a). The first row in 15(a) is the original waveform of ngang, the high level tone. Its filtered counterpart is in the second row. This tone has a modal voice, as indicated by the regular pulses and the relatively high intensity, i.e., the peak of the cycles is relatively high. Acoustic cues also show the presence of breathiness. Huyen, the low counterpart of ngang, is in the third row. Here there is a breathy tone characterized by periodic glottal pulses but with reduced amplitude after approximately 60 ms. The filtered counterpart in the last row shows the very low intensity of huyen compared with that of ngang. 15(b) shows spectrograms for these tones in [ta:]. It is more difficult to identify individual pulses in the spectrogram of the breathy huyen tone than in the waveform. Breathiness in huyen in 15(b) is clearly indicated by irregularly spaced pulses and reduced amplitude, especially between 60 ms and 100 ms. The research showed this be the typical pattern for breathiness. Breathiness is not apparent in the tone ngang where the pulses are periodic with moderate amplitude, represented here by darker glottal pulses. Figure 16 shows the waveforms and spectrograms for the rising tone sac1 and the falling tone nang1 in the open syllable [ta:]. In 16(a) sac1 has modal voice with regular glottal pulses and nang1 has creaky voice. However, in Vietnamese there are

VIETNAMESE TONE: A NEW ANALYSIS

45

Figure 13. Nga from Binh

two manifestations of creaky voice: one has a glottal stop and the other is very creaky, similar to the pattern of irregular, low frequency pulses shown in Figure 14. The presence of either of these phenomena was used to identify creakiness in the sound waves. Nang1 has a glottal stop in the middle of the tone, from approximately 100 to 200 ms, characterized by a very long non-periodic portion in the waveform, interrupted by two irregular pulses. In the spectrogram in 16(b) the corresponding portion of this period shows as irregular, widely spaced pulses. After this portion the tone continues with a few irregular pulses. This is a typical pattern for nang1 in various speakers. Figures 17 show the waveforms and spectrograms of the two curved tones hoi and nga. In 17(a) hoi has breathy voice like huyen in Figure 15(a). However, hoi is somewhat breathier than huyen. Its original (raw) waveform showed weaker amplitude and it was quite difficult to identify individual pulses in this tone. The reduced amplitude of hoi is clearly observable in the filtered form in the second row as well as in the spectrogram in 17(b) in the gaps between pulses, from 40 ms to 60 ms. Between 70 and 120 ms the tone is very breathy; the glottal pulses are also very weak and it is difficult to identify individual pulses. The other curved tone nga is broken in the middle with a glottal stop from 80 to 120 ms, interrupted by one or two widely spaced irregular pulses, the second row in 17(b). This break corresponds to the wide gaps between the vertical

46

VIETNAMESE TONE: A NEW ANALYSIS

striations in the spectrogram in the bottom row. The tone resumes regular cycles after approximately 130ms but with lower intensity. Figure 18 shows the last two tones sac2 and nang2, which occur in stopfinal syllables. The rising tone sac2 has modal voice like sac1 but is much shorter and ends after about 100 ms. The falling tone nang2 has a lesser degree of breathiness than sac2. In the waveform, i.e., 18(a) in the bottom row, the glottal pulses of nang2 are longer and weaker than those of sac2 in the second row. Nang2 is also very short and ends after about 100 ms. However, there is no glottal stop or creakiness in nang2 as there is in nang1. (67) summarizes the evidence from this visual inspection of the spectrographic characteristics of the eight Vietnamese tones. Three tones have modal voice: ngang, sac1, and sac2. Three tones have breathy voice: huyen, hoi, and nang2. Finally, two tones show creaky voice: nang1 and nga. (67) ngang huyen sac1 nang1

hoi nga sac2 nang2

modal voice (periodic, regular glottal pulses and moderate amplitude) breathy voice (regular glottal pulses and reduced amplitude) modal voice (regular glottal pulses and moderate amplitude) glottal stop or creaky portion close to the end of the tone (irregular widely spaced pulses, sometimes interrupted by one or two irregular pulses; reduced amplitude; complete closure of the vocal folds results in glottal stop and incomplete closure results in creakiness) breathy after approximately 40 ms, breathiest from 70 to 120 ms (regular pulses and reduced amplitude) glottal stop or creakiness in the middle of the tone (irregular widely spaced pulses from about 70 to 130 ms, and gaps in the spectrogram) modal voice (regular pulses and moderate amplitude), but very short because of to the final stop, ending before 120 ms some breathiness after 60 ms (regular pulses and reduced amplitude), but also very short because of the final stop, ending before 120 ms. The F0 of nang2 is very close to that of huyen

These characteristics are also quite consistent across speakers. See Appendix for the spectrographic data from other speakers. Although both hoi and huyen are breathy, hoi is breathier than huyen. However, breathiness in hoi decreases toward the end. Typically, in hoi the F0 of pitch rises after the breathy portion, giving the tone a curved shape. In the falling tone huyen, breathiness increases toward the end. Both ngang and sac1 have modal voice. Spectrograms do not easily show the difference between these tones. The difference lies in the value of the F0 of the ending: sac1 is a rising tone while ngang is a level tone. These results show that there is a relationship between fundamental frequency and phonation type in Vietnamese. Tones with creaky or breathy voice also have low F0, either throughout the whole tone, e.g., huyen and nang2, or at some point, e.g., nang1, hoi, and nga. This result is not surprising since the faster the vibration of the vocal folds the higher the tone. In breathy voice the vocal folds vibrate at a slower rate than in modal voice. (See also Chapter Five.) Although breathiness and creakiness in huyen, hoi, nang1, nang2, and nga are found in almost every token, acoustic measurements often show some differences in degree. For example, although both huyen and nang2 are breathy, nang2 may be less breathy than huyen for some speakers. This is a consequence of the shortness of nang2. Shortness prevents breathiness from being fully realized. Shortness does not cause breathiness because sac2 is also very short and has modal voice. Phonation type is usually observed in the spectrogram after about 60 ms. In nang2 the tone is very short. Breathiness is not very strong before the tone ends at about 110ms; therefore, there is not enough time for breathiness to be fully realized. Moreover, with sac2 and nang2, two tones that occur in stop-final syllables, the primary cue to perception is not breathiness (there is no breathiness in sac2) but the complete blockage of the air flow because of the sudden closure of the oral cavity at the end of the tone. Nguyen and Edmondson 1997 is the only study of Vietnamese tones that uses modern instrumental techniques to investigate phonation types. Using a Rothenberg mask to record the data and a method called Digital Inverse Filtering, they observed the size and shape of waveforms when airflow goes through the vocal folds. Examining glottograms (filtered waveforms that show only glottal vibration) from three speakers, they found a large difference between the waveforms of six tones in the adjustment of the vocal folds. Three tones, ngang, huyen, and sac1, were relatively stable whereas the production of nang1, hoi, and nga involved noticeable changes in the quality and quantity of the airflow. Nguyen and Edmondson found that ngang usually has modal voice. Huyen shows breathiness for some speakers but not for all. Sac1 also has modal voice. They describe a harsh voice in one speaker in hoi because of the vibration of both the vocal folds, and the false vocal folds; however, it is not clear what they mean by ‘harsh voice.’ Nga starts with modal voice, the airflow is very small in the middle (there is little or no voicing because of creakiness or glottal stop), and it ends with modal

VIETNAMESE TONE: A NEW ANALYSIS

Figure 15. The level tones ngang and huyen in [ta:] from Son

47

48

VIETNAMESE TONE: A NEW ANALYSIS

Figure 16. Sac1 and nang1 in [ta:] from Son

VIETNAMESE TONE: A NEW ANALYSIS

Figure 17. The curved tones hoi and nga in [ta:] from Son

49

50

VIETNAMESE TONE: A NEW ANALYSIS

Figure 18. Sac2 and nang2 in [ta:] from Son

voice. Nguyen and Edmondson say that the waveform looks like a horizontal hour glass. Nang1 also starts as modal voice but the amount of air going through the vocal folds decreases toward the end. These descriptions agree in large part with those reported here. One small difference is that Nguyen and Edmondson claim that breathiness is optional in huyen in some speakers. This study shows that there is a difference in the spectrograms of ngang and huyen in all speakers, with huyen always breathy and ngang always modal. The degree of breathiness also varies among speakers. Whereas Nguyen and Edmondson used only glottograms (filtered waveforms) to compare speakers, the waveforms and spectrograms in this study show differences in the phonation of tones in the same speaker. However, because there was variation in the degree of breathiness or creakiness among speakers, no comparison was made of phonation types across speakers. Just as each speaker has a different pitch range so each speaker uses phonation contrastively according to his or her physical characteristics, e.g., one speaker might have a creakier voice than another. Waveforms and spectral information therefore show that among the eight tones in Vietnamese, ngang, sac1, and sac2 have modal voice, huyen, hoi, and nang2 have breathy voice, and nang1 and nga have creakiness. These voice qualities are found consistently for all speakers. Sac2 and nang2 are also very short. (68) summarizes the phonation types using the term register for phonation type. (The motivation for this use will be discussed and justified in Chapter Five.) (68) shows the characteristics of features that remain stable in each tone. The parentheses around the feature (curve) in hoi shows that this feature can vary from speaker to speaker.

VIETNAMESE TONE: A NEW ANALYSIS

51

(68) Tone

ngang

huyen

sac1

nang1

hoi

nga

sac2

nang2

Contour Register

level modal

level breathy

rise modal

fall creaky

(curve) breathy

curve creaky

rise modal

fall breathy

The F0, length, and the linear portions of tones were also examined. The traditional literature on tonal pitch classifies huyen as a low-level tone and nang2 as a low falling tone. Figure 19 shows the fundamental frequency of these two tones from Dung, a female. The F0 difference between huyen and nang2 is very small in 19(b) and the two tones lie on top of each other. In both 19(a) and 19(b) huyen is a long tone and nang2 is a short tone that ends with a final stop consonant. There is also little difference in tonal height between the breathy curved hoi and the low falling nang2 in Figure 20. Huyen, hoi, and nang2 all have breathy voice and vibration of the vocal folds is at a lower frequency for breathy voice than for modal voice. This is a possible explanation for patterns such as those in Figures 19 and 20 in which all the tones are breathy. Consequently, there is very little F0 difference between them. Since both huyen and nang2 are classified as low register, it is not surprising that they are similar in F0. The similarity of hoi and nang2 is more surprising, given that they are in different registers phonologically (see Chapter Two). Even more surprising, sac1, traditionally a ‘high’ tone, and nang2, traditionally a ‘low’ tone, are almost identical in F0. Figure 21 shows the F0 of the high rising tone sac1 and of the low falling tone nang2 from Hoang. Sac1 is the long tone and nang2 is the short tone. The F0 in these cases provides no information for distinguishing the tones. In a number of studies, e.g., Doan 1977 and Vu 1999, length is used (among other features) to describe tones phonetically and even phonologically. Alves 1997 posits [length] as an inherent feature of tone. He suggests that sac and nang (he adopts the six-tone view so sac and nang in stop-final syllables are the same as in non-stop-final syllables) are inherently short, hoi and nga are inherently long, and ngang and huyen are neither inherently short nor long. Nguyen and Edmondson 1997 show some variation phonetically in most tones. This study shows even more variation. However, sac2 and nang2 are always short with that shortness almost certainly related to the final voiceless stop. In the traditional literature on tones, many studies, e.g., Doan 1977:116, Vuong and Hoang 1994:98, and Vu 1999:42, claim that nang1 ends with a glottal stop that makes the tone very short, even though this tone occurs in open syllables, e.g., /ma/ ‘rice seedlings,’ /ta/ ‘a hundred kilograms,’ and /la/ ‘strange.’ Therefore, nang1, sac2, and nang2 are classified as short tones, as in Vu 1999:46. Figure 22 with data from An shows the tone nang1 ending with a glottal stop. Nang1 is the falling tone in Figure 22 and it drops abruptly and stops within 30 ms, at 120 ms. As previously noted, except for sac2 and nang2, all tones can occur in sonorant-final syllables, which have equal length (see also Chapter Six.) The unusual shortness of the tone nang1 has puzzled researchers. Some, e.g., Vu 1999, ascribe phonemic status to this shortness. However, nang1 is not always short because of a glottal stop and sometimes there is no glottal stop but a creaky portion instead. If creakiness occurs, the tone is longer than when it ends in a glottal stop because the vocal folds continue to vibrate during creakiness. It is also unusual to end this tone after the glottal stop; more often, after the glottal stop nang1 either remains level, or falls, or even rises slightly. Thus this tone is as long as other tones. Figure 23 shows six tones in open syllables from Binh. In this figure nang1, represented by a line with crosses, is as long as other tones. Figure 24 shows nang1 in both short and long versions from another speaker An. Nang1 is short in [ma] in 24(a) just as it was in [ka] in Figure 22, but it is long in [sa] in 24(b). Therefore, the shortness of nang1 is optional and varies from form to form and from token to token. The shortness does not have distinctive status. It is generally assumed that since tone is realized on a vowel, it is a property of a vowel. (The phonological domain of tone is discussed further in Chapter Six.) The concern here is with the phonetic domain of tone. In closed syllables with non-stopfinal consonants, i.e., syllables that end with either a glide or a nasal consonant, the tone is spread over the whole rhyme not just over the vowel. Nga, the tone with most complex contour (it is creaky or has a glottal stop in the middle of the tone), can be used to illustrate such a contour. Figure 25 shows the tone nga in a form [ta:m] with a vowel and a final nasal consonant. The creaky portion of the glottal stop occurs approximately in the middle of the rhyme, i.e., between the end of the vowel and the beginning of the nasal (from 90ms to 150ms in the waveform). The rising part is realized on the nasal. Although there is only extremely weak voicing in the nasal portion (after 150ms), voicing is still essential if the tone is to be fully realized. If the nasal part is cut off, the rising part is missing and the tone becomes unrecognizable. If the vowel is short, as in [tam] in Figure 26, it is extremely short (ending before 60ms); the glottal stop occurs instead of the creaky portion and is quite long (from approximately 60 to 120ms); and the nasal part is lengthened (starting after 120ms, compared to 150ms in [ta:m] in Figure 25). Since the overall length is almost the same (210ms) in Figures 25 and 26, vowel length does not affect syllable length. Tonal length varies from token to token and from speaker to speaker. It can be distributed either over the entire vowel in open syllables or over the whole rhyme in non-obstruent-final syllables. Tonal length, therefore, is not distinctive. The final issue that must be addressed is that of the linear portion of tones, i.e., the portion that contains the most important information in a tone. Is it the onset, mid point, or end point of the tone that is important? Figures 27 and 28 using data from

