Professional summary of Chapter 12

An Introduction to Phonology

Francis Katamba

Longman London and New York

Addison Wesley Longman Limited, Edinburgh Gate, Harlow. Essex CM20 2JE, England and Associated Companies throughout the world.

Puhiished in the United States of America by Addison Wesley Longman Publishing, New York

© Longman Group UK Limited 1989

All rights reserved; no part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without either the prior written permission of the Publishers or a licence permitting restricted copying in the United Kingdom issued by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W i P 9HE.

First published 1989 Eigth impression 1996

British Library Cataloguing in Publication Data Katamba, Francis

An introduction to phonology I. Phonology 1. Tide 414

ISBN 0-582-2915O-X

Library o f Congress Cataloging-in-Publication Data Katamba, Francis, 1947-

An introduction to phonology/Francis Katamba. p. cm. — (Learning about language)

Bibliography: p. Includes indexes. ISBN 0-582-29150-X 1. Grammar, Comparative and general - Phonology. I. Title.

II. Scries. P217.K33 1989 4 1 4 - d c 19 8 8 - 2 8 1 4 9

CIP

Set in Linotron 202 11/12 pt Bembo

Produced through Longman Malaysia, CL

Contents

Preface ix

Acknowledgements xvi

Chapter 1 Introduction to phonetics 1 I . I I n t r o d u c t i o n i 1.2 T h e p r o d u c t i o n of speech 2

1.2. i The production of consonants 2 1.3 T h e p r o d u c t i o n of vowels .8

Chapter 2 The phoneme 16 2.1 Segments of s o u n d 16

2. I . I Distinctiveness: phonemes and allophones 18

2.2 Identifying p h o n e m e s 22 2.2.1 The minimal pair test 22 2.2.2 Contrast in analogous environments 23 2.2.3 Suspicious pairs 24 2.2.4 Recapitulation 24

2.3 Phonological s y m m e t r y 25

Chapter 3 Distinctive features 35 3.1 W h y are features needed? 35 3.2 J a k o b s o n i a n features 38 3.3 T h e SPE system of distinctive features 42

3.3.1 Major class features 42 3 - 3 3 - 3 3 - 3 3 - 3 3 - 3 3 - 3

2 Cavity features 43 3 Tongue body features 45 4 Tongue root features 47 5 Laryngeal features 48 6 Manner features 50 7 Prosodic features 51

3.4 S e g m e n t structure r e d u n d a n c y 56

vi Contents

Chapter 4 Phonological representations 60 4.1 Phonetics and p h o n o l o g y 60 4.2 T h e d o m a i n of p h o n o l o g y 66 4.3 Recapitulation: levels of representation 69 4.4 Phonetic and p h o n e m i c transcription 69 4.5 A gu i d e to p h o n e t i c transcription 72 4.6 W h y s t u d y p h o n o l o g y ? 73

Chapter 5 Phonological processes 79 5.1 I n t r o d u c t i o n 79 5.2 Assimilation 80

5.2.1 Direction of assimilation 84 5.3 Assimilation processes 86

5.3.1 Palatalisation 86 5.3.2 Labialisation 87 5.3.3 Voice assimilation 88 5.3.4 Place of articulation assimilation 89 5.3.5 Manner of articulation assimilation 91 5.3.6 Nasalisation 93

5.4 Dissimilation 94 >* 5.5 C o n c l u s i o n 96

Chapter 6 Naturalness and strength 98 6.1 I n t r o d u c t i o n 98 6.2 N a t u r a l s e g m e n t s , natural classes and natural

processes 98 6.2.1 Phonological strength hierarchies 103

6.3 Explanations of naturalness 108 6.3.1 Natural phonology 109

Chapter 7 Interaction between rules 117 7.1 I n t r o d u c t i o n to rule formalisation and

o r d e r i n g 117 7.2 Linear rule o r d e r i n g 122 7.3 A b a n d o n i n g extrinsic o r d e r i n g 128 7.4 Conclusion: w h y o r d e r i n g matters 132

Chapter 8 The abstractness of underlying representations 134

8.1 Abstractness 134 8.2 C o n c r e t e p h o n o l o g y ? 135 8.3 Absolute neutralisation 145 8.4 C o n c l u s i o n 150

Contents vii

Chapter 9 The syllable 153 9.1 T h e syllable 153 9.2 T h e representation of syllable structure 153 9.3 T h e C V - t i e r 156

9.3.1 A generative CV-phonology model of syllable structure 156

9.3.2 Syllabification 161 9.4 Functions o f the syllable 164

9.4.1 The syllable as the basic phonotactic unit 164

9.4.2 The syllable as the domain of phonological rules 166

9.4.3 The syllable and the structure of complex segments 169

9.4.4 Compensatory lengthening 171 9.4.5 The syllable as indispensable building block

for higher phonological domains 173 9.5 Syllable w e i g h t 175 9.6 Abstract s e g m e n t s 181 9.7 Extrasyllabicity 183 9.8 S u m m a r y 184

Chapter 10 Multi-tiered phonology 186 10.1 I n t r o d u c t i o n to tone languages 186 10.2 T h e n a t u r e of phonological

representations 188 10.3 T h e representation of tone 190

10.3.1 Contour tones 190 10.3.2 Tone stability 194 10.3.3 Melody levels 195

10.4 T h e a u t o s e g m e n t a l m o d e l and the representation of tone 196

10.5 T o n e and intonation 205 10.6 Pitch-accent 208 10.7 V o w e l h a r m o n y 211 10.8 Nasalisation 214 10.9 M o r p h e m i c tier 216

Chapter 11 Stress and intonation 221 11.1 I n t r o d u c t i o n : stress 221

11.1.1 What is stress? 221 11.1.2 Metrical phonology 225 11.1.3 Metrical trees and grids 229 11.1.4 Extrametricality 231

viii Contents

11.1.5 Quantity sensitivity 232 n . 1 . 6 English stress 234

11.2 Intonation 239 11.2.1 The form of English intonation 240 11.2.2 Accentuation function 242 11.2.3 Intonation and illocudonary force 243 11.2.4 The grammatical function of

intonation 244 11.2.5 Attitudinal functions 246 11.2.6 Discourse function 248

Chapter 12 Phonology in the wider context 254 12.1 T h e role of the lexicon 254 12.2 Lexical p h o n o l o g y 257

12.2.1 Level 1 of lexical phonology 258 12.2.2 Level 2 of lexical phonology 269 12.2.3 The elsewhere condition 270 12.2.4 The order of affixes 273 12.2.5 Post-lexical rules 274 12.2.6 Summary 279

12.3 P r o s o d i c d o m a i n s 280 12.3.1 Prosodic domains and French liaison 282

12.4 C o n c l u s i o n 283

Bibliography 288

Suggested answers to exercises 296

Language index 320

Subject index 322

Preface

This book is a hands-on introduction to P H O N O L O G Y for the absolute novice. Probably the best way to learn about phonology, i.e. to learn how speech sounds are used to convey meaning, is to do phonological analysis and confront theoretical issues as they get thrown up by the data.

With this in mind, this book has been written not only with exercises at the end of each chapter, but also with in- text problems and tasks which are separated from the discussion by a line drawn across the page. You should always attempt these problems before reading on. They are an integral part of the discussion. Suggested answers are included within each chapter. Answers to end of chapter exercises will be found at the end of the book.

Some remarks on presentation: technical terms are commented on and highlighted using capital letters when they are introduced for the first time or when it is important to emphasise them. The common convention of using an asterisk to indicate impossible or wrong forms is also observed (e.g. *tleg is 'starred' to show that it is not a poss- ible word in English). Examples discussed in the text are written in italics.

The model of phonology which I introduce you to is called GENERATIVE P H O N O L O G Y . It was given its first full and authoritative statement in Chomsky and Halle's 1968 book The Sound Pattern of English. As we shall see, since then it has moved on in various directions. In the next few paragraphs the objectives of this theory are explained.

Generative phonology is part of the theory of language called GENERATIVE GRAMMAR which has been devel-

x Preface

oped by Chomsky and his collaborators. The basic goal of generative grammar is to explore and understand the nature of linguistic knowledge. It seeks answers to questions like: what does knowing a language entail? How is linguistic knowledge acquired by infants? Are there any properties of language that are universal, i.e. is there such a thing as 'Universal Grammar'?

Chomsky believes that the answer to the last question, which he thinks holds the key to the other questions, is 'yes' and goes on to argue that Universal Grammar has a biological basis. Biologically determined characteristics of the brain pre-dispose humans to acquire grammars with certain properties. But this raises further questions: what are the properties of Universal Grammar? In attempting to answer this question, generative linguists have developed principles and posited rules of the kind we shall explore. They form part of their model of Universal Grammar.

Like other linguists, generative linguists know that some aspects of language are not universal. But still they raise the question whether some non-universal properties of language fall into certain well defined parameters. Are there any pre-set limits within which differences between languages occur? If the answer is 'yes' what are these limits and why do they exist? These are some of the main issues which this book addresses.

Besides being concerned with general patterns of language structure, linguistic theory must provide us with the tools for describing those idiosyncratic properties which are peculiar to a particular language. For instance, linguistic theory should enable us to write a grammar for English showing that the final / consonant of chief is pronounced / when the plural -s ending is present but the final / of thief is pronounced v when the same plural ending is present.

The grammar of a language can be regarded as a model of the C O M P E T E N C E (i.e. inexplicit knowledge of rules) that underlies a native speaker's overt linguistic PERFORMANCE as a speaker-hearer. We need to distinguish between the knowledge speakers have and the manner in which they put that knowledge to use in concrete situations as, sometimes, there is a difference between what one knows to be correct and what one actually says. This may be due to a number of factors such as slips of the

Preface xi

tongue or memory lapses. Linguistics is primarily concerned with linguistic competence (knowledge) rather than performance (use). This book is primarily concerned with P H O N O L O G I C A L C O M P E T E N C E .

Interestingly, knowing a language, say English, is not merely a matter of learning by rote a very large number of sentences. Native speakers of a language can always produce and understand completely new sentences which they have not previously encountered. No list, however long, could contain all the potential sentences of a language. Therefore a grammar of a language cannot be simply a list of words and sentences of that language. In view of this, Chomsky proposes that a grammar of a language should be a gener- ative algebraic system of formal, explicit rules that enum- erates a non-finite number of well-formed sentences and assigns to each one of them a correct analysis of its structure.

The motivation for using rules to account for the fact that there is no limit to the number of possible sentences that a language can contain is obvious: speakers produce and understand sentences using rules. They do not merely memorise long lists of sentences. However, the reasons for assuming that there are rules which underlie speakers' knowledge of the sound system of their language are perhaps less obvious, given the fact that a language only uses a finite set of sounds to form words.

Just a little reflection is enough to show that the sound system is also rule governed. Determine which of the following nonsense words (which you are probably seeing for the first time) is a possible English word: tpat, bender, bintlement and zvetsin. You no doubt have chosen bintlement as the only potential English word. This is because you know that the consonant sequences tp, Is and zv which occur in the other 'words' are not permitted at the begin- ning of an English word. On the other hand, all the sequences of sounds in bintlement are allowed by the rules of English phonology. You might indeed be tempted to look up bintlement in a good English dictionary - but not the other nonsense words.

The implicit knowledge of linguistic rules that speakers have is probably modular. Generative linguists have proposed that it can be represented using a model with a

xii Preface

n u m b e r of c o m p o n e n t s w h i c h represent semantic k n o w l - edge, syntactic k n o w l e d g e , k n o w l e d g e of s o u n d structure, and so on.

O v e r the years, various proposals h a v e been m a d e r e g a r d i n g the precise organisation and content o f a g e n e r - ative g r a m m a r . Y o u are n o t expected to have any prior k n o w l e d g e of these theories. N o t h i n g is presupposed. Essential aspects of the t h e o r y will be i n t r o d u c e d , w h e r e necessary.

T h e diagram b e l o w s h o w s the place of p h o n o l o g y in the general theory of language w h i c h w e shall be using:

Phrase structure i rules '

Lexicon

Initial phrase maker

Transformational rules (movement rules)

Surface structure

Phonological rules Semantic interpretation rules

Phonetic representation j ^ Semantic representation y

A Generative Grammar Model of Language (Based on Lightfoot 1982)

T h e S Y N T A C T I C C O M P O N E N T consists of the base s u b - c o m p o n e n t and the transformational rules. T h e P H R A S E S T R U C T U R E (PS) rules and the L E X I C O N found in the base s u b - c o m p o n e n t of the g r a m m a r generate the I N I T I A L P H R A S E M A R K E R ( D E E P S T R U C T U R E ) of a sentence. T h e lexicon lists the w o r d s of the language, t o g e t h e r w i t h their syntactic and phonological properties

Preface xiii

and the PS rules define the constituent structure (i.e. the structure of n o u n phrases, verb phrases etc.) and h o w they interact w i t h each other. T h e initial phrase mar ker enters the transformational c o m p o n e n t w h e r e it m a y be modified by various transformational rules w h i c h m o v e a r o u n d constituents. T h i s is d o n e to relate sentences like Money is what I need and What I need is money. T h e o u t p u t of the syntactic c o m p o n e n t is the S U R F A C E S T R U C T U R E .

Surface structures are the input to rules of L O G I C A L F O R M and S E M A N T I C I N T E R P R E T A T I O N . T h e rules of logical form explain, for example, w h y to pay in the sentence Jane ordered Bill to pay is u n d e r s t o o d to mean that Bill is the one that was expected to pay while in Jane prom- ised Bill to pay it is Jane w h o is expected to pay.

Rules of semantic interpretation are used, for instance to account for logical relations like entailment. A sentence like ' The Mayor of Lancaster switched on the Christmas lights last year' entails that t h e r e w e r e Christmas lights last year. It w o u l d be c o n t r a d i c t o r y to utter that sentence and continue 'but there were no Christmas lights last year because of budget cuts'.

P H O N O L O G I C A L R U L E S also apply to the surface structure and assign it a P H O N E T I C R E P R E S E N - T A T I O N (i.e. s h o w h o w it is p r o n o u n c e d ) . It is this final aspect of the g r a m m a r that w e are mainly concerned with in this b o o k .

This b o o k is a simple, practical i n t r o d u c t i o n to p h o n - ology within the m o d e l of generative p h o n o l o g y as it has evolved d u r i n g the last t w e n t y years or so. While in the early years the emphasis was on m a k i n g explicit the relationship b e t w e e n u n d e r l y i n g and surface phonological representations by investigating the n a t u r e o f formal p h o n o - logical rules, the w a y s in w h i c h rules interact and the distance b e t w e e n u n d e r l y i n g and surface representations in p h o n o l o g y , lately the focus has shifted to scrutinising the nature of phonological representations themselves and the relationship b e t w e e n p h o n o l o g y and o t h e r c o m p o n e n t s of the g r a m m a r .

This shift in focus is reflected in the contents of this b o o k . After a brief i n t r o d u c t i o n to articulatory phonetics, the o p e n i n g chapters deal w i t h distinctiveness, 'naturalness', the relationship b e t w e e n levels o f phonological represen-

xiv Preface

tation and rule interaction. These were the main issues explored in the 1960s and 1970s.

However, the latter part of the book is devoted to topics of current interest. One major trend in generative phonology today involves several 'non-linear' approaches to the nature of phonological representations. It is being developed through an examination of the nature of sound 'segments', syllable, tone, stress, and intonation in numerous languages. The other major current trend focuses on the relationship between phonology and other components of the grammar such as the lexicon, morphology and syntax. These two trends are complementary.

The exclusive concentration on generative phonology should not be taken as evidence of a belief on my part that nothing of value has been said about phonology in the other frameworks. Occassionally the contributions of other schools are mentioned in a footnote. But I have restricted the exposition to generative phonology for two reasons. In my experience, for the beginning student it is more bewil- dering than enlightening to be presented with several competing theoretical positions, with their different the- oretical concepts, analytical techniques and nomenclature. There is virtue in introducing students initially to one coherent theoretical' approach. The question that then arises is: which approach?

I have chosen tq .introduce. yp.tr. to generative phon- ology. Thi^&i&'t-feerely-'^i'mattef-bf my personal taste. Generative > phonology "isj cwFF8 t̂Py-lfner .dominant model of phonology.'; It is the mT)deLt;o,:ep.r4a&eao grips with, if eventu- ally you wish to read the- cufMenf'descriptive and theoreti- cal phonological literature. Much of it is written in some version of this framework.

However, should you wish to survey other past and present trends in phonology there are many books which you can turn to. If you wish to acquaint yourself with the history of phonology you can read excellent historical studies like Fischer-Jorgensen (i975) and the more recent Anderson (1985). If you want an eclectic, 'unbiased' intro- duction to phonological concepts and their philosophical underpinnings you can turn to Lass (1984).

This book has developed from phonology courses that

Preface xv

I have taught over the years at the University of Nairobi, Kenya and at the University of Lancaster. I am grateful to the generations of students who were subjected to earlier drafts of the book for the feedback I got from them.

In writing this book I have benefited immensely from the help of Professor Geoffrey Sampson. Very special thanks also go to my editors and colleagues Mr. Mick Short and Professor Geoffrey Leech whose critical comments and suggestions have made this a better book than it would otherwise have been. Those others who over the years have taught me directly or indirectly something about phonology deserve a special mention. Their scholarship is reflected in the theory presented here as well as in the data from the dozens of languages cited. And finally, I am grateful for the encouragement of my wife Janet during the long gestation of this book.

21 March ig88

Acknowledgements

We are grateful to the following for permission to repro- duce copyright material:

Cambridge University Press for extracts from NGIYAMBA (1980) by T. Donaldson & from 'Kimatumbi phrasal phon- ology' by D. Odden Phonology Yearbook 4 (1987); Harper & Row Inc for an adaptation based on pp 302-329 from The Sound Pattern of English by Noam Chomsky & Morris Halle (Copyright (E) 1968 by Noam Chomsky & Morris Halle); the Editor, Professor Larry M. Hyman for an adaptation of pp 112—115 'Noun tonology in Kombe' by B. Elimelech in Studies on Bantu Tonology SCOPIL 3 (1976). International Phonetic Association for a table from p. 10 of Principles of the International Phonetic Association 1949/84; the author, P. Kiparsky for a figure from 'From cyclic phonology to lexic- al phonology' in The Structure of Phonological Representations Part 1 (1982) by Van der Hulst & Smith; MIT Press for a figure from p. 39 of The Language Lottery by D. Lightfoot. Copyright © 1983 MIT Press.

CHAPTER 1

Introduction to phonetics

1.1 Introduction

The purpose of this book is to introduce you to the study of P H O N O L O G Y . Phonology is the branch of linguistics which investigates the ways in which sounds are used systematically in different languages to form words and utterances.

In order to understand phonology, one must have a grasp of the basic concepts of P H O N E T I C S , the study of the inventory of all SPEECH S O U N D S which humans are capable of producing. The term speech sound has been used advisedly since not all noises which we are capable of producing with our vocal apparatus are employed in speech: we can all snore; we can all cough and hiccup; we can all sneeze and we can all gnash our teeth. However, no linguist, has yet discovered a community that has a language in which noises produced by any one of these mechanisms are used to form words. It is almost certain that no such speech community exists. One reason for this is the fact that there are obvious disadvantages in letting communication depend on involuntary noises like hiccups which speakers cannot start and stop at will. Other methods like the gnashing of teeth may be easy to control, but have their drawbacks - the wear and tear which gnashing of teeth would entail must have ruled out that method. I am using these examples to underscore the point that speech sounds form a small subset of all the noises which humans can produce with their vocal apparatus. It is this subset that phoneticians focus on.

The study of speech sounds can be approached from

2 Introduction to phonetics

various angles. These are reflected by the three major branches of phonetics:

(a) A C O U S T I C P H O N E T I C S : the study of the physical properties of speech sounds using laboratory instruments;

(b) A U D I T O R Y P H O N E T I C S : the study of speech perception;

(c) ARTICULATORY P H O N E T I C S : the study of speech production.

1.2 The production of speech

It is articulatory phonetics that we shall concentrate on here because it is the branch of phonetics on which most phono- logical theories have been based in the past.

1.2.1 The production of consonants

Speech sounds are produced by interfering in some way with a body of moving air. Phoneticians use the term AIR- STREAM MECHANISM to describe a body of moving air used in speech production. It is important to describe how the air is set in motion and the direction in which it travels because that makes a difference in the sound produced. The commonest airstream mechanism used in the world's languages (and the only one found in English) is the P U L M O N I C EGRESSIVE mechanism. When this mech- anism is employed, air is expelled from the lungs, up the windpipe and gets out through the mouth, or through the nose, or through both.

There exist other airstream mechanisms, but we shall postpone discussion of these until Chapter 3. We shall provisionally assume that all speech sounds are made with air pushed from the lungs up the WIND PIPE (also called the TRACHEA, more technically). Continuing its outward journey, the air reaches the LARYNX (or voice box - that cartilaginous membrane at the top of the windpipe which is called the Adam's apple in everyday language). Once in the larynx, the air must pass through the GLOTTIS. This is the space between the vocal cords. If the vocal cords are apart, i.e. if the glottis is open, the air escapes unimpeded. Sounds produced in this way are said to be VOICELESS.

The production of speech 3

If, on the other hand, the vocal cords are very close together, the air will blow them apart as it forces its way through. In doing so, it will make them vibrate, producing a VOICED sound.

You can perform a little experiment to determine for yourself whether a given sound is voiced or voiceless. Say [ffffj and then [vvvv] with two fingers held firmly on your larynx. Repeat this four times. What do you observe?

N o w say [ffffvvvvffffvvvv] this time with your index fingers in your ears. What do you observe?

In the first experiment, you should be able to feel your fingers vibrating slightly when you say [vvvv] but not when you say [ffffj. And in the second experiment, you should be able to hear a low buzzing noise in your head when you produce [vvvv], but not when you produce [ffffj. The noise that causes the vibration in your larynx which you feel with your fingers as well as the low buzzing sound which you feel when you have fingers in your ears is called VOICING.

Voicing is linguistically important. The difference between voiced and voiceless sounds is functional. In many languages, English included, as you can see in [1.1], there are many consonants which come in pairs, with the two sounds in question differing in voicing:

A pull ten cot fast sink chew

B bull den got vast zinc Jew

The initial consonant of the words in column A is voiceless while the initial consonant of the words in column B is voiced.

Find five more words which begin with voiceless and voiced consonants respectively.

4 Introduction to phonetics

[ 1 . 2 ]

T h e O r g a n s of Speech

1 lips 2 teeth 3 alveolar ridge 4 (hard) palate 5 velum 6 uvula 7 tongue tip 8 tongue blade 9 front of the tongue

10 back of the tongue 11 mouth cavity 12 nose cavity 13 pharynx 14 epiglottis 15 oesophagus 16 glottis 17 larynx

T h e t e r m V O C A L T R A C T is used to refer to the air passages which the air enters o n leaving the l a r y n x . As y o u can see in the figure in [1.2], the vocal tract has t w o parts: the O R A L T R A C T , w h i c h is the air passage offered b y the m o u t h and the N A S A L T R A C T , w h i c h is the air passage p r o v i d e d by the nose.

C o n s o n a n t s are p r o d u c e d by o b s t r u c t i n g in s o m e way the flow of air t h r o u g h the vocal tract. W e can identify the P L A C E (or P O I N T ) w h e r e the obstruction takes place, and the organs involved. T h e parts of the oral tract such as the t o n g u e and lips w h i c h can be used to form speech sounds are called A R T I C U L A T O R S . A list of places of articulation is given in [1.3] t o g e t h e r w i t h the phonetic symbols representing s o m e of the sounds m a d e at each place.

As you read t h r o u g h [1.3] y o u should say aloud the w o r d s in the illustrative examples and observe h o w y o u use the articulators to p r o d u c e each sound. N e x t , y o u should find examples of y o u r o w n of each s o u n d . T h e examples can c o m e from any language you k n o w . If y o u are u n c e r - tain about the t e r m i n o l o g y , consult [1.2].

The production of speech 5

[1.3] Places of articulation

place

BILABIAL

LABIO- DENTAL DENTAL

A L V E O L A R

R E T R O - F L E X

P A L A T O - A L V E O L A R

P A L A T A L

V E L A R

L A B I O - V E L A R

G L O T T A L

articulators

both lips

lower lip and upper front teeth tongue tip and upper front teeth t o n g u e tip or blade and the alveolar ridge

t o n g u e tip curled back past the alveolar ridge

blade of the t o n g u e rising t o w a r d s the alveolar ridge and the front o f the hard palate front of the t o n g u e and the hard palate

back of the t o n g u e and the v e l u m simultaneously using b o t h lips and raising the back of the t o n g u e t o w a r d s the v e l u m vocal cords

examples

[p] peat

[f\fine

[0] thigh

[t] tip [s] sip [1] lip

[x] raft

[n_] pina

[J] sheep [tf] cheap

[j] yes

In] agneau [k] cot

[w] win [kp] kpakp (Nigeria))

[h] hot

b] beat m] meat v] vine

5] thy

d] dip z] zip r] rip n] nip in American English) lake' (in Wangkatja 'Australia)) 3] genre d3] jeep

c] ich T German) l a m b ' (French)

g]got 0] song

a 'field' ( Y o r u b

[•>] better (in many varieties of nonstandard British English, e.g. Cockney)

Besides describing the place where the obstruction occurs in the production of a consonant, it is also essential

6 Introduction to phonetics

to consider the M A N N E R OF ARTICULATION, i.e. the nature and extent of the obstruction involved because sounds made at the same place of articulation - and which are both voiced or voiceless, as the case may be - can still differ depending on the manner in which the airstream is modified. Take these words, which all begin with VOICED ALVEOLAR consonants: dine, nine, line and Rhine. Not only do they sound different, they also mean different things. The difference lies in the manner of articulation of their initial consonants.

In [1.4] below a brief survey of manners of articulation is given, together with English examples.

[1.4] STOP: The articulators come together and completely cut off the flow of air momentarily, then they separate abruptly. Examples: pin bin tin din kilt gilt. Stops like [ p b t d k g ] which are made with the pulmonic egressive airstream mechanism (as the speaker breathes out through the mouth) are called PLOSIVES. Sounds like [m n], as we shall see presently, are also stops but they are not plosives; they differ from plosives in that they are formed by completely blocking off the airstream in the mouth while at the same time allowing it to escape through the nose.

AFFRICATE: In the production of an affricate, first the articulators come together and completely cut off the flow of air, just as they do in a stop; then they separate gradually.

Examples: chain Jane [tf] [d3]

Say the following words very carefully char jar Describe as accurately as you can the ways in which the articulators are used to obstruct the flow of air in the first sound of each word. Which one of the affricates is voiced?

The production of speech 7

FRICATIVE: the articulators are brought very close together leaving only a very narrow channel through which the air squeezes on its way out, producing turbulence in the process. Examples: [f v 0 5 s z J 3]

Say the following words and listen for the turbulence of the initial consonant: fan van thin then sink zinc shrill genre

A P P R O X I M A N T : the articulators are brought near each other but a large enough gap is left between them for air to escape without causing turbulence. Examples: [r 1 w j] as in read lead

weed and yield respectively. NASAL: Nasal sounds are produced with air

escaping through the nose; the velum is lowered to allow access to the nasal tract. Examples [m n rj] The final sounds in all these words are nasal: sum sun sung. O n the other hand, to produce ORAL sounds like [p t k b d g s z] the VELUM is raised right up against the back wall of the PHARYNX, cutting off access to the nasal cavity and making air escape through the mouth only (see [1.2]).

LATERAL: to produce a lateral, the air is obstructed by the tongue at a point along the centre of the mouth but the sides of the tongue are left low so that air is allowed to escape over one or both sides of the tongue. Example: [1] If you say the following words slowly and carefully, you will be able to observe how the air escapes over the sides of the tongue in the last [1] sound:

peddle paddle huddle kettle battle cattle

8 Introduction to phonetics

Examine the examples in [1.4] above once again. State the place of articulation of each consonant and also deter- mine whether it is voiced or voiceless.

Next find three more words exemplifying each manner of articulation that has been described. Take your examples from any language which you know.

Summary

The production of consonants involves four major param- eters which can be varied independently of each other to create different kinds of consonant. The four parameters are:

(a) THE AIRSTREAM MECHANISM This refers to the way in which the moving body of air that provides the power for speech production is generated and the direction in which it moves.

(b) THE STATE OF THE GLOTTIS Voiceless sounds are produced when there is a wide open glottis, with a big space between the vocal cords; voiced sounds are produced when the vocal cords are close together so that the air has to force its way through them, making them vibrate in the process.

(c) THE PLACE OF A R T I C U L A T I O N This refers to the place in the vocal tract where the airstream is obstructed in the production of a consonant.

(d) T H E M A N N E R OF A R T I C U L A T I O N This refers to the way in which the airstream is inter- fered with in producing a consonant.

1.3 The production of vowels

Vowels are more difficult to describe accurately than consonants. This is largely because there is no noticeable obstruction in the vocal tract during their production. It is not easy to feel exactly where vowels are made. Moreover, in many cases, sounds perceived by hearers as the 'same vowel' may be produced using a number of substantially different articulatory gestures. The only reliable way of

The production of vowels 9

observing vowel production is using x-ray photography. But this is not only expensive, it is also dangerous and could not be carried out each time one wanted to describe a particular vowel (even if willing subjects eager to allow themselves to be exposed to radiation in the noble cause of phonetic inquiry could be found).

The account in this chapter will be restricted to vowels which occur in English but much of what is said is gener- alizable to many other languages. As we noted above, many of the parameters described above which are used in the description of consonants have not been found applicable to vowels. Vowels are typically voiced, but they have no place or manner of articulation. Traditionally, for the description of vowels a different set of concepts has been found necessary.

Say the words in [1.5] carefully. Observe in a mirror the position of the highest point of your tongue and your lower jaw.

[1.5] [i] seek [u] pool [1] sick juj pull [e] set [D] pot [ae] sat [a] part

Now repeat the exercise, just uttering the vowel sounds on their own. Again observe the position of your tongue and j a w .

In each column your tongue is high when you say the vowel in the first word on the list and gets progressively lower as you work your way through the list. If you watch yourself in a mirror, you will observe that there is a corre- lation between tongue height and jaw opening: when the tongue is high, the jaw is not lowered but when the tongue is low the jaw is also low and your mouth is wide open.

Vowels produced with the highest point of the hump in the tongue close to the roof of the mouth are said to be HIGH and those produced with the highest point of the hump in the tongue barely rising above the floor of the mouth are said to be LOW; the intermediate position is referred to as MID.

10 Introduction to phonetics

Up to now, we have considered the vertical axis in locating the highest point of the tongue. That is not enough. It is also necessary to determine the location of the highest point on the horizontal axis: the highest point of the tongue may be at the front, in the centre or at the back of the mouth. Depending on the location of the highest point of the tongue, vowels may be regarded as F R O N T , CENTRAL or BACK.

Say the words in [1.6] slowly a number of times and observe the position of your tongue in each case:

[1.6] [e] shed [a] should ' [D] shod

When you say the vowel [e] your tongue is in the front, palatal region; when you say [a] your tongue is in the centre of the mouth; when you say [D] your tongue shifts to the back of the mouth. That this is the case will be clear if you utter the vowels on their own one after the other.

Finally, the quality of a vowel is affected by the shape of the lips. For simplicity's sake, phoneticians assume that lips can assume only two positions: they are either R O U N D E D or U N R O U N D E D - intermediate positions are ignored. Lip rounding results in an elongated resonating chamber while lip SPREADING or U N R O U N D I N G (the position the lips assume if you put on that fixed grin when the voice behind the camera tells you to smile or say 'cheese') does not. Just as the notes which can be produced by different wind instruments partly depend on the shape of the instrument, the QUALITY of a vowel sound partly depends on the shape of the resonating chamber in the vocal tract resulting from adjustments in lip position.