52

VIETNAMESE TONE: A NEW ANALYSIS

Figure 19. Huyen and nang2 from Dung

two males show that the onsets of tones are not very different. It is even difficult to separate the onsets of some tones. (In these figures, the y-axis scale is different from that in females. For males, the pitch range is scaled maximally, i.e., 180Hz rather than 350Hz for females.) Figure 27(a) shows the eight tones in [ta] and [ta:k] from Hoang. Figure 27(b) shows only the first 60ms of each tone. In Figure 27(b) it is hard to make any distinction at all among the tones in the first 60ms, except to note that nga goes down noticeably. Figure 27(c) shows the same eight tones with the first 60ms omitted. 27(c) shows that all the tones are distinctive especially those that rise or fall immediately after the first 60ms. Nang2 does not fall steadily because, like huyen, it is breathy. Figure 28 shows eight tones from Son, another male. Figures 27 and 28 show that the tones of the two male speakers have similar shapes except for hoi and nang 1. Whereas hoi goes down then rises in Figure 28 it does not rise in Figure 27(a). Nang1 goes down in Figure 27(a) but it goes down and then rises in Figure 28. Furthermore, nga has a glottal stop in Figure 28 but not in Figure 27(a). There are many cases in which the fundamental frequency of the same tone varies randomly from token to token. For example, in Figure 27(a) the two level tones ngang and huyen have a steady fundamental frequency thoughout; however, nga

VIETNAMESE TONE: A NEW ANALYSIS

53

Figure 20. Hoi and nang2 from Phuong

Figure 21. Sac1 and nang2 from Hoang

falls abruptly after the onset. With the other tones, only after about the first 60ms does the difference in F0 become very noticeable, e.g., the fall of hoi or the rise of nang 1 in Figure 27(c). This is a consistent finding with the patterns in spectrograms. The laryngeal features of creakiness and breathiness become obvious after about the first 60 to 70ms. The spectrogram of huyen in (a) The entire tones (b) The first 60 ms Figure 15(b) shows that the pattern of breathiness in the waveform and spectrogram starts at about 60 ms. In Figure 16(b) the glottal stop in nang1 starts at 90 ms. Where do the crucial cues in a tone appear? In the two level tones ngang and huyen the information is given throughout the whole tone after approximately the first 60ms. The phonation type that the tone should have, e.g., breathiness in huyen, is consistent throughout the whole tone. For the other tones, the essential information is given either during the end point or in the middle of the tone depending on which feature the tone in question has. A close look at the rising tone sac2 and the falling tone nang2 in Figure 18(b) reveals that these tones are very short, approximately 90 to 100ms in duration. The most important cue in these tones is that they end very suddenly, segmentally with a final stop consonant. To distinguish between the two, nang2 has breathiness, here shown as longer and weaker pulses. Breathiness increases toward the end. The crucial portion of these two tones, therefore, is the end point.

54

VIETNAMESE TONE: A NEW ANALYSIS

Figure 22. Nang1 from An

Figure 23. Nang1 from Binh

With the rising and falling tones in non-stop-final syllables sac1 and nang1, the essential information comes in the second part of the tone. Sac1 usually does not rise until the second part of the tone, e.g., for Binh in Figure 23, or for most speakers in Figure 8(c), where the tone does not rise until after 120ms. However, it does not rise until two-thirds into the tone for Hoang in Figure 21, or even until the last 30ms for An in Figure 24. With the falling tone nang1, the glottal stop or creakiness comes in the second part of the tone. Although the tone usually goes lower before it ends, the most important cue is the glottal stop. That is why the tone is sometimes very long, as in Figures 23 and 24(b), and sometimes short, as in Figures 22 and 24(a). Most of the time the portion after the glottal stop is level, as in [ka] in Figure 24(b), or occasionally it rises, as in [ta] from Son in Figure 28. However, it is much more usual for nang1 to flatten or drop slightly after the glottal stop with its rising part optional. With the two curved tones hoi and nga, as the tonal shape suggests, the most crucial part is the middle of the tone (the curve) especially for tone nga. Hoi has breathiness and nga has either a glottal stop or creakiness. Nga rises after the glottal stop or creakiness in the middle and this rising part after the glottal stop is distinctive. Without this rising part, nga might be recognized as nang1, which also has a glottal stop. For example, in Figure 27(c), which omits the first 60ms, the creaky portion in nga is shown by the fact that the tone has the lowest starting point. With the breathy curved tone hoi, the curve is present in most speakers, although the actual curvature of the tone varies from speaker to speaker and from form to form, e.g., Figures 10(a) and 11 from Binh and 8(e) from eight speakers. The fact

VIETNAMESE TONE: A NEW ANALYSIS

55

Figure 24. Nang1 from An

that the curve in nga is much more obvious than the curve in hoi has a physiological explanation: subglottal air pressure builds during the glottal stop. The vocal folds open and the air is suddenly released, causing a burst of high frequency vibration. Hoang 1986:27 comes to a similar conclusion, i.e., the first part of a Vietnamese tone gives less information than the second part. However, he does not distinguish among tones with different contours. The inescapable conclusion is that the linear portion in Vietnamese tones is the mid point in curved tones and the end point in non-curved tones. Summary The phonation types of breathiness and creakiness are found in huyen, hoi, nang1, nang2, and nga. The F0 of pitch is also variable within and across speakers and provides no information to allow listeners to distinguish between tones when two different tones are compared. Furthermore, there is a correlation between phonation types and F0: where there is breathiness or creakiness the tone is low, either in the entire tone if breathiness or creakiness occurs throughout, or at the point where they are found. Tonal length and height also vary from speaker to speaker, from form to form, and from token to token.

56

VIETNAMESE TONE: A NEW ANALYSIS

Figure 25. Nga in a nasal-final syllable with a long vowel |ta:m|

Figure 26. Nga in a nasal-final syllable with a short vowel [tam]

A phonetic feature that is invariant across and within speakers is a primary feature. If the F0 of pitch is not a primary feature of Vietnamese tones, there must be something else that is stable and more reliable. The following chapter addresses this issue.

VIETNAMESE TONE: A NEW ANALYSIS

Figure 27. Eight tones in [ta] and [ta:k] from Hoang

57

58

VIETNAMESE TONE: A NEW ANALYSIS

Figure 28. Eight tones in [ta] and [ta:k] from Son

CHAPTER 5 The Phonetics and Phonology of Tone

The acoustic study of tone reported in the previous chapter has important phonological implications for an understanding of tone. It suggests that the mismatch problem between hoi and nga discussed in Chapter Two is an illusory one and that the laryngeal features proposed there are a key to that understanding. Tonal shape and the phonation types of creakiness and breathiness are primary features for differentiating tones with pitch height predictable from tonal shape and phonation types. Resolving the Mismatch Problem (69) shows the characteristic phonetic features of each of the eight tones. The parentheses around the curved feature in hoi show that this feature varies from speaker to speaker. The last row (Pitch) does not show pitch as it is classified traditionally. It shows how Register predicts pitch height, i.e., how breathiness predicts lowness in the tone either throughout, as in huyen, or at a certain point, as in hoi. (69) Tone

ngang

huyen

sac1

nang1

hoi

nga

sac2

nang2

Contour Register (Pitch)

level modal H

level breathy L

rise modal H

fall creaky L

(curve) breathy L

curve creaky L

rise modal H

fall breathy L

Hoi is classified traditionally as a phonetic low tone but patterns phonologically as a high tone; nga is classified traditionally as a phonetic high tone but patterns phonologically as a low tone. The phonetic characteristics and phonological patterning in reduplication of hoi and nga are given in (70) and (71) respectively. (70) High Low

ngang huyen

sac1 nang1

nga ; hoi

sac2 nang2

(71) Unmarked Marked

ngang huyen

sac 1 nang1

hoi nga

sac2 nang2

This study shows that hoi is higher than nga. In some cases hoi is phonetically higher than nga only in the middle, the most crucial point in these tones. This finding is consistent with the pattern found in the studies discussed in Chapter Four. In other cases hoi is phonetically higher than nga throughout. The glottal stop in the middle of nga is shown in the pitch tracks in Nguyen and Edmondson 1997 and in Vu 1982, which were discussed in Chapter Four (see Figures 5 and 6). This study also shows this clear break in nga. In Figures 29 and 30 there is a glottal stop in the middle of nga from Binh and Hung. In these figures nga is broken by a glottal stop, represented by a discontinuous portion in the middle of nga. When the glottal stop is not present, hoi intersects nga, as in Hoang 1986 (see Figure 7 in Chapter Three). This pattern also occurs here. Figure 31 shows two forms from two females, Binh and An, and Figure 32 shows two forms from two males, Hoang and Son. In these cases, although both hoi and nga go down in the middle, the lowest part of nga is always lower than that of hoi. Previous studies have failed to notice this fact and have treated nga ‘overall’ as a high tone and hoi ‘overall’ as a low tone (‘overall’ in traditional usage being the average of an equally-spaced sample). This view is inadequate.

60

VIETNAMESE TONE: A NEW ANALYSIS

Figure 29. Nga with a glottal stop in [ta] and [ta:j] from Binh

Two points need to be clarified concerning the F0 of tones. First, any comparison of the F0 of hoi and nga (if such a comparison of F0 is appropriate) must take into account the relationship of these two tones alone, i.e., whether hoi is lower or higher than nga. Other ‘low-high’ tone relationships are irrelevant. For example, if tones are divided into their registers according to overall height in the pitch span, e.g., Cu et al. 1977 and Huu and Vuong 1980, then both hoi and nga can be specified as low tones. In Figure 33, which omits sac2 and nang2, hoi and nga fall completely into the lower part of the pitch range along with nang 1. A similar observation can be made using [ta] and [ka] from Binh in Figure 34 (with sac2 and nang2 once again omitted). In these forms hoi and nga fall into the lower part of the pitch range. This finding suggests that tones should be compared within their pairs rather than in a general relationship with other tones. Previous studies confirm this observation. For example, speaker 6 in Figure 5 from Nguyen and Edmondson 1997 produces three tones that occur in the lower part of the pitch range: ngang, huyen, and hoi, and, if the glottal stop in nga is ignored, three that occur in the higher part: sac 1, nang 1, and nga. Traditionally, ngang is described as a high tone and nang 1 as a low tone, but nang1 shares the same pitch space with high tones and sac 1 shares the same pitch space with low tones. If a tone is classified as low or high according to which part of the pitch range it falls into, nang1 should be classified as a high tone and ngang as a low tone in Figure 5.

THE PHONETICS AND PHONOLOGY OF TONE

61

Figure 30. Nga with a glottal stop in [ma] and [ka] from Hung

Depending on which point in the tone is deemed relevant hoi can be regarded as low and nga as high, or vice versa. Hoi is breathy and nga is creaky. If nga is creaky in the middle, this part is lower in nga than in hoi, since creakiness makes the tone lower than does breathiness, e.g., Marasek 1997 and Ladefoged and Maddieson 1996. If nga has a glottal stop in the middle, then the F0 comparison is impossible, because there is no phonation during a glottal stop. Nga is usually treated as a high tone because it ends very high. This end is said to make the tone high ‘overall.’ However, this argument for treating nga as high is not convincing. First, Chapter Four, shows that the end of nga can be either high or low, varying as it does from speaker to speaker, from form to form, and from token to token. A high pitch ending for nga is clearly not a good criterion for classifying it as a high tone. While the height of the end point of nga can vary, nevertheless there is a constant: namely, contour (tonal shape), which is a distinctive feature in Vietnamese tones. For example, hoi and nga are the two curved tones. Nga without the middle part would be like sac1, a rising tone. Therefore, in a comparison of hoi and nga the middle part cannot be ignored; nga must be curved. All previous accounts classify sac1 as a high-rising tone and huyen as a low-level tone. However, Figure 35 shows that there are problems with such accounts. It shows the pitch trajectories of huyen and sac1 in [ka] and [sa]. Huyen starts higher than sac1 and gradually goes down. Sac1 starts lower than huyen, goes flat, and then rises to be higher than huyen only during the last 30ms. Therefore, if overall pitch height is the criterion needed to classify tonal height, the high-rising sac1 would be

62

VIETNAMESE TONE: A NEW ANALYSIS

Figure 31. Hoi intersects nga from Binh and An

low and the low-level huyen would be high. If tonal height is instead reconceived as ‘overall shape,’ it is possible to challenge the traditional classification of tones. In another pattern, the whole of hoi is higher than nga. This pattern is found not just in one speaker but in various speakers and even in different tokens from the same speakers. Figures 36, 37, and 38 provide data from Dung, Phuong, and Binh, respectively. In every token hoi is higher than nga throughout. Figures 36(a) and (b) show different tokens of [ta] and [ka] from Dung. In some cases the onset of nga is higher than that of hoi, but the onset is not crucial, as Chapter Four showed. Figure 37 shows [ka] in (a) and [ta] in (b) from Phuong. In this same pattern hoi is higher than nga throughout. Figure 38 shows data from Binh, again for [ta] but with different tokens. Although there is a glottal stop in the middle of nga, hoi is higher than all of nga. This result challenges the view that hoi is low and nga is high. If tonal height is defined according to the maximal low/high points, the lowest point of nga is always lower than that of hoi. The middle point is important in these two tones because without a curve nga is similar to sac1, a high-rising tone, and hoi would be confused with huyen, a low-level tone. Also challenged is the claim that there is a mismatch between the phonetics and phonology of hoi and nga. However, it may be the case that the phonological diagram in (71) also accurately represents the phonetics.

THE PHONETICS AND PHONOLOGY OF TONE

63

Figure 32. Hoi intersects nga from Hoang and Son

Tonal Shape and Phonation Types One of the criteria used to determine whether a tonal feature is phonological is that the feature should be useful in grouping tones into natural classes and help explain the tonal patterning found in a language. If features are also actual articulatory instructions (Rice and Avery 1991) and the theory of Contrastive Specification is correct, a phonological feature should be a reliable and constant part of the phonetic realization of a segment. A distinctive feature is not something that occurs optionally or varies freely. It should play a role in producing a contrast in the language and should have a regular mapping into the phonetics. None of the phonetic features proposed in the literature on Vietnamese properly groups tones into natural classes. For example, glottalization, i.e., glottal stop or creakiness, should pair sac1 with nga but these tones do not pattern together. Tonal height seems to be the most reasonable feature for grouping tones into two registers according to their patterning in reduplication. However, even with the mismatch problem between phonetics and phonology solved, the F0 of tone is unstable and appears not to be the appropriate phonological feature for classifying tones into natural classes in Vietnamese. The register feature [laryngeal] proposed in Chapter Two seems to be a better candidate. The F0 of tone varies from token to token and from speaker to speaker. For example, nga can have a very high end point for one speaker but a low one for another and the end point can even differ from token to token for the same speaker.

64

VIETNAMESE TONE: A NEW ANALYSIS

Figure 33. Hoi and nga from Dung

However, phonation type is quite consistent both within individuals and across speakers. The tones that have non-modal phonation are huyen, nang1, nga, nang2, and hoi. Huyen, hoi, and nang2 are breathy. Nang1 and nga are creaky. The breathiness in huyen is the only difference between the two level tones ngang and huyen. All speakers have breathiness in huyen. However, the degree of breathiness varies from speaker to speaker. Figure 39 shows the waveforms and spectrograms of ngang and huyen for a male, Hung. Huyen is breathy but only minimally. The spectrogram of huyen in the last row shows that although the glottal pulses are periodic in the waveform, breathiness appears as reduced amplitude after 70ms, increasing after 150ms. Because of the incomplete closure of the vocal folds, there is also some space between glottal pulses, especially after 150ms. However, these signals are weak. Figure 40 shows the waveforms and spectrograms of ngang and huyen for the same form [ma] for another male, Son. Huyen is very breathy. The breathiness in huyen is clearly seen in the waveform in the second row in the very weak cyclical pulses, especially after 150ms. There are gaps between pulses toward the end of the vowel in the spectrogram in the last row. A comparison of the spectrograms of ngang and huyen in Figures 39 and 40 with their corresponding pitch graphs shows that the difference in F0 between ngang and huyen for Son is much smaller than that for Hung. Figures 41 and 42 show the corresponding pitch graphs of ngang and huyen for Hung and Son. This finding suggests that a speaker may have a choice: higher F0 for ngang or stronger breathiness for huyen to the degree necessary to make a phonetic contrast. There is a trade-off between the two phonetic realizations. Whatever the choice, huyen always has some degree of breathiness. However, ngang has no breathiness; it is modal voice. Huyen with breathiness is lower than ngang. Nang2 is another tone with breathiness. Sac2 and nang2 occur only with final obstruents, which make the tone very short. Nang2 does not have the length necessary for breathiness to be fully realized, but there is still some breathiness. Figure 43 gives the spectrograms of sac2 and nang2 in the form [ta:k] for An, a female. Phonation types usually begin after approximately 60ms, as nang2 in the last row shows. The cycles are regular and periodic in sac2 in the third row since the tone has modal voice, whereas the cycles of nang2 become very weak with noise in the high frequencies toward the end. There is a glottal stop in nang1 but none in nang2. The fact that nang2 is breathy and does not have a glottal stop makes it very much like huyen, a level breathy tone. This similarity is very noticeable in the pitch graphs of these two tones, as Figure 19(b) in Chapter Four showed. The two tones are identical in height. In other pitch graphs these two tones are also very close together. This is further evidence that nang2 is as breathy as huyen. The only phonetic difference between nang2 and huyen is that in nang2 the vocal folds suddenly stop vibrating; there is oral closure and the tone ends abruptly. There are two tones with laryngealization, nang1 and nga. In the literature nang1 is said to end with a glottal stop. However, there are two possibilities: the glottal stop can end the tone or the vocal folds might continue to vibrate after the glottal stop. If the vocal folds stop vibrating at the glottal stop, the tone is very short. If the vocal folds still vibrate after the glottal stop, there is creakiness. This variation after the glottal stop in nang1 shows that the portion after the glottal stop is not distinctive in this tone. Figure 44 shows the pitch graphs of nang1 in [ka] for Binh and An. (Nang1 is written as na in the legend of the pitch graph.) Nang1 is short for An and ends at about 120ms. However, for Binh the tone goes further and ends at approximately 210ms. The difference in length is shown in the corresponding spectrograms in Figure 45 in which the spectrogram of nang1 for Binh is in the last row. For An, in the third row, the vocal folds stop vibrating after 120ms. For Binh, creakiness starts after 120ms. The distances between pulses increase to the longest and then decrease after 180ms toward the end of the vowel. There is a similar situation in the same speakers with different forms and even in the same form but with different tokens from the same speaker. Figure 46 shows the pitch graphs of nang1 in [ma] for two different tokens from An. In this graph

THE PHONETICS AND PHONOLOGY OF TONE

65

Figure 34. Hoi and nga from Binh

nang1 in token I, represented by a line with grey diamonds, stops at about 120ms. The tone has about the same F0 in token III, except that after the glottal stop it continues to about 200ms. The different lengths of nang1 in Figure 46 are shown in their corresponding spectrograms in Figure 47. The spectrogram of the first token of [ma] is in the third row. Nang1 is very short. The vocal folds stop vibrating for 60ms after the glottal stop. Before vibration stops fully the vocal folds vibrate very weakly but do not produce a clear cycle. In the spectrogram of the other token in the last row, the vocal folds vibrate intermittently after 120ms and produce increased distances between pulses, followed by decreased distances between pulses toward the end. If phonation predicts tonal height, then the glottal stop pulls the tone down suddenly, as Figures 46 and 47 show, and the creaky portion makes the tone fall and become low, as in the long [ka] from Binh in Figure 45. The prediction that creaky tones will be low explains why nang1 in [ma] in Figure 46 is very high during the first 90ms before the glottal stop occurs: this part is high (about 270Hz) because it has modal voice. It is even higher than ngang and sac1, traditionally high level and rising tones, respectively. Figure 48, which shows six tones from An, is a clear illustration of this phenomenon. In this figure, discounting the last 30ms in nang1 when the tone is pulled down by the glottal stop, the level part of nang1 is higher than the whole tones ngang and sac1. The result is a paradox: nang1, a traditional ‘low’ tone in every description, is higher than two unquestionably ‘high’ tones, ngang and sac1. A theory that employs pitch height as a distinctive feature cannot explain why

66

VIETNAMESE TONE: A NEW ANALYSIS

Figure 35. Huyen and sac1 in [ka] and [sa] from An

nang1, a low tone traditionally, can be higher than high tones unless it makes reference to the glottal stop. However, if phonation type is the main cue, then so long as the glottal stop/creakiness is present, the tone is perceptible regardless of the height of the portion before the glottal stop because nang1 can either stop after the glottal stop (a common description of nang1 in traditional literature) or continue with creakiness. Figure 5(a) from Nguyen and Edmondson also shows that nang1 is considerably higher than ngang, a high level tone. A theory that treats pitch height as the primary feature of tone cannot explain such facts. However, if the laryngeal feature of creakiness is a primary feature of tones, it provides an explanation. What is important is that there is a glottal stop (or creakiness) in nang 1. If a tone has a glottal feature, it is recognizable no matter how high it is. However, the glottal stop also tends to lower the pitch height. This problem of the inconsistent description of nang1 is discussed in early acoustic studies of Vietnamese tone. One of the earliest experiments was conducted by Andreev and Gordina 1957. Vuong and Hoang 1994 point out that there is an inconsistency between the pitch height and shape of nang1 in their work and also in the work of Haudricourt 1972 (cited in Vuong and Hoang 1994). The description of pitch height and the shape of tones in the research reported here is similar to that of Andreev and Gordina 1957. For Andreev and Gordina and also for Vu 1982 nang1 is a low falling tone. However, according to Haudricourt 1972, tone nang1 is 22, a low level tone, whereas huyen, the low level tone in all accounts, is 21. This fact suggests that Haudricourt’s description captures only the first part of nang1 before the glottal stop. Without the glottal

THE PHONETICS AND PHONOLOGY OF TONE

67

stop or the creaky portion, nang1 can have high pitch. In nang1 glottalization not pitch height is crucial. Unfortunately, glottalization cannot be represented in a pitch graph but only in a spectrogram. The curved tone nga is usually said to have a glottal stop in the middle. If there is a glottal stop, it breaks the tone in two, as shown in Figures 5 and 6. This research found a similar pattern, shown in Figures 29 and 30, where the glottal stop is indicated by a blank. In this case, there is no phonation at all because the vocal folds cease vibrating and there is no measurable F0 at that point. If instead of a glottal stop there is a creaky portion in the middle of the tone, the line representing this tone in the pitch graph remains unbroken. The creakiness makes the tone very low at that point and the tone has a clearly curved shape. After creakiness the vocal folds vibrate either at very high frequencies, producing a clear V-shape tone as in Figures 31 and 32, or at lower frequencies, creating a slight curve as in Figure 36. Consequently, nga can be very curved or slightly curved, but crucially in each case there must be some creakiness in the middle of this tone. Figure 49 shows nga for a male, Hung. The first and second rows show the waveforms of hoi and nga. Their corresponding spectrograms are the third and fourth rows. Nga does not have a glottal stop and it is not strongly creaky. There are neither irregular glottal pulses nor widely spaced pulses in the spectrogram. However, the waveform from approximately 50ms to 90ms shows that the pulses reduce in intensity. The corresponding portion of the spectrogram shows some slight weakening

68

VIETNAMESE TONE: A NEW ANALYSIS

Figure 36. Hoi and nga in different tokens from Dung

of energy between 3200 and 4100Hz. At low frequencies there is some distance between pulses from about 50 to 90ms. This is where creakiness occurs. Creakiness predicts lowness. Figure 50 shows the pitch graph of the spectrogram in Figure 49. When creakiness is not strong enough to make the tone go down in the middle, the tone is consequently not clearly curved (in Vietnamese curved tones are concave not convex). In Figure 50, nga goes down only a little at approximately 60ms and then rises, although this type of realization of nga is rare. Figure 51 shows that it is really not important whether nga is low or high, or how low/high it is. The crucial information is whether there is some creakiness in nga. Without a glottal stop or creakiness, nga would be confused with sac1, a rising tone. Figure 51 shows the pitch graphs of both nga and sac1 for Hung. The tones are very similar in shape: both go down very slightly and then rise gradually. Figure 52 provides the corresponding spectrograms. Although nga and sac1 are similar in shape in Figure 51, they are not exactly alike in their spectrograms. In Figure 52, the waveform of nga in the second row is of low intensity (between 50ms and 80ms). On the other hand, sac1 does not change at all in intensity throughout the tone. The waveform and spectrogram are stable throughout. A comparison of the waveforms and spectrograms of the two tones in Figure 52 between 50 and 90ms with the corresponding portions in Figure 51 shows that nga with a little creakiness is higher in F0 than sac1. It would appear that although creakiness is produced at low frequencies, low frequencies do not necessarily produce creakiness. This is very strong evidence that phonation predicts pitch height, not vice versa.