Get a mirror and once again observe your lips as you say the words in [1.7]. You will be able to verify that in each case the first word in each pair is said with spread (unrounded) lips while the second is produced with the lips rounded:

[1.7] [i] see [u] sue [i] tea [u] two

The production of vowels 11

[e] fen [o] fought [ae] lag [o] log

The various articulatory parameters are not entirely independent of each other. Note, for instance, that there is a systematic correlation between lip rounding and tongue height. For a rounded vowel, the higher the tongue is, the greater the degree of lip rounding.

The phonetic properties of vowels surveyed in this section are conventionally represented in this diagram:

[1.8] Front Central Back

High \ i u

Mid \ e o

Low \ a

Unround Round

As [i.8] shows, typically front vowels are unrounded and back vowels are rounded.

The parameters high, mid and low enable one to distinguish three degrees of vowel height as you can see in [i.8]. But this is not always sufficient. There are languages which make a four way distinction on this parameter. Many phoneticians use the terms CLOSE, HALF-CLOSE, HALF-OPEN and O P E N to reflect this:

I 1 - 9 ] Front Central Back

Close \ i u

Half-close \ e o

Half-open \ e o

Open \ .a

12 Introduction to phonetics

All the vowels w h i c h have been described so far are M O N O P H T H O N G S i.e. vowels w h o s e quality remains virtually u n c h a n g e d t h r o u g h o u t their d u r a t i o n . In addition to such vowels s o m e languages (English included) also have D I P H T H O N G S , i.e. v o w e l s w h o s e quality changes d u r i n g their p r o d u c t i o n .

Say the following w o r d s , concentrating on the v o w e l sound:

[ I . I O ] pie b u y m y g u y cry tie

die sigh shy high lie five

Identify the shifting positions w h i c h y o u r t o n g u e occu- pies as y o u p r o d u c e the v o w e l .

Initially the t o n g u e is l o w and front b u t it finishes up in a high, front position. T h e s y m b o l for this v o w e l is [ai].

N o w say the w o r d s in [ i . n ] and again observe the shifting quality of each v o w e l :

[ i . n ] a . c o w n o w s h o u t o u t [ao] b . toy b o y boil coin [DI] c. wait pay w e i g h t hay [ei] d. air fare w e a r chair [ea]

In [1.12] y o u can see the changing quality of the d i p h - t h o n g s exemplified a b o v e :

[1.12]

The production of vowels 13

The primary aim of this chapter has been to introduce you to the basic concepts of articulatory phonetics so that you can understand the nature of speech production. The secondary largely through familiar English data."

For a complete list of phonetic symbols see the Inter- national Phonetic Alphabet in the chart in [1.13]. You do not need to learn all the symbols in the chart right away. The chart is provided so that you can refer back to it when- ever you encounter unfamiliar phonetic symbols in the course of reading this book.

[i.i3]

B il

a b

ia l.

L a

b io

d e

n ta

l.

D e n

ta l

a n

d

A lv

e o

la r.

[ I . / J ] T h e International P h o n e t i c ,

Plosive . Nasal . Lateral . " fricative . Rolled . Flapped . Rolled fricative . Fricative .

Frictionless C o n t i n u a -

nts and S e m i - vowels .

V o w e l s

Close . Half- close . Half- open .

O p e n .

P b m

• P

W lj

R o u n d e d

(y « " ) ( 0 0 )

(CC 3 )

(D)

rt)

f v

V

t d n 1

i i r

r «

e 5

Alph

sz j

j

R et

ro fl

ex .

labet

t 4 a I

M

§ z.

P a la

to -a

lv e o

]

J 3

A lv

e o

lo -p

a la

t

C %

P al

at al

.

c J

fl A

£ j

J m

V e

la r.

k g

0

x Y

( w ) V

U v

u la

r.

q c N

R

R

X K

n

P h

a ry

n g

e a l.

h f

F r o n t C e n t r . Back

i y i e a

c ce

ae a

U 111 U

Y O

3

A 3

B

a o

O

•>

h A

14 Introduction to phonetics

Exercises

I . Write a one-page summary of this chapter. Your summary should include the following points:

(a) the domain of phonetics (b) the three main branches of phonetics (c) the description of speech sounds

(i) The four parameters used to describe consonants are:

(a) (b) (c) (d)

(ii) The three parameters used to describe vowels are: (a) (b)

(iii) The two major types of vowel sounds are: (a) (b)

2.(a) Write down the appropriate phonetic symbol for each one of the sounds described below,

(b) Give an example of a word containing the sound you have written down. Use examples from any language which you know. Underline the relevant sound,

(i) alveolar lateral (ii) voiceless alveolar fricative

(iii) alveolar nasal (iv) voiceless glottal fricative (v) voiced bilabial stop

(vi) voiceless bilabial stop (vii) velar nasal

(viii) voiced dental fricative (ix) voiceless alveo-palatal fricative (x) voiced alveo-palatal affricate

3. Circle each sound in the following words which matches the description given. Follow the example given in (i).

(i) a high front vowel feet fell pat wet full peat

(ii) a low front vowel what bad cat saw these eggs

The production of vowels 15

(lii)

(IV)

(v)

(Vl)

(vii)

(viii)

— (ix)

(x)

(Xl)

a high back vowel women suit pool fool blood flood

a front vowel weed word when hat card hit

a back vowel hut call guard sell soot mist

a central vowel skin her winter pertain doctor sir

a rounded vowel her good dumb ball pod cart

a high vowel we do see ten pan bin

a mid vowel send card keys school hall you

a low vowel man moon art cup knot teeth

a diphthong why he may boy tar house bird

Notes

1. Say 'should' as you would in a casual conversation when uttering a sentence like 'John should go' where there is no emphasis on should.

2. For a general introduction to phonetics I refer you to one of a number of textbooks such as Abercrombie (1967), Ladefoged (1982) or Knowles (1987).

Chapter 2

The phoneme

2.1 Segments of sound

As we hinted in the opening chapter, and are going to see in detail in this and subsequent chapters, in purely physical terms, any utterance is a continuum: articulatorily, it is a continuous flow of gestures that blend with each other; acoustically, it is a continuously fluctuating wave of sound. Yet, for several reasons, from a functional point of view, the fiction of discrete speech segments with which the last chapter ended is worth clinging to.

Firstly, an ALPHABETIC WRITING SYSTEM, with discrete letters written in a line one behind the other, seems to be capable of representing the sound system of any language in a non-arbitrary way. Indeed, native speakers, if they are literate in an alphabetic writing system, can usually determine without any difficulty, the 'distinct' sounds which a word consists of. O f course, to this you might reply that the idea of SEGMENTS is perhaps not inherent in the nature of the spoken language but is rather something inculcated by an alphabetic writing system - if people see words in terms of segments, that is simply attributable to the conditioning they have been subjected to by alphabetic writing. You might argue that it would not be surprising if it turned out that illiterate people or literate speakers of languages like Chinese and the American Indian language Cherokee, which have non-alphabetic writing systems, do not view speech in terms of segments.

The Chinese writing system represents the W O R D in its entirety. Thus the CHARACTER 0 represents the word 'sun' and the character A represents the word 'man'. This is analogous to way in which mathematical symbols like ^

Segments of sound 17

< etc. represent entire concepts or words which are pronounced differently in different languages. Turning to Cherokee writing we observe that each symbol represents an entire syllable. For instance, f, A, J respectively represent the syllables ga, go, and gu.

By itself^-alphabetic writing is not sufficient proof of the reality of sound segments. More persuasive evidence needs to be found. One source of such evidence is SPEECH ERRORS. Quite often, speech errors occur in which sounds are transposed in words, as though they were discrete items. Have we not all heard spoonerisms where word initial consonants are switched from one word to another, resulting in forms like polley trusher and sork pausages when the speaker really means to say trolley pusher and pork sausages'?

Pause for a moment and list at least three more spoon- erisms. Describe the transposition of sounds that takes place in each one of your examples.

Admittedly, spoonerisms are an aberration (the Rev. W.A. Spooner would not have become so famous if swop- ping consonants between words was the norm). What makes spoonerisms interesting is the way in which they clearly illustrate the treatment of speech sounds by speakers as discrete segments which can be shunted from one word to another.

The same kind of segmentalization of speech sounds is also manifested in normal spoken language - albeit usually in a less extreme form. There are numerous phonological operations (as you will discover as you work your way through the book) which re-arrange, add or delete indi- vidual consonant and vowel segments of a word rather than some larger pieces such as syllables. For instance, many non-standard varieties of British English have a rule which 'drops aitches' at the beginning of a word. The effect of this rule is to make words like at and hat sound exactly the same. Rules which drop sound 'segments' exist in languages regardless of whether they have an alphabetic writing system. Similarly, rules which insert segments exist in languages regardless of whether they have an alphabetic

18 The phoneme

writing system. I have in mind rules like the one which optionally inserts [r] between vowels within a word as in draw(r)ing (for drawing) or across a word boundary as in the idea (r) is (for the idea is).

The upshot of this discussion is that one important aspect of linguistic knowledge is the knowledge of the func- tional phonological units which occur in one's language - the knowledge of the segments which can be arranged in different patterns to form words. This knowledge is for the most part unconscious (just in the same way that knowledge of how to build boats which can float when fully laden has been in the minds of boat builders from time immemorial - long before they had any explicit understanding of the principles of physics involved). Most speakers of English do not spend half the day wondering whether to 'drop an aitch' in a word like hat or 'swallow a t' in a word like better. But that does not mean that they do not k n o w the rules which govern the sound pattern of their language. Phon- ology attempts to make explicit the nature of the knowl- edge that underlies their subconscious linguistic behaviour.

2.1.1 Distinctiveness: phonemes and allophones

A useful approach to the problem of how segments of sound are used in speech is to consider sounds from a F U N C T I O N A L point of view, to talk not in terms of individual physical sounds, but of FAMILIES of sounds which count as the same in the language in question. Such a family of sounds is called a P H O N E M E . 1 Members of the same phoneme family, i.e. the various physically distinct sounds which count as executions of a given phoneme are called the ALLOPHONES (or VARIANTS) of that phoneme. A phoneme is conventionally represented by a letter symbol between slant lines. Thus, in English, the infinitely physically diverse voiceless stops that can be made with contact between the tongue and the palate are grouped into two phonemes labelled / t / and /k/, each of which has a range of allophones which differ slightly from each other.

Which phoneme a particular sound heard in speech belongs to entirely depends on the language in which it occurs. In some languages, such as Slavonic ones, some of

Segments of sound 19

the physically attested sounds which in English would be regarded as somewhat off-target renderings of / t / or / k / would instead be members of a third, separate phoneme / c / , to which English has nothing corresponding.

Sounds are grouped together as members of the same phoneme when the very real physical differences between them happen to be functionally immaterial with respect to the language being described.

Utter the following words slowly: car keys. You will notice that in car, the back of the tongue touches the part of the soft palate near the uvula (see the figure in [1.2] on page 14), at the very back of the roof of your mouth, but in key, it is the more front part of the soft palate near the hard palate that the tongue makes contact with. The two varieties of 'k' are physically different. But they are not functionally different in English. They cannot be used to distinguish word meaning. Rather, they are allophones of the same / k / phoneme and which one is used on a given occasion depends on what the neighbouring sounds happen to be.

The two 'k' sounds are in C O M P L E M E N T A R Y DISTRIBUTION. When two sounds are in complementary distribution, they are barred from occurring in identical environments: there is a rigid division of labour, as it were, so that one sound appears in certain contexts and the other in some different, clearly defined contexts. Thus, in our example, the very back allophone (variant) [k] of the phoneme / k / found in car (allophones are written in square brackets) occurs before back vowels and the fronted [+], as in key occurs before front vowels (the + diacritic under a consonant indicates fronting). Another example of comp- lementary distribution in English should make the point clearer. The phoneme / t / has several allophones. It is poss- ible to predict the allophone that speakers produce in a given word if you know the sounds that are adjacent to it.

Say the words: tea, too and eighth. You will notice that the alveolar allophone [t] in tea [ti] is made with the lips spread but that in too [twu] is made with pursed or rounded lips (a little, raised 'w' is used to show lip rounding). However, in [eit0] 'eighth' the allophone of / t / is dental,

20 The phoneme

[t] [n] is the diacritic mark for dental sounds) not alveolar - in anticipation of the dental fricative [8] which follows, the tip of the tongue is positioned against the upper front teeth.

It is worth stressing that the relationship between a family of sounds is not like that between members of a human family. The phoneme is an abstraction. What actu- ally occur are the allophones - to be precise P H O N E S i.e. sounds which can be grouped together as allophones of a particular phoneme. We hear a range of physically different sounds in English, such as those described above, which we recognise as being functionally non-distinct. We use the theoretical construct of P H O N E M E to mean that function- ally speaking, a given set of sounds never contrast with each other although they may contrast with other sounds outside that set.

All sounds used in a language belong to some phoneme. Some phonemes have numerous allophones others may have a less diverse membership. Since we need some label to identify a phoneme, the obvious and usual thing to do is to represent the phoneme using the phonetic symbol for one of its allophones. Accordingly, for simplicity the phoneme that includes [t] and [t] and [tw] is labelled / t / ; it could equally well be labelled / t w / or, for that matter an arbitrary number like / 1 3 / . But, although in principle there is no significance in the label given to a phoneme, it would be perverse to make a determined effort to avoid mnemonic labels: / t / is preferable to / 1 3 / as the label for the English phoneme which we are discussing.

Some phonemes such as English / m / have a dominant family member which occurs in almost all contexts. In such cases the selection of the symbol of that allophone to represent the phoneme is uncontroversial. Other phonemes such as / t / have no clearly dominant family member and the choice of the symbol of one of the allophones to represent the phoneme is not as easily justified. Neverthe- less, in neither case should the sound chosen to represent the family be equated with the family itself.

The family metaphor is also useful in highlighting another principle of phoneme theory: namely, the

Segments of sound 21

P H O N E T I C SIMILARITY OF ALLOPHONES of the same phoneme. The phoneme is a suitable label for a group of sounds which are phonetically alike and which show certain typical patterns of distribution. This principle is intended to prevent the lumping together as allophones of the same phoneme's sounds which appear to be in comp- lementary distribution but have no phonetic similarity. A classic example of this is provided by [rj] and [h] in English. At first sight, these two sounds appear to be in comp- lementary distribution, with [h] occurring syllable initially followed by a vowel (as in hat and ahead) and [n] occurring in consonant clusters and syllable finally (as in longer and long). But because they lack phonetic similarity, [h] and [rj] cannot be grouped together as allophones of the same phoneme.

What are the articulatory differences which disqualify [h] and [o] from counting as allophones of the same phoneme? Take into account the following articulatory parameters: place of articulation, manner of articulation and the state of the glottis.

Recall from Chapter 1 that [h] is a voiceless glottal fricative and [rj] a voiced velar nasal.

There is an alternative approach to the phoneme which highlights the linguistic function of phonemes in DISTINGUISHING (or CONTRASTING) word meaning rather than their physical phonetic characteristics. The key notion in this approach is C O N T R A S T or DISTINC- TIVENESS. On this view, the phoneme is a minimal sound unit which is capable of contrasting word meaning. As we noted above, although in reality there is an infinite amount of variation in the sounds produced by speakers of a given language, not all these phonetic differences are pertinent. Some objectively noticeable variation in the production of sounds is not used to convey semantic differences. Consider the words in [2.1]:

[2.1] tip ~ dip bet ~ bed pat ~ pad bit ~ bid mate ~ made white ~ wide

22 The phoneme

The words in each pair have different meanings and this difference is signalled by the difference between [t] and [d]. When two sounds can be used to distinguish word mean- ings in a particular language they are said to be separate phonemes. On the basis of [2.1] we can say that in English / t / and / d / are separate phonemes.

Supply examples of English words with different meanings which only differ in that where one word has [1] the other has [r], all the other sounds being exactly the same. Are [1] and [r] separate phonemes in English?

2.2 Identifying phonemes

In this section I shall outline the basic procedures which are used to identify the phonemes, i.e. the functionally signifi- cant segments of a language.

2.2.1 The minimal pair test

When two words are identical in all respects, except for one segment, they are referred to as a MINIMAL PAIR. The pairs in [2.1] are minimal pairs (and so should be the pairs illustrating the contrast between [r] and [1] which you have written down). The MINIMAL PAIR TEST (i.e. the method of determining that a single sound difference distinguishes the meanings of two words) is a key principle of phonemic analysis. Sounds are classified as separate phonemes if they are responsible for a difference in meaning in a minimal pair.

Another way of saying this is to state that sounds are separate phonemes if they C O N T R A S T IN IDENTICAL E N V I R O N M E N T S , i.e. if either sound can occur in a given context and the choice of one or the other does alter the meaning of a word. The forms in [2.2] show minimal pairs in which [r] and [1], [m] and [n] and [k] and [g] contrast in identical environments and are therefore distinct phonemes:

Identifying phonemes 23

[2.2] read ~ lead rice ~ lice r o o m ~ loom m o w ~ k n o w mice ~ nice seen ~ seem buck ~ b u g cot ~ g o t card ~ guard

Before y o u read on, list as m a n y m i n i m a l pairs as y o u can think of w h i c h s h o w that [b] and [m], [s] and [z] and [i] and [e] are separate p h o n e m e s in English or s o m e other language w h i c h y o u k n o w . T o start y o u off, here are s o m e minimal pairs w h i c h s h o w that [b] and [m] contrast w o r d meaning in English: bat ~ mat, bet ~ met, meat ~ beat, robe ~ roam, cub ~ come.

2.2.2 Contrast in analogous environments

Sometimes it is n o t possible to find m i n i m a l pairs contrasting each single p h o n e m e . In such circumstances, the phonologist has to settle for s o m e t h i n g less r i g o r o u s : C O N T R A S T I N A N A L O G O U S E N V I R O N M E N T S . U s i n g this principle, sounds are isolated as belonging to separate p h o n e m e s if they occur in phonetically very similar, t h o u g h n o t identical e n v i r o n m e n t s provided that the differences b e t w e e n t h e m cannot be reasonably a t t r i b - uted to the influence of n e i g h b o u r i n g sounds (see C h a p t e r 5). A d m i t t e d l y , there are b o u n d to be occasional differences of opinion between phonologists as to w h a t constitutes adequate similarity to justify labelling e n v i r o n m e n t s as ' a n a l o g o u s ' . P h o n o l o g i c a l analysis is n o t an exact science. Here, I shall avoid getting bogged d o w n in controversies and simply illustrate the principle of contrast in analogous e n v i r o n m e n t s w i t h an example from E w e , a Ghanaian language.

W e can regard /f/ and / v / as separate p h o n e m e s in E w e because they contrast in analogous e n v i r o n m e n t s in w o r d s like [evlo] 'he is evil' and [efle] 'he split off ([e] symbolises a nasalised [e] v o w e l ) . T h e difference b e t w e e n nasalised [e] on the o n e hand and oral [e] and [o] on the other cannot be the reason for the difference b e t w e e n voiceless [f] and voiced [v] since nasal [e] as well as oral [e] and [o] are all voiced and should n o t affect in different w a y s the voicing of n e i g h b o u r i n g s o u n d s .

24 The phoneme

2.2.3 Suspicious pairs

Since, as we observed above, only those sounds which show considerable phonetic similarity can be grouped together as allophones of the same phoneme, it would be pointless to comb through the entire phonological system of a language in search of minimal pairs and examples of contrast in analogous environments. A practising phonol- ogist can normally safely assume that sounds like [n] and [x] or [1] and [p] which show no phonetic resemblance are distinct phonemes. As a rule, methods for determining whether or not sounds belong to the same phoneme are only employed where SUSPICIOUS PAIRS have been identified, that is to say, pairs of words containing sounds which only differ slightly and which can plausibly be members of the same phoneme.

A few examples should clarify this point. If you look back at the data in [2.2] which were used to illustrate the minimal pair test, you will notice that the sounds in each pair are quite similar and could conceivably be allophones of the same phoneme: [r] and [1] are both voiced alveolar approximants, the main difference between them being the fact that [1] is lateral and [r] is not; [m] and [n] are both voiced nasal stops, only differing in place of articulation - the former is labial and the latter alveolar; [k] and [g] are both velar stops - the only difference between them is that [k] is voiceless which [g] is voiced. Likewise, in the Ewe example, the contrast in analogous environments test is worth attempting because [fj and [v] are both labiodental fricatives, only distinguished by voicing: they could easily be allophones of the same phoneme.

2.2.4 Recapitulation

While admitting that the tacit 'theory' of clear-cut speech- sound distinctions implied by the nature of our alphabet script is physically quite wrong, nevertheless we have gone on to argue that functionally it is largely right: a phonemic transcription will indeed treat a language as containing a small number of discretely-different unitary sounds, even though the phonetician knows that it does not. Many of the physical details of a sound are functionally irrelevant within

Phonological symmetry 25

a given language. Hence sounds that are physically different may be regarded as functionally the same in a particular language. For example, speakers of English who are not linguists are unlikely to be aware of the variety of allo- phones of / t / which we noted above; they are only likely to be interested in knowing whether a word contains / t / as opposed to some other phoneme - they will only want to know whether the word uttered by their interlocutor is cart_ or card, tin or thin etc.

Furthermore, as we shall see presently, functionally it makes sense to think of the changes in the various phonetic variables which make up a sound as happening in an abrupt and synchronised fashion, even though physically they are gradual and they overlap each other. That is to say, although in a word like inn, it is true that the soft palate will begin to be lowered before the tongue makes contact with the alveolar ridge, there is no functional significance in the fact that the latter part of [i] in inn will be progressively nasalised while the [i] of if is wholly oral. The English- speaking hearer will not perceive the difference between the two [i]'s. To such a hearer inn consists of just two sharply- separated sounds: oral [i] followed by nasal [n]. In view of this, psychologically we are justified in thinking of speech as segmented along the time dimension, even though physi- cally the segments blend into one another - the situation is very much analogous to that of cursive handwriting.

2.3 Phonological symmetry

There remains one aspect of the 'tacit alphabet theory' of speech-sound which has not yet been justified even in func- tional or psychological terms, and which indeed cannot be so justified. I refer here to the way that alphabetic transcrip- tion (traditional orthography and phonemic transcription alike) treats successive segments of speech-sound as single atomic elements, transcribed with one symbol each, rather than as bundles of simultaneous elements, each element being a value of a particular phonetic variable. That is: let us forget for a moment that the [n] of inn overlaps the [i], and pretend instead that there is a sharp division between the [i] part of the word and the [n] part; it still

26 The phoneme

remains true that an utterance of [n] is not one unitary action, but the simultaneous realisation of a range of separate phonetic properties. These include: a tongue/palate contact belonging to the family that count as 'alveolar' rather than 'palatal', 'stop' rather than 'fricative' closure, voice, a lowered soft palate that results in a sound that is nasal rather than oral (see [1.4]); and, if we are thinking about dissimilarities not only with other English phonemes but with non-English sounds as well, we would need to add 'pulmonic egressive airstream' and perhaps other features of sound. Clearly, speech sounds are not indivisible atoms.

However, for convenience in reading and writing, it is advantageous to use a code which represents a whole bundle of such phonetic properties using a single symbol, rather than having to represent a word like inn in a two- dimensional notation, with a table showing articulatory parameters as in [2.3]:

[2-3]

Segment 1: [1] Segment 2: [n]

vocal cords: soft palate: place of articulation: manner of articulation: degree of aperture:

vibrating up (not applicable) (not applicable) close etc.

vibrating down alveolar stop (not applicable)

This aspect of alphabetic writing is only a matter of practical convenience and long tradition, and there seems to be no good theoretical justification for it; functionally as well as physically, sounds are 'bundles of phonetic properties' rather than unitary lumps.

One reason for saying that the internal structure of phonemes is in all probability relevant to the way in which they function phonologically has to do with the tendency to SYMMETRY found in the phonological systems of numerous languages.

Consider vowel systems, for instance. It is possible to classify vowel systems into two common categories: TRIANGULAR and Q U A D R A N G U L A R systems. They may be very basic, or more elaborate as you can see in [2.4]

Phonological symmetry 27

[2-4]

Front Back Front Back Front Back unrounded rounded unrounded rounded unrounded rounded

Close i u i u i u Half-close e o e o Half-open e o Open a a a

(e.g. Arabic, (e.g. Spanish, (e.g. Italian, Aleut Eskimo (Alaska) Czech, Burmese, Kikuyu (Kenya) Dyirbal (Australia)) Swahili) Efik (Nigeria))

and [2.5] which illustrate triangular and quadrangular vowel systems respectively.

About the simplest system of vowel phonemes common- ly found (it occurs in Aleut Eskimo and Arabic, for instance) has just three members. Many other languages, such as Spanish, Czech, Burmese, and Swahili, distinguish three degrees of aperture and have a five-member vowel system. Yet others, like Italian, Kikuyu and Efik have four degrees of aperture and a seven-member vowel system. All these systems are triangular: there is a lone low (open) vowel and at least one pair of non-low vowels.

Azerbaijani, British Columbian French, Persian and English exemplify (with different embellishments) the other common type, namely a quadrangular vowel system which may be represented as [2.5]:

[2-5]

Front unrounded Back rounded

Close i u Half-close e o Open a D/O

(rounded/unrounded respectively)

Asymmetrical systems are logically possible but occur less commonly than symmetrical ones. Thus, for instance, four-member vowel systems such as those in [2.6] seem to be rare in comparison with the symmetrical ones.

28 The phoneme

[2.6] (a), i u (b). i e

a E

a a (Cocopa (Arizona, U S A )) (Marshalese,

(Marshall Islands))

Likewise, c o n s o n a n t systems also tend to be s y m m e t r i - cal. If w e consider j u s t stop consonants, for simplicity, English has:

[2.7] bilabial alveolar velar voiceless p t k voiced b d g

while Czech has:

[2.8] bilabial alveolar palatal velar voiceless P t c k voiced b d j g

A system like that of Siriono, a Bolivian language, w h i c h has holes in the pattern is less c o m m o n . In Siriono there are n o voiced alveolar, velar and palatalised velar stops c o r r e - s p o n d i n g to voiceless stops at these places of articulation.

[2.9] bilabial alveolar palatalized velar velar

voiceless p t kJ k voiced b

If a language uses a particular phonetic p r o p e r t y (such as ' b a c k ' or ' s t o p ' or 'voiced') at all, rather than use it to m a k e a o n e off contrast, it will n o r m a l l y exploit it fully by e m p l o y i n g it t o form several p h o n e m e s . C o n s e q u e n t l y phonological systems tend to be s y m m e t r i c a l . Hence, s y m m e t r y is an o u t w a r d sign of the u n d e rl ying dimension of contrast in p h o n o l o g y .

A n o t h e r c o m m o n factor e n c o u r a g i n g phonological s y m m e t r y has to d o w i t h permissible c o m b i n a t i o n s of p h o n e t i c properties in particular languages. In the v o w e l systems surveyed a b o v e , all the vowels w e r e either front and u n r o u n d e d (spread) or back and r o u n d e d . In all the languages in o u r examples front vowels m u s t be u n r o u n d e d

Phonological symmetry 29

and back vowels (with the exception of [a](/) must be rounded. The prohibition of the combination of rounded- ness with frontness and backness with unroundedness in vowels is by no means universal. There exist languages such as French and Swedish in which it is permissible for front vowels to be rounded and other languages such as Ivatan, spoken in the Philippines, and Vietnamese in which back vowels may be unrounded.

Thus, in French, the vowel in words like tu, vu and lu which is represented by the phonetic symbol [y] is both front and round. And again, as you can see in [2.10], this combination of frontness and roundedness is not restricted to the close vowel [y] but is also found with other degrees of aperture (other than fully open - the reason for this gap being that the physical distinctions between spread and rounded lip-position are very slight for fully-open vowels, so that lip rounding is seldom used contrastively for such vowels in most languages):

[2.10] Unrounded Rounded Front Front Back

i (as in six) y (as in tu) u (as in fou) e (as in ete) 0 (as in peu) o (as in peau) e (as in pere) ce (as in ocuf) 0 (as in monopole)

In [2.10] I have not shown the degree of aperture. Before you proceed, rank the vowels on a scale indicating the relative height of the highest point of the tongue during the production of each one of these vowels. You may use [2. n ] to check your answer.

I do not mean to exaggerate the extent to which languages exhibit P H O N O L O G I C A L SYMMETRY, it is only a tendency, and often a chart of phonemes will include one or two gaps which spoil the symmetrical pattern. For instance, among the consonants listed above for Czech, / g / is quite marginal, being found only in words borrowed from other languages; in this case, one might be tempted to argue that it was the existence of this gap in an otherwise balanced pattern that enabled words containing [g] to be borrowed into Czech without this sound having to be

30 The phoneme

modified in the way that foreign sounds that do not exist in the borrowing language are normally modified.

This account might look plausible for Czech, but it would not do in those situations where asymmetry is unre- lieved, even by borrowing. Thus, while French permits front rounded vowels although they are unusual, Viet- namese forbids them but permits back spread (unrounded) ones which are no less unusual. Interestingly, there are only two back unrounded vowels in Vietnamese where symmetry would lead one to expect three:

[2.11] Unrounded Rounded Front Back Back

Close i ui u Half-close e o Half-open Open e (A) 0 Open a

There 'ought' to be a phoneme / A / (I have filled it in and circled it in the chart), but in fact, Vietnamese has no such phoneme, not even in borrowed words.

Such counter-examples notwithstanding, the import- ance of phonological symmetry as a tendency rather than an absolute rule is very well established. Yet, if phonemes were regarded as unitary, indivisible atoms, with their internal phonetic make-up irrelevant to their functioning, this tendency would seem quite inexplicable. If a language, say English, were to possess twenty-four consonant phonemes, one would suppose that any selection of twenty- four consonant sounds would do as well as any other, and it would be a remarkable coincidence if the twenty-four actually used in the language manifested any particular symmetrical patterning. It would be even more remarkable if the types of symmetrical patterns accidentally stumbled upon by one language, say English, just happened to be the same as those stumbled upon by numerous unrelated languages. As we have seen, this is indeed the case: the symmetrical patterns found in the phonological systems of the languages of the world appear to come from a very restricted range.

If, on the other hand, instead of regarding phonemes as indivisible, atomic entities we think of phonemes and the phonemic symbols that represent them as merely a short- hand way of indicating bundles of distinctive phonetic

Phonological symmetry 31

features, then the apparently mysterious recurrence of the same parameters in phonological symmetry can be explained. Members of a language-community have to learn to produce and perceive the psychological 'target-value' of various phonetic dimensions which their language uses to distinguish word meanings: for instance, the Arabic speaker has to learn to make two distinctions on the dimension of vowel aperture and the Spanish speaker three.2 For a language to be phonologically symmetrical, with simple rules for combining phonetic properties, is a way of optim- izing the use of phonetic parameters; it is a way of getting a relatively small number of different bundles of phonetic properties to do the job of distinguishing word meanings.

Referring to the International Phonetic Alphabet in [1.13] on page 31, study the following consonants systems and write down your answers to the questions that follow:

[2.12] (a) b n x t f g z (b) p r j t x d m k g b y n f v s z

Which of these two systems is symmetrical? What advan- tages has the symmetrical system got over the asymmetrical one?

I hope you chose the system in [2.12b] as the symmetri- cal one. There, every stop has a corresponding fricative, every voiced oral stop has a corresponding nasal one, and every voiced stop and fricative has a corresponding voiceless one:

[2.13] labial alveolar velar voiced voiceless voiced voiceless voiced voiceless

stop b p d t g k fricative v f z s y x nasal m n n

The system in [2.12a] is the asymmetrical one. As [2.14] shows it is riddled with holes:

[2.14] labial alveolar velar voiced voiceless voiced voiceless voiced voiceless

stop b t g fricative f z x nasal n

32 The phoneme

The consonant inventory in [2.14] is extremely un- economical, using, as it does, three places of articulation, three manners of articulation and two values of the voicing variable (eight phonetic parameters in all) to define a mere seven consonants. O n the other hand, the consonant inven- tory in [2.13] represents a much better use of resources: a consonant system with fifteen sounds is created using the same eight parameters.