THE PHONETICS AND PHONOLOGY OF TONE

69

Figure 37. Hoi and nga from Phuong

Figure 53 gives the complete pitch graph of eight tones for [sa] and [sa:k] for Hung to show where sac1 and nga are in the pitch range of this speaker, who has the widest pitch range of the three males. The value of F0 on the y-axis is usually from 200 to 250 Hz; with other males it is 180 Hz. Figure 53 shows that nga does not fall as expected. However, the other tones are distributed as expected, i.e., nang1 falls, hoi falls, and sac1 rises and is higher than ngang. There is also a further example of the unreliability of F0: it is difficult to separate the two tones huyen and nang2 in Figure 53. If phonation types are distinctive and breathiness lowers the tone, then because both huyen and nang2 are breathy, they are very similar in F0. The difference is that nang2 occurs in stop-final syllables, which shorten the tone. Phonologically the feature [obstruent] captures this characteristic of nang2. Without creakiness in the middle nga would lose its most important cue. Pitch height is not critical because nga can have either a very high ending, making it partially higher than hoi, or a very low ending, making it lower than hoi throughout. Treating the rising part of nga as phonetic resolves the problem of which is higher: nga or hoi. The phonetic height varies randomly. All that is required is that the F0 rises a little to make a curve. These results show that in nang1 and nga, glottal stop or creakiness is consistently, and thus distinctively, present. In huyen and nang2 there is breathiness in the tone. Nang2 is distinguished from huyen by the feature [obstruent] that shortens the tone. All these tones belong to the marked register. In the traditional literature these tones are classified as low, except for nga, whose glottal stop is ignored in order to allow it to be grouped with high tones. In the account given here phonation types are

70

VIETNAMESE TONE: A NEW ANALYSIS

Figure 38. Hoi and nga in [ta] from Binh

Figure 39. Huyen in [ma] from Hung

distinctive; therefore, the glottal stop or creakiness in nga must be taken into account. Creakiness and breathiness lower the tone so it is not surprising that marked tones are often low because each tone has either breathiness or creakiness at some point or throughout the tone. However, lowness is derived from the phonation types of creakiness or breathiness, i.e., without creakiness or glottal stop, nang1, the traditionally low tone, appears as a high tone in terms of pitch range. Pitch height, therefore, is not distinctive in this analysis. The fact that there is always breathiness and creakiness in marked tones is phonetic motivation for the phonological feature [laryngeal] for these tones. This feature groups tones into a natural class, i.e., the marked register. Hoi is the only tone that is breathy but unmarked for register. It is usually described as a curved tone that goes down and then rises higher than the onset. The spectrograms in Chapter Four and this chapter show that hoi is breathy and nga is creaky. Both breathiness and creakiness lower the tone; however, creakiness is produced at very low frequencies and makes a tone lower than does breathiness (Marasek 1997). This is a clear indicator that hoi cannot be lower than nga because nga has creakiness. The fact that hoi is not as curved as nga also has a physiological explanation: in nga subglottal pressure is built up after the glottal stop and the sudden release of the pressure results in strong turbulence causing high frequency noise, but in hoi breathiness does not cause high pressure because the vocal folds do not close completely. Therefore, after the breathiness there is no strong air turbulence to cause high frequency noise. The curved shape of hoi is represented with three points under the contour node in 28(c). The pitch graphs in Figure 54 show the consistent curved shape of hoi in [ta], [ka], [ma], [za], and [sa] for Dung from two different tokens. Figure 55 shows a similar pattern for VanKhanh from two different tokens. The tone is less curved than that for Dung, rising very little at the

THE PHONETICS AND PHONOLOGY OF TONE

71

Figure 40. Huyen in [ma] from Son

Figure 41. Ngang and huyen in [ma] from Hung

end. The breathiness in the middle of hoi relates to the curved shape of that tone. This study shows cases where the curve is very breathy, as the spectrogram in Figure 57 from Van demonstrates. In this figure hoi is in the first and third rows. Breathiness starts at about 90ms and is shown by very weak glottal pulses in the waveform. Breathiness increases to a peak level from between 150 and 200ms. In the corresponding portion in the spectrogram, there are wide gaps between pulses. Figure 56 is the corresponding pitch graph. Figure 56 shows that hoi is too breathy in the middle for actual cycles to be measured. However, more typically, breathiness spreads toward the end of hoi and the tone goes down and then rises slightly shortly before it ends, as Figure 32 shows for Son. Figure 58 is the corresponding spectrogram. In some cases hoi is breathy throughout the tone and the tone does not rise again toward the end. The curve disappears and the tone flattens toward the end or even falls. When hoi goes down at the end, its contour is similar to that of huyen, a level tone without creakiness or glottal stop. Others have also noted this phenomenon in modern Northern dialects, i.e., that hoi sometimes loses the rising part, e.g., Figure 5 from Nguyen and Edmondson. Vu 1999 says that the loss of the rising part is found more in females than males. This study shows that it occurs equally in males and females. When hoi occurs without the rising part, i.e., is breathy and slightly falling, it is like huyen. How is the missing information recovered to avoid confusion between hoi and huyen? The answer lies in phonation: breathiness must be stronger in hoi to compensate for the lost curve. Therefore, in order to be distinguished from huyen, hoi is much breathier than huyen, as Figure 59 shows. In this figure the waveform of hoi is in the second row and its spectrogram is in the last row. In the waveform, the tone is very breathy between 70 and 130ms so that it is difficult to separate individual pulses. The glottal pulses in hoi are much weaker than those in huyen. In the corresponding portion in the spectrogram of hoi, the spaces between cycles are wider than those of huyen. Between 70 and 130ms individual pulses are hard to identify and the cycles are irregular.

72

VIETNAMESE TONE: A NEW ANALYSIS

Figure 42. Ngang and huyen in [ma] from Son

Figure 43. Sac2 and nang2 in [ta:k] from An

The breathier the tone the lower it is in F0. Figure 60 shows that the pitch graph of hoi is lower than that of huyen. A theory that relies solely on pitch height would need two different types of ‘low’ here: huyen would be something like ‘mid-low’ and hoi ‘low-low.’ A theory that employs only phonation as a way to distinguish tones also needs two types of breathiness here: breathy and very breathy. However, whereas [breathy] is underlyingly specified as a register feature in huyen, it is not specified in hoi. Instead the breathiness associated with hoi is phonetic only. A tonal-height-based classification is also less preferable than a laryngeal-based classification because there are many cases in which pitch height does not pattern as expected under the pitch-height hypothesis. Listeners who are not very familiar with Northern dialects and even listeners who speak a Northern dialect easily confuse hoi without its rising portion with huyen because phonetically hoi and huyen become similar. Hoang 1986 reports a result from a relevant perception test. Participants were asked to recognize tones individually. Only 25% of instances of hoi were recognized correctly; the rest were confused with huyen. The fact that hoi can sometimes appear without the rising part might suggest a change in progress in Northern dialects, a neutralization of contour features, i.e., loss of an edge point in the contour feature of hoi, a possibility discussed later in this chapter. Since F0 varies to the point of not distinguishing a register pair, it is not a reliable cue to differentiate tones. The laryngeal features of breathiness and creakiness are stable and provide a more reliable auditory cue. These features, along with tonal shape, account for tonal height: where breathiness or creakiness occurs, the tone is low. In curved tones, breathiness and creakiness signal the lowest part. The features of breathiness and creakiness are found in all marked tones, i.e., huyen, nang1, nga, and nang2. The tone hoi is breathy but is unmarked for register. The breathiness in this tone satisfies a requirement of the contour feature [curve]. All marked tones except nga are described as low in the traditional literature. Nga is classified as high because of its high ending. However, the high ending of nga varies from token to token and should be classified as phonetic rather than

THE PHONETICS AND PHONOLOGY OF TONE

73

Figure 44. Nang1 in [ka] from An and Binh

Figure 45. Nang1 in [ka] from An and Binh

distinctive. Therefore, it is difficult to claim that phonation predicts low but not vice versa. However, in the case of nga, creakiness predicts curve but not vice versa. (72) summarizes all the phonological and phonetic properties discussed so far. The rows Contour and Register show phonological properties except that the phonation of breathiness in parentheses in hoi is phonetic. The feature [obstruent] occurs only in sac2 and nang2 and distinguishes both sac2 from sac1 and nang2 from nang1. The last row shows pitch height, treated here as a phonetic property. For example, huyen is a level tone, although phonetically it may fall a little at the end. It has breathy voice that predicts lowness. Nga is a curved tone with creakiness regardless of how high it ends or how low its middle part is. Nang1 is a falling tone with creakiness regardless of how high it is before falling because of creakiness or of how short or long it is. Breathiness in hoi is a phonetic feature. (72) Tone

ngang

huyen

sac1

nang1

hoi

nga

sac2

nang2

Contour Register (Pitch)

level

level breathy L

rise

fall creaky L

curve (breathy) L

curve creaky L

rise [obst] H

fall breathy [obst] L

H

H

Although there exists no account of Vietnamese tone that postulates phonation types as phonological features, researchers have recognized the important role such types play in the tonal systems of Asian languages. For example, Hoang Cao Cuong 1989 states that phonation types cannot be ignored in the phonology of Vietnamese tones. He criticizes a common assumption that in Vietnamese phonological features of the tonal system can be generalized from pitch alone. For example, if only F0 is taken into account, it is impossible to differentiate the tones huyen, hoi, nang1, and nang2. Moreover, he points out how the

74

VIETNAMESE TONE: A NEW ANALYSIS

Figure 46. Nang1 as short or long from An

Figure 47. Nang1 in [ma] in diffent tokens from An

F0 of tone can vary among speakers and claims that the phonology of the Vietnamese tonal system involves a trade-off between the fundamental frequency and the remnants of segments that still remain following the development of tones from segments, i.e., phonation types. However, he produces no evidence for such claims nor does he show how the phonation of laryngeal features can be a distinctive feature, how it is represented, or what the structure of tones is. Alves 1997 also recognizes that phonation features should be taken seriously as distinctive features. However, he likewise fails to produce any evidence for this claim nor does he show how such features can be distinctive in the phonology. The use of the phonation types of breathiness and creakiness distinctively is not unusual in languages (see Gordon and Ladefoged 2001). Ladefoged and Maddieson 1996 provide detailed examples of languages that use phonation types distinctively in both consonants and vowels. A large number of MonKhmer languages such as Phalok, Wa, Chong, Mon, Bru, Kui, So, and Nyah lexically contrast modal voice versus breathy voice (Thongkum 1988). Thongkum reports rare cases in which languages have three- or four-way contrasts among phonation types, e.g., Chong with the four-way contrast of clear voice, clear voice followed by creaky voice, breathy voice, and breathy voice followed by creaky voice. In Nyah and Kui, Thongkum reports that breathy voiced vowels and clear voiced vowels systematically differ from each other only in the laryngeal dimension (phonation types); the other differences (length of the vocal tract and vocal tract shape) are minor. In these two languages breathy vowels always have lower F0 than clear voice (modal voice) in all types of syllable structures (1988:321). It is now quite clear that the mismatch problem between the phonetics and phonology of hoi and nga is resolvable. Vietnamese uses phonation distinctively and it is phonation that predicts tonal height. F0 varies from token to token but phonation types remain constant. Although there is variation from speaker to speaker (more or less breathy, glottal stop or creakiness), for each speaker the presence of breathiness and creakiness in tones with non-modal voice is clearly evident. Specifically,

THE PHONETICS AND PHONOLOGY OF TONE

75

Figure 48. Six tones from An

Figure 49. Nga with little creakiness from Hung

creakiness is contrastive in nang1 and nga. Breathiness is contrastive in huyen and nang2. The traditional low tone nang1 can be higher than the high tone ngang in most parts of the tone so long as creakiness or glottal stop follows. Without creakiness nang1 is not perceptible. Nga can have a high or low ending because the middle portion is crucial to the tone. Therefore, without creakiness in the middle, nga would be confused with sac1. The level tone huyen without breathiness would also be confused with ngang, another level tone. Nang2 without breathiness is not Vietnamese (there is no such tone). The phonation types [creaky] and [breathy] motivate the laryngeal features for tonal register and are present in all marked tones. The Phonetics of Tone (28) in Chapter Two shows that the two level tones ngang and huyen are characterized with one point, which indicates that the tones do not change direction thoughout their duration. The height of the tones will prove to be predictable from laryngeal features. The rising and falling tones (sac1, nang1, sac2, and nang2) are specified with two points, which indicate that these tones change direction, i.e., up or down or falling-rising. Sac2 and nang2 have the feature [obstruent], which indicates that these tones occur only in obstruent-final syllables (see Chapter Six for additional discussion). The curve in hoi and nga is also important. It is characterized by three points, which indicate that these tones change direction twice, i.e., they go down and then rise. Chapter Two proposed that [laryngeal] is a register feature that is underlyingly present in all non-modal voice tones. This term refers to modes of vibration of the glottis in which the vocal folds are further apart or are drawn further apart than in

76

VIETNAMESE TONE: A NEW ANALYSIS

Figure 50. Nga from Hung

Figure 51. Nga and sac1 from Hung

modal voice: [constricted glottis] refers to sounds that are produced with the vocal folds drawn together, preventing normal vocal fold vibration, and [spread glottis] refers to sounds that are produced with the vocal folds drawn apart, producing a nonperiodic (noise) component in the acoustic signal (Halle and Clements 1983). Creakiness/glottal stop and breathiness are always found to some degree in marked tones. [constricted] is a distinctive feature for a tone that is produced with creakiness or glottal stop, and [spread] is a feature for a tone that is produced with breathiness. The phonological features of register, therefore, are very close to the physiological features. Both [constricted] and [spread] are grouped as features that further specify [laryngeal]. (73) provides the phonetic representations of all eight Vietnamese tones. A comparion of (28) with (73) shows that on the contour side all points are realized phonetically with the value ‘h’ or ‘l’ present in the phonetic representations. On the register side the unmarked tone hoi is realized with the [laryngeal] feature and [spread] as a dependent. In the phonetic representations in (73) all contour features are realized as level, rising, falling, or curved. In Vietnamese phonation types, i.e., creakiness and breathiness, lower a tone. Tones with non-modal voice have at least a portion realized as ‘l’ by default. (74) states the two rules that are required. (74) (a) A non-modal voice tone is realized as ‘l’ for Contour on the final point. (b) A modal voice tone is implemented with the feature ‘h’ on the final point for Contour by default. The two level tones ngang and huyen have one point specified in (28) for the contour feature. Neither ngang or huyen is underlyingly specified for the value ‘h’ or ‘l’ under the contour node. Ngang with modal voice is realized phonetically as ‘h’ Huyen with the register feature [spread] is realized as ‘l.’ The default rule in (74) applies. Sac1 is a rising tone and is specified in (28) with two points to indicate the change in direction. These two points at the phonological level must be interpreted as having different values and the end point is most crucial in the recognition of a tone. Since sac1 has modal voice, the end point is realized as ‘h’; the onset has the value ‘l’ and generates the contour ‘lh.’ The rising part can come early or late in the

THE PHONETICS AND PHONOLOGY OF TONE

77

Figure 52. Nga and sac1 from Hung

Figure 53. Eight tones for [sa] and [sa:k] from Hung

tone and it does not matter how late the tone rises. If sac1 rises only a little at the end, it is not confused with ngang, a high level tone. The late rising part in sac1 is a typical pattern in Hanoi speakers. Sac2 is a rising tone with modal voice. Its contour, therefore, is ‘lh.’ Sac2 is very short; the tone must rise early before it can end. Nang1 is a falling tone and has creakiness captured by the register feature [constricted]. Because of the register feature the end point is realized as ‘l.’ The first point is realized as ‘h,’ a different value from that of the first point, thus generating the falling contour ‘hl.’ It is also unimportant whether the tone is ‘high’ or ‘low’ before it is pulled down by creakiness or glottal stop. Likewise, nang2 has breathiness, represented by the register feature [spread], which lowers the tone. The end point is realized as ‘l’ from [spread], generating the falling contour ‘hl.’ Without the feature [obstruent] that shortens the tone, nang2 would be identical with huyen in pitch height. The curve is very important in hoi and nga. The tone changes direction twice to make the curve; therefore, it is specified with three points. If a tone is specified for more than one point, the adjacent points have different values. For example, if two points are specified, the contour must be either ‘x y’ or ‘y x,’ as in sac1 or nang1, but not *‘x x’ or *‘y y.’ If three points are specified, the first and last points have a different value from the middle point. Consequently, the contour is ‘hlh’ in hoi and nga because curved tones in Vietnamese are concave, i.e., falling-rising. Nga also has the register feature [constricted], which is motivated by creakiness or glottal stop in the middle of the tone. The degree of creakiness can vary from speaker to speaker but it always occurs to make a curve, even a slight curve, as in Figures 49 and 50. The rising part after creakiness is crucial. In nang1 the tone sometimes rises a little after creakiness; however, this rising part is not crucial in nang1 because nang1 is a falling tone. Therefore, variation is found in nang1: sometimes it ends suddenly after the glottal stop and sometimes it goes further, flattening out or rising a little. Consequently, whether the starting point in nga or nang1 is very high in the speaker’s

78

VIETNAMESE TONE: A NEW ANALYSIS

Figure 54. Hoi in two different tokens from Dung

pitch range or quite low, the tone is equally perceptible if there is creakiness in the middle of nga to make the curve, or in nang1 without making a curve. Like nga, the curved tone hoi is also specified with three points for contour but without an underlying laryngeal feature. The reason for the contour ‘hlh’ is clear: like tones cannot occur adjacent to each other and non-modal voice is present phonologically so ‘hlh’ is predictable. However, this tone has breathiness phonetically. It is the low mid point of this tone that forces it to be breathy. Both creakiness and breathiness are compatible with low F0. Because the contour is curved with three specified points, this contour predicts either creakiness or breathiness in the middle. Since creakiness is phonologically a laryngeal feature in nga, it is a default feature in hoi. Because the curved contour is phonological in hoi, this tone acquires breathiness by default in order to satisfy a requirement from the contour side that the curve requires a laryngeal feature in the middle of the tone. Breathiness, therefore, is a phonological feature in huyen and nang2 but a phonetic feature in hoi. Why is the feature [laryngeal] realized as [spread] rather than as [constricted] in hoi? A spread glottis is less marked than a constricted glottis. In glottalization the vocal folds must make a complete closure for a glottal stop or they must be tense enough to produce creaky voice. In breathy voice the vocal folds are lax to open. Therefore, it requires less effort to produce breathy voice than creaky voice. According to the Gestural Theory of Markedness (Hamilton 1996:8), articulatory gestures that deviate minimally from configurations of least effort are highly valued. Using evidence concerning the frequency of occurrence of unmarked features in various languages, Greenberg 1966 claims that non-glottalized consonants, including aspirated consonants, occur more frequently than their glottalized counterparts, a not unexpected finding on the assumption that glottalized consonants involve more effort than non-glottalized consonants. (75) modifies (74) through the addition of a third rule.

THE PHONETICS AND PHONOLOGY OF TONE

79

(a) First token

(b) Second token Figure 55. Hoi in two different token from VanKhanh

(75) (a) A non-modal voice tone is realized as ‘l’ for Contour on the final point. (b) A modal voice tone is implemented with ‘h’ on the final point for Contour by default. (c) Hoi acquires [spread] for Register by default to fulfil a requirement for the curved contour. As previously noted many Hanoi speakers realize hoi as a falling tone without the curve. In this case it is easily confused with huyen. This is a neutralization of the contour feature and (76) shows the phonetic process. The last point under the contour node, which constitutes the curved shape for the contour in this tone, is delinked, leaving only two points for the contour. It is breathiness under register that lowers the tone to create the falling contour ‘hl.’ Phonetically, hoi is very much like huyen: it shares its breathiness but has a falling contour. In (a), the simplified hoi, the contour is ‘hl’ whereas in huyen it is ‘11.’ However, the contour node in huyen is not underlyingly specified for ‘l’ or ‘h.’ With such lack of specification some variation is possible (Archangeli 1988, Pulleyblank 1986, 1988, and Rice 1996) and huyen can be realized as a level tone or can go down a little at the end as a falling tone. Hoi can be confused with huyen when the rising part is neutralized; consequently, in order to distinguish hoi from huyen, Hanoi speakers increase the breathiness in hoi after it loses the rising part. The result is that hoi is much breathier than huyen. Some dialects have another neutralization pattern for hoi. Vuong and Hoang 1994:103 observe that in parts of Quang Binh province in Central Vietnam nga neutralizes to nang1 and hoi neutralizes to sac1. Hoi neutralizes to sac1 when it loses the falling part and only the rising part remains: the result is sac1 with the rising contour shown in (77). This pattern is similar to one found in the Mai Ban dialect discussed in Chapter Two. (77) shows this last neutralization process at the phonological level.

80

VIETNAMESE TONE: A NEW ANALYSIS

Figure 56. Breathiness in hoi from Van

Figure 57. Hoi from Van

There are still other neutralization phenomena in other dialects of Vietnamese (see also Chapter Two). For example, in Southern dialects, the tones hoi and nga are merged phonologically and realized as hoi. What is the phonetic process? The analysis proposed here suggests that nga is structurally more complex than hoi and that it is phonation rather than pitch height that is distinctive in Vietnamese tones. Greenberg 1966:16–34 observes that when the contrast between two segments is neutralized in a particular context, the neutral segment that occurs is often phonetically closest to the unmarked member of the set. Neutralization favors unmarked features. Consequently, in the neutralization of nga, the distinctive feature in nga is lost and the result is a tone that is similar to hoi, the unmarked member of the set. Delinking of register in nga results in hoi. What is particularly interesting here is the phonetic contour of the tone that emerges. It is phonetically half hoi and half nga. When hoi and nga merge in Southern dialects, one often disappears and the resulting tone has a very interesting contour feature. Hoi and nga share the same contour feature: they are both curved. The register feature [constricted] in nga is delinked and the result is a new tone, here labeled ‘hoi’ With its curved shape, the new tone ‘hoi’ sounds like the Northern hoi and most speakers of Northern and Southern dialects recognize it as hoi. In order to compensate for the lost register feature hoi rises quite high and clearly occupies the full range of pitch, i.e., between minimal and maximal F0 values, and with this shape it cuts across both the lower and higher parts of the pitch range. This variation in the resulting hoi in Southern dialects shows that the contour feature of hoi can vary to produce a very high ending as a trade-off for the marked laryngeal feature lost in nga.

THE PHONETICS AND PHONOLOGY OF TONE

81

Figure 58. Hoi from Son

Figure 59. Huyen and hoi from Son

Figure 60. Huyen and hoi from Son

It is now clear how the phonological features of tone are motivated, realized, and implemented. The contour feature is realized as ‘l’ if the tone has a marked register feature, as in huyen, nang1, nang2, and nga. The phonation types of creakiness and breathiness in these tones generate a low contour in level and falling tones or in the middle of the curved tone nga. Otherwise, the contour is realized as ‘h’ in tones with modal voice such as ngang, sac1, and sac2. Hoi is special in that, as an unmarked register tone, it acquires breathiness by default in order to satisfy the contour requirement. Thus, the realization of tone is generated from the configuration of both tonal shape (contour) and phonation types (register).

82

VIETNAMESE TONE: A NEW ANALYSIS

(73)

Summary The phonation types of creakiness and breathiness can be used reliably in Vietnamese to group tones into natural classes. Instead of pitch height, the laryngeal features of creakiness and breathiness are distinctive as the register feature in Vietnamese tones. Tonal height is derived from the configuration of contour and register, i.e., tonal shape and phonation types. Using phonation types distinctively not only explains the various tonal patterns in the language but also shows the mismatch problem to be illusory. The relationship between phonetics and phonology in Vietnamese is very close.

THE PHONETICS AND PHONOLOGY OF TONE

(76)

(77)

83

CHAPTER 6 The Domain of Tone

A further issue is the domain of Vietnamese tone. Does the evidence point to the nucleus, the syllable, or the rhyme? Both phonological and phonetic evidence point in the same direction: the rhyme. The Vietnamese Syllable The Vietnamese syllable has four components: tone, initial consonant, vowel, and final consonant; while the initial or final consonant can be absent the vowel and tone are necessary. There is evidence from reduplication, poetry, and language games that tone is independent of segments. Tone patterns independently of segmental features (see Chapter Two). For example, in total reduplication whole syllables are retained but tones change. In partial reduplication either the initial consonant or rhyme is retained but the tone changes. In poetry syllables in the rhymal position (where only words having the same rhyme can occur) keep a constant rhyme but the initial consonant and tone are changeable. Finally, many researchers, e.g., Doan 1977 and Vuong and Hoang 1994, also identify a language game in Vietnamese in which onsets, tones, and rhymes can be switched around within a morphologically complex word. (78) illustrates this game using words with non-stop finals. (78) lo (nang1) tuong (ngang) ‘a bottle of sauce’ > > > >

> (b) (c) (d) (e)

(a) luong (nang 1) to (ngang) luong (ngang) to (nangl) to (nang1) luong (ngang) to (ngang) luong (nang1)

tuong (nang 1) lo (ngang) ‘great merit’

‘a worried statue’

‘great merit’

In the original syllables ‘lo tuong’ (‘a bottle of sauce’) in (78) the first syllable bears the tone nang1 and the second syllable bears the tone ngang. This phrase can occur in five new and different ways. (78) shows all possible combinations although (c) is nonsensical. In (a) the two original syllables ‘lo’ (nang1) and ‘tuong’ (ngang) exchange their initials and rhymes but retain their tones: ‘lo’ with the original nang1 becomes ‘tuong’ and keeps nang1; ‘tuong’ with the original ngang becomes ‘lo’ and keep ngang. In (b) the two syllables exchange non-tonal rhymes only, e.g., ‘lo’ keeps the initial ‘l’ and nang1 but has the rhyme ‘uong.’ In (c) the syllables exchange both tones and rhymes, i.e., ‘lo’ keeps the initial ‘1’ but has ngang and the rhyme ‘uong.’ The vowel can switch independently of tone, as in (a) and (b), and vice versa, as in (a), (b), and (e). The fact that the initial, rhyme, and tone can be detached freely from the original syllable provides an argument for the independence of these elements in the Vietnamese syllable and for tone being independent of segments (Doan 1977). The independence of tone is also recognized in autosegmental theory, which places tone on an independent tier, e.g., Leben 1973, Williams 1976, and Goldsmith 1976. Some researchers, e.g., Le 1948 and Hoang et al. 1962, do not consider tone to be a constituent of syllable structure because tone is not a segment and cannot therefore be treated as a phoneme. Other researchers, e.g., Mkhitarian 1959, cited in Doan 1977, emphasize the function of each component in the syllable. However, all researchers treat tone on a different level from the level of segments. For example, Doan 1977:88 proposes that the Vietnamese syllable has the structure in (79) in which tone is on a separate level. Others adopt a similar structure. The fact that tone can be separated from segments in several patterns in the language shows its independence from segments.