Furthermore, you should note that while the asym- metrical inventory can be defined only by imposing complicated regulations to ensure that only segments with permissible combinations of phonetic properties are gener- ated (e.g. in [2.14] only the labial and velar stops, the alveolar fricative and nasal can be voiced), no such ad hoc restrictive statements are needed for the symmetrical system.

One probable consequence of this is that a symmetrical system is easier to learn than an asymmetrical one - even when, as in our example the latter contains fewer elements. Predictably, for the most part phonological systems tend to be symmetrical, using in an optimum way phonological properties such as voice, alveolar and fricative which speech- sounds are made up of.

As we have already noted, it is unusual, but not unheard of, for a language to have an asymmetrical conso- nant system like the one in [2.14]. An example is given below in [2.15] of such a skewed consonant system in Palauan, a Filipino language:

[2.i$]Palauan consonants t k •?

b 5 s

m rj

1 r

We shall conclude this chapter by emphasising the point that while the practical convenience of alphabetic tran- scription is so great that it is used much of the time even in the theoretical study of phonology, we should nonethe- less bear in mind the fact that such a transcription is no more than a practically convenient short-cut. What is

Phonological symmetry 33

'really' happening in phonology is happening in terms of separate phonetic properties, not of 'unitary sounds'. It is these phonetic properties rather than phonemes that are the basic building blocks of phonology. Before we delve any deeper into phonology we shall explore the nature of these phonetic and phonological properties. That is the subject of the next chapter.

Exercises

1. The pairs of consonants listed below are separate phonemes in English. Provide minimal pairs showing how these sounds contrast word meaning.

word initially word finally /p b / e.g. ^at feat cup cub / b m / ~ - As/ / z s / / P t / / d n / A g / AJV AW AJ d3/ / I d /

2. In French, the following pairs of sounds are in C O M P L E M E N T A R Y DISTRIBUTION: [m m], [1 1], [r r]. Study the data below and identify the phonetic context in which the voiced and voiceless member of each pair occurs. State the rule that governs the distri- bution of each sound.

maternite mekonetr matinal preta:dr prezatabl metr tordr ratabl i f el

maternity' to fail to recognise' morning (adj.)' to claim as a right' 'presentable' to put' to wring' profit-earning' island' she'

34 The phoneme

Poepj tapl 5:kl tabl rym rymatism film " r5fle kasabl rali Je:r Jifr limite me:r pers etr ekri:r term li:r levr

people' temple' uncle' table' cold (with a running rheumatism' film' to snore' breakable' race-meeting' fare, living' number, figure' limited' mother' Persian' to be' to write' term' to read' »P'

nose)'

Notes

1. There exists a very rich literature on the phoneme. Key works include Bloomfield (1926, 1933), Sapir (1925, 1933), Jones (1931), Bloch (1941), Pike (1947), Hockett (x955) and Trubetzkoy (1939). Standard textbooks like Hyman (1975: 59-98) and Sommerstein (1977: 16-53) also contain good introduction to the literature. To those who want a more detailed account of the devel- opment of phoneme theory (and phonological theory in general), I recommend Fischer-J0rgensen (1975) and Anderson (1985).

2. This view is held by those who regard the phoneme as a psychological entity. Badouin de Courtenay is probably the best known holder of this view. He defined the phoneme as 'a mental reality, as the inten- tion of the speaker or the impression of the hearer or both' (Twaddell 1953: 56) (see page 67 below).

CHAPTER 3

Distinctive features

3.1 Why are features needed?

We saw at the end of Chapter 2 that phonological systems tend to be symmetrical and that a limited number of phonetic parameters, taken from a fairly small universal set recur in a variety of combinations in different languages. It makes sense to look beyond the phoneme and focus on those basic phonological ingredients, called DISTINCTIVE FEATURES, which phonemes are made of.

Besides introducing you to distinctive features, this chapter will also serve as a restatement in a slightly different form of the principles of phonetics which were outlined in the first chapter. There is nothing mysterious about the fact that there is a relatively small inventory of phonetic features from which languages select different combinations to construct their individual phoneme systems. As all members of the human race are endowed with very similar articula- tory and auditory capabilities, it is only to be expected that they will only be able to produce and utilise speech sounds built up from the set which is pre-determined by their biological endowment.

The position presented in this chapter regarding distinctive features has not always been accepted by all students of phonology. At one time some scholars, like Bloomfield (1926), would not agree with the claim that the phoneme is not the most basic phonological element. Bloomfield thought that there were no phonologically relevant particles more elementary than the phoneme, although he was aware that on purely phonetic grounds the phoneme can be decomposed into more basic particles.

The belief that the phoneme is the atomic, basic

36 Distinctive features

building block of p h o n o l o g y is untenable, n o t only because of the a r g u m e n t s w h i c h h a v e already been advanced concerning s y m m e t r y , b u t also for other reasons which I e x p l o r e b e l o w .

Firstly, even a cursory inspection of the p h o n o l o g y of any language will reveal that the phonological behaviour of p h o n e m e s is largely d e t e r m i n e d by the p h o n e t i c features w h i c h they are m a d e u p of. C o n s i d e r the b e h a v i o u r of / r / in English. It u n d e r g o e s partial devoicing w h e n i m m e d i - ately preceded b y voiceless stops but n o t w h e n it is preceded b y voiced ones:

[3.1] pray [ p r e i ] brain [brem] train [trem] drain [drem] crane [krem] grain [grem]

If y o u look at it in t e r m s of the i m p l e m e n t a t i o n of distinct articulatory gestures, it is clear that the signal to start voicing is delayed until well after the b e g i n n i n g of the / r / d u e to imprecise adjustment of the articulatory apparatus in the transition from one s o u n d to the next. L o o k i n g at sounds in t e r m s of the individual parameters w h i c h they consist of allows an insightful expression of A S S I M I L A T I O N P R O C E S S E S , i.e. phonological processes w h e r e b y one s o u n d changes to b e c o m e m o r e like s o m e o t h e r sound in its e n v i r o n m e n t (see C h a p t e r 5). By highlighting each ar- ticulatory parameter and singling o u t the s u b - p h o n e m i c particles (i.e. distinctive features) w h i c h p h o n e m e s are m a d e up of, this approach is capable of treating assimilation insightfully as an instance of S P R E A D I N G of distinctive features.

E x a m i n e the A m e r i c a n English data in c o l u m n s A and B in [3.2] and suggest reasons w h y a distinctive feature approach to these data is preferable to an indivisible, a t o m i c p h o n e m e approach:

[3-2] A B / p a d / [paed] ' p a d ' / p a n / [psen] ' p a n ' / p a s / [pass] 'pass' / p a m / [paem] ' P a m ' / p a k / [paek] 'pack' / p a g / [paerj] ' p a n g '

Why are features needed? 37

When solving [3.2], you will have discovered that if you treat phonemes as unanalysable entities, you have no straightforward way of showing that the vowel only assim- ilates the property of nasality from the following consonant if that consonant is nasal as in [3.2B]. The SPREADING of nasality to the preceding vowel is due to the premature, anticipatory lowering of the velum, as the vowel is being produced, to let air escape through the nose during the articulation of the nasal consonant. Distinctive features thus facilitate the statement of assimilation processes by high- lighting the various separate gestures involved in the production of speech. No equally natural way of stating assimilation processes is available if phonemes are treated as unanalysable units.

An added advantage of the feature approach is that it enables us to highlight the internal structure of a sound. When we do that, it soon becomes obvious that p h o n o - logical segments have internal structure. Sounds are not bundles of unordered, unstructured phonetic properties. A simple example like [3.3] shows that distinctive features can be arranged one before another within a single phoneme:

[ 3 3 ]

[non-nasal] [nasal]

[p * n] _

Whereas [p] is wholly oral and [n] is wholly nasal, the vowel [se] occurring between them is oral to begin with but subsequently becomes nasalised in anticipation of the following nasal consonant. The properties NASAL and N O N - N A S A L occur together in sequence in the same phoneme.

Similar evidence of internal structure is to be found in diphthongs. In words like way [wei] and why [wai] the vowel sound has two distinct vowel qualities, a fact which is reflected in this case in the way in which the two phases of the sound are transcribed. The same point can be made about affricates like [ts], [dz], [tj] and [d3]. They are

38 Distinctive features

composite consonants starting with a stop phase and ending with a fricative phase. In many languages, affricates behave in part as though they were stops and in part as though they were fricatives. Thus we can see that the phoneme is not an indivisible phonological unit.

If we revisit the American English example above, a further reason for the espousal of a feature approach should become apparent: distinctive features bring out the fact that, in general, phonological rules apply to N A T U R A L CLASSES of sounds i.e. sounds which share certain phonetic properties. Thus, it would be bizarre for any language to have a rule nasalising vowels before the following assortment of consonants: [d s k t h], This is a ragbag of sounds which are phonetically very different from each other and which all lack the crucial property of being nasal which is passed on to an adjacent vowel during nasal- isation. The chances of such an arbitrary nasalisation rule existing in any language are extremely remote. O n the other hand, a rule which nasalises vowels in the neighbourhood of nasal consonants (as in [3.2]) is phonetically plausible and is found in numerous languages. The nasal consonants [m n rj] which condition the nasalisation form a natural class and they all contain the crucial feature of nasality which triggers off the nasalisation of the preceding vowel. Normally, sounds which are phonetically similar display similar phonological behaviour. In order to state the basis of the similarity between a group of phonemes, it is necessary to penetrate beyond the phoneme and scrutinise the phonetic features which they share.

3.2 Jakobsonian features

Current distinctive feature theory has its roots in the work of the Russian scholars Trubetzkoy and Jakobson who were based in Prague between the wars. When the Nazis occu- pied Czechoslovakia, Jakobson fled to New York and later moved to Harvard where he introduced some of the ideas of the 'Prague School' to the Americans.

The key publications are Trubetzkoy (1939) and Jakobson, Fant and Halle (1952). Trubetzkoy was mainly interested in devising a system of classifying the phonemic

Jakobsonian features 39

OPPOSITIONS (i.e. contrasts) in the use of common phonological parameters like voicing and aspiration. His approach was TYPOLOGICAL, i.e. concerned with the classification and comparison of the sound systems of different languages in order to show that there is a limited number of ways in which phoneme inventories can be organised. He regarded the exploration of the constraints on the structuring of phonoloigcal systems as one of the tasks of the phonologist

To take one example, Polish and English have only two bilabial stops, namely / p / and / b / ; the two sounds are almost identical except for the fact that the latter has the additional property or 'mark' of being voiced. It can there- fore be referred to as being MARKED while the former, which lacks that additonal property, can be said to be U N M A R K E D . Note that the question of MARKEDNESS only arises when two sounds are in a two-way opposition. In the case of a multi-dimensional opposition like the place of articulation contrast between / p t k/, where no one sound is in any sense more basic than the others, the ques- tion of markedness does not arise.

Significantly, if in a particular language a phonemic contrast is NEUTRALISED or SUSPENDED, i.e. if a phonemic contrast generally observed is not found in a given environment, it is invariably a two-way opposition that is involved and normally it is the unmarked member of such an opposition that occurs in the PLACE OF NEUTRALISATION. Thus, in Polish / k / and / g / are distinct phonemes and contrast between vowels in minimal pairs like [ro^i] 'horns' and [sofei] 'juices'. But, in word-final position, [k] and [g] do not contrast. There, the opposition between them is neutralised or suspended and only the voiceless [k] occurs. That is the situation in the singular form of the noun : 'horn' and 'juice' are pronounced as [rufe] and [sofe] respectively (Kenstowicz and Kisseberth 1979:72-3). Neutralisation is one of the key concepts in phonology. We shall come back to it in Chapter 8.

While Trubetzkoy was primarily concerned with phono- logical typology, Jakobson and his collaborators concen- trated their investigations on phonological oppositions that occur UNIVERSALLY. Jakobson hypothesised that although languages show an almost infinite amount of phonetic

40 Distinctive features

variation, the range of phonemically contrasting features is severely restricted by universal principles. The initial proposal was that just a dozen acoustically defined pertinent contrasts (other than prosodic ones, involving such things as tone and stress which are usually associated with an entire syllable or word) would be found in all languages.

Jakobson and his co-workers further hypothesised that the presence of certain oppositions in a language precludes the existence of other oppositions. For instance, they suggested that no language phonemically contrasts labialised consonants (i.e. consonants produced with rounded lips) with pharyngealised consonants (i.e. consonants made with a constriction in the pharynx). Although labialization and pharyngealisation are distinct as far as articulation is concerned, they are phonologically merely implementations of the same acoustic distinctive feature FLAT.1

Another aspect of the Jakobsonian feature system was its BINARISM. Jakobson et al. (1952) insisted on a binary interpretation of all features. This was done by pushing the phonemicist's principle of distinctiveness to its logical conclusion: in order to distinguish between the meanings of words, they argued, what counts is either the presence or absence of a given feature (respectively indicated by a ' + ' or ' —' before the feature in question). For example, to discriminate between the words in [3.4] what is crucial is the presence or absence of voicing in the first segment:

[3. 4][+ voice] [— voice] bet pet den ten ghoul cool zinc sink

Admittedly, binarism works well where there exists a two-way opposition, but it does not yield entirely satisfac- tory results where multilateral contrast is involved. Take the case of a language which distinguishes four degrees of vowel height and has the vowels [i e e a] or a language that distinguishes four places of articulation and has the bilabial, alveolar, palatal and velar nasal consonants [m n ji rj]. In neither case can a binary characterisation of opposition be employed without some degree of arbitrariness, or at the

Jakobsonian features 41

very least, without obscuring the fact that what is involved is a cline rather than a dichotomy.

Binarism has remained controversial and there has developed a powerful lobby (Ladefoged 1971, 1975; Venne- mann 1972) for multivalued features where GRADUAL OPPOSITIONS (like those found in vowel contrasts) or MULTIVALUED OPPOSITIONS (involving say, places of articulation) are concerned. Nevertheless, the dominant view among phonologists, and the the view adopted in this book, is that although the case for multivalued features is not without merit, from a phonological point of view, distinctiveness is the paramount consideration and its inter- ests are best served by binary distinctive phonological features; for on every occasion, the hearer has to make a decision as to whether the speaker has said either this word or that word. True, the phonetic features which specify the phonetic realisation of a given feature, say voicing of stops in English, could be put on a multivalued scale like:

[3.5] [d]: [3 voice] as in ladder where [d] is surrounded by voiced segments

[d]: [2 voice] as in date where [d] occurs at the beginning of a word

[d]: [1 voice] as in laid where [d], which is word final, is almost completely devoiced

Phonetically, English voiced stops are fully voiced between vowels; less fully voiced word-initially and minimally voiced word-finally. In principle we could indicate the amount of voicing by using an infinite number of points on the voicing scale. Phonologically, however, this kind of low level detail is overlooked. Thus, for instance, although acoustically the final sound in a word like late may be virtually identical to the final consonant of laid, the two sounds are phonologically different: the former belongs to the / t / phoneme and the latter to the / d / phoneme. Phonetic features may be multivalued but phonological features need not be. A binary distinctive feature system is adequate because, what is relevant linguistically, is simply whether or not the sound uttered by a speaker is a realis- ation of phoneme / t / or / d / . As a rule, it is a dichotomous choice.

42 Distinctive features

3.3 The SPE system of distinctive features

Various shortcomings of the Jakobsonian features came to light in the 1950s and 1960s. It was discovered that the model was too parsimonious. The dozen or so features which it allowed were insufficient to account for all phono- logical contrasts found in the languages of the world. Furthermore, it was criticised for using the same phono- logical feature to characterise phonological oppositions which in some cases were manifested by different phonetic properties. For instance, if a sound was described as GRAVE (which means that most of the acoustic energy used in its production is concentrated in the lower part of the spectrum) you would not be able to tell whether it was a labial like [p], or a velar like [g], since the acoustically defined phonological property GRAVE could be correlated with either labial or velar articulation.

Because of these and other inadequacies, Chomsky and Halle (1968) in their book The Sound Pattern of English (henceforth SPE) proposed a major revision of the theory of distinctive features. They replaced acoustically-defined phonological features with a set of features that have, in most cases, articulatory correlates. Furthermore, the number of features was also substantially increased. But, like their original Jakobsonian precursors SPE features remain binary. They have only two coefficients or values, plus (+) indicating the presence of a feature and minus (—) its absence, so that, for example, among other things, a sound like [p] is said to be [—voice] and [ — nasal] while [m] is [ + voice] and [ + nasal].

The list of distinctive features given below is based on SPE in the main, but it incorporates some of the modifi- cation that have been proposed since 1968.

It is not important to 'master' all the details of distinc- tive feature theory at this stage. They are described mainly in order to show how the system works. But you should come back to this chapter ot refresh your mind as the need to use features arises in later chapters.

3.3.1 Major class features

The major class features define the major classes of sounds

The SPE system of distinctive features 43

that are relevant in phonological analysis. The major classes include C O N S O N A N T S and N O N C O N S O N A N T S , SYLLABICS and NONSYLLABICS, S O N O R A N T S and N O N S O N O R A N T S (OBSTRUENTS).

i. C O N S O N A N T A L - N O N C O N S O N A N T A L [± cons]

Consonantal sounds are produced with a drastic stricture along the centre-line of the vocal tract; nonconsonantal sounds are made without such obstruction.

Obstruents, nasals and liquids are consonantal; vowels and glides" are nonconsonantal.

2. SYLLABIC - NONSYLLABIC [± syllabic] Syllabic sounds are sounds which function as syllable nuclei; nonsyllabic sounds occur at syllable margins. Normally, syllabic sounds are auditorily more salient than adjacent nonsyllabic sounds. Vowels are syllabic and so are syllabic consonants such as [J] in bottle and candle or the nasal [n] in cotton and [m] in bottom.

3. S O N O R A N T - N O N S O N O R A N T (OBSTRUENT) [± sonorant]

Sonorants are produced with a vocal cavity disposition which makes spontaneous voicing easy while nonsonorants (obstruents) have a vocal cavity disposition which inhibits spon- taneous voicing. In other words,the unmarked (normally expected and natural) state for sonorants is to be voiced, while for obstruents the unmarked state of affairs is to be voiceless.

Vowels, nasals and liquids are sonorant; stops, fricatives and affricates are obstruents.

3.3.2 Cavity features

These features refer to place of articulation. They specify where in the vocal tract modifications of the airstream take place in the production of particular sounds.

44 Distinctive features

4. C O R O N A L - N O N C O R O N A L [± coronal] To produce a coronal sound, the blade of the tongue is raised towards the front teeth, the alveolar ridge or the hard palate; for noncoronal consonants the blade of the tongue remains in a neutral position.

Dental, alveolar, alveo-palatal, retroflex and palatal sounds are coronal; labial, velar, uvular and pharyngeal consonants are noncoronal.

5. ANTERIOR - N O N A N T E R I O R [± anterior] In the production of anterior sounds, the main obstruction of the airstream is at a point no farther back in the mouth than the alveolar ridge; for nonanterior sounds the main obstruction is at a place farther back than the alveolar ridge.

Labials, dentals and alveolars are anterior while all other sounds are not.

6. LABIAL - NONLABIAL [± labial] A sound is labial if it has a stricture (narrowing) made with the lips; if there is no such stricture, the sound is nonlabial.

In the literature the alternative feature R O U N D is often used to refer to many of the sounds which can also be described as labial. Rounded sounds are produced with a pursing or narrowing of the lip orifice. There is a considerable degree of overlap between the groups of sounds covered by the features [ + round] and [+labial]. Rounded sounds like [o] [u] and [w] are a subset of labial sounds; consonants like [p b m] are labial but not round. Labial sounds include bilabial and labiodental consonants as well as rounded vowels. All other sounds are nonlabial.

7. D I S T R I B U T E D - N O N D I S T R I B U T E D [ ± distributed]

Distributed sounds are made with an obstruction extending over a considerable area along the middle-line of the oral tract; there is a large area of contact between the articulators. In nondistri-

The SPE system of distinctive features 45

buted sounds, there is a smaller area of contact. T h i s feature is primarily used to distinguish A P I C A L sounds from L A M I N A L sounds. In apical sounds the tip of the t o n g u e makes contact w i t h the front teeth or the alveolar or alveo- palatal regions of the r o o f of the m o u t h while in laminal sounds it is the blade of the t o n g u e that makes contact w i t h those same areas. This feature also distinguishes labial from labiodental s o u n d s .

T h e following sound types are distributed: bilabial fricatives like [(j) 6] (lamino-)alveolar fricatives like [s z] (lamino-)alveo-palatal fricatives like [J 3].

T h e following sound types are n o n d i s t r i b u t e d : labiodental fricatives like [f v] (apico-) dental fricatives like [0 3] retroflex fricatives like [§ z j .

3.3.3 Tongue body features

In SPE the neutral position of the b o d y of the t o n g u e is said to be the position w h i c h it assumes in the p r o d u c t i o n of a mid front v o w e l . O t h e r t o n g u e configurations are regarded as departures from that n o r m .

8. H I G H - N O N H I G H [ ± high] H i g h sounds are m a d e w i t h the t o n g u e raised from neutral position while n o n h i g h sounds are m a d e w i t h o u t such raising of the b o d y of the t o n g u e .

H i g h sounds include v o w e l s like [i u], the glides [w j ] , alveo-palatal, palatalized, palatal and velar consonants. All o t h e r sounds are n o n h i g h .

9. L O W - N O N L O W [ ± l o w ] L o w sounds are p r o d u c e d w i t h the t o n g u e depressed and lying at a level b e l o w that w h i c h it occupies w h e n at rest in neutral position; n o n l o w sounds are p r o d u c e d w i t h o u t depressing the level of the t o n g u e in this m a n n e r .

46 Distinctive features

Open vowels like [a a D] are low and so are the pharyngeal consonants [h] and [?]. All other sounds are nonlow. (MID vowels are both N O N H I G H and N O N L O W . )

10. BACK - N O N B A C K [± back] Sounds produced with the body of the tongue retracted from neutral position are back. Sounds produced with the body of the tongue either in neutral position or pushed forward are nonback.

This feature distinguishes between back vowels like [u o a] and front vowels like [i e e]. (Note that because in SPE the position of the tongue in the production of a mid front vowel is taken as the neutral position, it is not only back vowels that are [+ back], but also central ones like [4- a-tt]. (Because the feature system is binary, there is no way of showing that certain vowels are neither front nor back; central vowels are grouped together with back vowels. This does not appear to be entirely satisfactory on purely phonetic grounds.) Of the consonants, velars, uvulars and pharyngeals are back while labials, dentals, palatals as well as glottals are nonback.3

I I . VELAR S U C T I O N - NONVELAR S U C T I O N [± velar suction]

This feature is used to characterise CLICK S O U N D S like the Tut tut\ used in English to show disapproval.

While in English clicks are not fully-fledged speech sounds which can combine with other sounds to form words, in some languages (which happen to be almost exclusively found in Southern Africa) clicks are used as regular speech sounds.

The production of a click involves two crucial phases. First there must be a stricture made using the back of the tongue and the velum; this is called VELIC C L O S U R E . Secondly there must be some constriction in the labial, dental, alveolar or alveo-palatal area. The

The SPE system of distinctive features 47

feature [ + velar suction] itself describes the pulling back of the b o d y of the t o n g u e which creates a partial v a c u u m b e t w e e n the velic closure at the back and the forward closure, say at the alveolar ridge. T h e resulting sound is a click.

Clicks are produced w i t h an I N G R E S S I V E V E L A R I C A I R S T R E A M M E C H A N I S M (which is different from the m o r e c o m m o n p u l m o n i c egressive airstream m e c h a n i s m described in C h a p t e r i which is used in English). T h e soft palate and the back of the t o n g u e are the ini- tiators of the m o v e m e n t of air and the direction of the airflow is i n w a r d .

Clicks are velar suction s o u n d s ; other sounds are not. Z u l u has a dental click [j], an alveolar click [rj] and an alveo-palatal click [c].

3.3.4 Tongue root features

The vocal tract is a l o n g tube with holes at b o t h the lip end and the throat end. T h e shape of this tube can be modified by r o u n d i n g the lips and m a k i n g t h e m p r o t r u d e - and thus elongating the tube. Alternatively, the t o n g u e r o o t position can be adjusted by p u s h i n g it forward or retracting it so that the vocal tract is either lengthened or shortened. Either of these actions has the effect of modifying the shape of the resonating c h a m b e r in the vocal tract in m u c h the same way as differences in size and shape of w i n d i n s t r u m e n t s affect the notes w h i c h they p r o d u c e .

12. A D V A N C E D T O N G U E R O O T - N O N A D - V A N C E D T O N G U E R O O T [ ± A T R ]

T h e t o n g u e root is pushed forward in the p r o d u c t i o n of advanced t o n g u e r o o t sounds, thus e x p a n d i n g the resonating c h a m b e r of the p h a r y n x and possibly p u s h i n g the t o n g u e b o d y u p w a r d ; if the t o n g u e root is n o t advanced, it remains in a neutral position.

V o w e l s like [i e o] in m a n y West African languages are m a d e with the t o n g u e root pushed forward while [i c o] are m a d e w i t h the t o n g u e r o o t in neutral position.

48 Distinctive features

13. TENSE - LAX [± tense] The validity of the feature TENSE has always been controversial. I quote Halle and Clements (1983:7) who give this cautious definition: 'Tense vowels are produced with a tongue body or tongue root configuration involving a greater degree of constriction than that found in their lax counterparts; this greater degree of constriction is usually accompanied by greater length. (Tense vowels vs. lax vowels.) We note that this feature and the last [ATR] are not known to co-occur distinctively in any language and may be variant implementations of a single feature category.'

The English 'long' vowels and diphthongs [i, a, 0, u, ei, ai, oi, au, ju] are tense while the 'short' vowels [1, e, se, u, A, D] are lax.

3.3.5 Laryngeal features

14. SPREAD GLOTTIS - N O N S P R E A D GLOTTIS [± spread]

Pushing the vocal cords wide apart augments the airflow through the glottis and inhibits voicing. This gesture, which is associated with voiceless- ness and aspiration, is absent in nonspread sounds.

Spread sounds include aspirated stops; murmured and breathy voice sounds, voice- less vowels and voiceless glides. All other sounds are nonspread.

15. C O N S T R I C T E D GLOTTIS - N O N C O N S T R I C T E D GLOTTIS [± constr]

Constricted sounds are GLOTTALISED. They are produced with a severe obstruction of the glottis which is made using the vocal cords. This inhibits or prevents the free vibration of the vocal cords. No such gesture occurs in the production of nonconstricted sounds.

Constricted sounds include implosives, ejec- tives, glottalised and laryngealised consonants

The SPE system of distinctive features 49

as well as creaky voice and glottalised vowels and glides. All other sounds are nonconstricted.

In the opening chapter of this book we observed that English only uses speech sounds made with the pulmonic airstream mechanism, but there are languages which employ additional mechanisms. Earlier in this chapter (see section 3-3-3) y ° u wiH recall that I remarked on the use of clicks in many Southern African languages when another airstream mechanism, namely the velaric airstream mechanism was introduced. Here I shall briefly describe the remaining airstream mechanism found in languages. It is called t h e G L O T T A L I C A I R S T R E A M MECHANISM.

To understand how it works, I suggest that you begin by making a big constriction in your throat - pretend that you are just about to cough. Get your vocal cords firmly together and shut off the glottis. At this stage, an upward movement of the larynx will push out the air above the larynx while a downward movement of the larynx will suck in air and loosen the constriction; the vocal cords will vibrate vigorously as the air flows in.

Stops made with the GLOTTALIC AIRSTREAM MECHANISM, with the air above the larynx being expelled, are called EJECTIVES. Ejectives are voiceless. An example of an ejecdve is the [t'] in Tlingit, a language of Alaska, in the word [t'ij] 'elbow'.

Stops made with the GLOTTALIC INGRESSIVE AIRSTREAM MECHANISM are called IMPLOSIVES. The air drawn into the larynx as it is pulled downward normally causes heavy voicing. An example of this is the bilabial stop, in the word [6a6a] 'father' in Shona (Zimbabwe).

16. VOICED - VOICELESS [± voice] Voiced sounds are produced with the vocal cords vibrating at regular intervals; voiceless sounds are produced without such periodic vibration.

50 Distinctive features

3.3.6 Manner features

T h e s e characterise the w a y in which the airstream is obstructed in the p r o d u c t i o n of a consonant.

17. C O N T I N U A N T - N O N C O N T I N U A N T [ ± c o n t ] C o n t i n u a n t s are p r o d u c e d by i m p e d i n g , but n o t completely blocking, the flow of air t h r o u g h the glottis, or the p h a r y n x or t h r o u g h the centre of the oral tract; n o n c o n t i n u a n t s are m a d e by completely blocking the flow of air t h r o u g h the centre o f the vocal tract.

Affricates, nasals and oral stops and laterals are noncontinuant. All other sounds are continuant.

18. L A T E R A L - N O N L A T E R A L [ ± lateral] A lateral s o u n d is produced if the airflow t h r o u g h the centre of m o u t h is blocked and air only escapes over o n e or b o t h sides of the t o n g u e . In nonlateral sounds air flows o u t t h r o u g h the centre of the m o u t h . T h e English [1] is an e x a m p l e of a lateral L I Q U I D . (The t e r m 'liquid' is conventionally used to refer to T and 'r'-like sounds.)

Languages m a y have lateral s o n o r a n t s , frica- tives and affricates m a d e at various places of articulation.

19. N A S A L - N O N N A S A L ( O R A L ) [ ± nasal] In the p r o d u c t i o n of a nasal s o u n d the v e l u m is lowered to allow air to escape t h r o u g h the nasal cavity. O r a l sounds are produced w i t h the velum raised so as to block access to the nasal cavity and to allow air to g o o u t only t h r o u g h the m o u t h .

Nasal sounds include nasal stops like [m n H, ji rj] (which are m a d e with c o m p l e t e blockage of air at the place where the articulators meet) as well as nasalised consonants, glides and v o w e l s . All o t h e r sounds are oral.

20. S T R I D E N T - N O N S T R I D E N T [ ± strident] O n l y fricatives and affricates can be strident. Acoustically, strident sounds are characterised by

The SPE system of distinctive features 51

m o r e r a n d o m noise than their n o n s t r i d e n t c o u n t e r p a r t s .

In the SPE system, w h e r e features are generally defined in articulatory t e r m s , the feature [strident], w h i c h is acoustically-defined is different from the rest - it is a relic retained from the original J a k o b s o n i a n s y s t e m . Fricatives w h i c h have high pitched strident noise are referred to as S I B I L A N T S . T h e feature stri- dent distinguishes fricatives as follows:

Sibilant

n o n s i b i l a n t

[ + s t r i d e n t ]

Voiceless voiced

s

ts t j f Pf X

z

dz

d 3 V

bv B

[-strident]

Voiceless voiced

6 6 c J

2 i . D E L A Y E D R E L E A S E - I N S T A N T A N E O U S R E L E A S E [ ± del rel]

This feature is only applicable to sounds p r o d u c e d in the m o u t h cavity and distinguishes stops from affricates. In stops, the closure is released a b ru p t l y while in affricates it is released gradually: the initial h o l d p h a s e of an affricate is similar to that of a stop b u t in the later release phase an affricate is like a fricative.

O n l y affricates can have the p r o p e r t y [ + del rel]; all other sounds are [-del rel].

3.3.7 Prosodic features

Prosodic features such as tone and stress, w h i c h are usually associated w i t h an entire syllable or w o r d , are difficult to describe. A l t h o u g h their phonological relevance is clear, their phonetic properties are not easy to specify satisfactorily.

T h e account of p r o s o d i c features given in this chapter

52 Distinctive features

is provisional. It is based on the SPE position w h i c h was d o m i n a n t for a long time, and is reflected in m u c h of the literature that appeared d u r i n g the 1970s, b u t has been superseded by recent theories w h i c h will be outlined in last four chapters of this b o o k .