THE DOMAIN OF TONE

85

(79) Tone Initial

Rhyme

/w/

Vowel

Final

The initial also forms one constituent and the rhyme forms another. In the language game just described, the initials of the two syllables can be switched as in (c) and (d) so the onset is independent of the rhyme. In reduplication, there are several patterns in which either the initial or the rhyme is reduplicated. (80) provides the data with the bases underlined. (80)

In (a) the initial is reduplicated and in (b) the rhyme is reduplicated in a process which is no longer productive. The rhyme is unpredictable. There are other processes of reduplication in which only the initial remains. For example, (81) shows a sitution in which the reduplicant is a combination of the initial of the base and the rhyme [i*****k]. The meaning is derogatory (Doan 1977 and Ngo 1984). (81) (82)

(81) is a very productive process. These examples show that the initial can be separated very easily from the rhyme. There is also a reduplicative process in which the initial, final, and tone reduplicate but not the vowel, e.g., [lak] sac2 ‘to shake’ > [luk] [lak] sac2 - sac2 ‘to wag.’ Although this process is not productive, it demonstrates the independence of the nucleus. The previously cited rhymal position in poetry also supports the claim that the relationship between the vowel and the final consonant is much closer than that between the initial consonant and the rhyme. The traditional literature reveals two views concerning the internal structure of the syllable. In the first (Le 1948 and Hoang et al. 1962) the initial, the vowel, and the final have an equal and independent status in the syllable, represented by the plus signs: C1+ V+C2. In the second, a syllable is represented as a unit with constituents. Gordina’s experimental work in phonetics led her to postulate a close relationship between the vowel and the final consonant. Phonetically, the rhyme in a Vietnamese syllable occupies a constant amount of time regardless of segmental content. By recognizing the relative independence of the initial consonant from the rhyme and the close relationship between the vowel and final consonant, other researchers, e.g., Doan 1977, Vuong and Hoang 1994, and Cu et al. 1977, divide a Vietnamese syllable into two major components: the initial and the rhyme. Vuong and Hoang 1994: 78 propose the structure shown in (82). (82) Syllable Initial Tone

Rhyme /w/

vowel

final

In (82) the syllable has three major components: tone, initial, and rhyme. The vowel and final consonant are dependents of the rhyme. There are also phonological processes that take place in the nucleus but not in the whole rhyme, further evidence that the rhyme and nucleus are two different domains in Vietnamese. If the nucleus and final consonant form a constituent, then the nucleus is a constituent of the rhyme. Components of the nucleus have a special relationship to each other: in all dialects of Vietnamese the realization of final velar consonants depends on the quantity and quality of the vowel. Furthermore, only a limited number of consonants can occur in final position. (83) shows the final consonant inventory of the Hanoi dialect. The glides /j/ and /w/ are omitted since they behave differently from the final consonants (see Pham 1997). (a) shows the phonological inventory (see Pham 1998) and (b) the phonetic inventory.

86

VIETNAMESE TONE: A NEW ANALYSIS

(83)

(84)

In Vietnamese the quantity and quality of the vowel determine the surface form of a final velar consonant. For example, in the Northern dialects if the vowel is short, the final velar consonant is realized as a labio-velar consonant after a round vowel and as a palatal consonant after a front vowel. It is realized as a plain velar consonant after a central vowel, long vowel, and diphthong (Pham 1998). (84) shows the surface distribution of the final velar consonants. Here a nasal is used but the consonant can also be an obstruent. In (84) V stands for a short vowel and V: stands for a long vowel. Only occurs after a short front vowel. Only [′ m] occurs after a short back vowel. Only [′ ] occurs a after central vowel, a long vowel, or a diphthong. The realization of a final velar consonant depends on vowel length and which feature it receives from the vowel. The consonant shares the feature of the preceding front or back vowel if the vowel is short. The central vowel has no feature to share (see Pham 1998). If the vowel is long, there is no feature sharing between the vowel and final consonant. If a short vowel contributes one mora and a long vowel or a diphthong contributes two moras and if the nucleus is minimally and maximally bimoraic, the consonant is affected by the vowel only if it is moraic or is within the nucleus domain, i.e., when the vowel is short. The consonant is not affected by the vowel if it is non-moraic or is outside the nucleus domain, i.e., when the vowel is long or is a diphthong. Only nuclear consonants are subject to feature sharing with the vowel. (85) is an example with F being some feature. In (a) the vowel is short and the final consonant is inside the nucleus so it receives a place feature from the vowel. In (b) the vowel is long and the final consonant is outside the nucleus domain. It cannot therefore receive a place feature from the vowel (see Pham 1998). If the vowel in the rhyme is short, it is centralized. For example, in the Saigon dialect a short front vowel surface as a central vowel of the same height before [n] or [t], i.e., /i/ surfaces as /e/ surfaces as [ ]; and /′ / surfaces as [a] (see Pham 1998). The relationship between a short vowel and a following consonant is very close. If VC forms a nucleus in the Vietnamese rhyme, then the fact that feature sharing and centralization of vowels do not occur if the vowel is long shows that the nucleus and the rhyme are not the same domain for certain phonological processes. The nucleus obligatorily contains two moras with the second of these filled by either a vowel or a consonant. Additional material is found in the rhyme. The processes illustrated here have the nucleus as their domain. The Significance of Rhyme The literature on tone contains a variety of views on what the tone-bearing unit is. Goldsmith 1976 suggests it is the vowel but Clements and Ford 1979 argue for the syllable as the unit. Odden 1995 discusses various views. Bao 1999b, using Chinese dialects, argues that the rhyme is the tone-bearing unit. There are two views concerning the tone-bearing unit in Vietnamese. The first (Cu et al. 1972, 1977) makes tone a component that associates with the rhyme, as in (86). This view is based primarily on the acoustic work of Andreev and Gordina 1957, which shows that tone is realized only on the rhyme and

THE DOMAIN OF TONE

87

(85)

that the initial does not contribute to the value of tones. The second view (Doan 1977) places tone over the whole syllable, as seen in (79). Huu and Vuong 1980 share this view but note that the typical contour of each tone lies on the rhyme (1980:66). (86) Syllable Tone Rhyme

Final

Initial

Vowel

Since tones do not start until late in the vowel, it appears that tone has no effect on the onset. For example, tones should affect the laryngeal features of onsets but they do not: any onset can occur with any tone. Bao 1999b: 10 shows that tones do not affect onsets in some Chinese dialects, and Vietnamese behaves likewise. The domain of tone is not the whole syllable. Evidence from vowel quantity and tonal distribution shows that the domain cannot be the mora either. All tones can occur with both short and long vowels, as (87) shows. (87)

sa:w da:′

fa:j ha:j

long vowel (ngang) (ngang) (sac1) (sac2) (nang2) (hoi) (nga)

‘star’ ‘paint’ ‘deserve’ ‘drop’ ‘light (color)’ ‘correct, right’ ‘be scared’

saw da′

faj haj

short vowel (ngang) (ngang) (sac1) (sac2) (nang2) (hoi) (nga)

‘behind’ ‘backyard, frontyard’ ‘bitter’ ‘very’ ‘turn upside down’ ‘comma’ ‘imperative’

Syllables in the first column have long vowels and form minimal pairs with those in the same row in the second column. These forms differ from their long vowel counterparts only in that their vowels are short. A moraic analysis for the placement of tone in which the tone occupies only a single mora cannot account for the fact that tones take either the long or the short vowel as their domain.

88

VIETNAMESE TONE: A NEW ANALYSIS

Acoustic evidence also shows that the nucleus is not the domain of tone. This study shows that tone is not always located at a particular point within the vowel, e.g., onset, middle, or end point, but rather that it is distributed over the entire rhyme. Moreover, in sonorant-final syllables the location of the distinctive feature of a tone within the rhyme depends on vowel length. Some researchers, e.g., Doan 1977 and Vuong and Hoang 1994, describe tone as distributed over the rhyme with the distribution of tonal features varying according to rhyme type. For example, Doan 1977:111 describes the distribution of nga in non-obstruent-final syllables as follows: the tone goes down suddenly and rises up or it is broken in the middle. In an open syllable the lowest or broken portion usually falls at the beginning of the second part of the rhyme. In a syllable with a final nasal, this portion can fall on the final nasal, especially if the vowel is short. Chapter Four showed that creakiness or glottal stop in the middle of the curved tone nga is a distinctive feature of this tone and that it can be located anywhere in the tone except for the beginning and ending so long as the tone is curved. Doan notes that in a sonorant-final syllable the broken part falls on the final consonant and claims that tone is spread over the whole rhyme, which contains at least a sonorant-final syllable, and not just over the vowel. Doan also emphasizes that when the vowel is short, the broken portion falls on a final nasal. Tone is therefore sensitive to syllable length and adjusts the location of its features as it is realized over the rhyme. Doan also gives a similar distribution for hoi over the rhyme, i.e., the lowest part of this tone falls on the middle of the rhyme and on the final sonorant if it is a closed syllable. The evidence in Chapter Four supports this view. It shows that nga is a curved tone with either a glottal stop or creakiness in the lowest part of the tone. The part with glottal stop or creakiness is usually in the middle of the vowel; however, it can be distributed anywhere from right after the beginning of the vowel to just before the end of a final sonorant. Figure 61 shows the tone nga in an open syllable. Because the syllable ends with a vowel, the broken portion (a wide gap in the spectrogram) is in the middle of the vowel, from approximately 100 to 120ms. Figure 62 shows nga in a nasal-final syllable with both a long vowel [ta:m] and a short vowel [tam]. The arrows show the approximate beginning and end points. When the syllable ends with a sonorant consonant, the nasal part shows very little voicing in the spectrogram. The broken part is not in the middle portion of the vowel since it is in an open syllable; it is in the middle of the rhyme, which is the last part of the vowel and the beginning part of the final nasal consonant. The broken part (or creakiness) is from approximately 100 to 150ms into the syllable with a long vowel (the first and third rows). In the

THE DOMAIN OF TONE

89

Figure 61. Nga in an open syllable from Son

Figure 62. Nga in nasal-final syllables with a long and short vowel from Son

syllable with a short vowel (the second and fourth rows), this part, as the gap in the spectrogram shows, is much longer and comes much earlier, i.e., from where the vowel ends at about 50ms to the beginning of the nasal at 150ms. In the open syllable (Figure 61) the laryngeal feature in nga ends before the end of the vowel. In nasal-final syllables (Figure 62) the laryngeal feature in nga continues to the end of the vowel. The spectrogram in Figure 62 does not clearly show whether the creakiness in nga continues into the final nasal; however, this is not crucial because a glottal stop or creakiness occurs in the middle of the tone. The important point is that when the vowel is long in nasal-final syllables, the laryngeal feature does not occur in the middle of the vowel. The fact that this tonal feature is not always in the middle of the vowel but moves toward the final consonant supports the claim that tone is not a feature of the vowel alone but is realized over the entire rhyme. What happens if the distinctive feature of the tone is at the end of a vowel? In this case, a similar pattern emerges: regardless of rhyme type, a tone always distributes its features over the whole rhyme. With sac1, a rising tone, the rising part is distinctive and this makes it different from nang1, a falling tone that goes down because of the creakiness. The rising part of sac1 is usually in the second half of the rhyme. In some speakers, however, it comes very late, just before the rhyme ends. According to Doan, the rising part in nasal-final syllables occupies the second half of the rhyme, i.e., the final sonorant. Figure 63 shows the pitch graph of the tone sac1 from An, with and without the final nasal consonant. In both the open syllable [ta] and the closed syllable [ta:′ ] the tone has the same fundamental frequency through the whole rhyme: the flat portions lie on top of each other, and, strikingly, in both syllables the tone starts to rise at approximately 150 ms. The corresponding spectrograms in Figure 64 show that the final nasal starts at about after 150ms. In [ta] the rising part is in the second part of the vowel, but in [ta: ′ ] it does not occur until the beginning of the nasal. Tone sac1 in the two syllables [ta: ′ ] and [ta′ ] in Figure 65 has the rising part located in the same place in both syllables, i.e., on the final consonant (‘tang’ stands for [ta:′ ] and ‘ta(ngg’ for [ta′ ]). In this pitch graph the rising part in both syllables starts after 150ms. Figure 66 shows the corresponding waveforms of the tone sac1 in these syllables. In the syllable with the long vowel, i.e., the third row of the waveform, the rising part starts just before the end of the vowel at about 150ms. In the syllable with the short vowel (the last row) the vowel ends at approximately 90ms. The rising part does not start from the end of the vowel but from the second half of the final consonant at about 150ms. It is therefore not important which component in the rhyme the tone links to because the tonal features are distributed flexibly over the whole rhyme. Consequently, it is possible to predict that if the final consonant were omitted in sonorant-final syllables, the distinctive feature would not be clearly

90

VIETNAMESE TONE: A NEW ANALYSIS

Figure 63. Sac1 in [ta] and [ta:′ ] from An

Figure 64. [ta] and [ta:′ ] from An

present and the tone would be unrecognizable. Long vowels show that the tone is on the final consonant, further evidence that the domain of tone goes beyond the nucleus. Obstruent-final syllables show still another distribution of tonal features over the rhyme. Only sac2 and nang2 can occur in this type of syllable. Sac2 also has modal voice, the distinctive feature in sac2 being the rising contour, which is identified though the rate of vibration only. Nang2 has some breathiness, which is not very clear in the spectrogram because of the shortness of the tone. Figure 67 shows the pitch graphs of sac2 with [ta:k] and [tak]. The syllable with the short vowel is very short and the tone rises immediately and finishes at approximately 70ms. The syllable with the long vowel is longer and rises after the first 60ms. Because the tone is so short and there is no voicing after the oral closure for place of articulation, i.e., -p, t, -c, -k, -kp, the distinctive feature must be realized before that point. Figure 68 shows the corresponding spectrogram. Although the rising part is not seen in the spectrogram, it is still apparent that the tone has modal voice (regular cycles) and that the short vowel makes the rhyme shorter. The same situation is found with nang2. Figure 69 shows nang2 in [ta:k] and [tak] from An. The tone is shorter in the syllable with the short vowel. Tonal features are realized over the whole rhyme not just on the vowel alone. This finding is especially clear for non-stopfinal syllables ending in V:C, the test case required to differentiate between nucleus and rhyme. In stop-final syllables the tone is too short for its features to adjust according to vowel length. However, it still shares a tonal feature with the final consonant, namely the feature [obstruent]. Rhyme is said to have a relatively constant length in Vietnamese regardless of its composition. However, Figures 68 and 69 show that such equal length is not found with stop-final consonants. Short vowels occur only in closed syllables. An acoustic study carried out by Gordina and Bystrov 1970 (cited in Hoang Thi Chau 1989:153), claims that Vietnamese rhyme