22. L O N G - S H O R T [ ± long] This feature refers to the d u r a t i o n of a s o u n d . Clearly, in purely physical t e r m s this feature cannot be binary since length is always relative. In the SPE system, and in its J a k o b s o n i a n predecessor, length is n o t regarded as a basic phonetic p r o p e r t y b u t rather as an incidental attribute of the feature [tense], w h i c h is binary. T h u s in English, for example, the contrast between [i] and [1] (as in / s i k / 'seek' and / s i k / 'sick';) is regarded as being essentially a tense versus lax opposition; the fact that / i / is longer than / i / is viewed as secondary.

B u t even if this were the correct analysis, using the feature [tense] to s u b s u m e length w o u l d be questionable w h e r e differences in duration i n v o l v e consonants since t h e feature [tense] is m e a n t to specify v o w e l s o n l y .

If w e use the feature [ ± l o n g ] , h o w e v e r , these difficulties are avoided because it is equally applicable to v o w e l s and consonants. N o p r o b - lems arise w h e n w e encounter a language like Luganda ( U g a n d a ) which distinguishes long consonants from short ones in w o r d s like / t a / [ta] 'release! (imperative)' and / t t a / [t:a] 'kill! (imperative)' as well as long and short vowels as in / k u l a / [kula] ' g r o w up! (imperative)' and / k u u l a / [ku:la] ' u p r o o t ! (imperative)'.

In principle, the feature [long] is m u l t i - valued. A language could contrast m o r e than t w o d e g r e e s of l e n g t h . K i k a m b a (Kenya) contrasts three degrees of v o w e l l e n g t h . 4 In p r a c - tice m o s t languages w h i c h have distinctive v o w e l length only distinguish between l o n g and short vowels. A b i n a ry approach, while n o t ideal is n o t altogether inappropriate.

The SPE system of distinctive features 53

23. S T R E S S [ ± stress] Stress is an elusive concept. All stressed (or accented) syllables in a w o r d are m o r e salient than their unstressed counterparts b u t the p h o n e t i c manifestation of stress varies. T h e p r o m i n e n c e of stressed syllables is due to an a d m i x t u r e of raised pitch, greater length and increased intensity of the signal, which is perceived as loudness. In addition, in s o m e languages, e.g. English, stressed vowels retain their full v o w e l quality while unstressed vowels are reduced to s o m e muffled v o w e l sound like schwa [a].

It is generally agreed in principle that stress need n o t be binary. In SPE, for instance, three degrees of stress are recognized for English. In practice, h o w e v e r , phonologists often treat stress as binary for convenience.

24. T O N E It is i m p o r t a n t to distinguish b e t w e e n tone and pitch. T h e pitch of an utterance depends on the rate of vibration of the vocal cords, the higher the rate of vibration, the higher the resulting pitch b e c o m e s . A n y time a voiced s o u n d is p r o d u c e d , the vocal cords m u s t vibrate at a certain rate: all languages have sounds which s h o w pitch differences. In a T O N E language those pitch differences are used phonemically either to differentiate b e t w e e n w o r d meanings or to c o n v e y g r a m m a t i c a l distinctions.

C o m m o n l y used tone features (which are n o r m a l l y indicated by w r i t i n g the appropriate diacritic above the a p p ro p ri a t e tone-bearing element) are:

[ ± ] H i g h : (marked by (')) f ± ] Mid: (marked by ( - ) ) [ ± ] L o w : (marked by (')) [ ± ] Rising: (marked by (—')) [ ± ] Falling: (marked by (*)) [ ± ] Fall-rise: (marked by ("))

54 Distinctive features

The use of tone to distinguish lexical meanings can be illustrated with this example from Igala, a Nigerian language (Welmers 1973):

[3.6] awo 'guinea fowl' awo 'a slap' awo 'an increase' awo 'a comb' awo 'hole (in a tree)' awo 'star'

The use of tone to signal grammatical distinctions is illus- trated with this example from Engenni (Nigeria), appearing in Bendor-Samuel (1974), where the distinction between declarative and negative sentences is marked using tone:

[ 3 . 7 ] Positive Negative

amo na dire eda amo na dire eda

'the child should 'the child should not cook beans' cook beans'

I shall sum up the discussion so far by showing charts of distinctive feature matrices which contain arrays of features necessary for the representation of various sounds.

[3.8] Distinctive feature matrix for vowels

i 1 y u u e e 3 3 + U A 0 o e o D a e a c i D u i Y

high + + + + + - - + + - - + - low + + + + back + + + + + - + + + + + tense + - + + - + - + - - + - + - + - + - + + round + + + - - - + - + + + + - +

[3.9] Distinctive feature matrix for sonorants

m n n. Ji 3 N 1 [ A r t j w

cons. + + + + + + + + + + + - - cont. - - - - - - + + + + + + + nas. + + + + + + - - - - - - - lat. _ _ _ _ _ _ + + + _ _ _ _ lab. + _ _ _ _ _ _ _ _ _ _ _ + ant. + + _ _ _ _ + _ _ + _ _ _ cor. - + + + - - + + + + + + - high - - - + + - - - + - - + + back _ _ _ _ + + _ _ _ _ _ _ +

The SPE system of distinctive features 55

c 3

O

I* 3

>

o

+J "=• I + I I I + + I I I

-a •£. I + I + I +

j ' o . l + I + +

* J3 + I I

o - <C + I I

>. X + I I

. _ , u - + I I

M ^ + + +

isf M , + + I

N * + + + + I +

w o + I I + I +

>< ,̂ + + I + +

I

I

I

I

I + +

I + +

I + +

ca •©• +

a a-

" a -4-.

+ + +

I

I +

I +

+

+

+ +

+

+

+ I + + I

I + + I

I I I

I I I

+ 1 1 +

+ + I

+ I

+ I

I

I

I

I

I

I

I

+ I +

+ + I

+ +

+ I

I

I

+

I

+

I

. * S - 2 o o ^ -o i: •S *ts 'i

56 Distinctive features

3.4 Segment structure redundancy

O n e striking thing a b o u t the distinctive feature matrices presented in the previous section is the a m o u n t of R E D U N D A N C Y w h i c h they reveal: n o t every single phonetic p r o p e r t y of a M O R P H E M E (i.e. m i n i m a l distri- butional or meaningful part of a w o r d , such as un-, kind, and -ly in unkindly) is distinctive. M a n y properties are R E D U N D A N T . This means that they are predictable by general principles.

O n the one hand, there are feature c o m b i n a t i o n s that are predictable by universal S E G M E N T S T R U C T U R E C O N S T R A I N T S . These feature combinations (e.g. [+nasal, + sonorant, + voice]) go t o g e t h e r hand in glove: they are U N M A R K E D , they are the n o r m . O n the o t h e r hand, there are feature c o m b i n a t i o n s that are n o r m a l l y i n c o m p a t - ible (e.g. [ + high, + l o w ] or [— sonorant, +syllabic]). It is unusual for a language to have segments w i t h c o m b i n a t i o n s of features o t h e r than those predicted b y universal s e g m e n t s t r u c t u r e constraints. T h o u g h n o t absolute, (e.g. a language can have syllabic o b s t r u e n t s or voiceless v o w e l s or voiceless nasals) such restrictions o n the c o m b i n a t i o n of features play a key role in delimiting the i n v e n t o r y of possible speech s o u n d s . O d d sounds like voiceless v o w e l s , voiceless nasals or syllabic o b s t r u e n t s are e x t r a o r d i n a r y . T h e y are M A R K E D .

In addition to c o n f o r m i n g to universal s e g m e n t s t r u c - ture constraints languages m a y have their o w n idiosyncratic rules regulating feature c o m b i n a t i o n s in their s e g m e n t s . A n English e x a m p l e will clarify the point. In English (and indeed in m o s t other languages), the feature [ — back] cannot be c o m b i n e d with the feature [-round]. R o u n d e d front vowels like [y] are n o t p e r m i t t e d in English (but they are allowed in French, o c c u r r i n g in w o r d s like [ty] tu [vy] vu). T o take another example, [1] is the only lateral allowed in English. If you k n o w that a s e g m e n t is lateral, then you k n o w that it is also consonantal, sonorant, alveolar and voiced. H o w e v e r , a l t h o u g h voiceless [1] is disallowed in English and m o s t o t h e r languages, it does occur in a few languages. For instance, the s o u n d spelled w i t h '11' in Welsh w o r d s like Llandudno is a voiceless lateral fricative [i].

Segment structure redundancy 57

Besides s e g m e n t s t r u c t u r e r e d u n d a n c y , there also exists S E Q U E N T I A L R E D U N D A N C Y . As w e shall see in the c o m i n g chapters correct inferences can often be m a d e about a s e g m e n t if the context in which it occurs is k n o w n . T h u s , in spoken English, w h e n e v e r there is a cluster of three consonants o c c u r r i n g syllable initially, the first consonant in the sequence is the fricative [s]. N o o t h e r consonant can occur in that position: splash and scream are English w o r d s but *Jilash and *zgream are n o t - and could n o t be. W e shall explore this p r o b l e m m o r e fully in C h a p t e r 9.

Clearly, n o t all the phonetic properties of a sound are crucial in every case in distinguishing w o r d m e a n i n g . S o m e properties are R E D U N D A N T , i.e. predictable and n o n d i s - tinctive. W h e n w r i t i n g phonological rules, the convention of o m i t t i n g r e d u n d a n t properties is n o r m a l l y observed. T h u s [p] [b] and (ejective) [p'] m a y be represented simply as in [3.11]:

[ 3 - I I J

son. voice cont. lab. constr high back nasal

Write d o w n the values of r e d u n d a n t features which have been o m i t t e d in [3.11]. Check w i t h table [3.10]

Exercises

I. Circle the s e g m e n t s w h i c h are: e x a m p l e (i) [ + s y l l ] w p © ( e ) m h ( u ) g v (a)

(ii) [ + a n t ] d z ? p b n N B 3 f (iii) [—cons] a f v l r h 0 j x u (iv) [ + c o r ] v t r n j J r j x d k (v) [ + r o u n d ] w t i u 0 y o e 1 ae

(vi) [—voice] o w s p i t g m i q

p

-

+ —

b

+

+ —

p'

-

+ +

58 Distinctive features

2. For each segment alter the value of the feature indi- cated, keeping all the other features unchanged. Write down the phonetic symbol for the resulting segment. Original segment Feature value New segment

changed example (i) i [ + syll] j [—syll]

(ii) u [+back] (iii) b [—nasal] (iv) e [—round] (v) d [+voice]

(vi) g [—nasal]

3.(a)Write down the phonetic symbol for the initial seg- ment of each one of the words below:

that cat band wet write philosophy shy June knee tea

(b) Fill in the feature values of the initial segment of each of the words above for features listed below. Follow the example which is given, e.g. [5] f/iat

cons + son — syll voice cont nas ant cor lab back high

-

+ + — + + — — -

Notes

FLAT consonants are contrasted with PLAIN ones. A flat consonant is produced by rounding the lips or by pharyngeal contraction. The resulting modification of the oral tract produces the acoustic effect of flattening, i.e. downward shift in the formants in the spectrum (Jakobson et al. 1952).

Segment structure redundancy 59

In SPE the class of glides includes the semi-vowels [j] and [w] as well as the glottal stop [?] and the glottal fricative [h]. However, the case for formally recog- nising a class of glides which includes [w j h ' ] is not altogether overwhelming. Some writers like Fromkin (1970) and Lass (1984) only recognise [w] and [j] as glides and argue for treating the glottal fricative [h] and the glottal stop ['] as obstruents on articulatory grounds.

Although they are not made in the front of the mouth, glottals are not back because, by the definition given in SPE, only sounds produced with the tongue pulled back from neutral position are back. In the production of glottals the tongue is in neutral position, and does nothing. The vocal cords are the articulators.

Ladefoged (1971), citing Whiteley and Muli (1962), reports that Kikamba phonemically contrasts four degrees of vowel length: short, half-long, long and extra-long /V V. V: V::/ as in: kwelela 'measuring' ku/a 'start' kwele.la 'moving backward

and forward' kuele:la 'aiming at' ku/a: 'giving birth'

kuja:: 'giving birth frequently'

CHAPTER 4

Phonological representations

4.1 Phonetics and phonology

P H O N O L O G Y is the b r a n c h of linguistics w h i c h investi- gates the w a y s in w h i c h speech sounds are used s y s t e m a t i - cally t o form w o r d s and utterances. B y n o w it will be o b v i o u s that in o r d e r to u n d e r s t a n d p h o n o l o g y one m u s t have a grasp of the basic concepts of phonetics, w h i c h as w e saw in the first chapter, is the study of the p r o d u c t i o n of speech sounds by speakers, their perception b y hearers and their acoustic properties. M o s t speakers of a language like English, w i t h a l o n g tradition of w r i t i n g , w h o h a v e n o t m a d e a study of phonetics and p h o n o l o g y tend to think of the sounds of speech t h r o u g h the m e d i u m of the o r d i n a r y alphabet. B e c o m i n g literate involves, a m o n g other things, mastering one particular r o u g h - a n d - r e a d y scheme of analysis for the phonetics of a given language.

In the particular case of the English language, it happens that the analysis implied by the o r d i n a r y spelling is very ' r o u g h ' and inaccurate indeed. P r o b a b l y m o s t people w h o are literate in English are aware of this fact a l t h o u g h few of t h e m could state explicitly and systematically w h a t these s h o r t c o m i n g s are. S o m e o n e w h o has n o t studied phonetics has scarcely any vocabulary for describing spoken English other than the n a m e s of the letters of the alphabet (and possibly t e r m s like ' c o n s o n a n t ' , ' v o w e l ' and 'syllable'; b u t these are often interpreted as graphological, i.e. pertaining to the s t u d y of the w r i t t e n language). So, the m o s t that can be d o n e by the average speaker t o w a r d s p o i n t i n g o u t the s h o r t c o m i n g s of English spelling is to cite particular instances of 'illogicality', such as the various p r o n u n c i a t i o n s w h i c h c o r r e s p o n d to the spelling -oo- in foot,

Phonetics and phonology 61

fool, and blood or the various spellings that correspond to [ei] in maid and made, break and brake, weigh and way. Although people are aware of many individual cases of this kind, their general understanding of what speech is like is moulded almost wholly by the nature of the writing system that we use to record speech. Much of the point of a study of phonetics lies in acquiring the ability to penetrate behind the rough-and-ready analysis of speech which is implied by the standard orthography and to grasp what is really going on, especially in cases where the standard orthography is misleading. Moreover, as we shall see in the final section of this chapter, the ability to analyse and record speech in a more sophisticated way than that provided by the standard alphabet is not only of theoretical interest, it also has many practical applications.

Answer this question before you read on: if traditional English orthography inculcates, in everyone who masters it, a crude theory about the nature of speech sounds, what would that theory be like, if it were stated explicitly?

Most obviously, the English alphabet tells us that spoken utterances consist of linear sequences of elements: UNITARY S O U N D S occurring one after another. It is a fundamental characteristic of alphabetic writing to be one- dimensional; letters follow one another across the page, it never happens that two letters occupy the same position in the left-to-right sequence as in [4.1].

[4.1] h or m or f1

And, although in handwriting individual letters blend each into the next without sharp boundaries between them, this is just regarded as a hasty, convenient way of transcribing what are intended as a row of separate marks and appear as such in print.

Secondly, the writing system tells us that these unitary elements differ from each other discretely, like beads on a string or chemical elements, rather than continuously, like shades of colour. If you ask me what colour a dress is, it makes perfectly good sense for me to say it is between green and yellow - perhaps more towards yellow than green; but

62 Phonological representations

if you asked a geologist what metal a sample of moon rock consisted of, it would make no sense at all if the reply you got was that it was a metal somewhere between aluminium and molybdenum, a bit more molybdenum-y than aluminium-y. (Of course, it might be a mixture or alloy, containing both aluminium and molybdenum atoms; but each individual atom would be either aluminium or molyb- denum: atoms do not shade into each other as colours do.)

Do letters of the alphabet shade into each other as colours do or do they differ discretely like chemical elements?

The answer, as we have already seen, is that letters of the alphabet resemble chemical elements rather than colours in this respect: a given word, say key either begins with the letter k or it does not, and there would be no sense what- soever if, for instance, when playing a game of 'I spy with my little eye' one child said to another that she spied some- thing beginning with a letter halfway between k and t.

In brief, alphabetic writing teaches us:

(i) that speech consists of linear sequences of unitary, discretely-different sounds;

(ii) that there only a small number of types of sound in a language;

(iii) that a sound occurring at a given place in one word can be identified with certain sounds occurring at different places in other words.

When children learn to read in the early years at school, the above statements (i)-(iii) about the nature of speech sounds are not, of course, taught to them in so many words — indeed they are rather abstract statements which small children could hardly understand if they did hear them stated explicitly. But the fact that these propositions about how speech works are left implicit does not make them any less influential in moulding adults' perception of the nature of their language. Rather, the reverse: precisely because these propositions are not explicitly articulated, it does not, on the whole, occur to people to question them. As you

Phonetics and phonology 63

work your way through this book, you will realise that each of these assumptions is, if not down-right false, at best a very partial and crude approximation to the truth.

Of the points (i)-(iii) that I have listed, the one that is often called into question by adult English speakers without special knowledge of phonetics is (ii). People are well aware that the inventory of English sounds is not to be identified with the twenty-six letters of the English alphabet. They may recognise that, for instance, the two letters sh in words like shop and push stand for one rather than two sounds while the single letter x stands for two sounds in words like tax and box; and they may also be aware that the sound spelled c in rice is not the same as the sound spelled c in coal, but that the latter sound is the 'same' as the sound spelled by the different letter k in koala or by the letters ck in back. This does not mean that people generally question the exist- ence of a limited inventory of English sounds - on the contrary, the general principle that speech is appropriately represented by means of an alphabet of a few dozen letters is usually taken for granted; but people think of English as a language whose alphabet regrettably fails to represent the spoken language accurately, partly because it needs a few more letters than it has got.

The student of phonetics, however, knows that even the fundamental assumption of the linearity of sound segments in (i) is misleading as a model of what happens when we speak. A speaker does not articulate just one holistic speech sound at a time. Speech sounds are the result of many separate articulatory gestures which are made simultaneously. As we saw in the last chapter the speaker changes the position of the vocal cords as the utterance progresses; at the same time the soft palate is raised and lowered; the shape and position of the tongue and of the lips is also modified while all that is going on. A writing system faithful to this aspect of the nature of speech would record speech not as a linear sequence of letters, but more after the fashion of an orchestral score which includes separate staves, one above the other, for the activities of the various instruments (in the phonetic case, the various vocal organs). The 'score' for the articulation of the word mat, for instance might look like this:

64 Phonological representations

[4-2]

LIPS close., open T O N G U E neutral . . lower front raise blade return

make closure blade with alveolar to neutral ridge position

VOCAL together and vibrating stop vibrating CORDS VELUM down up

[ m x t ]

As we saw in the last chapter, sounds can be thought of as bundles of distinctive features; they are the result of more or less simultaneous articulatory activities of various organs which imperceptibly merge with each other. Sounds are not neatly separated from one another in the way suggested by the self-contained, separate letters in a line of print. True, some of the articulatory events in a stretch of speech are fairly abrupt; during the production of [m], for instance, the lips stop the escape of air from the mouth until the bilabial closure is abruptly released. But even in cases like this where events on one stave of the orchestral score occur at precise moments, it is the exception rather than the rule for events on all the various staves to be synchronised. You can see in [4.2] that all the gestures involved in the production of a given sound do not necessarily start and stop at exactly the same time. Moreover, many of the events happening on the different staves will, by their nature, be gradual and not abrupt. For example, the raising of the velum takes time, and consequently, the first section of the vowel in mat will be somewhat nasalised during the transition from [m] to [ae]; movements of the body of the tongue are gradual and even a bilabial stop closure is not released as suddenly as all that. To take another example, conventionally, the English word hour is phonetically tran- scribed as a T R I P H T H O N G (i.e. a vowel with three vowel qualities) [aua] but it would be a travesty to suggest that an utterance of this word consists of an initial period during which the tongue-body is depressed and the lips spread; a middle period in which the tongue is humped up in the back of the mouth and the lips pursed; and a final period in which the tongue has an intermediate shape and the lips are spread

Phonetics and phonology 65

again. We humans do not have tongues and lips which are controlled by ratchets which click instantaneously from one configuration to another in that fashion; rather, throughout the utterance of hour the configurations of the tongue and lips are constantly changing, in complex ways, without stable intervals.

Furthermore, the fact that the vocal organs are controlled by muscles rather than clicking mechanical ratchets, means that speech sounds (or, to be precise, the separate articulatory gestures of which speech sounds are composed) are much more similar to colours which shade into each other gradually than to chemical elements which are sharply, discretely different. In terms of English orthog- raphy it makes no sense to talk of a letter halfway between a k and a t; phonetically, however, it is perfectly reasonable to talk of a sound somewhere in between [k] and [t]. For instance, a palatal stop [c] could be so described. But even the three phonetic symbols [k c t] do no more than crudely indicate three broad areas at which the tongue may make contact with the roof of the mouth.

This is the reason: between the ends of the upper teeth and the uvula (the fleshy bit of your anatomy hanging down from the middle of the soft palate at the very back of the roof of your mouth beyond the velum (see [1.2])) there is an infinity of slightly different areas against which the tongue can make a total closure and, at the same time, the areas of the tongue involved in contact vary continuously rather than discretely.

The implication of the continuously, rather than discretely varying nature of articulatory parameters is that, even if stretches of speech could be chopped up into separate 'sounds', there would not be a limited inventory of possible sounds. Rather, the range of pronounceable sounds is as infinitely variable as the range of possible shades of colour, and the number of different objects one chooses to label with distinct colour-names - or phonetic symbols - depends purely on how finely one wishes or is able to discriminate.

Let us ignore for one moment the problems of segmenting an utterance into a sequence of sounds. Admit- tedly, although the range of physically possible sounds is infinite, within any one language only a small, finite set of sounds is used. For instance, in English there are no speech-

66 Phonological representations

sounds, such as clicks, implosives or ejectives, which are made otherwise than with the pulmonic egressive airstream mechanism; there are no front rounded vowels like the [y] in French words such as lu, and so on. But equally clearly, within the range of sounds that do occur in a given language, all the same arguments about variation will apply - so that the range of slightly different sounds produced by English speakers will be enormous, even though that range is only a small part of the more extensive range of sounds which are produced by speakers of all the languages of the world.

Finally, we come to point (iii). The English version of the Roman alphabet has just twenty-six letters; but, of course, speakers of English produce a lot more that twenty- six physically distinct sounds. Often, sounds that are different are represented using the same letter. In learning to spell, we learn to think of the first sound of say, car as the 'same' as the first sound of keys although they are not phonetically identical (see section 2 . I . I ) . This is mis- leading.

4.2 The domain of phonology

In terms of the physical properties of speech, then, sounds are endlessly diverse, and the appearance of a few discrete units suggested by the Roman alphabet is seriously misleading. To study phonetics is, partly, to learn to penetrate beyond that fiction to the chaotic wealth of slightly different sounds that we actually use. However, if we turn from considering the physical properties of speech- sounds to thinking about how they F U N C T I O N in a language as a communicative system, the picture suggested by the traditional orthography becomes rather more appro- priate. It is this alternative approach to speech-sounds, examining them from the point of view of how they are actually USED in different languages, which is adopted by the student of P H O N O L O G Y .

What I mean by saying that the alphabet-based analysis of speech-sounds is reasonably adequate from a functional point of view is that, if we think of phonetic variation in connection with the work it does in language, then it ceases

The domain of phonology 67

to be true that variation is continuous and not discrete. The phonetic symbols [t c k], we noted earlier, impose a three- way classification on what is physically an infinitely graded series of slightly different areas of the palate where the tongue can make contact. But the English language, as a tool of communication, imposes an even cruder classifi- cation on that series, treating it as if it had only two members t and k, with no in-between sound.

That does not mean that the initial sound in every utterance of a word such as tea, time, tell, etc., involves exactly the same area of contact between the tongue and the roof of the mouth; it does not mean that the tongue makes contact with exactly the same part of the roof of the mouth in the articulation of the k sound at the beginning of key, coal etc. Speakers could no more achieve that sort of consist- ency in their speech than darts players could manage to place the dart at exactly the same point on the dartboard as the dart before it.

To discover in a practical way what this means, find a dozen different words beginning with a t and a k sound respectively. Identify the exact points of contact between the tongue and the roof of the mouth when you say each one of the words which you have chosen.

You will notice that the sounds tend to cluster into two groups, rather in the same way that the holes made in the dartboard would cluster into two groups if the players were attempting to aim for one or other of two points selected as targets. Furthermore, speakers of English have learned as hearers to listen for the phonetic 'targets' and not the precise phonetic properties of each sound, so that if a speaker uttered in the middle of an utterance a 'word' which is transcribable as [cul] or [tul], with an initial sound inter- mediate between [k] and [t], and if it were perceived as such by the hearer, it would be dismissed as a mistake. What is more likely is that the 'word' would be perceived as a rather unsuccessful or inelegant way of saying [tul] tool or [kul] cool, whichever seemed appropriate in the context.

The view that the phoneme is a TARGET S O U N D in the minds of speakers and hearers was originally

68 Phonological representations

proposed by the Polish linguist Baudouin de Courtenay in 1894. He suggested that much of the precise phonetic detail that is physically present in speech (from an articulatory or acoustic point of view) is irrelevant from a F U N C T I O N A L point of view.

In the intellectual climate of the first half of this century, especially in America, de Courtenay's mentalist definition of the phoneme was rejected by most linguists. Because they considered themselves empirical scientists, they saw no value in speculating about the workings of the invisible mind (Twaddell 1935:57).

While objections to the mentalist approach to the phoneme cannot be ignored, it does nonetheless seem to offer some useful insights. If you say tea and key, for example, there will inevitably be some very slight difference in the way you produce the first consonant on various occasions but all that matters is that each time these words begin with an articulation whose target is either / t / or /k/. Another language may well have three targets on this particular dimension (indeed, many languages have).

Variation in the physical execution of a given phonetic target need not be random. For instance, of the two English clusters of tongue/palate contact-areas just discussed, the area I am calling the / k / cluster, if it could be plotted, would turn out to involve a scatter not round a single point but round a line. This is because, typically, velar consonants preceding front vowels tend to be made with contact farther forward on the soft palate than velar consonants preceding relatively back vowels (section 2.1.1).

You can verify this by pronouncing the following words very carefully and observing the movement of your tongue:

[4.3] car key King Cole keep cool Ken Carr

I hope you observed that the k's in car, cool, Cole and Carr are made farther back than those in key, keep, King and Ken.

Although this variation in the k sound is systematic

Phonetic and phonemic transcription 69

rather than random, it is not functionally significant. The distinction between a fronted k and a more back k cannot be used to contrast word meaning in English. It is not phonemic. Fronted [k] (see [4.7] below) and nonfronted [k] are allophones of the same phoneme. Phonology focuses on functional differences. Phonology is sometimes referred to as F U N C T I O N A L P H O N E T I C S .

4.3 Recapitulation: levels of representation

To sum up the discussion so far, we recognise at least two distinct levels of representation of speech: the P H O N E T I C LEVEL and the P H O N E M I C LEVEL. At the P H O N E T I C LEVEL, our task is to provide an accurate description of the characteristics of the sounds that occur in speech; at the P H O N O L O G I C A L LEVEL the emphasis is on those prop- erties of sounds that are functionally significant in the formation of words and utterances.

4.4 Phonetic and phonemic transcription

We have seen in the preceding sections the ways in which the standard orthography fails to provide an adequate representation of speech. In this section we are going to see how the use of P H O N E T I C and P H O N E M I C T R A N - SCRIPTION might improve matters. In addition to intro- ducing you to transcription, this section is intended to familiarise you with aspects of the consonant system of English. The section should also serve as a practical demonstration of the use of phoneme theory which was introduced in Chapter 2.

Phonemic transcription (also called B R O A D T R A N - SCRIPTION) only shows functional differences, i.e. differ- ences between sounds which are used to distinguish word meaning. It only uses enough symbols to represent each phoneme of the language in question with a symbol of its own. Phonetic transcription (also called N A R R O W T R A N - SCRIPTION) on the other hand, is much more detailed and attempts to provide a more faithful representation of speech. It normally represents the allophones of a phoneme that

70 Phonological representations

occur in various contexts, but can be made to show even finer detail, if necessary. To do this it uses a much larger number of symbols together with diacritic marks to distinguish subtle nuances of pronunciation. In [4.4] and [4.5] I exemplify broad and narrow transcription respectively:

[4.4] Broad transcription attend /atend/ two / t u / eighth / e i t / seat /sit/ nutmeg /nAtmeg/

[4.5] Narrow transcription [athend] the raised (h) shows ASPIRATION i.e. there is a slight delay between the release of the stop [t] and the onset of voicing in the vowel [e]. An aspirated stop is released with greater force than a unaspirated one.

[twu] the raised (w) shows lip-rounding (the speaker starts rounding the lips before the production of [t] is completed in anticipation of the vowel [u] which is made with rounded lips.

[eitG] the subscript (_) marks dental sounds; here, in anticipation of the dental fricative [8], the tongue makes contact with the upper front teeth rather than the alveolar ridge.

[si?t] the symbol (9) marks pre-glottalisation. There is glottal reinforcement - with a glottal stop coinciding with or slightly anticipating the allophone of ft/ that occurs in word final position. The airstream is simultaneously obstructed at two points like a river dammed at two points. In some accents the word ends in a glottal stop, with no [t] element at all.

[nA^tmeg]1 the symbol (?) again marks pre-glot- talisation; that is glottal reinforcement when [t] is followed by another consonant.

The various t sounds are in complementary distri- bution, each one has got its special contexts in which it occurs. They are allophones of the phoneme / t / . Below I have written informal rules stating the distribution of these allophones:

Phonetic and phonemic transcription 71

[4.6] / t / is realised as: [th], an aspirated voiceless alveolar stop when

it occurs at the beginning of a stressed syllable as in [athend];

[tw], a labialised (rounded) voiceless alveolar stop before rounded vowels; The word two also has a voiceless aspir- ated alveolar stop and we could have written it phonetically as [tw h]. But, as you can see, diacritics can crowd each other out. For the sake of clarity we may use only those diacritics that are necessary to clarify the point we wish to focus on;

[t], a dental alveolar voiceless stop before dentals;

[?t], a glottalised alveolar stop in word final position or before another consonant.

In order to describe the allophones of a phoneme or to make a narrow phonetic transcription you will need to know various DIACRITICS devised by phoneticians for this purpose. An annotated list of common diacritics is provided in [4.7].

[4.7] o VOICELESS: examples (voiceless) [ g a] This indicates that a segment type that is normally fully voiced is fully or partially devoiced (e.g. in English, word-final voiced stops like / g / are realised as [g] as in bag [baeg].

h ASPIRATED: examples (as°pirated) [ph th kh] This indicates that a segment is aspirated (e.g. voiceless stops in English are aspirated when they occur at the beginning of a stressed syllable in words like [athend] attend, [phet] pet and [khaet] cat).

w LABIALISED: examples (labialised) \pw tw kw\ It indicates labialisation (lip-rounding) (e.g. non-labial consonants followed by round vowels are labialised in English words like [twu] two and [kwul] cool). SYLLABIC: examples (syllabic) [1] and [n]

72 Phonological representations

It indicates that a consonant functions as a syllable nucleus (e.g. in English the nasals and [1] are syllabic when they occur at the end of a word if they are preceded by another conso- nant as in [ketj] kettle and [kDtn] cotton).

r RHOTACISED (i.e. r-coloured) It indicates that a vowel has 'r-colouring' (e.g. the pronunciation of words such as [kaJ] car and [kaJt] cart in American and English west country accents).

+ shows that a segment is F R O N T E D (e.g. velar is made with the back of the tongue moved forward close to the hard palate when it is followed by a front vowel as in [ki] key).

~ NASALISED +

It indicates that air escapes through the nose as in American [psen] pan or French [b5] bon.

: L O N G It shows that a segment is long (e.g. the vowel in the word see [si] can alternatively be tran- scribed as [i:]2).

The list in [4.7] is not exhaustive but it contains most of the diacritics you are likely to need frequently. For a more comprehensive list refer to the IPA chart in [1.13] in the first chapter.

4.5 A guide to phonetic transcription

This section contains a few practical hints about transcription.

(i) Start by reading the word or passage aloud several times in a manner which is as close as possible to natural conversation. If possible, make a recording of your reading so that when doing the transcrip- tion you work from the recording. This is because it is virtually imposisble to repeat an utterance consistently from one time to the next.

When working with informants, do not start transcribing until you can mimic their pronunci- ation to their satisfaction.

Why study phonology? 73

(ii) Write on alternate lines. (iii) Avoid using letters of the alphabet (such as the

capital letter A) which are not part of the phonetic alphabet. Avoid using capital letters (at the begin- ning of sentences or in proper names) because they may have a different phonetic value from that of lower case letters, e.g. [n] is an alveolar nasal but [N] is a uvular nasal.

(iv) Do not confuse the spelling of a word with its pronunciation.

(v) Mark off intonation group boundaries using vertical lines like these | | e.g. | The aim of this exercise \ is to identify some of the phonemes in your speech \

(vi) Avoid confusion. Do not use any of the standard punctuation marks like question marks, because many of them have a different value in the phonetic alphabet e.g. the colon [:] represents length; the symbol for a glottal stop ['] is like a question mark [?] without the dot, etc.

(vii) If you are transcribing English, remember that unstressed vowels are in most cases realised as [a] (or [i] or [u]). E.g. [banaena], not *[baenaen£e], for banana.

(viii) You may mark off those syllables which carry stress by a raised, small vertical line before the syllable carrying the main stress of a word, e.g. ['enta] enter. Where in addition to a main stress there is a secondary stress, you can mark it with a lowered vertical line, e.g. [,dAbled3d] double-edged.

4.6 Why study phonology?

In this chapter we have seen that it is necessary to recognise at least two levels of speech representation: the phonetic and the phonemic. We have further seen that corresponding to each level there is a system of transcription, namely phonemic (broad) and narrow (phonetic) transcription. But you might still wonder why anyone should be interested in working out ways of representing speech, other than the standard orthography which, in spite of its limitations, has

74 Phonological representations

served us so well. In the closing r e m a r k s of this chapter, I briefly address this question.

First there is the theoretical m o t i v a t i o n : theoretical linguists are interested in exploring the w a y s in w h i c h language uses the m e d i u m of s o u n d , w h i c h is in itself m e a n - ingless (a s o u n d like [t] or [e] in itself means n o t h i n g ) to build w o r d s w h i c h c o n v e y an infinite r a n g e of m e a n i n g s . T h e y are interested in the restrictions on the w a y s in w h i c h sounds can c o m b i n e to form w o r d s ; t h e y are interested in the w a y s in w h i c h p h o n o l o g y relates to g r a m m a r and m e a n i n g . T h e y are interested in b o t h the language-specific and universal d i m e n s i o n s of these questions.

T h e answers to these questions are relevant to other areas of linguistics.

(a) Psycholinguists investigate the acquisition of p h o n o - logical k n o w l e d g e b y infants and the disintegration of that k n o w l e d g e in aphasia (the loss of linguistic ability t h r o u g h brain damage) - these are w i n d o w s t h r o u g h which the n a t u r e of h u m a n cognition in general can be glimpsed. (b) Historical linguists investigate the evolution of s o u n d systems because it provides the surest evidence of languages sharing c o m m o n descent. (For example, / p / in Latin w o r d s like pater and piscis corresponds to /f/ in the English w o r d s father and fish w h i c h have the same m e a n i n g ) . T h e r e tends to be a significant degree of s o u n d - m e a n i n g c o r r e s p o n d e n c e b e t w e e n languages descended from a c o m m o n ancestor. (c) Sociolinguists investigate the correlation b e t w e e n geographical dialect, social s u b - g r o u p i n g s and p r o n u n c i - ation. (For examp l e , in England, middle-class speakers of English are very m u c h less likely than their working-class counterparts to d r o p their aitches initially in n o u n s , adjec- tives and verbs and p r o n o u n c e the w o r d s heat, and hat, for instance, as [it] and [ast] respectively, rather than [hit] and [haet]. B o t h sociolinguists and historical linguists, for instance, a t t e m p t to explain w h y look and luck r h y m e in the N o r t h of E n g l a n d , b o t h being p r o n o u n c e d [luk], b u t n o t elsewhere, and w h y [hsum] h o m e and [staon] stone are p r o n o u n c e d as [he:m] and [ste:n] in Scotland.

For all these g r o u p s of linguists it is essential to be able

Why study phonology? 75

to understand the workings of the spoken language. At the very least, they all need to be able to make adequate broad and narrow transcriptions of spoken language data so that they have a proper representation of the subject matter which they examine.

Phonetics and phonology are not esoteric academic disciplines with no relevance to the real world. Beyond the confines of linguistics, there are many kinds of specialists working with language who need to be able to analyse and represent speech in a more sophisticated way than that provided by the standard orthography. They include literacy experts, especially those devising orthographies for unwritten languages (a famous book on phonemics by Kenneth Pike has the sub-title 'a technique for reducing languages to writing'). An orthography ensuring that there is just one symbol for every contrastive sound in a language is ideal for writing down a language and introducing literacy to a speech community. Even for a written language like English, literacy workers wishing to understand the problems of their pupils or clients need to have the grasp of spoken English and its relationship with the standard orthography which is derived from a study of phonology. Not least because often spelling mistakes - such as misspelling feel as feeyal - can be understood and remedied if the instructor realises that phonetically that word is frequently pronounced as [fijsl]. The incorrect spelling is, in fact, phonetically well motivated.

For teachers of foreign languages, too, a good under- standing of phonetics and phonology is desirable, for it facilitates the diagnosis of learners' errors and provides the concepts and notation needed to represent accurately both the learner's speech and the target pronunciation.

Actors need to be convincing mimics of a wide range of accents. An understanding of the phonological charac- teristics of social and geographical dialects hinted at above is invaluable in that profession. Phonetics (in some form) is usually one of the subjects studied at drama school.

Speech therapists treat patients with speech defects which sometimes are due to inability to produce certain sounds, or a failure to produce certain allophones of phonemes, or again a failure to master some principles that govern phonological patterning. Not only do they need to

76 Phonological representations

be able to transcribe their raw data phonetically and analyse aspects of their patients' pathological speech case histories, but they need also to have a thorough knowledge of normal phonological behaviour which their patients are exepcted to approximate. For the same reasons, specialists in speech hearing science and teachers of the deaf need to have a good understanding of phonetics and phonology in order to enable their speech-impaired clients to communicate using the spoken language.

Neurologists and neurolinguists often have to treat aphasics (brain-damaged patients) who have lost the ability to produce or process the spoken language but who may, in some cases, have retained the ability to communicate through the written word. They need to know how speech is produced and heard, they need to know about the neuro- logical wiring responsible for speech, and they need to know how sounds are organised in speech, in order to help restore speech in their patients.

Speech is a very important personality trait: in everyday life we all make judgements about people on the basis of how they talk. We can judge whether they are lugubrious, excitable or depressed on the basis of their voice quality, volume, tempo, and so on. In a professional context, psychiatrists can diagnose certain mental abnor- malities by focusing their attention on these and other aspects of speech.

The police, too, in recent years have increasingly used 'forensic phonetics' in order to track down suspects on the basis of their geographical dialect, occupation and social class as well as physical characteristics such as sex, age and size which can be established on the basis of their speech. But although the popular press speaks of 'voice prints' which are supposedly as individual as finger prints, phone- ticians are much more cautious. The contribution of phonetics in this area is still modest.

It is probably in the development of modern communi- cations that phonetics has had the greatest impact. Alex- ander Graham Bell, the inventor of the telephone, was a Scottish phonetician, like his father and grandfather before him. Initially, like his father and grandfather, he was inter- ested in clinical applications of phonetics, but later he realised that phonetics had other uses. He emigrated to

Why study phonology? 77

America and there his invention, the telephone, was devel- oped and marketed by the Bell Telephone Company. To this day, the Bell Telephone Laboratories in America remain at the forefront of acoustic phonetics research.

Communications engineers are interested in increasing the efficiency of telephone systems. To a great extent this means cutting out redundant, non-distinctive aspects of speech so that more messages can be carried on their networks without reducing the level of clarity and intelli- gibility. To do this effectively, a good model of the physics of speech has to be complemented by an equally sophisti- cated understanding of phonological systems.

More recently, there has been a surge of interest in speech processing by computer. Acoustic phonetics and phonology are both central disciplines in the production of synthesised speech and the decoding of spoken language using computers. Both have great future promise. Once computers are capable of inputting and transcribing spoken language and producing good quality synthesised speech, they will possibly become as easy to use and as ubiquitous as the telephone. The commercial possibilities of talking computers which give verbal responses to questions, in naturalistic plain speech, without the user needing to spend hours tied to the keyboard, have attracted massive invest- ment in phonetics research in the last few years.

But even if the study of the spoken language had absolutely no practical applications it would still be an important discipline because it is central to understanding who we are. More than anything, the ability to speak, the ability to use language is what sets us apart from the rest of the animal kingdom.

Exercises

Read aloud the phonetic script:

(i) [den] (n) [ p a n ]

(iii) [saians] (iv) [sikstin] (v) [a=0g!]

following

(vi) (VI.)

(viii) (IX)

(x)

w o r d s w h i c h are w r i t t e n in

[lirjgwistiks] [hausbrcikazl [individjual] [naithud] [pr3fju3n]

78 Phonological representations

2. Make a broad transcription of the following words: (a) full fool foot coot cut but boot tuck took (b) glass path mast plastic bath last laugh (c) stair stare fare rare fair fur where wear were (d) philosophy finish enough fish caution (e) ringing wringing bringing longer long

3. Using the symbols for allophones introduced earlier in this chapter, and where appropriate using diacritics, make a narrow phonetic transcription of your relaxed pronunciation of the following words:

(a) apart attention atmosphere (b) button cotton bottom bacon baking (c) kettle little medal metal (d) drain train strain play splay sweep scream

Notes

1. Speakers of Northern English may have the vowel [u] in [nu'tmeg].

2. There is some controversy surrounding English vowels. Some regard the basic distinction as that between long and short vowels and hence transcribe them as /i: u: a: o:/, the set of long vowels, which is opposed to /i u ae D/, the set of short vowels. Others regard the quality difference as more basic and think the features tense/lax bring this out more appropri- ately. They would prefer to use the symbols /i u a D/ for the tense vowels and /i u ae D / for the lax vowels. It is the latter position which is adopted in this book.

CHAPTER 5

Phonological processes

5.1 Introduction

The phonological systems of different languages are obvi- ously different. No two languages have exactly the same inventory of phonemes which are realised by the same set of allophones; no two languages have exactly the same phonological rules regulating the deployment of their sounds. However, while all that is true, it would be wrong to ignore the similarites between languages because they are no less impressive than the differences. In the discussion of phonological symmetry in Chapter 2, we noted that languages tend to exploit the same phonetic parameters in building their phonological systems and that there are certain patterns that recur frequently.

In Chapter 3 we observed that it is feasible to use a relatively small number of DISTINCTIVE FEATURES like [± back], [± high], [± low], [± round], f± voice] etc. to characterise the phonological contrasts found in all the world's languages. One possible explanation for this is the fact that human anatomy and physiology impose limits on the range of sounds which people can produce as speakers and discriminate as hearers. Thus, for instance, since no human is endowed with a tongue which is so long that the tip can curl all the way back to the throat, it is safe to predict that no language has apico-uvular consonants made with the tip of the tongue and the uvula as the articulators (the uvula is the fleshy bit that hangs down from the centre of the soft palate at the very back of the mouth). On the other hand, given the ease with which the tip and the blade of the tongue can be raised towards the upper front teeth and the teeth ridge, it is not surprising to discover that all

80 Phonological processes

languages have either dental or alveolar sounds, if not both. Distinctive feature theory claims that there is a universal inventory of phonological construction materials from which various languages chose different elements which they use in building their phonological systems. Alterna- tively, distinctive features can be likened to cooking ingre- dients on a supermarket shelf. The selection of ingredients that a particular language puts in its shopping basket depends on the recipe which it wishes to concoct. (Of course, I do not mean this in a literal sense. Languages cannot be credited with intentional behaviour!)

It is significant, but not unexpected, that the phono- logical recipes which are available fall within the range permitted by human biology. What is intriguing is the fact that not everything that is biologically possible is equally likely to occur. Within the range of possible sounds, certain articulatory parameters are exploited by languages much more commonly than others.

Furthermore, besides exhibiting similarities in the features they use in structuring their sounds, languages also show other phonological similarities. For example, although the phonological systems of different languages are governed by different rules, the variation which occurs does, for the most part, fall within certain parameters. Similar phonological processes turn up, in language after language.

5.2 Assimilation

In Chapter 2 we saw that often in language a phoneme has several allophones, with the allophone selected in a particular position being dependent on the other sounds that are adjacent to it. The commonest phonological process responsible for this is ASSIMILATION. Assimilation, as you will recall from the discussion in Chapter 3, is the modification of a sound in order to make it more similar to some other sound in its neighbourhood. The advantage of having assimilation is that it results in smoother, more effortless, more economical transitions from one sound to another. It facilitates the task of speaking. The speaker usually tries to conserve energy by using no more effort

Assimilation 81

than is necessary to produce an utterance. (Guardians of linguistic good taste in a speech community might view many instances of assimilation as nothing less than culpable sloppiness. But their admonitions 'to uphold standards' tend to go largely unheeded.)

Usually, the alternation in the phonological realisations of a M O R P H E M E (i.e. minimal meaningful or distribu- tional unit in a language, which may be a simple word like dog, or a suffix like the -s plural marker in dogs, or the nega- tive prefix un- in un-do) can be accounted for in terms of assimilation.

Consider the regular plural ending in English. It is written as s, but it may, in fact, be pronounced as [-s] as in [pet-s] pets, or as [-z] as in [bel-z] bells or even as [-iz] as in [rauz-iz] roses. The choice is not random. The principle that determines the shape of the suffix is VOICE ASSIMI- LATION: this suffix must always agree in voicing with the preceding sound. Turning voicing on and keeping it on throughout, or alternatively, not turning voicing on at all is much easier than the alternative of turning it on or off part-way through a sequence of sounds.

More data is provided in [5.1] to illustrate the pattern of alternation of the plural ending. Study the data and answer the questions which follow:

[5.1] English Plural Formation

A

Singular

dog bid rib love sea

Plural

dogs bids ribs loves seas

B

Singular

dock bit tip giraffe moth

Plural

docks bits tips giraffes moths

C

Singular

witch nose marsh badge bus

Plural

witches noses marshes badges buses

(a) Add five more examples of your own to each column.

(b) Suggest a rule to predict the shape of the ALLO- M O R P H (i.e. alternative realisation of the morpheme) that occurs with the nouns in each column.

82 Phonological processes

If you have answered [5.1] correctly, you have estab- lished that the English plural suffix has the following allomorphs:

[5-2l W [_z] occurring with the words in column A

(ii) [-s] occurring with the words in column B

(iii) [-iz] occurring with the words in column C

Typically, as these examples show, the alternation in the shape of a morpheme is not arbitrary. Rather, it is PHONOLOGICALLY C O N D I T I O N E D . This means that the allomorph of a morpheme that occurs in a given context is partly or wholly determined by the sounds found in the allomorphs of adjacent morphemes. It is not merely coinci- dence that is responsible for the allomorphs of the plural morpheme and the third person plural being [-z -s -iz]. The suffix AGREES IN VOICING with the preceding sound. We shall provisionally state the rule thus:

the plural suffix is realised by a voiced or voiceless alveolar fricative depending on whether the noun ends in a voiced or voiceless segment. (The rule will be embel- lished presently to take into account the forms in column C.)

Note that this voice assimilation is not an idiosyncratic property of the plural morpheme. It is a rule that applies more generally to English suffixes consisting of plosives or fricatives (i.e. obstruents). For instance, the / - z / suffix, whether it is the noun plural marker as in [5.1], or the third person singular present tense suffix, shows the same pattern of alternation. You can verify this by examining [5.3]:

[5.3] English third person singular present tense suffix

sec love rub mend come call know

A

sees loves rubs mends comes calls knows

paint hate talk make laugh wipe keep

B

paints hates talks makes laughs wipes keeps

teach wish judge kiss rise lose catch

C

teaches wishes judges kisses rises loses catches

Assimilation 83

Instead of listing separately the three a l l o m o r p h s of the plural m o r p h e m e and those of the third person singular present tense m o r p h e m e (and for that m a t t e r, of any other regular, phonologically conditioned m o r p h e m e in any language), w e can posit a single U N D E R L Y I N G R E P R E - S E N T A T I O N (or B A S E F O R M ) w h i c h is entered in the ' D I C T I O N A R Y ' . T h e various a l l o m o r p h s can be derived by rule from that u n d e r l y i n g representation.

In this case, if w e assume that the base form of the n o u n plural suffix is / - z / , and if w e further assume that the base form of the third person singular present tense m o r p h e m e is also / - z / , w e can derive the various allo- m o r p h s of each m o r p h e m e using the informal V O I C E A S S I M I L A T I O N rules in [5.4]:

[5.4] English voice assimilation (i) T h e u n d e r l y i n g , dictionary representation

of the plural and of the third person singular present tense m a r k e r s is / - z / .

(ii) Insert [1] (or [a]) after any r o o t ending in a sibilant (i.e. a s o u n d such as [s z J" tf d3] w h i c h has high pitched fricative noise). See w o r d s like witch in c o l u m n C in (5.1).

(iii) M a k e sure that the suffix agrees in voicing w i t h the s e g m e n t preceding it. After a voiceless segment, like the final [k] of dock, the u n d e r l y i n g / - z / is changed t o [s] so that the r o o t and the suffix b e c o m e m o r e similar: / d o k - z / —> [dr>k-s]. See w o r d s in c o l u m n B in (5.1).

(iv) After a r o o t that ends in a voiced s e g m e n t like a v o w e l or like the voiced stop / g / in dog the suffix is realised as [z], which is also voiced, i.e. the u n d e r l y i n g / - z / in that case surfaces unmodified: / d D g - z / - > [dDgz] Sec the w o r d s in c o l u m n A in (5.1).

(v) Likewise, after the v o w e l [1] has been inserted after roots ending in a sibilant, the u n d e r l y i n g / - z / is realised as [z] so that it agrees in voicing w i t h [1]: / w D t f / —> [ w D t f - l / - > [ w D t J - l z / . See w o r d s in c o l u m n C in (5.1).

84 Phonological processes

In the next section we shall explore some general issues regarding assimilation before returning to voice assimilation in 5.3.3 below.

5.2.1 Direction of assimilation

When discussing assimilation, it is useful to look at processes in terms of DIRECTIONALITY, i.e. we can say whether a sound becomes more like either the sound that precedes it or the sound that follows it. If a sound becomes more like the sound that precedes it, the process is called REGRESSIVE assimilation; if, on the other hand, a sound is modified so that it becomes more like the sound that follows it, the process is called PROGRESSIVE assimilation.

In all the English examples above, the suffix consonant is made to agree in voicing with the root segment that precedes it. Hence, the rule of suffix voice agreement is an instance of regressive assimilation.

N o w consider the following examples from Luganda (Uganda) and determine the direction of assimilation.

[5. 5]m-bala m - p a m - m a l a n-daga n-sika n-nejia

I c o u n t I give I finish I s h o w I pull I blame

n - t e m a ji-Jagala ji-jiumya ji-coppa rj-kola 0-gula

I cut I like I converse I b e c o m e destitute I w o r k I b u y

You will have observed how, in these data, the nasal always shares the place of articulation of the consonant that follows it. Anticipating the place of articulation of the following consonant, the speaker adjusts the place of articu- lation of the nasal. The direction of assimilation in [5.5] is PROGRESSIVE (also called ANTICIPATORY):

[5.6] The nasal is realised as (i) [m] before bilabial consonants (e.g. when

one of [p b m] follows)

Assimilation 85

(ii) [n] before alveolar consonants (e.g. when one of [t d n s] follows)

(iii) [ji] before palatal consonants (e.g. when one of [c J ji] follows)

(iv) [rj] before velar consonants (e.g. when [k or g] follows)

Stop reading and write down at least two fresh exam- ples from English (or any other language you know) which exemplify progressive and regressive assimilation.

Examine the data [5.7], again taken from Luganda, and determine whether the direction of assimilation is progressive or regressive.

[5.7] A Pakulagrab! (imp.) Palacount! (imp.) ka-Puzismall goat ka-Pogo small buffalo

B m-bakula m-bala m-buzi m-bogo

I grab I count goat buffalo

The process whereby the nasal is altered to agree in place of articulation with the f o l l o w i n g consonant must be viewed as anticipatory (i.e. progressive) assimilation. However, if you examine the data closely, you will observe that there is another process going on at the same time. The bilabial fricative [P] alternates with a plosive: the fricative occurs word initially (see column A) while the plosive [b], which shares the same places of articulation, occurs after a nasal (see column B). The fricative, which is a continuant, assimilates the property of being noncontinuant (i.e. being a stop) from the nasal immediately preceding it.

What [5.7] shows is BIDIRECTIONAL assimilation. The nasal prefix assimilation rule which ensures that all nasals have the same place of articulation as the f o l l o w i n g consonant exemplifies progressive assimilation; but the S T R E N G T H E N I N G of continuants so that they are real- ised as stops when they are preceded by a nasal stop is an example of regressive assimilation.

86 Phonological processes

5.3 Assimilation processes

Another way in which assimilation processes can be seen is in terms of whether a vowel or consonant acquires vowel or consonant features of a neighbouring segment. Various patterns are examined in turn below. The coverage is not intended to be exhaustive. It is only meant to show some of the commonest assimilation processes found in the languages of the world.

5.3.1 Palatalisation

Say the following words and describe the position of your tongue during the production of the first consonant in each one of them:

[5.8] key [ki] car [ka] then say: [ka kiz] keep [kip] calm [kam] [kip kam] get [get] garlic [galik] [get gahk] give [giv] guns [gAnz] [giv gAnz]

Observe that, in each case when a velar consonant is followed by a front vowel, there occurs some slight antici- patory fronting of the part of the tongue that makes contact with the roof of the mouth. This fronting is indicated by a subscript (+) under the consonant. The effect of the fronting is that the velar consonant is made partly in the palatal region. This process is called PALATALISATION. Velar consonants often have slightly palatalised allophones which occur after front vowels because the tongue is raised towards the hard palate in the production of front vowels and speakers anticipate that gesture and start making it before they have completed the articulation of [k] or [g].

Palatalisation is not limited to velar consonants. It is equally possible to palatalise anterior consonants. In fast, causal spoken English, for instance, alveolar consonants are usually palatalised when they occur at the end of a word and are followed by another word which begins with an alveo- palatal consonant:

[5.9] his shoes [hiz Juz] —» [hi3 Juz] nice shirt [nais J3t] —» [naij" /3t] miss Ure [mis jua] —» [mif jua] John's shorts [d3Dnz Jots] —* [d3Dn3 Jots]

Assimilation processes 87

N o w use the notion of palatalisation to help account for the alternative pronunciations of the words in [5.10].

[5.10] A B C issue [ISJU] ~ [ijju] ~ [iju] consume [kansjum] ~ [kanjjum] ~ [kanjum]

Add at least two more examples of your own.

If the forms in A are taken as the C I T A T I O N FORMS that would be listed in a dictionary, B can be explained as a case of palatalisation. The speaker, anticipating the palatal approximant which follows, places the tongue in the alveo- palatal region instead of the alveolar region. In C the speaker once again anticipates the palatal approximant and has the tongue making contact with the alveo-palatal region and drops the [j] sound which triggered off the palatalis- ation in the first place.

5.3.2 Labialisation

Palatalisation is not the only vowel feature which can be acquired by consonants. Say the words in [5.11] and then get another person to say them. Watch your partner's lips very carefully. Describe the lip position adjustment process which you observe as the initial consonant of each pair of words in A and B is said.

peel tea she leek get

A [Pil] [«] [Ji] [Hk] [get]

pool two shoe Luke got

B [pwul] | t w u | [;wui [lwukl [gW°t]

In each case, the word in B is said with some degree of secondary lip rounding. Anticipating the next segment, which is a rounded vowel, the speaker starts rounding the lips before the articulation of the consonant is completed. This assimilation process is called LABIALISATION or ( R O U N D I N G ) . It can be indicated in a phonetic transcrip- tion by using the raised w after a consonant [Cw].

88 Phonological processes

Find two examples of labialisation in any language you know and write them down using a narrow phonetic transcription.

5.3.3 Voice assimilation

You will remember that in section 5.2 of this chapter we saw that in English suffixes agree in voicing with the last segment of the stem to which they are attached. That is a classic example of VOICE ASSIMILATION: whatever happens to be the specification for the feature [voice] of the preceding segment of the root is automatically carried over into the suffix. The relevant examples are not repeated here. If you want to refresh your memory, please read section 5.2 again.

The phonetic cause of voice assimilation is well-under- stood. Given the fact that speech is a continuum, the process of putting the vocal cords close together to produce voicing or keeping them wide apart to produce voicelessness is not always perfectly synchronised with other articulatory gestures. This may mean voicing spilling over into an adjacent segment. This frequently happens where a voiceless consonant occurs between two (voiced) vowels. In many languages, in that position, 'voiceless' consonants acquire a certain amount of voicing. This happened historically as Spanish developed from Latin. One of the changes that occurred was the voicing of voiceless stops between vowels. For instance, Latin fata 'fate' became fada in Spanish. The converse situation is also attested. In some languages a vowel is devoiced when it occurs between voiceless conson- ants. That is the case in Japanese in a word like [kita] 'came'.

Cast your mind back to the French problem in Chapter 2. The data is reproduced below for convenience. Recall that in French, at the end of a word the LIQUIDS (i.e. / I / and /r/) as well as the nasals agree in voicing with the preceding consonant: they are voiced after voiced conson- ants (as in [5.12]), and voiceless after voiceless ones (as in [5.13]):

Assimilation processes 89

[5.12] French voice assimilation: word final nasals and liquids are voiced after a voiced segment,

matinal 'morning' (adj.) li:r 'to read' film 'film' tabl 'table'

[5.13] French voice assimilation: word final nasals and liquids are voiceless after a voiceless segment, metr 'to put' tapl 'temple' Jiff 'figure' rymatisrn 'rheumatism'

Examine the data in [5.14] and describe the rule respon- sible for the voice assimilation in Kalenjin (Kenya) (Toweett 1975)-

[5.i4]kep to notch kebe:t is notching nap to sew nabe:t is sewing luk to fight luge:t is fighting ku:t to blow ku:te:t is blowing

There is a twist in the data. While labial [p] and velar [k] acquire voicing in intervocalic position, alveolar [t] does not. It remains voiceless between vowels.

5.3.4 Place of articulation assimilation

We saw in [5.6] that in Luganda the place of articulation of a nasal is predictable from the place of articulation of the consonant that comes after it.

N o w apply the same kind of analysis to the Malay data in [5.15] and state the distribution of the allomorphs of the agentive nominalising prefix /pan/. This prefix can be attached to most verbs to form a noun with a meaning similar to that of a noun derived from a verb in English using the -er suffix.

90 Phonological processes

[5.15] Malay (Dodds 1977)

baca bali baranan danar dakwa cakap cun gosok gali

read buy swim hear prosecute speak steal polish dig

pambaca pambeli pambaranan pandanar pandakwa pajicakap pajicun pajigosok pajigali

In [5.15], the nasal is H O M O R G A N I C with the consonant that goes after it, i.e. the nasal shares the place of articulation of the following consonant. Hence the use of the label H O M O R G A N I C NASAL ASSIMILATION to refer to this assimilation process. The homorganic nasal assimilation rules which were suggested for Luganda in [5.6] would also cover Malay. In both languages the assimi- lation is automatic. It applies wherever a nasal is followed by another consonant in the same word.

Homorganic nasal assimilation also applies in English, albeit somewhat sporadically. Compare the two columns in [5.16] which show the negative prefixes in- and i r - respectively:

[5.16] A [in] in-appropriate [im] im-plausible [in] in-decent [in] in-gratitute

B [An] un-exciting [An] un-pretentious [An] un-deserving [An] un-grateful

Add three fresh examples to each column, choosing a different consonant after the prefix, but avoiding the prefixes ir- and il- (see next section). Say carefully each word that you add to the list, noting the changes in the place of articulation of the nasal in column A in anticipation of the place of articulation of the consonant that follows. Assume that the underlying representation of the prefix in A is in- since that is the form that appears before vowels, a position where there is no phonetic motivation for

Assimilation processes 91

modifying the place of articulation (given the fact that vowels have n o place of articulation).

H o m o r g a n i c nasal assimilation is n o t an a u t o m a t i c and obligatory rule of English p h o n o l o g y . It applies selectively to certain forms and is n o t triggered by phonological infor- m a t i o n alone. T h e r e are w o r d s w h i c h contain n o n h o m o r - ganic nasals. T h e h o m o r g a n i c assimilation rule n o r m a l l y applies to in- b u t n o t to un-, except in casual speech w h e r e , for example, unkempt and unpleasant m a y be p r o n o u n c e d [Ankempt] and [Amplezant] respectively.

It is interesting that across w o r d boundaries, in fast speech, consonants (especially alveolar ones), can be o p t i o n - ally h o m o r g a n i c w i t h the following c o n s o n a n t . T h e r e , n o grammatical restrictions seem to apply:

[5.17]bad m a n [baed maen] —> [baeb0 maen] ten m e n [ten m e n ] —> [tern m e n ] w h a t car [wDt ka] —> [wDk° ka] top ten [tDp ten] —» [tDt° ten] (C° = unreleased stop: n o r m a l l y there is n o audible release of a stop followed by another stop.)

Find t w o m o r e examples of assimilation similar to those in [5. 17].

5.3.5 Manner of articulation assimilation

In the last section w e established that the u n d e r l y i n g r e p r e - sentation of the negative prefix w h i c h occurs w i t h adjectives in [5.16] m u s t be in-. If that a s s u m p t i o n is correct, h o w can w e explain the historical d e v e l o p m e n t s h o w n in [5.18]?

[5.18] in-legal in-licit in-rational in-revocable

—* illegal —» illicit —» irrational —» irrevocable

92 Phonological processes

The answer seems to be that before roots whose first consonant is a LIQUID (/l/ or /r/) the nasal of the negative prefix assimilates the manner of articulation features of the liquid so that / n / becomes [1] before /l/-commencing roots or [r] before /r/-commencing roots. However, this alter- nation is not purely phonetically conditioned since [n] can be followed by [1] or [r] in words such as unloved or unreason- able. In fact, this is a historical assimilation which took place in Latin, and is simply reflected in modern English through Latin borrowing.

An example of a manner of articulation assimilation rule which is still active in a contemporary language is given in the Cairo Arabic data in [5.19]. The underlying repre- sentation of the definite article is /il-/. However, it is not realised as [il-] in all contexts. Whether it is realised as [il-] or as something else depends on the first consonant of the root to which it is prefixed.

Study [5.19] and suggest a rule to account for the realisation of the definite article.

Here is a clue: the definite article is realised as [il-] unless the noun to which it is attached shares some place of articulation features with / l / .

[5.19] Cairo Arabic (Harms 1968) kursi 'chair' ilkursi 'the chair' dars 'lesson' iddars 'the lesson' innimra 'the grade' ilmudarris 'the teacher' issatr 'the line' ilwa:gib 'the assignment' ilba:b 'the door' ilqism 'the section' issanta 'the bag' ilge:b 'the pocket'

We could state the rule this way informally:

the [1] of the definite article takes on the manner of articu- lation features of the initial root consonant and becomes indistinguishable from it, where like [1], that consonant is anterior and coronal (i.e. made at the alveolar place of articulation).