THE DOMAIN OF TONE

91

Figure 65. Sac1 in [ta:′ ] and [ta′] from An

Figure 66. [ta:′ ] and [ta′] from An

has inherent length. As (88) shows, if the vowel is long, the final consonant is short, and if the vowel is short, the final consonant is long. (88) Length a: a

m m

In the traditional literature such inherent length is used as an argument for postulating an intimate relationship between a vowel and a final consonant. This study also shows that syllables with short and long vowels and the same final consonant are usually equally long. Figures 70 and 71 show ngang and huyen in syllables with long [ta:′ ] and short [ta′] vowels from Phuong. The waveforms of the two syllables with ngang and huyen are shown in the first two rows. The corresponding spectrograms of the two syllables are in the last two rows. In Figure 70 the final nasal part has very little voicing and weak energy in the spectrograms. The long vowel ends at about 190ms; the short vowel ends much earlier, at approximately 70ms. .However, the syllable with a long vowel ends at approximately 270ms, only about 20ms longer than the syllable with a short vowel, which ends at about 250ms. In Figure 71, the long vowel ends at approximately 160ms while the short vowel ends at about 90ms. However, both syllables end at almost the same time, at approximately 260ms. Thus, the final nasal part in these figures, shown to have very little voicing, is short if the vowel is long and very long if the vowel is short. The final sonorant consonant lengthens to

92

VIETNAMESE TONE: A NEW ANALYSIS

Figure 67. Sac2 in [ta:k] and [tak] from An

Figure 68. Sac2 in [ta:k] and [tak] from An

compensate for the shortness of the short vowel. This phenomenon was also observed for Son and An in Figures 62 and 66. In those figures the syllable with a short vowel had a very long final nasal and its length was almost equal to that of the rhyme with a long vowel. A final nasal is long if the vowel is short. Sometimes syllables with short vowels are shorter than syllables with long vowels but such cases are rare. With stop-final syllables, a short vowel makes the rhyme shorter than does a long vowel (Figures 68 and 69). The final voiceless stop is just silence without any voicing; therefore, the syllable length relies on the vowel. Sonorant-final syllables also tend to be equal in length. The preference for a long final sonorant in the syllable if the vowel is short shows that members of the rhyme do ‘observe’ each other and adjust themselves to preserve a certain fixed length in different syllable types. Tonal features are realized over the whole rhyme not just on the vowel, particularly in syllables with a sonorant-final consonant. Moreover, Vietnamese syllables tend to be of a constant length regardless of vowel length. Such facts support the claim that the nucleus is not the domain of tone. They suggest instead that rhyme is the domain because of the way tonal features are distributed in the rhyme. However, there is also evidence from phonological processes such as feature sharing and neutralization that these occur in the nucleus only and not in the rhyme as a whole.

THE DOMAIN OF TONE

Figure 69. Nang2 in [ta:k] and [tak] from An

Figure 70. Length in [ta:′ ] and [ta′] with ngang from Phuong

93

94

VIETNAMESE TONE: A NEW ANALYSIS

Figure 71. Length in [ta:′ ] and [ta′] with huyen from Phuong

CHAPTER 7 Conclusion

The research on Vietnamese tones reported here proposes a particular phonological model of tonal representation which assumes that markedness is reflected structurally. This model was able to account for the phonological patterning of Vietnamese tones and predicts the neutralizations that occur. Phonetic evidence also shows that the well-established view of Vietnamese as a pitch register language must be abandoned; what has been called pitch is really the laryngeal features of breathiness and creakiness. An analysis of the phonetic patterning of the tones allows a simplification of the tonal phonology found in reduplication in Vietnamese; the ad hoc and unnatural flip flop rule that was required in previous models is no longer necessary. The phonological claims made in the model also prove to be well-grounded in the phonetics of Vietnamese. Vietnamese tones are organized in a hierarchical structure, the laryngeal features of phonation are distinctive, pitch height is not distinctive, and the features are grounded phonetically. Particularly to be noted is that replacing pitch height with phonation types and tonal shape provides a natural and elegant account of reduplication, the most widely discussed evidence for the patterning of tones in the traditional literature. When the assumption that tone is solely pitch height is abandoned, there also emerges a new way to examine tonal languages. It may even be the case that if ‘tone’ equates to ‘pitch,’ Vietnamese — perhaps along with many other ‘tone languages’—is not a ‘tone language’ after all!

Appendix

I. Pitch graphs of other speakers

1. Eight tones in [ka] and [ka:k] from Phuong (female)

2. Eight tones in [ka] and [ka:k] from Khanh (female)

97

II. Spectrographs in [ka] and [ka:k] from other speakers

3. Eight tones in [ka] and [ka:k] from Dung (female)

4. Eight tones in [ka] and [ka:k] from Van (female)

98

5. Eight tones in [ka] and [ka:k] from An (female)

6. Eight tones in [ka] and [ka:k] from Hung (male)

99

7. Eight tones in [ka] and [ka:k] from Hoang (male)

8. Eight tones in [ka] and [ka:k] from Son (male)

100

101

1. Eight tones from Phuong

102

103

2. Eight tones from Van Khanh

104

105

3. Eight tones from Dung

106

107

4. Eight tones from Van

108

109

5. Eight tones from An

110

111

6. Eight tones from Hung

112

113

7. Eight tones from Hoang

114

115

8. Eight tones from Binh

Bibliography

Alves, Mark. 1997. Tonal features and the development of Vietnamese tones. Ms. University of Hawaii. Anderson, Stephen R. 1978. Tone features. In Victoria A. Fromkin (ed.), Tone: A Linguistic Survey. New York: Academic Press, 133–73. Andreev, N.D. and M.V. Gordina. 1957. Sistema tonov vjetnamskogo jazyka. Leningrad: Vestnik Leningraskogo gosudarstvennogo Universiteta: Serija Istorii, Jazyka i Literatury, 8:132–48. Archangeli, D. 1988. Aspects of underspecification theory. Phonology, 6.2:179–200. Avery, Peter. 1983. Evidence for a laryngeal tier in Vietnamese. Toronto Working Papers in Linguistics, 4:1–26. Avery, Peter and Keren Rice. 1989. Segment structure and coronal underspecification. Phonology, 6.2:179–200. Bao, Zhi-ming. 1999a. Tonal contour and register harmony in Chaozhou. Linguistic Inquiry, 30.3:485–93. —— 1999b. The Structure of Tone. New York: Oxford University Press. Benedict, Paul K. 1972. Sino-Tibetan: A Conspectus. Cambridge: Cambridge University Press. Brockhaus, W. 1995. Final Devoicing in the Phonology of German. Ph.D. dissertation. Tübingen: Niemeyer. Burton, Strang. 1992. Reduplication and the representation of tone in Vietnamese. Ms. Brandeis University. Cao Xuan Hao. 1998. Hai cach mieu ta he thong thanh dieu tieng Viet (Two ways to describe the tonal system in Vietnamese). In Tieng Viet —May Van De Ngu Am, Ngu Phap, Ngu Nghia (Some Issues in Vietnamese Phonology, Syntax, Semantics). Hochiminh City: Nha xuat ban Giao duc. Chao, Yuen Ren. 1930. A system of tone letters. Le Maître Phonetique, 45:24–47. Cho, Y.Y. 1991. Universality of the coronal articulator. In C.Paradis and J.F.Prunet (eds), The Special Status of Coronals: Internal and External Evidence. San Diego: Academic Press, 159–80. Clements, G.N. 1981. The hierarchical representation of tone features. In I.R.Dihoff (ed.), Current Approaches to African Linguistics, Vol. 1. Dordrecht: Foris, 145–76. —— 1985. The geometry of phonological features. Phonology Yearbook, 2:223–50. —— 1990a. The role of the sonority cycle in core syllabification. In J. Kingston and M. Beckman (eds.), Papers in Laboratory Phonology 1: Between the Grammar and Physics of Speech. New York: Cambridge University Press, 283–333. —— 1990b. Place of articulation in consonants and vowels: a unified approach. NELS 21, U.Q.A.M., Montreal, 1990, 77–123. Clements, G.N. and K.C. Ford. 1979. Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry, 10:179–210. Clements, G.N. and Elizabeth Hume. 1995. The internal organization of speech sounds. In John Goldsmith (ed.), The Handbook of Phonological Theory. Oxford: Blackwell, 245–306. Cu Dinh Tu, Hoang Van Thung, and Nguyen Nguyen Tru. 1972. Giao Trinh Tieng Viet Hien Dai (Lectures on Modern Vietnamese). Hanoi: Giao duc Publisher. —— 1977. Ngu Am Hoc Tieng Viet Hien Dai (Phonology of Modern Vietnamese). Hanoi: Publisher of Education. Dilley, L, S. Shattuck-Hufnagel, and M. Ostendorf. 1996. Glottalization of word-initial vowels as a function of prosodic structure. Journal of Phonetics, 24:423–44. Doan Thien Thuat. 1977. Ngu Am Tieng Viet (Vietnamese Phonology). Hanoi. Dresher, B.Elan and Harry van der Hulst. 1998. Head-dependent asymmetries in phonology: complexity and visibility. Phonology, 15: 317–52. Duanmu, San. 1990. A Formal Study of Syllable, Tone, Stress and Domain in Chinese Languages. Ph.D. dissertation. MIT. Ferguson, Charles A. 1963. Assumptions about nasals: a sample study in phonological universals. In Joseph H. Greenberg (ed,), Universals of Language. Cambridge, Mass: MIT Press, 53–60. Goldsmith, John. 1976. Autosegmental Phonology. Ph.D. dissertation. MIT. Also New York: Garland Press, 1979. Gordina, M.V. and I.S.Bystrov. 1970. Mot so tieu chuan am vi hoc de phan vung cac phuong ngu tieng Viet (Some phonological criteria in classification of Vietnamese dialects). In Languages in Southeast Asia. Moscow, 203–11. Gordon, Matthew and P.Ladefoged. 2001. Phonation types: a cross-linguistic overview. Journal of Phonetics, 29. 4:383–406. Greenberg, Joseph H. 1966. Language Universals. The Hague: Mouton. Halle, Morris and G.N.Clements. 1983. Problem Book in Phonology. Cambridge: MIT Press. Hamilton, Philip. 1996. Constraints and Markedness in the Phonotactics of Australian Aboriginal Languages. Ph.D dissertation. University of Toronto. Han, Mieko. 1969. Studies in The Phonology of Asian Languages VIII: Vietnamese Tones. California: University of Southern California Press. Hashimoto, Mantaro J. 1985. The interaction of segments and tones in the Be language. In Graham Thurgood, James A.Matisoff, and David Bradley (eds), Linguistics of the Sino-Tibetan Area: The State of the Art. The Australian National University Press, 90–3. the Sino-Tibetan Area: The State of the Art. The Australian National University Press, 90–3. Haudricourt, André G. 1954. Sur l’ origine du ton de Vietnamien. Journal Asiatique, 242: 69–82.