The motivation for this kind of change is not difficult to see. Ensuring that segments made at the same place also

Assimilation processes 93

agree in manner of articulation is a way of minimising articulatory effort. Instead of making two articulatory gestures the speaker only makes one and holds it for a longer period.

5.3.6 Nasalisation

NASALISATION is a process whereby an oral segment acquires nasality from a neighbouring segment. Again, the articulatory motivation for this is self-evident. In order to produce a nasal segment, it is necessary to lower the velum (soft palate) and allow air to escape through the nose (the lower the soft palate is, the higher will the degree of nasal- isation be); to produce an oral sound, it is necessary to completely block off access to the nasal cavity by raising the velum as high as it can go. Any leakage of air past the velum will cause some nasalisation. To maintain an absolute distinction between oral and nasal consonants would require perfect synchronisation of velic closure with the other articu- latory parameters of (a) P H O N A T I O N (i.e. production of voicing), (b) the PLACE OF ARTICULATION and (c) the M A N N E R OF ARTICULATION. This is not always possible. Typically some nasalisation seeps through and affects an oral segment which is adjacent to a nasal. In many languages the nasalisation is prominently audible. Examples of nasalised vowels (V) are the sound [e] and [a] in the French words [pe] pain 'bread' and [mama] maman 'mum'.

In Kikuyu (Kenya), vowels have nasal variants which occur in the neighbourhood of nasal consonants, as you can see in [5.20]:

[5.20] Kikuyu nasalisation (Leakey 1959)

mondu person moanake young man ny5ni bird rjgolo heart kehembe drum

tato ihiya 163 kovolo oholo

three stone father foot news

Historically it is almost certain that nasalisation is always a consonant feature which is assimilated by vowels. However, in a synchronic description of a language it is possible to find vowels which are always nasal and which

94 Phonological processes

must be presumed to be underlyingly nasal. That is the case in Yoruba (Nigeria) where nasal vowels occur in the absence of nasal consonants in words like [odu] 'year'. At one time there would have been a nasal consonant con- ditioning the vowel nasalisation but it has disappeared.

5.4 Dissimilation

We have seen in the preceding section of this chapter that assimilation processes typically have a transparent phonetic basis that can be stated in terms of ease of articulation. But not all phonological processes can be plausibly explained in terms of assimilation. If we recognise the fact that phono- logical systems have to meet the needs of language users both as speakers and as hearers, we can easily appreciate that while assimilation (by making sounds more similar to each other) facilitates speech production, it does also have the undesirable effect of making the hearer's task of discrimi- nating between sounds somewhat more difficult.

(' Phonological processes which ensure that differences between sounds are enhanced so that sounds become more auditorily distinct make speech perception easier. DISSI- MILATION is the term used to refer to processes of that kind. The effect of dissimilation is to make sounds more distinct from other sounds in their environment. After a dissimilation rule has applied, phonological elements are less like each other than they were before the rule applied.

Let us begin by examining a very limited set of data from English which illustrates dissimilation. In English, the adjective forming suffix -al has two phonetic manifestations. Sometimes it is -al, as in column A in [5.21] below and sometimes it -ar, as in column B. Bearing in mind the fact that dissimilation plays a role in this, study the data and describe the factor which determines the allomorph that occurs in any particular instance. Pay special attention to the last segment of the noun which is adjacent tp the adjective ending. Say the words aloud and transcribe the final segment phonetically.

Dissimilation 95

[5.21] A B noun electric region orbit baptism genitive culture

adjective electrical regional orbital baptismal genitival cultural

noun angle circle table circle title single

adjective angular circular tabular circular titular singular

The pattern is clear. The shape -al is the base form. It is the form which you add in column A where the last consonant of the noun is a sound other than [1]. Verify this by looking at the transcription which you have made. The shape -ar is the alternant which is normally added where the last consonant of the noun is [1]. The addition of -al after a root ending in [1], which would have resulted in two [1] sounds merely separated by a schwa, is thus avoided. But like many other rules, this rule has exceptions, e.g. linear not *lineal although line does not end in -/. I (Note also in passing that in column B, [ju] is inserted between the final [1] and the consonant preceding it.)

The English dissimilation process is another Latin relic. It survives in words borrowed from Latin but is not a productive rule. It is often difficult to know where to draw the line between a synchronically relevant phonological process and a historical relic which is no longer relevant. We shall return to this problem in Chapter 8.

For our next example we shall consider a productive synchronic dissimilation process. In many Bantu languages there is a rule which requires a consonant in a prefix to DISAGREE in voicing with the first consonant of the root to which it is attached:

a vbiced stem initial segment requires a voiceless conson- ant in the prefix and a voiceless stem-initial segment requires a voiced consonant in the prefix.

Consider the Kirundi (Burundi) examples in [5.22]:

[5.22] (Kirundi dissimilation (Kenstowicz and Kis- seberth 1977)

a. Imperative 1st person singular present rya eat tu-rya

96 Phonological processes

m w a va b o n a s o m a te:ka seka kubita

shave c o m e see read cook laugh hit

from t u - m w a tu-va t u - b o n a d u - s o m a du-te:ka du-seka d u - k u b i t a

This dissimilation rule in Bantu is called Dahl's law, after the scholar who first described it.

5.5 Conclusion

In this chapter a number of common phonological processes have been examined. It has been shown that there are often good phonetic reasons for phonological processes. For the most part, phonological alternation in the shape of a mor- pheme has a phonetic motivation and that motivation tends to be similar in different languages. This largely explains the considerable degree of similarity between the phonolo- gical patterns found in different languages.

Exercise

I . Before attempting this question read [5.2] again. Next do the following:

(a) Make a broad phonetic transcription of the English data in example [5.1] above.

(b) Propose a rule which predicts the shape of the genitive suffix which goes with each one of the nouns in [5.1].

S t u d y the 1 regular past

present w a l k l o o k trap wish laugh w a t c h launder a r m

data b e l o w s h o w i n g : tense ending

past w a l k e d looked trapped w i s h e d l a u g h e d w a t c h e d laundered a r m e d

the in English: present paint w a n t part fit court land s o u n d fade

realisation of the

past painted w a n t e d parted fitted c o u r t e d landed s o u n d e d faded

Conclusion 97

warn warned pad padded sue sued row rowed

(a) Make a broad phonetic transcription of the data above. (b) Suggest a rule to account for the realisation of the

regular past tense ending. (c) Make a single statement to account for the realisation

of both the plural suffix and the past tense suffix.

CHAPTER 6

Naturalness and strength

6.1 Introduction

It will have b e c o m e clear in the last chapter that the alter- nation in the p h o n o l o g i c a l realisation of m o r p h e m e s is for the m o s t part n o t arbitrary. T h e same s e g m e n t types t u r n u p in similar processes found in diverse languages. P h o n - ologists h a v e used the t e r m N A T U R A L N E S S to refer to the fact that there is, for the m o s t part, a phonetically w e l l - m o t i v a t e d relationship n o t only b e t w e e n the allophones of a p h o n e m e , b u t also b e t w e e n the various phonological m a n i f e s t a t i o n s of a m o r p h e m e . N a t u r a l n e s s can be a p p r o a c h e d in t e r m s o f M A R K E D N E S S . W h a t is N A T U R A L can be said to be U N M A R K E D , and w h a t is n o t natural can be said to be M A R K E D , i.e. in s o m e sense unusual. T h e p u r p o s e of this chapter is to explore this p h e n o m e n o n .

6.2 Natural segments, natural classes and natural processes

It is n o t only classes of sounds which are affected b y the same phonological processes that tend to be m a d e u p of s e g m e n t s w h i c h are p h o n e t i c a l l y n a t u r a l . I n d i v i d u a l s e g m e n t s themselves also tend to contain phonetic features w h i c h are natural. Partly due to physiological constraints, n o t every conceivable c o m b i n a t i o n of features results in permissible s e g m e n t s . At a very o b v i o u s and trivial level, the fact that n o b o d y has lips long e n o u g h to m a k e contact w i t h the uvular precludes labio-uvular consonants. M o r e significant, h o w e v e r , is the fact that n o t all physically p o s s -

Natural segments, natural classes and natural processes 99

ible feature combinations are equally probable. Certain feature combinations are more likely; they are the ones that recur again and again in various languages. They are the unmarked combinations.

For instance, voiceless sonorants such as the nasals [m n ji rj] are much less common than their voiced counterparts [m° n ji rj]. Likewise, voiceless approximants like [w j 1 r] are less common than voiced ones. Sounds produced "with the velaric airstream mechanism (i.e. clicks) and those produced with the glottalic airstream mechanism (i.e. implosives and ejectives) are less common than sounds produced with the pulmonic airstream mechanism. Front rounded vowels are rare (the main concentration of front rounded vowels is north-western Europe where they occur in languages such as French, German and Swedish) but front unrounded vowels are not. Nasalised vowels, though widespread, are still much less frequent than their oral coun- terparts both in the world's languages and in those languages where they occur. That certain combinations of features are more favoured, more natural than others is beyond dispute. We shall consider some possible reasons for this at the end of this chapter.

I invite you now to determine which of the segments in each pair below is marked (less natural). State your reasons.

[6.1] A B u u 6 o ce e

In each case the sound in column A is marked. Vowels are normally voiced like [u] and not voiceless like [u]; vowels are normally oral like [o] and not nasal like [6]; front vowels are normally unrounded like [e] and not rounded like [oe].

Languages typically have both natural segments and natural phoneme inventories. In Chapter 2 we noted that phoneme inventories tend to be SYMMETRICAL. That observation can be restated in terms of naturalness. As a

100 Naturalness and strength

rule, creating symmetrical phoneme inventories entails maximising the use of a few phonological parameters. Such an arrangement is economical and has the merit of reducing the burden on memory during language acquisition: a small number of features is learned and is reused many times. This is preferable to having phonemes which have little in common with each other and which entail mastering numerous distinctive features.

Contrast the inventories in [6.2] with those in [6.3]:

[6.2] (a) Swahili vowels i u e o

a (b) Turkana (Kenya) consonants (Dimmendaal

1983) p t c k b d J g m n ji n

s 1 r

J w

[6.3] (a) i (b) p t * 1 d g e m ji e as

A p h o n e m e i n v e n t o r y like that in [6.3b], w i t h its m a n y holes in the pattern, is less likely to be attested in natural languages than the s y m m e t r i c a l T u r k a n a i n v e n t o r y in [6.2b], w h i c h distinguishes voiced stops, voiceless stops and nasals at each one of the three places of articulation exploited. Likewise, the balanced Swahili s y s t e m in [6.2a] w h e r e front vowels are paired w i t h back ones is a m o r e likely v o w e l i n v e n t o r y than the s k e w e d one containing only front v o w e l s w h i c h is s h o w n in [6.3a].

While recognising the i m p o r t a n c e of s y m m e t r y , w e need to constantly bear in m i n d the fact that it is n o t an a b s o l u t e i m p e r a t i v e . I n e l e g a n t , s k e w e d p h o n o l o g i c a l

Natural segments, natural classes and natural processes 101

systems are not unheard of. The Kikuyu plosives system, for instance, contains no bilabial voiceless stop:

[6.4] Kikuyu (Kenya) t k

b d g

We shall end this section by re-examining data first introduced in the last chapter, which are reproduced below as [6.5] for convenience. These data illustrate the role of NATURAL CLASSES in the phonology of Luganda:

[6.5] m-bala I count m-pa I give n-daga I show n-sika I pull ji-^ela I sweep rj-gula I buy rj-kuba I hit

The segments [m n ji rj] which acquire the place of articulation of the following consonant are not a random collection of segments; rather, they are a coherent class of phonetically similar sounds. They form a natural class. The homorganic nasal assimilation rule affects only the natural class consisting of nasal consonants.

The other phonological processes discussed in Chapter 5 also affect natural classes. The segments which condition or undergo a phonological process do share in each instance some phonetic characteristic. Thus, typically, palatalisation of velar consonants occurs in the context of front vowels (especially high ones like [i]) which are themselves produced with the tongue approximating the hard palate. Labialis- ation occurs in the neighbourhood of labial vowels like [u] which are themselves produced with rounded lips. Nasal- isation of vowels occurs when they are adjacent to nasal consonants. Voiceless consonants may acquire some voicing when juxtaposed with inherently voiced segments like vowels or sonorants (such as nasals), and so on.

Important though they are, naturalness and markedness are not absolute concepts. Rather, they are both relative. What is marked on unmarked will often depend on the circumstances. A nonlinguistic analogy should help to clarify this: wearing a bikini or a kilt and sporran are in a

102 Naturalness and strength

sense marked modes of dress. Most people most of the time do not go about their business so attired. However, a woman in a bikini on the beach on a hot, sunny day or a Scotsman in a kilt at a Burns Supper would respectively be very 'unmarked'. So it is in phonology. Nasal vowels, to take one example, are marked. Indeed, we would be extremely surprised if we found a language which had only nasal vowels and no oral ones. However, between two nasal consonants, or before a nasal plus consonant cluster like [nd], nasalised vowels would be unmarked. It would be somewhat unusual for vowels occurring in those contexts to have no nasalisation. Using an oral [ae] in [maen] is marked but using a nasalised [ae] and saying [mien] is unmarked. The same would apply to palatalisation of velars before high front vowels or labialisation before rounded vowels. Markedness cannot be interpreted with total disre- gard for context.

For the next example look back at the discussion of voice assimilation in English in section 5.2 of the last chapter. The consonants in the suffixes -s, -z, -iz; -t, -d, -id that figure in English voice assimilation are either frica- tives or stops. This is not accidental. In many languages fricatives and stops (together with affricates) form a natu- ral class to which the label O B S T R U E N T (or N O N - S O N O R A N T ) is given. These sounds share the phonetic characteristics of having very significant obstruction in the oral tract and of being typically voiceless. They also tend to display similar phonological behaviour.

A bat goal vine zoo gin

B tab log safe as edge

C pat coal fine sue chin

D tap lock safe ass etch

Say the words in [6.6] aloud again and again. Compare the voicing of the words on the same line carefully and see if there is some kind of pattern.

Natural segments, natural classes and natural processes 103

What you will discover is that in English obstruents are more heavily voiced at the beginning of a word than they are word-finally. Generally word final voicing is so atten- uated that it is barely detectable. In fact, the distinction between 'voiced' and 'voiceless' obstruents word-finally is less important in discriminating between words than the lengthening of the vowel that precedes a 'voiced' consonant.

It is natural for obstruents to be voiceless. Many languages have more voiceless obstruents than voiced ones and some have no voiced obstruent phonemes at all. That is the situation in many Australian languages. And numerous languages, with voiced obstruents, have a rule which devoices them in syllable final position. Such an obstruent devoicing rule can be seen at work in Turkish:

[6.7] Turkish obstruent devoicing (Kenstowicz and Kisseberth 1977:50)

objective absolute plural gloss ip-i ip ip-ler rope dib-i dip dip-ler bottom at-i at at-lar horse ad-i at at-lar name

The stops / b / and / d / are devoiced in syllable-final position, i.e. when they occur in preconsonantal or word final position.

6.2.1 Phonological strength hierarchies

In the last chapter we saw that assimilation and dissimilation are useful concepts for elucidating phonological alternation. Many natural phonological processes involve some kind of assimilation or less commonly, dissimilation. But assimi- lation and dissimilation are not the only concepts in terms of which naturalness can be discussed.

Many phonological processes can be fruitfully exam- ined using the notions of S T R E N G T H E N I N G (also called FORTITION) and WEAKENING (also called LENI- TION). These two concepts are not independent of each other. They are merely two poles of the same gradient. As in everyday life, strength and weakness are relative. In [6.8]

104 Naturalness and strength

I have reproduced a commonly accepted phonological strength hierarchy (> indicates a step towards a 'weaker' pronunciation):

[6.8] (a) VOICELESS > VOICED (b) STOP > AFFRICATE > FRICATIVE > A P P R O X I M A N T > Z E R O

Before you proceed, I suggest that you work out the phonetic parameter(s) on which this hierarchy is based. Is it place of articulation, manner of articulation, airstream mechanism, phonation or something else? N o w use your answer to rank the following sounds on the hierarchy which you have established. In some cases you might need to have more than one sound occupying the same place:

[6.9] t r o P z s k d g m w 3 8 p b a

In [6.8a] S O N O R I T Y is the parameter in question. As a first approximation, sonority is related to voicing. The greater the propensity a sound has of spontaneous voicing, the more sonority it has:

[6.10] Sonority hierarchy least sonority

1 voiceless obstruents (e.g. t s k) 2 voiced obstruents (e.g. d g P z) 3 nasals (e.g. m) 4 liquids (e.g. r) 5 glides (e.g. w) 6 ,, vowels (e.g. a 0)

greatest sonority

The sonority hierarchy is an inverse restatement of the strength hierarchy (see also [9.6] page 158 below). In view of this, some phonologists question the need to distinguish between the strength hierarchy and the sonority hierarchy. Under the heading of strength they would prefer to include not only stricture (i.e. obstruction of the airstream) but also sonority.

Grouping together stricture and sonority under the rubric of the strength hierarchy might strike you as odd.

Natural segments, natural classes and natural processes 105

On the face of it, the two phenomena appear unrelated. However closer examination reveals an interesting relation- ship, especially where language evolution is concerned. To understand this relationship we need to draw a distinction between S Y N C H R O N I C and D I A C H R O N I C approaches to the study of language. Synchronic linguistics studies the state of a language during one period in its history. A grammar of mid-twentieth century English is a synchronic description. Diachronic (or historical) linguistics on the other hand, studies the evolution of a language during successive periods.

Frequently, both synchronically and diachronically, voiceless consonants change into voiced ones in environ- ments similar to those where the reduction in the strength of the obstruction in the production of consonants takes place. Indeed, one process may facilitate the other. It is significant that the sounds at the weak end of the strength hierarchy are typically voiced, while those at the strong end are normally voiceless. Voiced sounds are weaker than their voiceless counterparts - [d] is weaker than [t], [z] is weaker than [s] and so on. When a voiceless sound like [t] becomes voiced, we can speak of 'weakening'.

If you are in any doubt as to what the answer to [6.8b] ought to be, turn back to the first chapter and work out the manner of articulation of each sound. The strength hier- archy in [6.8b] is based on manner of articulation i.e. the way and the extent to which the airstream is obstructed in the articulation of a particular sound. Stops involve the strongest obstruction and approximants the weakest, with the remaining types of sound falling in between. Of course, having no obstruction at all and dropping a sound alto- gether is the ultimate form of weakening.

Many phonological concepts like assimilation, strengthening, etc., refer to processes that can be under- stood from either a synchronic or diachronic point of view (see Chapter 8). Strengthening and weakening are exempli- fied below using some data from historical phonology.

Compare the forms in [6. n ] (Hooper 1976):

[6.11] (a) Latin Italian Spanish French vita vita vida vie 'life'

(b) Latin Italian Spanish French

106 Naturalness and strength

voiceless *-voiceless ^-voiced—•- zero stop stop stop t +• t *- d *- 0

(where *- = becomes)

The changes exemplified involve movement down the sonority hierarchy.

In some cases, before being deleted a voiced stop may go through a fricative phase. The word meaning 'loyal' in modern Spanish is derived from Latin legale whose [g] weakened to the velar fricative [y] in Medieval Spanish when it was pronounced as leyale. The velar fricative was eventually lost, resulting in the modern Spanish form leal. In this case the progression down the sonority hierarchy is from a voiced stop to a voiced fricative before deletion takes place.

The examples presented so far have been of weakening. But sounds can move up the strength hierarchy. In Luganda [l], an approximant, becomes a stop [d] when it is immedi- ately preceded by a nasal stop:

[6.12] [laga] 'show' [ndaga] 'I show' [kaliga] 'lamb' [ndiga] 'sheep'

Some linguists have added PLACE OF A R T I C U - LATION as a parameter of the universal strength hierarchy:

[6.13] LABIAL > ALVEOLAR > VELAR

This dimension of the hierarchy seems to be valid for the languages of Western Europe, like Danish, where in inter- vocalic position, velar / g / undergoes the most extreme form of lenition (i.e. weakening), being deleted altogether; alveolar / d / is moderately weakened, being changed to the fricative [3] but labial / b / remains unchanged.

However, the universal validity of a place of articu- lation hierarchy is doubtful. In many Bantu languages, for instance, it is the labial place of articulation that is weakest. Many languages in this family historically weakened and completely dropped the labial stops / p / or / b / while retaining the alveolar and velar ones. Thus in Kikamba (Kenya) the reflex of the Proto-Bantu noun class prefix *ba- is a-. The labial [b] was deleted except where it was preceded by a nasal. After a nasal it survives as prenasalised

Natural segments, natural classes and natural processes 107

[mb]. Proto-Bantu / p / is equally prone to lenition *p—»h in Sukuma (Tanzania), Rundi (Burundi), Pare (Tanzania) etc.; *p > ({> in Pokomo (Kenya), and Rimi (Tanzania) etc. Typically velars and alveolars do not undergo lenition to the extent that labials do. Given this evidence, it would be unwise to insist on the universality of a strength hierarchy based on place of articulation.

It is arguable that the sonority and manner of articu- lation strength hierarchies in [6.8] should be replaced by the single hierarchy shown in [6.14]. This, for the reasons given above excludes the dimension of place of articulation.

[6.14] VOICELESS STOP > VOICED STOP > VOICELESS AFFRICATE > VOICED AFFRICATE > VOICELESS C O N T I N U A N T > VOICED C O N T I N U A N T > NASAL > A P P R O X I M A N T

One type of consonant not included in [6.14] which there is strong evidence for is the GEMINATE consonant like [t:] or [d:], usually represented by doubling consonant letters ([tt] or [dd]). Gemination occurs when two identical consonants are adjacent to each other in the same syllable as in English penknife [pen:aif]; in other words, gemination occurs when a particular segmental articulation is prolonged to cover what would otherwise be two distinct segments. Geminate consonants occupy the highest rung of the hierarchy.

The strength hierarchy, re-stated now in [6.15], is manifested in synchronic phonological alternation and in historical sound change in numerous languages:

[6.15] GEMINATE VOICELESS STOP > GEMI- N A T E VOICED STOP > VOICELESS STOP > VOICED STOP > VOICELESS FRICATIVE > VOICED FRICATIVE . . . .

The relative strength remains the same with other manners of articulation.

Synchronically, in many languages in positions of weakening, for instance between vowels, geminate segments like [t:] (spelled with tt in the examples which follow) alternate with plain segments like [t]. Thus in Luganda the singular of the word 'branch' is ttabi [t:aBi] but

108 Naturalness and strength

the plural is matabi [mataBi]. An historical example of the same kind is provided by the word for 'drop' in Romance languages where the reflexes of Latin gutta are gota in Spanish and goutte in French. The original geminate has changed to [t] in both languages.

6.3 Explanations of naturalness

The assimilatory phonological processes of the kind which we explored in the last chapter all have a firm basis in articulatory phonetics. We can point to a good articulatory reason for a process like palatalisation, labialisation or nasal- isation. Where possible, adjacent sounds are made similar to each other so that one avoids using any more effort than is required to ensure that one is understood by the addressee. But while assimilation makes the task of speech production easier, it can make speech perception more difficult. It is easier to discriminate between sounds if they are very different from each other than it is to distinguish them when they are very alike. To counterbalance assimi- lation, there are natural processes which have the effect of enhancing differences between sounds. These facilitate the task of the hearer. In Chapter 5, voice dissimilation in Kirundi was introduced as an example of dissimilation. When this rule applies, the prefix and the root end up not having the same value for voicing, which makes them more different from each other than they would otherwise be.

Vowel patterns also frequently obey the principle of M A X I M U M PERCEPTUAL DIFFERENTIATION. The set in [6.16] turns up in language after language with a three vowel system (see section 3 of Chapter 2).

[6.16]

front back high

low

Explanations of naturalness 109

The choice of these vowels is not accidental. These three vowels occupy the most peripheral positions in vowel space: [i] is the highest front vowel, [u] is the highest back vowel and [a] is the lowest vowel. Perceptually they are maximally distinct. If a language has only three vowels, in order to avoid hearers getting them confused (and as a consequence getting the meanings of words in which they occur confused), it is almost invariably the three vowels [i a u] that are selected. Languages as diverse as Greenlandic Eskimo and Australian Pitta-Pitta have /i a u/ as their only vowel phonemes.

Languages with a five vowel phoneme system consisting o f / i e a o u/, if they have rules which neutralise vowel distinctions in certain environments, tend to maintain the opposition between /i a u / in the places of neutralis- ation. An example should clarify this. In many Bantu languages with the five vowel phonemes /i e a o u/, only the peripheral vowels /i a u/ occur in noun class prefixes which mark the class (or gender) of a noun, as you can see in the Luganda forms below:

[6.17] wM-sota snake mi-sota snakes mw-ntu person Ba-ntu people ki-ntu thing J3i-ntu things fefl-tale market fin-tale markets

6.3.1 Natural phonology

Pholonogists do not agree on the place of naturalness in phonological theory. Those who belong to a school called NATURAL P H O N O L O G Y which was pioneered by Stampe (1973) argue that the phonological component of language is not merely the result of C O N V E N T I O N , rather it is the way it is for very good reasons. It reflects, on the one hand, the auditory, articulatory and speech processing capacity of humans, and on the other hand the functions which language serves. They maintain that in all languages, phonemes and phonological processes are the residue of UNIVERSAL PHONOLOGICAL PROCESSES. Typically these are assimilation processes like homorganic nasal assimilation, patalisation, nasalisation, voice assimi-

110 Naturalness and strength

lation, and so o n . It is claimed that in acquiring the p h o n o - logical system of its language, a child has to learn to S U P P R E S S in different w a y s and to different l a n g u a g e - specific degrees various innate universal phonological processes. A r u m p of those processes survives, tailored to fit w h a t e v e r quirks a particular language m a y have acquired o v e r the centuries. It is suggested that the striking similarity that exists b e t w e e n the phonological processes found in unrelated languages can be largely explained in t e r m s of their h a v i n g preserved the same or similar aspects of universal natural processes.

In this approach, a sharp distinction is d r a w n between rules and processes. T h e t e r m R U L E is used to refer to phonetically w h o l l y or partially u n m o t i v a t e d alternations, like those exemplified in the next paragraph, w h i c h are g o v e r n e d b y the c o n v e n t i o n s of a particular language. P R O C E S S E S are alternations w h i c h are regulated by universal phonetic or functional factors. U n l i k e processes, rules are idiosyncratic properties of particular languages and d o n o t form part of h u m a n k i n d ' s c o m m o n phonological inheritance. N a t u r a l processes are m o r e c o m m o n than i d i o - syncratic rules. D o n e g a n and S t a m p e (1979:127) claim that natural p h o n o l o g y 'follows naturally from the n a t u r e of t h i n g s ' because essentially phonological patterning is n o t merely a m a t t e r of c o n v e n t i o n .

N o n e t h e l e s s , even the m o s t ardent natural p h o n o l o g i s t w o u l d a d m i t that n o t e v e r y t h i n g in s y n c h r o n i c p h o n o l o g y is natural. T h e alternation in the shape of s o m e m o r p h e m e s either has a tenuous p h o n e t i c basis or is entirely arbitrary: there are n u m e r o u s U N N A T U R A L alternations w h i c h are m o t i v a t e d b y lexical or g r a m m a t i c a l considerations rather than p h o n e t i c factors. For instance, in English the plural of a n o u n is n o r m a l l y f o r m e d by suffixing an alveolar fricative w h i c h agrees in voicing w i t h the last s e g m e n t of the n o u n r o o t . B u t in a few irregular forms like ox, w h i c h becomes oxen instead of the expected *oxes in the plural, or sheep, w h i c h remains sheep in the plural instead of the expected *sheeps, the regular natural process of assimilation does n o t apply.

A case has been m a d e in the literature for restricting p h o n o l o g y to processes w i t h a genuine phonetic basis and reserving for the m o r p h o l o g i c a l c o m p o n e n t of the g r a m m a r

Explanations of naturalness 111

any rules regulating alternations that are determined by nonphonetic factors. This dichotomy between processes and rules is justified on the grounds that natural processes are easier to master in language acquisition than rules, which are not natural. Furthermore, when the language faculty disintegrates in aphasia, control of idiosyncratic rules, vanishes before control of natural processes. The theory predicts, for example, that an English-speaking aphasiac is more likely to continue correctly forming regular plurals like ships and boxes which obey the natural process of voice assimilation than irregular plurals like sheep and oxen which obey an ad hoc rule.

But there are complications. There is not always a simple dichotomy between natural processes and unnatural rules. With the passage of time, natural processes may have severe constraints imposed on their application without their natural basis becoming entirely lost. Re-examine the treatment of place of articulation in the last chapter. In section 5.3.4 we considered the process of place of articu- lation assimilation in English and in various African languages. You will recall from section 5.2.1 that in a language like Luganda all nasals always take the place of articulation of the following consonant, this being the universal natural process. In English too, this principle does apply, albeit in a very restricted way, as we saw in the examples in [5.16]. It automatically affects the prefix hi- but not the prefix un-. In Luganda the universal natural process of homorganic assimilation is preserved intact but in English it is inhibited, though not completely suppressed. It applies not across the board, but rather in restricted, morphologically defined environments. But this is still less arbitrary than the rule which says add nothing to form the plural of sheep.

Continue exploring the notion of 'degrees of natural- ness' by examining some more data, this time taken from Swahili.

[6.18] noun class 9/10 noun class 1 noun class 3 mbuzi goat mtu person mvi grey hair rjguruwe pig mke wife mferejc gutter jvfugu peanut mze old person mlarjgo doorway

112 Naturalness and strength

mbuni ostrich rribuni inventor mbuni coffee plant ndizi banana rppya new person mkono arm ngao shield mgarjga doctor mjale arrow ntji country ntja point Oge scorpion

State the circumstances in which nasals are:

(a) homorganic (b) syllabic

Be warned! You will need to take into account some of the information provided about M O R P H O L O G Y . Information about noun class membership is important as different morphemes may behave in different ways in identical phonetic contexts.

I hope you have been able to see that in Swahili, a word-medial nasal or a nasal prefix marking classes 9/10 must be homorganic with the following consonant but that a nasal prefix marking class 1 or 3 need not be homorganic. While assimilation of a nasal to the place of articulation of the following consonant is itself a natural process, its implementation in Swahili is sensitive to nonphonetic factors - it depends on the noun class which the nasal represents. The homorganic nasal assimilation rule is MORPHOLOGISED. It is not automatically triggered by phonetic information. It requires morphological information.

Likewise, the distribution of syllabic nasals, is morphologised. Any class 1 or class 3 nasal prefix is syllabic but a class 9/10 prefix is syllabic only if the root to which it is attached is monosyllabic. The distinction between monosyllabic and longer roots is clearly phonological but information concerning noun class membership is morphological.

Rule morphologisation represents a cline rather than a dichotomy. Some rules may be more morphologised than others; the extent to which the natural basis of a phono- logical alternation is subverted by nonphonetic factors varies greatly.

Besides morphologisation, another common cause of

Explanations of naturalness 113

the loss of naturalness is T E L E S C O P I N G . Telescoping occurs w h e n s o m e intermediate stages in a series of natural historical changes get eclipsed or completely lost, leaving behind a phonetically bizarre set of synchronic alternations. N o r m a l l y alternants of the same m o r p h e m e are related by phonological processes that are plausible. Occasionally, h o w e v e r , w e find phonetically triggered, a u t o m a t i c alter- nations which are arbitrary from a synchronic point of view.

In Luganda, for instance, we find / p / alternating s o m e - times w i t h |j] and s o m e t i m e s w i t h [w], which is e x t r a o r - dinary. It is very o d d for a bilabial voiceless stop to have palatal and labio-velar a p p r o x i m a n t s as its allophones. Allo- phones o f the same p h o n e m e o u g h t to be phonetically similar; these s o u n d s have very little in c o m m o n .

Study the data in [6.19] and state the distribution of the allophones o f / p / .

[6.i9]kuwa to give kujita to call kuweta to bend wa give! (imp.) tujita we call tuweta we bend mpa I give mpita I call mpeta I bend

T h e fact that [p j w] are in c o m p l e m e n t a r y distribution is strange. B u t w e can easily write a rule stating their distribution:

/ p / becomes (i) [j] w h e n followed by [i] and n o t preceded by a nasal;

(ii) [w] w h e n followed b y any other vowel provided it is n o t preceded by a nasal;

(iii) w h e r e it is preceded by a nasal, it is realised as [p] regardless of the vowel that follows it.

When Luganda is c o m p a r e d with another very closely related U g a n d a n language, R u n y a n k o l e , the missing stages in Luganda can be reconstructed thus:

[6.20] p > p h > h > y or w (depending o n the following vowel)

114 Naturalness and strength

Evidence for [6.20] is contained in s o u n d correspondences in forms like:

[6.21] Runyankole Luganda kuha to give k u w a kuhaisa to cause to give kuweesa kuheijera to pant kuwejjeela o m u h i i g i h u n t e r omujizzi

N a t u r a l p h o n o l o g y opens n e w perspectives on the p r o b l e m s of phonological analysis. H o w e v e r , there remain s o m e nagging d o u b t s . T h e explanation of phonological alternation w h i c h it offers is essentially F U N C T I O N A L , emphasising as it does natural assimilation processes, and implicitly ease of articulation. But a functional account in t e r m s of ease of articulation which worjks well for speech p r o d u c t i o n will n o t be as satisfactory w h e n the focus shifts to speech perception, since w h a t makes the articulation of sounds easier tends to m a k e their discrimination m o r e difficult. T h e t w o v i e w p o i n t s are in conflict. In concrete t e r m s , w h a t this means is that w e cannot always explain w h y in a given case assimilation rather than dissimilation is the preferred solution. T h e explanation proferred by a natural p h o n o l o g i s t m i g h t appear to a sceptical observer to be a s o m e w h a t arbitrary, post hoc rationalisation w h o s e explanatory value is doubtful.

C o n s i d e r the claim, for instance, that the ideal syllable type is C V , i.e. the kind of syllable that ends in a v o w e l (see section 9.5, page 175). It is said that the function of m a n y rules is to m a x i m i s e preferred C V syllables. It is s o m e t i m e s suggested that this explains w h y , in m a n y languages, a w o r d final consonant not followed by a v o w e l is deleted. For exa m p l e , in French, the final [t] is d r o p p e d in w o r d s like petit [pari] 'little' (masculine); in the feminine w h e r e it is followed by [a] as in petite [patita] 'little' the consonant is kept. Similarly, the desire to maintain the preferred C V syllabic type is said to explain the insertion of extra vowels in English l o a n w o r d s like hospital w h e n they are b o r r o w e d by languages that require consonants to be followed by v o w e l s . T h u s hospital is often rendered as hosipital[ in Swahili and m a n y African languages.

At present there is n o w a y of d e t e r m i n i n g which natural process takes precedence w h e r e alternative natural

Explanations of naturalness 115

processes are available. Even if it turned out to be true that naturalness determines rule application, we would be none the wiser when it comes to providing an explanatory account of a specific phonological event. It is not yet poss- ible to explain why, to create a CV sequence, in one instance a vowel is added and in another a consonant is deleted. Intuitively naturalness seems to be a valid concept. However, its predictive value and hence its place in a rigorous theory of phonology is still uncertain.

Exercises j

l-.- All me segments; except one; in each set below form a natural class. Circle the odd one out and state the phonetic property that makes it different from the rest.

(a) r i m © example [k] is an obstruent [r 1 m] are sonorants

(b) p t x g k s f (c) 1 r n ji d w j t (d) f J t p g b 0 (e) s z f v x J t (f) pf d ts t j bv d3 (g) a 6 i u y (h) p m t s b d (i) b cf g d (j) e i e 0

Tairora (Papua N e w Guinea) (SIL: b u ' r a u k a I w e n t bi'lo bi'Ba he w e n t bi'rera 'tei'rima I u n d e r s t a n d bu'araBa 'bulo I'm g o i n g bi'reBe 'iria listen! 'binaBu

1980:39) he goes I will go y o u w e n t will you go? w e w e n t

aluke'loma he killed it 'biri 'timilo He brought it down and gave it to me (Note: ' marks stress)

(a) Are [1] and [r] distinct phonemes? What is the evidence? (b) State the distribution of [b] and [B]. They are in

complementary distribution. (c) Explain the natural basis of the alternation of [b] with

[6].

116 Naturalness and strength

Account for the realisation of the final consonant of the noun root in the data below.

singular genitive plural wife wife's wives knife knife's knives hoof hoof's hooves thief thief's thieves

CHAPTER 7

Interaction between rules

7.1 Introduction to rule formalisation and ordering

Previous chapters have shown the need for more than one level of phonological representation. If we assume that there are underlying representations from which surface (phonetic) representations are derived, we shall often need more than one rule to map underlying representations on to surface representations. The situation can be compared to rush hour city traffic at a road junction, with rules queueing up to be applied. Police officers or traffic lights are needed to regu- late the flow of traffic. Likewise, in phonology a mechanism is needed to determine which one of several rules affecting a piece of data has precedence over other rules. Obviously, where rules do not interfere with each other, where they are like traffic on parallel one way streets, the question of regulation does not arise. I illustrate this with English examples in [7.1]:

[7.1] [phit] pit [pwhul] pool [thik] tick [twhuk] took [k^l] keel [kwhul] cool [atVnd] attend [3kwhustik] acoustic

The rules we needed to account for [7.1] are listed infor- mally in [7.2]

[7.2] (a) voiceless stops are aspirated at the beginning of a stressed syllable;

(b) consonants are labialised (rounded) before rounded vowels;

(c) velar consonants are fronted (palatalised) before high front vowels.

118 Interaction between rules

Rules [7.2b] and [7.2c] do not affect each other. There is no interaction problem since they do not have the same structural description, i.e. the phonetic properties of their inputs are different. Rule [7.2b] requires the presence of a rounded vowel, and all such vowels are back in English while rule [7.2c] stipulates that there must be a high front vowel. The two rules are like traffic on different highways. There is no possibility of their interfering with each other.

Determine whether rule [7.2a] interferes with either of the other two rules.

The answer must be 'no' since the voiceless stop at the beginning of a stressed syllable is aspirated regardless of the vowel that follows. Here again no traffic control system is needed. It would make no difference to the final output whatsoever which order the rules were applied in. Regu- lation of rule interaction is required only if one rule affects in some way the potential input to another rule.

Before we examine more closely the problem of rule interaction, I shall introduce you to the basic formal conventions used by generative phonologists because normally rules are written using distinctive features and formal notation. The motivation for rule formalisation is that it increases the explicitness of linguistic descriptions and makes it easier to expose woolly or incoherent analyses. I shall introduce the basic formal conventions of GENER- ATIVE P H O N O L O G Y , by restating formally in [7.3] the rules outlined above in [7.2]:

[7.3]

00

input

(b)

—cont —voice_ eg- A /

[ + c o n s l

e.g. A/

becomes

becomes

becomes

output

[ + s p r e a d ] / -

[kh]

[ + r o u n d ] / _ [tw]

in the environment

-(C) "—cons + s t r e s s

before a stressed vowel (as in acoustic)

Y—cons _ |_+round

before a r o u n d e d vowel (as in too)

Introduction to rule formalisation and ordering 119

(c) [ + cons"] - h i g h *- [ — back] / |_—back

+ cons + back + high e.g. M becomes [+] before a

front v o w e l (as in key)

A formal rule consists of the following: (a) the input, which states the s o u n d or sounds affected

by the rule; (b) the arrow, which means ' r e - w r i t e as' or 'is realised as'

or ' b e c o m e s ' (but n o historical d e v e l o p m e n t is implied);

(c) w h a t occurs to the right of the a r r o w is the output of the rule;

(d) following the o u t p u t , there is a diagonal line ' / ' to the right of that line is the environment, the line which forms part of the e n v i r o n m e n t s h o w s precisely w h e r e the changed segment is located;

(e) brackets r o u n d an element (like (C) in [7.3a]) indicate that a given element is optional - the rule applies regardless of the presence or absence of any optional element. In this instance it indicates that a voiceless plosive is still aspirated even w h e r e a consonant inter- venes, as in prayer.

W e shall t u r n t o French for our next example of rule interaction and rule formalisation.

Usually in French, an u n d e rl y i n g w o r d final consonant is deleted unless it is followed by a v o w e l . This produces alternations like [triko] (le) tricot 'knitted w e a r ' and [trikotc] tricoter ' t o knit'. A g o o d place to look for examples of this is the alternation b e t w e e n the masculine and feminine form of n o u n s and adjectives:

[7.4] F I N A L C O N S O N A N T D E L E T I O N Masculine /bos/ [ba] / / a t / [Ja]

/ / o d / [Jo]

bas chat chaud

Feminine / b a s s / [bas] /Jaw/ [Jat] /Joda/ [Jod]

basse chatte chaude

'low' 'cat' 'hot'

In the u n d e r l y i n g representation the feminine form ends in [a]. Phonetically, h o w e v e r , that final vowel is n o r m a l l y deleted in c o n t e m p o r a r y French — it is the s o -

120 Interaction between rules

called 'silent e. In fact, this v o w e l has n o t always been 'silent'. It was weakly p r o n o u n c e d as [a] in O l d French. E v e n today, w h e n French p o e t r y is recited, 'silent e' is scru- pulously p r o n o u n c e d as [a] in certain contexts. T h e rules regulating the p r o n u n c i a t i o n o r omission of [a] are c o m p l e x . T h e i r main function is t o d e t e r m i n e w h e t h e r [a] contributes to the syllable c o u n t . See G r a m m o n t (1961). T h e rule of final c o n s o n a n t deletion w h i c h applies to u n d e r l y i n g r e p r e - sentation can be formalised as [7.5]:

[7.5] [ + cons] *- 0 /

Remarks on the notation: (i) 0 means zero i.e. the s e g m e n t is r e w r i t t e n as zero - in

o t h e r w o r d s it is deleted, (ii) # = w o r d b o u n d a r y

(iii) the curly brackets { } indicate alternatives; here deletion of a c o n s o n a n t occurs e i t h e r before a c o n s o n a n t o r before a b o u n d a r y at the end of a w o r d . N e x t consider V O W E L N A S A L I S A T I O N , another

phonological process which occurs in French:

[7.6] [fe] [da] [robe] [fe] [sa] [5] [5] [ta] [mule]

fin dans robin faim sans on en tant moulin

'end' 'in' 'lawyer (pejorative)' 'hunger' 'without idef. pron. 'one' 'in' 'so much' 'mill'

T h e rule to account for [7.6] is given in [7.7]. C

[7.7] V » - [ + n a s a l ] / [+nasal]

Strictly speaking distinctive features should always be used in the formal statement of rules. But in practice, for convenience, instead of using distinctive features, w e often a d o p t the c o n v e n t i o n of using V to represent any v o w e l , C to represent any c o n s o n a n t , G to represent any glide and N to represent any nasal.

Rule [7.7] states that a vowel is nasalised if it is followed by a nasal c o n s o n a n t w h i c h in turn is followed by

C #

s

Introduction to rule formalisation and ordering 121

either another consonant or a word boundary in the under- lying representation.

If you turn now to [7.8], you will see that rule [7.7] does not apply to the vowels in [7.8b] because in the underlying representation, none of them is followed by cither a nasal consonant which is in turn followed by another consonant or appears in word final position. But nasalisation does apply to the vowels in [7.8a] because their underlying representations satisfy the requirements of the nasalisation rule. (We shall use the abbreviations UR and PR to stand for Underlying Representation and Phonetic Representation respectively):

[7.8a] UR PR [b5] bon (masc.) 'good' [amerike] americain (masc.)

'American' [da] dans 'in' [mare] marin 'marine' (masc.)1

[ta] tant 'so much' [a] an (masc.) 'year' [mule] moulin (masc.) 'mill' [f5] fond (masc.) 'bottom' [fodamatal] fondemental

'fundamental' /fin/ [re] fin (fern.) 'end'

/bon/ /ameriken/

/dans / /marin / /tant / / a n / /mulin/ /fond] /fondamantal/

[7.8b] UR / b o n o m i /

/ a m e r i k e n o /

/ d a d a /

/ m a r i n a / / t a /

V , / a n e / / m u l i n a : 3 /

/ f w a / /finalite/

PR [bonomi] b o n h o m i e (fern.)

' g o o d n a t u r e ' [ameriken] americaine (fern.)

' A m e r i c a n ' [dada] dada (masc.) ' h o b b y - h o r s e ' [marin] marin ' m a r i n e ' (fern.) [ta] ta ' y o u r ' (sing.) [a] a prep, ' t o ' [ane] annee (fern.) 'year' [mulina:3] m o u l i n a g e 'milling

( n o u n ) ' [fwa] fois (fern.) 'time, occasion' [finalite] finalite (fern.)

'finality'

122 Interaction between rules

Before moving on, list separately words in [7.8a] which have nasalised vowels: (a) due to being followed by a nasal in word final position: (b) due to being immediately followed by a nasal which is

in turn followed by another consonant.

The words bon and Jin are obvious examples of the effects of rule (a) and Jondamental of rule (b).

7.2 Linear rule ordering

The French data in [7.8] pose a rule ordering problem. A word like an [a] is subject both to the rule that deletes final consonants which are not immediately followed by a vowel and also to the nasalisation rule. To determine the order in which the two rules should be applied, we can try out the two derivations in [7.9] and [7.10] and see which one yields the correct output.

[7.9] Underlying representation / a n / First apply

Rule [7.5]: Final consonant deletion [+cons] *- 0 / JC

1*1 (This deletes the / n / of an as it is word-final.)

Then apply Rule [7.7]: Vowel nasalisation

[-cons] ^[+nasal] / [+nasal] | ^

The nasalisation rule fails to apply because the nasal which triggers off the nasalisation process is absent, having been deleted by [7.5]. Consequently the final output is *[a]. This is obviously incorrect.

Let us now try applying the two rules in reverse order, assuming again that the underlying representation is /an/.

[7.10] First apply Rule [7.7]: Vowel nasalisation [-cons] *- [+nasal] / [+nasal] C I

Linear rule ordering 123

This yields [an] (and a lowering rule affecting nasalised vowels subsequently turns it into /an/).

Then apply Rule [7.5]: Consonant deletion

f+consl »- 0 / ^ I l*l

By this rule / a n / loses its final / n / and becomes [6], which is the correct phonetic form.

This example shows that there are situations where the order in which rules apply makes a difference. That being the case, principles regulating the order in which rules apply are needed.

In SPE Chomsky and Halle proposed that rules should be LINEARLY O R D E R E D . To illustrate this, suppose in the phonological component of the grammar of a given language there are thirty rules. By this principle, the rules would be arranged in a list and it would be stipulated that each rule applies after all the rules that precede it and before all the rules that follow it on the list. It would not be poss- ible, for instance, to apply rule two after rule ten, where both rules are applicable. Once a particular order has been established, it is strictly observed in every derivation in a language. Furthermore, a rule can only apply once in a derivation. This means that a rule cannot re-apply in the same derivation either to its own output or to the output of another rule ordered after it which satisfies its structural description.

If the application of one rule has absolutely no effect on the potential input to another rule, the question of the sequence of application of those rules simply need not be raised. This was established at the beginning of the first section of this chapter. However, the question of linear ordering arises in a very interesting way where one rule can affect another, as in the French data above. We shall explore this again by examining some data from Luganda.

[7.11] Luganda glide formation Column A Column B UR PR UR PR / m u + a n a / [mwa:na] / m u + t i / [muti] 'tree' 'child*

124 Interaction between rules

/ m u + o j o / [mwo:jo] 'soul' / m u + k a z i / [mukazi] 'woman'

/li+anda/ [lja:nda] 'coal' / l i + n o / [lino] 'this' / m i + a k a / [mja:ka] 'years' / m i + t i / [mid] 'trees' (+ is a sign for a morpheme boundary.)

Contrast the realisation of the underlying high vowels / i / and / u / in columns A and B. Write in your own words the rule which captures the relevant generalisation.

In [7.12] I have written the rule out formally (ignoring for the moment the fact that the vowel that follows the glide is lengthened).

[7.12] V [+high] *- [-syllabic]/ _ V

[7.12] states that a high vowel becomes a nonsyllabic glide when it is followed by another vowel.

Luganda has another rule which is of interest in this context. It is an optional rule which deletes a root-initial / j / which is preceded by certain CV prefixes. The effect of that rule is shown in [7.13].

[7.13] UR PR /tu+jagala/ [twargala] or [tujagala]

'we like, want' /ku+jaka/ [kwa:ka] or [kujaka]

'to blaze' / m u + j o l a / [mwo:la] or [mujola]

'you (pi.) carve' /tu+jela/ [twe:la] or [tujela]

'we sweep' / m u + j i k o / [mujiko] or [mwi:ko] 'trowel'

Where a prefix is represented by a lone V, root-initial / j / is not deleted:

[7.14] UR PR /a+jagala/ [ajagala] 'he/she likes, wants' /e+jaka/ [cjaka] 'it blazes' /a+jola/ [ajola] 'he/ she carves' /a+jela/ [ajela] ' h e / she sweeps'

Linear rule ordering 125

The / j / deletion rule, can be written as [7.15]:

[7-i5] / j / - 0 / C V l + P r e f . , + _ V

These Luganda examples show that in order to derive the correct output in [7.13] for words like [twe:la] 'we sweep', it is necessary to apply the / j / deletion rule in [7.15] before the glide formation rule in [7.12].

Work out the reason for this before you read on.

If we assume that rules apply linearly, until [7.15] / u / .

has been applied to a form like /tu+jela/, the vowels / u / and / e / are separated by / j / and the input to glide formation is not available and so glide formation cannot be applied. For glide formation to apply, a high vowel must be adjacent to another vowel. The deletion o f / j / creates the input to glide formation. The implications for rule ordering are obvious: / j / deletion must precede glide formation if both rules are to apply. Technically, this kind of rule relationship where one rule opens the door to the application of another rule is called FEEDING ORDER.

The reverse is also possible. One rule may prevent or pre-empt the application of another. In that case the rules are said to be arranged in a BLEEDING ORDER. We shall explore this type of rule interaction using data from Swahili.

Swahili has a rule of homorganic nasal assimilation. Here it is illustrated applying to forms in noun class 9. This class contains many nouns referring to non-humans:

[7.16] UR I N+boga/ / N + b u / / N + d i z i / / N + dama/ / N + j u g u / / N + guruwe/ / N + goma/

PR [mboga] [mbu]2

[ndizi] [ndama] [jrfugu] [rjguruwe] [ngoma]

vegetable' mosquito' banana' calf peanut' P>g' drum'

The nasal prefix / N / undergoes homorganic nasal assi- milation. If a noun root begins with a voiced consonant, the nasal class prefix adjusts to its place of articulation so that it is labial, or alveolar or velar depending on whether

126 Interaction between rules

the first consonant of the n o u n r o o t is labial, alveolar or velar (see section 5.3.4).

T o w r i t e this rule formally w e need to use Greek letter variables like a (alpha), P (beta) and y (gamma) w h i c h range over b o t h plus and m i n u s values of a given feature. T h u s ,

N C [aant] [aant] means that the nasal and the consonant after

N C it are either b o t h [ + a n t ] or b o t h [—ant]; [P back] [P back] means that the nasal and the consonant after it are either b o t h [ + b a c k ] or b o t h [—back] etc. Building o n that, w e can account for [7.16] using this formal h o m o r g a n i c nasal assimilation rule:

[7-I7] N [+nas]

"a ant " P cor _Y back_

/ -

"a ant P cor .Y back_

Where, h o w e v e r , the n o u n r o o t begins w i t h a voiceless stop, the assimilation rule in [7.17] is blocked. In that case, the stop following the nasal is aspirated and the nasal itself is d r o p p e d , as y o u can see:

[ 7 . 1 8 ] / N + p a n g e / [phange] 'gadfly' / N + taa/ [thaa] ' l a m p ' / N + k u b w a / [k h ubwa] 'big (adj.)'

T h e rules at w o r k here are s h o w n in [7.19] and [7.20].

[7.19] Voiceless s t o p aspiration ( [ + s p r e a d ] means 'aspirated')

-*-[ + s p r e a d ] / N + -cont. -voice

(Voiceless stops are aspirated after the class 9 nasal prefix.)

[7. 20] Nasal deletion N *. 0 / + T—cont.

—voice + s p r e a d

(The nasal prefix is deleted before aspirated voiceless stops.) T h e nasal deletion rule, it turns out, applies n o t only

w h e r e a nasal is followed by a voiceless stop, but also w h e r e any o t h e r o b s t r u e n t (that is, a fricative or affricate or stop) follows a nasal:

Linear rule ordering 127

[7.21] / N + f i m b o / [fimbo] 'stick' (not *[mfimbo]) / N + s i k u / [siku] 'day' (not *[nsiku]) / N + tJumvi/[tJhumvi] 'salt' (not *[ntjumvi])

If we turn our attention to rule interaction, we observe that the nasal deletion rule BLEEDS the homorganic nasal assimilation rule. Having been deleted by [7.20], the nasal is no longer available to the assimilation rule in [7.17].

N o w work out how nasal deletion [7.20] interacts with aspiration [7.19].

Since it is the presence of a class 9 nasal prefix which triggers off aspiration, the aspiration rule in [7.19] must precede nasal deletion [7.20], if we assume that these rules apply in a linear order. Having conditioned the aspiration, the nasal can be subsequently deleted. If the order were reversed, and [7.20] applied first, deleting the nasal, the input to [719] would be destroyed and the correct surface form would not be derived. The upshot is that given two obligatory rules, one of which can POTENTIALLY BLEED the other, they must be ordered in a manner that ensures that bleeding is avoided.

N o w return to the French consonant deletion and vowel nasalisation examples above and see how these rules affect each other. You will recall that in order to get the right output, it is important that vowel nasalisation precedes consonant deletion; if the order is reversed, vowel nasalis- ation is blocked because the deletion of the consonant robs nasalisation of its input.

When two rules interact in such a way that one of them would be blocked if their order were reversed, those rules are said to be in a C O U N T E R - B L E E D I N G relationship. The French rules [7.5] and [7.7] are in a counter-bleeding relationship: unless vowel nasalisation precedes consonant deletion the derivation aborts because nasal deletion would destroy the input to vowel nasalisation. There are a number of other rule relationships which we need not go into in an introductory book of this kind.

128 Interaction between rules

7.3 Abandoning extrinsic ordering

T h e linear o r d e r i n g hypothesis is presented a b o v e w i t h o u t considering any possible alternatives. It is assumed that phonological rules m u s t be ordered so as to ensure that their interaction is correctly handled. For instance, w h e r e one rule w o u l d destroy the i n p u t to another (i.e. w h e r e rules are in a B L E E D I N G R E L A T I O N S H I P ) or w h e r e the prior application of one rule opens up the possibility of applying another (i.e. w h e r e rules are in a F E E D I N G R E L A T I O N - SHIP) it is claimed that the linguist needs to o r d e r the rules in the appropriate linear sequence. In fact, this linear o r d e r i n g position is controversial.

Before w e see w h y the a b o v e o r d e r i n g hypothesis is controversial w e need to clarify t w o i m p o r t a n t concepts. We shall do that b y distinguishing b e t w e e n t w o kinds of o r d e r i n g relationships: I N T R I N S I C and E X T R I N S I C linear o r d e r i n g . Rules are said to be e x t r i n s i c a l l y o r d e r e d if their interaction is g o v e r n e d by tailor-made o r d e r i n g statements designed for that specific set of rules in a particular language. B u t rules are i n t r i n s i c a l l y o r d e r e d if the order in which they apply follows automatically from the w a y in w h i c h they are formulated. For instance, as w e saw above, if t w o obligatory rules are in a feeding relationship, for b o t h of t h e m to apply the feeding rule m u s t necessarily apply before the rule that is fed. In such a case an ad hoc decision needs to be m a d e as to w h i c h rule applies first: w h e n rules are intrinsically ordered, the order in w h i c h they apply follows automatically from universal principles. We shall consider s o m e of these principles presently.

T h e approach presented in the last section allows b o t h extrinsic and intrinsic linear o r d e r i n g . S o m e linguists (like K o u t s o u d a s et al. 1974 , R i n g e n 1972, H o o p e r 1976, P u l l u m 1978) have argued s t r o n g l y against extrinsic linear o r d e r i n g , p r o p o s i n g instead that intrinsic ordering is the only kind of rule interaction that should be allowed. Essentially, the case against extrinsic linear o r d e r i n g is that it gives the linguist m o r e latitude than is w a r r a n t e d . H o w e v e r , if extrinsic o r d e r i n g is disallowed, the range of legitimate rule inter- actions is reduced. This has the merit of m a k i n g it difficult to set u p u n d e r l y i n g representations w h i c h differ greatly

Abandoning extrinsic ordering 129

from phonetic representations. This seems to be desirable. The greater the distance between surface and underlying representations, the greater the likelihood of having a very involved set of rules interacting in a complex or idiosyn- cratic way with each other in the mapping of underlying onto surface representations. A ban on linear rule ordering effectively means that only those underlying representations which require rules interacting in a straightforward way can be successfully mapped on phonetic representations. Deri- vations requiring complex rule interaction would abort.

Let us now see how the phonological rule system might operate without extrinsic linear ordering. One proposal that has been made involves S I M U L T A N E O U S RULE APPLICATION. It is assumed that the question whether or not a given rule applies can be answered by inspecting the underlying representation: rule application is solely dependent on whether the underlying represen- tation satisfies the structural description of a given rule.

We shall return to the Swahili example in [7.18] which is reproduced here as [7.22] to illustrate this point:

[7.22] UR PR / N + pange/ [phange] 'gadfly' / N + taa/ [thaa] 'lamp' / N + k u b w a / [khubwa] 'big (adj.)'

The underlying representations of these data satisfy the structural descriptions of both the nasal deletion rule in [7.20] and the obstruent aspiration rule in [7.19]. So, both these rules apply directly and simultaneously to the under- lying representations:

[7.23] UR / N + t a a / / N + thaa/ (by [7.19]) and / thaa/ (by [7.20])

PR [thaa]

Extrinsic linear ordering is unnecessary, in cases of this kind.

Another situation in which simultaneous rule appli- cation could also be used is where rules are MUTUALLY NON-AFFECTING. In such a situation any ordering would be arbitrary. You can verify this by looking back at the English examples and rules in [7.1] to [7.3]. The rules

130 Interaction between rules

aspirating voiceless stops, labialising consonants before r o u n d e d vowels and fronting velars before front vowels d o n o t affect each o t h e r ' s i n p u t . A n y o r d e r i n g of these rules w o u l d m a k e n o difference to the final o u t p u t . T h e y can directly apply at the same time to the u n d e r l y i n g represen- tation w h e n e v e r their structural description is satisfied.

O u r next e x a m p l e of rule interaction will also be taken from English. T h e relevant rules are [7.24] and [7.25]

[7.24] Shortening: C

[ + s e g m e n t ] • • [ — l o n g ] / [ — voice] (Vowels and consonant segments are shortened w h e n followed b y voiceless segments.)

T h e existence of this rule means that the v o w e l in sat is s o m e w h a t shorter than that in sad; it also means that the [1] in colt is shorter than that in cold.

[7.25] Obstruent devoicing [ —sonorant] *- [ — v o i c e ] / #

( N o n s o n o r a n t s (obstruents) i.e. stops, affri- cates and fricatives are (partially) devoiced in w o r d final position.)

T h e effect of [7.25] is to m a k e the last s e g m e n t of each of the following w o r d s less fully voiced than the first one: bib, did and gag.

P r o p o n e n t s of extrinsic linear o r d e r i n g w o u l d suggest that the fact that the v o w e l s e g m e n t preceding the devoiced / g / s e g m e n t in a w o r d like gag does n o t b e c o m e shorter even w h e n , after the application of [7.25], the w o r d ends in a voiceless or partially voiceless velar, is evidence that [7.24] applies before [7.25] so that b y the t i m e [7.25] applies, the o p p o r t u n i t y of s h o r t e n i n g has already been missed. In other w o r d s , rule [7.25] on the face of it, could P O T E N T I A L L Y F E E D [7.24] if the o r d e r i n g of these t w o rules was reversed. B u t the reality is that it does n o t . T h u s , o r d e r i n g [7.24] before [7.25] ensures that a potentially feeding rule inter- action is avoided.

In fact, w h e r e rules are in a potentially feeding (also called c o u n t e r - f e e d i n g relationship), an approach w h i c h forbids extrinsic linear o r d e r i n g w o u l d also yield the right

Abandoning extrinsic ordering 131

result if it stipulated that rules apply directly to underlying representations which satisfy their structural descriptions. Thus, it is possible to show that [7.24] applies since its structural description is met by the underlying represen- tation while [7.25] fails to feed [7.24] because the voiceless obstruents produced by this rule miss the application of [7.24] since both rules are assumed to apply directly to underlying representations. N o further ordering statements are needed.

Another proposal that has been made by critics of extrinsic ordering is R A N D O M SEQUENTIAL RULE APPLICATION. This principle states that rules apply one at a time, rather than simultaneously. But they are not strictly regimented in a fixed order. They apply, whenever a string that satisfies their structural description arises in a derivation. In cases of potential ambiguity, rules are arranged in that order which ensures that all obligatory rules are applied.

This principle is needed because the claim made above that rules only apply directly to underlying representations is not always correct. Earlier in this chapter we established that there are rules which cannot apply directly to underlying representations and which only become appli- cable after their input has been created by another rule during the course of a derivation. That is the case when rules are in a feeding relationship (see [7.11] - [7.15] above). The feeding rule must precede the fed rule. But this ordering need not be done on an ad hoc basis. The random sequential application principle, by ensuring that rules apply at any stage in a derivation when their structural description is met provides a simple mechanism for arriving at the correct order: the structural description of the feeding rule must, by definition, be met before the conditions for the application of the fed rule are created. So, feeding rules must always precede the rules that are fed.

A further universal principle which renders extrinsic ordering unnecessary is the ELSEWHERE C O N D I T I O N (Kiparsky 1973). The elsewhere condition states that where the input to two rules partially overlaps, the more specific rule applies before the more general rule. Discussion of this principle will wait until section 12.2.3.

132 Interaction between rules

7.4 Conclusion: why ordering matters

It would be wrong to think that in this chapter too much fuss has been made over an abstruse theoretical point. This is not the case. Rule interaction is not a fringe, esoteric issue. The importance of rule ordering lies in the fact that it offers us a way of constraining the power of the model. A phonological theory, which incorporates powerful rules which can insert, move and delete elements enables the linguist to perform a very wide range of operations, some of which may not be possible in human language.

It is therefore necessary to find ways of reducing the power of the model so that only those operations that are possible in human language are catered for. Restrictions on rule interaction have the effect of reducing the range of possible outputs of phonological rules. Furthermore, restric- tions on rule interaction also indirectly curtail the distance between underlying and surface representations. Very complex extrinsic rule ordering is required where a large number of rules are needed to translate abstract underlying representations into phonetic representations.

The theme of constraining the remoteness of under- lying representations from phonetic representations is devel- oped further, from a different angle, in the next chapter.

Exercises

1. In fast, casual speech the words in the left-hand column may be realised as indicated in the right-hand column. Column A Column B handball [haembol] handbag [haembaeg] hand-made [haemmeid]

(a) Formulate the rules needed to state the processes involved.

(b) Should these rules be extrinsically ordered?

2. Study the data below and answer the questions which follow. Assume that the morphemes of the two dialects have identical underlying representations. The differ- ences in the phonetic shape of morphemes are due to

Conclusion: why ordering matters 133

differences in their phonological rules. Lumasaaba (Uganda) (based on Brown Dialect A im-piso im-pale im-fula in-temu iji-cese ig-kafu in-sami im-beba in-dali in-zu ig-gwe im-bululuka in-dima iji-cina iji-Jo:la ig-kuba

needle trousers rain snake sheep cow fly rat beer house leopard I fly I run I dance I grow I hit

(1972)) Dialect B i:-piso i:-pale irfula i:-temu i:-cese i:-kafu i:-sami im-beba in-dali in-zu ig-gwe im-bululuka in-dima i:-cina iji-Jo:la i:-kuba

(a) Determine the underlying representation of each prefix. (b) State formally the rules that account for the alternation

in the shape of the prefixes in the two dialects. (c) Describe the differences in rule interaction which you

have observed in the two dialects.