BIBLIOGRAPHY

117

—— 1972. Two-way and three-way splitting of tonal systems in some Far Eastern languages. In Jimmy G. Harris and Richard B. Noss (eds), Tai Phonetics and Phonology. Bangkok: Central Institute of English. Hoang Cao Cuong. 1986. Suy Nghi Them ve Thanh Dieu Tieng Viet (More thoughts about Vietnamese tones). Ngon ngu, 3:19–38. —— 1989. Thanh Dieu Viet qua Giong Dia Phuong Tren Cu Lieu F0 (The F0 values of Vietnamese tones in dialects). Ngon ngu, 4:1–17. Hoang Dung. 1998. Ban them ve van de nhan dien tu lay tieng Viet (On the classification of Vietnamese reduplication). Ms. Hanoi University. Hoang Thi Chau. 1989. Tieng Viet Tren Cac Mien Dat Nuoc (Vietnamese Dialectology). Hanoi: Khoa Hoc Xa Hoi Press. Hoang Tue, Le Can, and Cu Dinh Tu 1962. Giao Trinh Ve Viet Ngu (Lectures about Vietnamese Language). Hanoi: Giao duc Publisher. Hockett, Charles F. 1955. A Manual of Phonology (Part I, International Journal of American Linguistics, 21:4. Indiana University Publications in Anthropology and Linguistics, Memoir 11). Baltimore: Waverley Press. Huu Quynh and Vuong Loc. 1980. Khai Quat ve Lich Su Tieng Viet va Ngu Am Tieng Viet Hien Dai (Introduction to the History and Phonology of Modern Vietnamese}. Hanoi: Giao duc Publisher. 1980. Hyman, Larry, Francis Katamba, and L. Walusimbi. 1987. Luganda and the strict layer hypothesis. Phonology, 4:87–108. Jessen, M. 1998. Phonetics and Phonology of Tense and Lax Obstruents in German. Studies in Functional and Structural Linguistics, 44. Amsterdam/Philadelphia: John Benjamins Publishing Company. Kaye, Jonathan, Jean Lowenstamm, and Jean-Roger Vernaugd. 1985. The internal structure of phonological elements: a theory of charm and government. Phonology Yearbook, 2. 305–28. Keller, Eric. 1994. Signalyze: Signal Analysis for Speech and Sound, User’s Manual. Kirk, P.L., P.Ladefoged, and J.Ladefoged. 1984. Using a spectrograph for measures of phonation types in a natural language. UCLA Working Papers in Phonetics, 59:102– 13. Ladefoged, P. 1988. Discussion of phonetics: a note on some terms for phonation types. In O.Fujimura (ed.), Vocal Physiology: Voice Production, Mechanisms and Functions. New York: Raven Press, 373–6. Ladefoged P. and I.Maddieson. 1996. The Sounds of the World’s Languages. Oxford: Blackwell. Laver, John. 1994. Principles of Phonetics. Cambridge: Cambridge University Press. Le Kha Ke. 1988. Tu Dien Phap-Viet (French-Vietnamese Dictionary). Hanoi: NXB Khoa hoc xa hoi. Le Van Ly. 1948. Le Parler Vietnamien. Paris: Huong Anh. Leben, Will. 1973. Suprasegmental Phonology. Ph.D. dissertation. MIT. McCarthy, John. 1988. Feature geometry and dependency: a review. Phonetica, 43:84– 108. Maddieson, I. 1984. Patterns of Sounds. Cambridge: Cambridge University Press. Marasek, K. 1997. Automatic classification of voice quality using the EGG signal. Universitat Stuttgart. Mkhitarian, T.T. 1959. Fonetika Vjetnamskogo jazyka. Moskva: Izdatelstvo Vostochnoj Literatury. Mester, R. Armin and J. Itô. 1989. Feature predictability and feature specification. Language, 65:258–93. Ngo Thanh Nhan. 1984. The Syllabeme and Pattern of Word Formation in Vietnamese. Ph.D. dissertation. New York University. Nguyen, Nhu Y. 1998. Dai Tu Dien Tieng Viet (Vietnamese Dictionary). Hanoi: NXB Van hoa thong tin. Nguyen Tai Can. 1977. Ngu Phap Tieng Viet: Tieng-Tu Ghep-Doan Ngu (Vietnamese Syntax: Word-Compound-Phrase). Hanoi: Dai hoc va trung hoc chuyen nghiep. Nguyen Van Loi and Jerold Edmondson. 1997. Thanh dieu va chat giong trong tieng Viet hien dai (The tones and voice quality in modern Northern Vietnamese: instrumental case studies). Ngon ngu, 1:1–16. Nguyen Van Tao. 1975. Tan Dai Tu Dien Viet-Anh (New Vietnamese-English Dictionary). Saigon: Tao Dan. Odden, David. 1995. Tone: African languages. In John Goldsmith (ed.), The Handbook of Phonological Theory. Oxford: Blackwell, 444–75. Ohala, John. 1973. The physiology of tones. In Larry Hyman (ed.), Consonant Types and Tone. Southern California Papers in Linguistics, 1:1–14. Paradis, C. and J.-F.Prunet. 1989. On coronal transparency. Phonology, 6:317–48. Pham Hoa. 1997. Clitics: evidence for Vietnamese phonological words. Ms. University of Toronto. —— 1998. The coronal-velar relationship in Vietnamese dialects: a prosodic account. In Paul Watter (ed.), Asia-Pacific Language Reasearch, Vol 1. Australia: Cassowary Computing Publisher. Phan Khoi. 1997. Viet Ngu Nghien Cuu (Study Vietnamese). Danang: Danang Publisher. Pike, Kenneth. 1948. Tone Languages. Ann Arbor: University of Michigan Press. Pulleyblank, D. 1986. Underspecification and low vowel harmony. Studies in African Linguistics, 17:119–53. —— 1988. Vocalic underspecification in Yoruba. Linguistic Inquiry, 19:233–70. Rice, Keren. 1992. On deriving sonority: a structural account of sonority relationships. Phonology, 9:61–99. Pulleyblank, D. 1986. Underspecification and low vowel harmony. Studies in African Linguistics, 17:119–53. —— 1988. Vocalic underspecification in Yoruba. Linguistic Inquiry, 19:233–70. Rice, Keren. 1992. On deriving sonority: a structural account of sonority relationships. Phonology, 9:61–99. —— 1993. A reexamination of the feature [sonorant]: the status of ‘sonorant obstruents.’ Language, 69. 1:308–44. —— 1996. Default variability: the coronal-velar relationship. Natural Language and Linguistic Theory, 14:493–543. Rice, Keren and Peter Avery. 1989. On the interaction between sonorancy and voicing. Toronto Working Papers in Linguistics, 10:65–82. —— 1991. On the relationship between coronality and laterality. In C. Paradis and J.F. Prunet (eds), The Special Status of Coronals: Internal and External Evidence. San Diego: Academic Press, 101–24. —— 1993. Segmental complexity and the structure of inventories. Toronto Working Papers in Linguistics, 12.2:131–53.

118

BIBLIOGRAPHY

Snider, Keith. 1999. The Geometry and Features of Tone. Texas: The Summer Institute of Linguistics. Steriade, Donca. 1987. Redundant values. In A.Bosch, B.Need, and E.Schiller (eds), CLS 23: Parasession on Autosegmental and Metrical Phonology, 339–62. Stevens, Kenneth N. 1988. Mode of vocal fold vibration based on a two-section model. In O.Fujimura (ed.), Vocal Physiology: Voice Production, Mechanisms and Functions. New York: Raven Press, 357–67. Thompson, Laurence C. 1965. A Vietnamese Grammar. Seattle: University of Washington Press. Thongkum, Theraphan. 1988. Phonation types in Mon-Khmer languages. In O.Fujimura (ed.), Vocal Physiology: Voice Production, Mechanisms and Functions. New York: Raven Press, 319–34. Tran Tri Doi. 1999. Nghien Cuu Ngon Ngu Cac Dan Toc Thieu So Vietnam (Studies of Languages of Minority Groups in Vietnam). Hanoi: Dai hoc quoc gia Hanoi Press. Truong Van Chinh. 1970. Structure de la Langue Vietnamienne. Paris: Imprimerie nationale. Vo Xuan Hao. 1997. Nghien Cuu ve Chuc Nang cua Thanh Dieu Tieng Viet Theo Phuong Phap Dinh Luong (A Statistical Study of the Function of Vietnamese Tones). Ph.D. dissertation. University of Education, Hanoi. Vu Ba Hung. 1988. Hien tuong tac hong va thanh dieu cua tieng Viet (A solution for the phenomenon of glottalization in the Vietnamese tones nga and nang). Ngon Ngu, 2: 40–8. —— 1999. Ve dac trung co ban cua thanh dieu tieng Viet o trang thai tinh (Tren co so cac cu lieu phan tich thuc nghiem) (On characteristics of Vietnamese tones—based on experimental analyses). Ngon ngu, 6:35–53. Williams, Edwin. 1976. Underlying tone in Margi and Igbo. Linguistic Inquiry, 7:463– 84. Woo, Nancy. 1969. Prosody and Phonology. Ph.D. dissertation. MIT. Wu, Y. 1994. Mandarin Segmental Phonology. Ph.D. dissertation. University of Toronto. Yip, Moira. 1980. The Tonal Phonology of Chinese. Ph.D. dissertation. MIT. Also New York: Garland Press, 1991. —— 1989. Contour tones. Phonology, 6:149–74. —— 1995. Tone in East Asian languages. In John Goldsmith (ed.), The Handbook of Phonological Theory. Oxford: Blackwell, 476–94. Zipf, G.K. 1949. Human Behavior and the Principle of Least Effort. Cambridge, Mass: Addison-Wesley.

Index

Andreev, N.D., 115, 140 African tone languages, 19, 27 Archangeli, D., 5, 133 Asian languages, 127 Avery, P. 5–6, 16, 34, 36, 107

eight-tone system, 15, 37, 42–54 English, 34–6, 52 Excel, 61 female/male speech, 62–8, 126 French, 34–6, 52 frequency of occurrence, 14, 16, 25–6, 29–30, 34 52–3, 132 fundamental (F0), 8, 55, 60–2, 67–8, 73–4, 80–1, 87, 120, 127, 143

Bao, Z., 17, 19–20, 27, 140 Be-Tai, 40 borrowings, 26, 34–6, 51–2 breathiness, 3–4, 8, 60, 68, 74, 79–81, 91–3 107–8, 117, 120, 126– 8, 131–4 Bru, 128 Burton, S., 16–8, 36, 40, 42, 52 Butterworth filter, 61

game, language, 135–6 German, 28 Gestural Theory of Markedness, 132 glottalization, 12–3, 47, 107, 115, 132 glottalic, 4, 13, 15 glottograms, 80–1 Goldsmith, J., 19, 136, 140 Grodina, M.V., 115, 137, 140, 146 Greenberg, J., 25, 28–9, 132–3

Cao, X.H., 37, 44–5 Central Vietnamese, 4, 31–3, 38–9, 133 Centralization, vowel, 139 Chinese, 19, 28, 40, 140 Chaozhou, 19 Cantonese, 40 Chong, 128 Clements, G.N., 5, 19, 48, 129, 140 clitics, 26, 33–4, 48–50 complexity, 5–7, 24–5, 29, 38, 40, 42 concave tone, 22, 115, 131 Concave Tone Reversal, 16, 18 consonants, final, 5, 13, 37, 40, 44–5, 47- 9,60, 138, 146 contour, 7, 10, 12, 13, 15, 17–9, 34, 36, 40 44, 50–1, 55–6, 60, 62, 67, 81, 84, Contrastive Underspecification, 23 87, 93, 101, 120, 126–7, 130–4 contour tone, 18, 22, 38, 40, 42 contour tone system, 18 Contrastive Specification, theory of, 5–6, 107 convex tone, 22, 115 creakiness, 3, 4, 8, 47, 61, 68, 79–81, 84, 87, 91, 101, 107–8, 112, 115, 117, 120, 126–32, 141, 143 Creakiness Acquisition, 17–8 Cu, T., Hoang, Nguyen 4, 94, 137, 140 curved tone, 21–3, 40, 56, 67–8, 74, 91- 2, 101, 115, 120, 126–7, 131, 141

Hanoi, 4, 38, 40, 42–3, 56, 131–3, 138 Haudricourt, A.G., 16, 46–7, 115 Hoang, C.C., 4, 10, 38–41, 127 Hoang, T.C., 4, 12, 27, 32, 37–8, 46–7 Hue, 38–40 Italian, 48 Jalapa Mazatec, 71, 73 Karen, 40 Keller, E., 155 Kui, 128 Ladefoged, p., 71, 73, 101, 128 Laver, J., 18, 73 Luganda Mai Ban, 32, 133 Marasek, K., 733, 101, 120 markedness, 4–9, 24–36, 51–4, 67–8, 94, 132 register, 10–36, 37–8, 42, 44–5, 47, 50–1, 60, 81, 93–4, 107, 117, 120, 126–34 Mester, R., Itô, 5 Mon-Khmer, 4, 128 Muong, 4 mora, 8, 139–41

diachronic issues, 13–4, 16, 18, 31, 46- 7 dialects, 4, 27, 31, 33, 38, 56, 61, 126, 133–4, 138 Digital Inverse Filtering, 80 digital sound spectrography, 73–4 Doan, T.T., 4, 10, 12, 15–6, 31, 37–8, 42, 47, 84, 135–7, 140–1

119

120

INDEX

Nam Dinh, 56 Nasal Default, 49 natural class, 5, 13, 18, 107, 134 neutralization, 4, 7, 9, 18, 20, 25–6, 28- 36, 126, 132–3, 150–1 Nghe Tinh, 31–3, 36 Ngo, T.N., 15–6, 26, 31, 137 Nguyen, V.L., Edmondson, 10–1, 16, 55–7, 60, 68, 80–1, 84, 94, 101, 112, 120 North Vietnamese, 4, 38–9, 55–61, 126, 132–4, 138–9 nucleus, 8, 135–9, 141–3, 146, 150 Nyah, 128 Odden, D., 19, 140 Ohala, J., 40 orthography, 10, 27, 35, 37, 44, 47 Phalok, 128 Pham, H., 4, 28, 33–4, 138–9 phonation types, 8, 13–4, 19, 47, 55–6, 60, 68–70, 73, 80–1, 92–3, 107 128, 130, 134 poetry, 12, 26, 30, 33, 36, 42, 45–6, 47, 135, 137 Pwo, 40 Quang Binh, 133 reduplication, 3–4, 8, 12, 16–8, 31, 33, 36, 45–6, 51, 53, 94, 107, 135–7 Register Flip-Flop, 17–8 register tone system, 18 rhyme, 8, 30, 38, 40, 84, 87, 135–50 Rice, K., 5–7, 23, 25, 28, 34, 49, 107, 133 Saigon, 28, 38, 40–1, 139 Sgaw, 40 SignalyzeTM, 61–2, Sino-Tibetan, 40 six-eight verse, 30 six-tone system, 37–8, 46–7, 52, 84 Snider, K., 18–9, 27 So, 128 South Vietnamese, 4, 27–8, 31, 33, 36, 38 133–4, 139 Spanish, 48 Spectrograms, 61–2, 73–4, 79–81, 87, 91, 107–8 Steriade, D., 5 syllable, 3, 4–5, 8–9, 60–1, 135–50 final, 28, 49 initial, 28 Tai-Kadai, 40 Thanh Hoa, 56 Thompson, L., 4, 13, 26 Thongkum, T., 128 tone sandhi, 3, 19, 26 tone linear portion, 8, 55, 81, 87, 92 tone-bearing unit, 19, 140 Vo, X.H., 4, 29, 52 vowel length, 87, 139–46, 150 Vu, B.H., 37, 47, 55, 68, 84, 120 Vu, T.P, 4, 38, 42–3, 58, 68, 94, 115

Yip, M., 177–9, 27–8, 40