Notes

1. Note, incidentally, that vowel nasalisation tends to induce lowering. Oral [i] corresponds to nasalised [e], oral [o] to nasalised [o], and so on. We do not have the space to explore this phenomenon in depth here.

2. This nasal prefix is both homorganic and syllabic before monosyllabic roots like /N+bu/[mbu] 'mosquito'.

CHAPTER 8

The abstractness of underlying representations

8.1 Abstractness

So far it has been assumed without much discussion that normally the alternation in the realisation of a morpheme is most appropriately stated by positing a single underlying form from which its various alternants are derived by rule.

Thus, for instance, the forms sign [sain], signal [signal], signature [signatja] and signify [signifai] share the underlying representation /sign/. It can be argued that in sign the word final / - g n / consonant sequence makes underlying / i / change to [ai], and then / g / disappears. But in the rest of the examples, where various affixes are present, the under- lying representation /sign/ remains unchanged because in those cases the string / - g n / is not word final.

Note that the underlying representation /sign/ is ABSTRACT, i.e. it is distant from the phonetic form [sain] which is actually uttered by speakers. Support for the abstract underlying form is drawn not only from the phono- logical and semantic relatedness of sign, signal, signature and signify, but also from the fact that several other sets of forms such as malign [malain] ~ malignant [malignant] and benign [binain] ~ benignant [binignant] show the same alternation. This suggests that the alternation is rule governed. It is not a unique characteristic of the form /sign/. To capture the relevant generalisation rules changing word final underlying / - i g n / into surface [-am] are required.

The position which I have just outlined appears reason- able, but it is not entirely uncontroversial. The controversy centres round the extent to which phonetic realisations can differ from underlying representations. Is there a principled way of determining when underlying representations are

Concrete phonology? 135

too far removed from surface manifestations of a morpheme to be credible? Is there a principled way of determining when semantic relatedness is too dilute and phonological similarity too tenuous to justify deriving different forms from the same base form? Is there a principled way of determining when relatedness between different forms is only of diachronic interest and has ceased to be synchron- ically valid?

These questions are too important to be left to the linguist's common sense. Phonological theory would become empty if any sound was allowed to change arbi- trarily into any other sound at the whim of the analyst. Principles are needed which would outlaw a maverick phonological gambit in which, for instance hundred and centenary, or foot and pedal, were derived from the same abstract synchronic base. Notwithstanding their semantic similarity and faint phonological resemblance, these forms are not closely enough related synchronically to be derived from the same base form. This begs the question of the degree of relatedness that is sufficient to allow the derivation of phonetic representations from the same base form. Can effective mechanical procedures that determine which forms are derivable from the same synchronic source ever be devised? Unfortunately, the answer is 'no'.

In the last chapter we saw one perspective from which the problem can be dealt with: restrictions can be placed on rule systems, making it impossible to derive phonetic representations from distant underlying representations where extrinsic linear rule ordering is required. But even if it were desirable, banning linear ordering in itself would not be enough. A two-pronged attack is what is needed: one prong being the incorporation in the model of as restrictive a theory of rule interaction as the facts of language allow and the other prong being the development of restrictive principles for the setting up of synchronic underlying representations.

8.2 Concrete phonology?

A knee-jerk reaction might be to simply resort to draconian measures like banning all abstract underlying represen-

136 The abstractness of underlying representations

tations (i.e. underlying representations that are remote from surface (phonetic) representations). The dichotomy between underlying and surface representations would be abandoned and we would have ' C O N C R E T E P H O N O L O G Y ' with only one level of representation, namely the phonetic. But, for a number of reasons, this would not be a satisfactory solution.

Abandoning underlying representations would be unsatisfactory because, as we saw in section 3.4 there is redundancy in language. Many phonetic properties of a segment are predictable because certain feature combi- nations either presuppose or preclude each other (e.g. in English any nasal consonant is also noncontinuant, nonstri- dent and voiced). Other properties of a sound may be predictable, given the context in which it occurs (e.g. in English if we know that a syllable begins with a consonant cluster whose second segment is a stop, we can always correctly predict that the first consonant of that cluster is [s]). Having two levels of representation allows us the possi- bility of extracting all the predictable phonetic features of a morpheme so that we only include the idiosyncratic prop- erties as part of the underlying representation. All predict- able features are filled in later by redundancy rules. There would be no adequate way of dealing with this redundancy in a model of phonology which did not have more than one level of representation.

The second objection is that 'concrete phonology', in a literal sense is a chimera (and nobody has ever proposed it as a viable alternative) since even so-called 'concrete' phonetic representations do not exist. The phonetician, like all other scientists never deals with 'brute facts'. Raw data is always seen through the lenses provided by the analyst's theory. The measurements which laboratory instruments are calibrated to make depend on what sort of data the analyst considers relevant to testing certain theoretical claims. As for the phonetic transcriptions on which phono- logical descriptions are based, they can never be totally faithful to the brute facts. Phonetic transcription is a subtle art which involves a degree of interpretation. In view of this, a one-level phonological representation identical to a phonetic representation could never be concrete, except in a metaphorical sense.

Concrete phonology? 137

The third objection to an approach which only recog- nises phonetic representations is that it fails to capture significant generalisations concerning relationships between allomorphs of a morpheme. You will recall that in Chapter 5 we saw that usually the relationship between allomorphs of a morpheme is not arbitrary. The realisation of allo- morphs tends to be PHONOLOGICALLY C O N - D I T I O N E D . For example, listing all the allomorphs of the past tense would fail to show the difference between the realisation of the regular endings [-t -d -id] as in walked, loved and wanted, which are governed by voice assimilation, and the totally arbitrary past tense formation of went from go. If, on the other hand, underlying representations are set up we can show the non-arbitrary relationship between the various allomorphs by deriving them from a single base form as we did in [5.2].

However, the search for underlying representations cannot be allowed to proceed without any constraints. It is always essential to ensure that putative underlying repre- sentations are synchronically valid. This is by no means an easy matter because for better or for worse, languages have to live with undigested bits of their history. Often there is no clear division between synchrony and diachrony. Phon- ological processes which might have been regular, natural and motivated at an earlier period in the life of a language may only survive as sporadic alternations in a few isolated forms. That is true, for example, of the vowel alternation in English nouns like:

[8.1] Singular Plural goose [gus] geese [gis] foot [fut] feet [fit] tooth [tu0] teeth [tie]

During the Middle Ages, many Germanic languages underwent changes known as U M L A U T (or i-mutation). This is partial assimilation involving the fronting of a stressed back vowel in anticipation of a high front vowel or glide [i or j] in the next syllable.

In Old English (which was spoken circa 449-1100 and from which modern English is descended) umlaut probably began during the sixth century. Originally the nominative singular and the nominative/accusative plural of the noun

138 The abstractness of underlying representations

foot w e r e fit [fo:t] and fitiz [fo:tiz] respectively, b u t from about the sixth century, the plural form changed to fitiz [fe:tiz]. Later the ending was d r o p p e d and [fe:tiz] became

fit [fe:t]. (Later still, a r o u n d the fifteenth c e n t u r y l o n g / e : / changed to long / i : / ; b u t that change was never reflected in the o r t h o g r a p h y ) . After the loss of / i / in the plural suffix / - i z / , the alternations survived as w a y s of m a r k i n g plural in feet and a few other w o r d s . B u t there was n o longer a phonetic reason for the u m l a u t alternation and it became fossilised.

If a n e w p r o d u c t was launched on the m a r k e t , and it was called a voot [vut], there is n o likelihood w h a t s o e v e r that t w o of those objects w o u l d be called veet [vit]: this internal v o w e l change in the r o o t is n o longer a p r o d u c t i v e process, i.e. it cannot be extended to n e w forms. Yet, at the same time, as [8.1] s h o w s , the alternation b e t w e e n [u] and [i] still exists in English and it cannot be disregarded. A r g u a b l y , to s h o w the relationship b e t w e e n the different shapes of the w o r d s in [8.1], w e could set u p a single u n d e r l y i n g representation for each pair from w h i c h the surface representations are derived. H o w e v e r , in view of the lack of p r o d u c t i v i t y of the alternation b e t w e e n [i] and [u], there is a s t r o n g case for n o t setting u p a single synchronic u n d e r l y i n g form for each one of these pairs of w o r d s and instead s h o w i n g the relationship b e t w e e n t h e m directly using a rule which is part of the lexicon.

This is o n e w a y in w h i c h the search for u n d e r l y i n g representations could be restricted: a single u n d e r l y i n g representation is only w a r r a n t e d w h e r e a given p h o n o - logical alternation can be s h o w n to be alive and well in a language. This can be d o n e b y s h o w i n g that it is p r o d u c t i v e l y extended to n e w forms entering the language. T h e fronting of [u] (or [u]) to [i] as in [ 8.1] is u n p r o d u c t i v e while the addition of -5 to voot to form voots [vuts] is p r o d u c t i v e . A d d i t i o n of -5 rather than fronting of a back v o w e l (umlaut) is the rule w h i c h m u s t be included in a g r a m m a r w h i c h aims to m o d e l speakers' k n o w l e d g e of their language.

U n f o r t u n a t e l y , o u r troubles do n o t necessarily end if w e take that stance. T h e difficulty of d e t e r m i n i n g w h a t the native speaker k n o w s still remains. If the g r a m m a r is a

Concrete phonology? 139

model of linguistic knowledge, we must be wary of assuming rashly that on every occasion when an ingenious linguist can spot a regularity in the alternation in the phonetic shape of a morpheme, for which a base form can be posited and a rule written, the native speaker too will necessarily have such a rule. This was the subject of a heated debate in the early seventies. The debate centred on data like these from English showing alternative shapes of a root, depending on whether a suffix is present:

[8.2] tense vowel [ei] lax vowel [ae] vain vanity sane sanity opaque opacity profane profanity inane inanity grave gravity

Write down in words the rule which determines the quality of the italicised vowel.

Note that no vowel change occurs in words like fame [feim] ~ famous [feimas], base [beis] ~ basic [beisik], fate, [feit] ~ fatal [feitl], safe [seifj ~ safety [seifti], etc.

From your solution can you see why this rule is called TRISYLLABIC LAXING? It is a rule which changes a tense vowel into a lax vowel when a short word is lengthened by adding a suffix so that it ends up having at least three syllables. Trisyllabic laxing applies in [8.2] but fails to apply to famous, basic, etc. because these contain only two syllables after the addition of a suffix.

This rule is not only responsible for the [ei] ~ [ae] alter- nation, but also for the [i] ~ [e] alternation in words like serene ~ serenity; the [au] ~ [A] alternation in words like pronounce ~ pronunciation, the [ai] ~ [1] alternation in words like divine ~ divinity, etc.

Find two other words exemplifying each of these patterns.

140 The abstractness of underlying representations

These alternations are a synchronic relic of the momen- tous upheaval in the English vowel system that took place during the late Middle English period, around the fifteenth century, to which the name the English GREAT VOWEL SHIFT is given. Trisyllabic laxing is part of the undigested history of the language.

Not having some kind of rule to show the regularity of these M O R P H O P H O N E M I C ALTERNATIONS (i.e. alternative phonological realisations of a morpheme) which we have observed would result in a failure to capture obvious generalisations. It would mean failure to show that the pairs of words in [8.2] are related. But, on the other hand, it would be somewhat misleading to claim that the rule that relates, say, profane to profanity is every bit as synchronically relevant and phonologically well-motivated as the allophonic rule aspirating a voiceless stop occurring initially in a stressed syllable that relates the aspirated [kh] of cake [kheik] to the unaspirated first [k] of [thikeik] tea-cake. The issue is that not all regularities that can be identified by the linguist have the same status for speakers of a language.

In view of this, phonologists belonging to the Natural Generative Phonology school (e.g. Hooper 1976; Venne- mann 1974a) have argued that the phonological component need only deal with transparent, phonetically motivated, regular, productive processes like the aspiration of voiceless stops occurring initially in stressed syllables in English. All other regularities should be handled by the morphological component. Where psychologically valid synchronic relationship is doubtful, though some semantic and phono- logical link exists, LEXICAL VIA RULES located in the lexicon should be used to indicate the link.1

Hooper (1976:9), examines the Spanish forms in [8.3] which Harris (1969), who favoured a more abstract approach, had claimed to be synchronically related:

[8.3] A B leche [letje] 'milk' lactar [laktar] 'to lactate' ocho [otfo] 'eight' octavo [okta|Jo] 'eighth' noche [notfe] 'night' nocturno [nokturno] 'nocturnal'

Harris used phonological rules, closely reflecting the historical development of Spanish from Latin, to derive the words in each pair from a single underlying representation

Concrete phonology? 141

in c o n t e m p o r a r y Spanish. His u n d e r l y i n g representations were respectively / l a k t e / , / o k t o / and / n o k t e / . His p r o p o s e d rules are illustrated w i t h a derivation of noche:

[8.4] UR / n o k t e /

n o y t e (a) k —* y /

n o y t j e (b) t -^ t f / y _ n o t j e (c) y —» 0 / t j

P # [notjc]

T h o u g h historically valid, the u n d e r l y i n g representation / n o k t e / and the t w o intermediate forms [noyte] and [notje] do n o t occur in m o d e r n Spanish. Harris is rebuked by H o o p e r for being too abstract.

In N a t u r a l Generative P h o n o l o g y w h e r e the p r e m i u m is on avoiding abstract solutions, the relationship between the pairs of w o r d s in [8.3] is s h o w n by H o o p e r using a via rule in the lexicon. This is a statement to the effect that in certain forms x corresponds to y:

[8.5] k t - » t

A via rule is not p r o d u c t i v e . T h o s e few items linked b y a via rule have to be individually m a r k e d . F u r t h e r m o r e , a particular via rule is n o t assumed to be part of every native speaker's linguistic competence.

You can see w h y this should be so by considering relationships b e t w e e n s o m e English w o r d s , e.g. stink and stench, drink and drench, break and breach. T h e w o r d s in each pair are semantically and phonologically related. B u t p r o b - ably m o s t speakers of English are u n a w a r e of this fact. T h o s e w h o arc a w a r e of the connection m a y have a via rule linking each pair.

T h e t r e a t m e n t o f sounds in language acquisition, language change and dialectal variation w o u l d p r o v i d e w i n d o w s t h r o u g h which speakers' phonological k n o w l e d g e could be glimpsed. For instance, the fact that the p r o d u c t i v e rule of plural formation in m o d e r n English is the addition of / - z , - I Z , - s / can be seen in the w a y children s o m e t i m e s ovcrgeneralise / - z / and form *mans (instead of men) as the plural of m a n . T h e y d o n o t analogise in the other direction m a k i n g the plural of e.g. pan and van *pen and *ven (instead of the correct pans and vans). A fundamental c o n t e n t i o n of

142 The abstractness of underlying representations

these scholars is that linguists should posit as underlying representations only those forms that can be plausibly learned by induction on the basis of the phonetic evidence which a child is exposed to.

The sceptic might retort that the effect of these proposals is not to solve the problem but rather to shunt it into a morphological or lexical siding. Proponents of Natural Generative Grammar are aware of this possible crit- icism. They provide a battery of principles to be used in deciding whether or not a transparent phonological relation- ship exists between forms which the linguist wishes to relate using synchronic phonological rules so that problem cases are not dispatched too hastily to the lexical via rules.

We could simply insist that the underlying represen- tation must be IDENTICAL to one of the phonetic repre- sentations of a given morpheme, i.e. one of the allomorphs should be selected as the underlying representation. Let us return to the data in [8.2] to illustrate this. If we choose /vein/ as the underlying representation of vanity [vaeniti], there is no problem since all the segments of the root morpheme do surface in a single shape of the word vain [vein].

Unfortunately the principle stated above fails in some- what more complex situations like that described below in Russian where the underlying form would need to be cobbled together from more than one surface form. Consider the data in [8.6] and [8.7] (based on Kenstowicz and Kisseberth:i979).

[8.6] vowel neutralisation Pes 'forest' Pisa 'forests' gorat 'town' garada 'towns'

[8.7] Final obstruent devoicing gorat 'town' garada 'towns' sapok 'boot' sapaga 'boots' ras 'time' razi 'times' ruka 'hand' rukJi 'of the hand'

There is a rule which has the effect of neutralising vowels in [8.6]: in stressed positions Russian has the vowels / i a e o u/ but in unstressed position the contrast between underlying / i / and / e / is suspended and both are realised as / i / . Likewise, when unstressed, / o / and / a / do not

Concrete phonology? 143

contrast: they are both realised as [a]. O n l y / u / appears in both stressed and unstressed syllables. (As a result of vowel neutralisation only the maximally perceptually distinct peripheral vowels / i a u / occur in unstressed position.) T h e vowel neutralisation rule is presented in [8.8]:

[8.8] V o w e l neutralisation

(a)

(b)

-high —back

e

- h i g h — back + r o u n d

[+high]

+ l o w — r o u n d

/ [—stress] (i.e. / e / occurs unstressed)

/ [ - s t r e s s ] (i.e. / o / occurs unstressed)

T o account for the devoicing of o b s t r u e n t s (d —» t, z etc.) seen in [8.7] the following rule is required:

[8.9] Final o b s t r u e n t devoicing [—son] —» [—voice] / #

s,

Explain w h y , given the existence o f rules [8.8] and [8.9], establishing the u n d e rl y i n g form of w o r d s like g o r a t / g a r a d i ' t o w n s ' is p r o b l e m a t i c if you assume that one of the a l l o m o r p h s of a m o r p h e m e is the base form from which other alternants are derived.

T h e p r o b l e m is that n o single phonetic representation contains all the s e g m e n t s which m u s t form part of the base form. Given [8.8] the unsuffixed forms in the left-hand column p r o v i d e the correct representation of the underlying vowel which reveals itself u n d e r stress; but given [8.9], the forms in the r i g h t - h a n d c o l u m n contain the underlying voiced o b s t r u e n t which is devoiced in w o r d final position.

This p r o b l e m is solved by a weaker proposal which merely requires that S E G M E N T S P O S I T E D AS U N D E R - L Y I N G A P P E A R I N A T L E A S T O N E A L L O M O R P H of a m o r p h e m e . B u t all the properties of the underlying representation need n o t occur t o g e t h e r in o n e of the allo- m o r p h s . T h e p h o n o l o g i s t is allowed to posit as underlying a c o m p o s i t e form containing segments w h i c h surface in

144 The abstractness of underlying representations

distinct a l l o m o r p h s of the m o r p h e m e . T h e under lying representations of the Russian forms are Ijes, gbrat, sapbg and ruk, w h i c h are n o t necessarily attested as such in the super- ficial phonetic representations (Kenstowicz and Kisseberth

1979)- T h e s e Russian data also illustrate another i m p o r t a n t

principle which has to be observed in setting u p underlying representations. It is the principle of P R E D I C T A B I L I T Y . If t w o analyses appear to describe the data adequately, that analysis w h i c h has greater generality, i.e. greater predictive p o w e r , is preferred.

In o r d e r to account for the data in [8.6] and [8.7], we could have regarded the roots as either e n d i n g in voiced or voiceless o b s t r u e n t s in the u n d e r l y i n g representation. W e could have posited / l e s / , 0 g 6 r a t / , / s a p o k / , / r a z / and / r u k / as the underlying representations. A rule w o u l d have been needed to voice o b s t r u e n t s w h e n they occur b e t w e e n vowels.

Write u p that hypothetical o b s t r u e n t voicing rule formally.

T h e o b s t r u e n t voicing rule could be stated as:

[8-9] -son -voice

['+voice] / V V

Rule [8.9] w o u l d correctly predict [garada], [sapaga] and [razi] but it w o u l d w r o n g l y predict the voicing of the o b s t r u e n t in *[lJiza] and *[rug J i]. By ad hoc measures, like using a diacritic to distinguish those o b t r u e n t s which u n d e r g o voicing rule [8.9] from those which defy it, embarrassing exceptions m i g h t be s w e p t u n d e r the carpet. But there is n o t h i n g to c o m m e n d such a solution.

T h e r e is a better alternative. If w e posit underlying voiced o b s t r u e n t s as the last r o o t consonants in / g o r a d / , / s a p o g / , and / r a z / , w e can apply the devoicing rule [8.9] to all forms which satisfy its structural description. Base forms like / P e s / and / r u k / which end in voiceless consonants are unaffected by the rule. T h e r e w o u l d be no need to resort to ad hoc measures to explain away exceptions.

Absolute neutralisation 145

8.3 Absolute neutralisation

In the SPE version of generative p h o n o l o g y a high degree of abstractness was tolerated if it enabled the linguist to capture generalisations - so long as it was not too 'costly'. Very abstract solutions and c o m p l e x derivations t h o u g h not encouraged, w e r e n o t prohibited. It was considered legit- imate to posit u n d e r l y i n g abstract ' g h o s t s e g m e n t s ' which never occur in the p h o n e t i c realisation of any m o r p h e m e of a language b u t w h i c h nonetheless can be inferred on the basis of regular phonological alternation.

I shall use s o m e facts about French which have received considerable attention in the literature to explore the p r o b l e m of abstractness and particularly the issue of ghost segments:

[8.10] (a) le pic [la pik] 'the pickaxe' l'abbe [labe] 'the a b b o t '

(b) le hasard [ b aza:r] 'the accident' la hirondelle [la ir5del] 'the s w a l l o w '

As [8.10] s h o w s , French has a rule of v o w e l deletion

[8.11] V -* 0 / + V

which, in this case, deletes the v o w e l of the definite article w h e n the article is followed by a n o u n b e ginning with a v o w e l . This rule applies to yield [labe] from / b abe/, b u t does not apply to / b p i k / w h e r e the n o u n begins w i t h a consonant.

In [8.10b], the n o u n s are spelled with consonant h as their first letter, b u t that h is not p r o n o u n c e d . Phonetically, the first sound of b o t h [le aza:r] and [la ir5del] is a v o w e l . Therefore one w o u l d expect the vowel deletion rule to apply to the article. In fact, as you can sec, vowel deletion fails to apply in these and m a n y other w o r d s beginning w i t h the h in the o r t h o g r a p h y a l t h o u g h that h is ' m u t e ' e.g. le hale [ b al] ' t o w - l i n e ' , la harengere [la arfi3e:r] 'fish-wife' le hall [ b D1] '(hotel) l o u n g e ' , le harem [ b a r e m ] ' h a r e m ' , le hideur [ b idce:r] 'hideousness', le hongre [ b 5:gr] 'gelding', la Iwue [la u] ' h o e ' and even le hold-up [ b oldoep] ' h o l d - u p ' and le home [ b o:m] ' h o m e (as in "children's h o m e " ) ' .

It has been p r o p o s e d in the literature that a l t h o u g h such

146 The abstractness of underlying representations

stems phonetically begin with a vowel, they should be deemed to begin with an abstract / h / segment in the under- lying phonological representation which is peculiar in that it always fails to surface phonetically (Schane 1968).

But others have suggested that it is unnecessary to insit that the sound that blocks the vowel deletion rule is indeed / h / . It is sufficient to assume that there is a very UNDERSPECIFIED consonant which is only represented by the residual feature [ +consonant]. That is enough to block the vowel deletion rule since that rule is only appli- cable where the noun begins with a vowel. Furthermore, only being specified for the feature [+consonant] means that the abstract segment cannot surface phonetically (i.e. be uttered). Much fuller phonetic specification is required before pronunciation is possible (Clements and Keyser 1983). This line of argument is persuasive. We shall return to it in [9.6] below.

While with regard to French positing an abstract segment is plausible, there are situations where the argu- ments for abstract non-surfacing segments are less per- suasive. That is the case with respect to Luganda vowels.

In this language the five vowels /i e a o u/ are the only ones which appear on the surface; vowel length can be distinctive:

[8.12] Short vi [sima] [tema] [mala] [wola] [tuma]

owel 'dig a hole!' 'cut!' 'finish!' 'lend' 'send'

Long vowel [si:ma] [te:Ba] [ma:la] [wo:la] [tu:ma]

'be grateful!' 'guess!' 'smear!' 'scoop!' 'heap!'

Although phonetically there are two high vowels, namely [i] and [u] (which can be long or short), they condition different phonological processes. Sometimes a stop followed by [i] or [u] undergoes SPIRANTISATION (i.e. becomes a fricative) and sometimes it does not. Since sounds normally display consistent phonological behaviour, we would expect the same sound to behave in the same way whe"n it occurs in identical phonetic contexts. Consider the data in [8.13] where no spirantisation occurs before high vowels:

Absolute neutralisation 147

[8.13] No spirantisation before [i] and [u] [pima] [mitima] [kiki]2

[kiguli] [kuginga]

'measure!' [kupuluka] 'to escape' 'hearts' [kutu] 'ear' 'what?' [mukutu] 'canal' 'cage' [mudumu] 'pipe' 'to notch' [mugugu] 'burden'

In contrast with the [i] and [u] vowels in [8.13], those in [8.14] and [8.15] trigger off the spirantisation of a preceding nonlabial consonant. (Note that labials are exempt from spirantisation.)

After examining the data below, suggest a rule to account for spirantisation.

[8.14] Spirantisation before [i] [le:ta] [6u:ka] [lo:nda] [jiga] [kola]

'bring!' j u m p ! ^ 'choose' 'learn!' 'work!'

[8.15] Spirantisation before [kukwa:ta] 'to turn (of milk)'

[kugo:nda] 'to be soft' [laBuka] 'be alert!'

[jiga] iearn!'

[mule:si] [muBu:si] [mulo:nzi] [mujizi] [mukozi]

'bringer' jumper' 'elector' 'student' 'worker'

M [makwa:fu] 'turned (of milk)' [Bugonvu] 'softness' [BulaBufu] 'alertness' [Bujivu] 'learning'

In [8.16] I have stated the spirantisation rule

[8.16] (a)

(b)

r t k

Y d g

- s /

- z /

i

i

informally:2

148 The abstractness of underlying representations

(c) t k f/-

(d) 1 d -> v / . _gj

T h e account of spirantisation so far is s o m e w h a t inex- plicit. T h e w o r d 'certain' is t o o vague to characterise the vowels which condition spirantisation. T o m a k e it m o r e explicit o n e possible m o v e that could be adopted is to use a D I A C R I T I C F E A T U R E . An advocate of this less abstract solution could a r g u e that since the vowels triggering spiran- tisation are phonetically indistinguishable from those which do not, the best solution is to use an arbitrary diacritic, w h i c h is totally devoid of phonetic content, to identify the relevant v o w e l s . In the dictionary m o r p h e m e s which cause spirantisation w o u l d be m a r k e d w i t h a 'flag' saying ' [ + s p i r a n t i s a t i o n ] r u l e ' . 3 T h a t w o u l d be d o n e to m o r p h e m e s like the agentive nominalising suffix / - i / (similar in m e a n i n g to / - e r / in English w o r d s like maker) or the stative n o u n and adjective suffix / - u / .

O n the other hand, a . p r o p o n e n t of a m o r e abstract approach m i g h t interpret the evidence differently. For instance, H e r b e r t (1974) suggests that the fact that s o m e high vowels cause spirantisation while others d o not can be attributed to an u n d e r l y i n g P H O N O L O G I C A L difference between the v o w e l s . T h e vowels which cause spirantisation could be represented as j'\j and / u / respectively. T h e y have different distinctive feature matrices from those of / i / and / u / which do not cause spirantisation.

T h e phonological feature that distinguishes the vowels which cause spirantisation from those which d o n o t m i g h t be [ + tensc] (or [ + extra high]). Essentially, this position a m o u n t s to a claim that a l t h o u g h only five v o w e l s surface phonetically, there are seven vowels in the underlying phonological representation. T h e surface, phonetic vowels are [i e a o u] (which can be long or short) b u t the u n d e r - lying vowels arc / i i c a o u u / . T h e v o w e l s / i / and / u / are A B S O L U T E L Y N E U T R A L I S E D i.e. they occur in u n d e r - lying representations b u t the o p p o si t i o n b e t w e e n t h e m and

Absolute neutralisation 149

/ i / and / u / respectively is suspended or neutralized in all contexts in the phonetic representation. In other words, the extra-high sounds are ABSTRACT SEGMENTS which never occur phonetically but they are said to be nonetheless phonologically real. Their existence is inferred, as we have seen, from the phonological alternations in the language.

Such an abstract analysis is problematic. First, sanc- tioning absolute neutralisation means countenancing claims about neutralisation which are not normally regarded as legitimate. The orthodox view is that when a phonological opposition is neutralised, the opposition between sounds which contrast elsewhere in the phonetic representation is suspended in a specific context (Kiparsky 1968).

For instance, as you will recall from Chapter 2, in English voiced and voiceless stops are distinct phonemes. There are words like pull and bull which show that they contrast word meaning. In fact, they can contrast in any position - except after / s / . There is no possibility of [p] and [b] contrasting after / s / in putative words like [spul] and *[sbul]. The latter is not a possible English word. In the Luganda example the opposition between j\j and / i / (or between / u / and /u/) is suspended not just in one specific context, but everywhere in the phonetic representation. The essence of the criticism is that absolute neutralisation over- extends the concept of neutralisation.

A further objection to positing abstract segments which are absolutely neutralised concerns LEARNABILITY. As our phonological theory attempts to model speakers' knowledge of the sound systems of their languages, we should not assume underlying segments which infants acquiring a language would not be able to infer from the phonetic input to which they are exposed. So, although the solution which allows two extra-high vowels which never surface phonetically is historically sound, it will not be warmly embraced here. True, Proto-Bantu, Luganda's ancestor, had the seven vowels /i i c a o u y/; but at some point in the evolution of Luganda the vowel mergers shown in [8.17] took place:

[8.17] Pronto-Bantu i i e a o u u \ / \ /

Luganda i u

150 The abstractness of underlying representations

The abstract solution, while recognising the mergers at the phonetic level, denies their existence at the underlying level in present-day Luganda. This implies that phonological contrasts can be maintained long after they have ceased being realised phonetically. But how could Luganda- speaking children ever acquire the two extra-high vowels / i / and / u / which they never hear?

We can restrict the degree of abstractness of under- lying representations by insisting that only those segments that could be plausibly learned by infants on the basis of the phonetic input should be considered possible underlying segments. Thus, abstract segments which never occur phonetically would be excluded. That would mean that the linguist might on occasion have to concede that some of the regularities in the phonetic realisation of morphemes are not phonologically relevant in a synchronic grammar.

8.4 Conclusion

Underlying representations need to be posited in order to show relationships between morphemes. But, in setting them up, care must be taken to avoid positing base forms which are at worst arbitrary or at best historically valid but synchronically unmotivated. For the working phonologist, the difficulty is that abstractness is a scale rather than a dichotomy. It is not possible to establish the absolute cut off point beyond which the degree of abstractness becomes intolerable: while it might be relatively uncontroversial to proscribe ghost underlying extra-high vowel segments in Luganda, it would not be as easy to proscribe some form of ghost segment in cases like that of the French h- aspire. But we should not complain. That is what makes phon- ology so fascinating.

Exercises

I. Luo (Kenya) (Okoth 1979) Study the following data and answer the questions that follow:

Conclusion 151

Singular a b t luB got ko8 g u o k lep lak adit

'vegetable' 'stick' 'hill' 'rain' ' d o g ' ' t o n g u e ' ' t o o t h ' 'basket'

Plural a b d e luSe gode koQe g u o g i k p c leke adite

(The data is simplified. Many of the complications have been left out.)

(a) List the allomorphs of the plural morpheme. Do not attempt to predict their distribution.

(b) Determine the underlying representation of each noun root. Focus on the consonants. You should always select a base form which allows you to make the most general and consistent predictions about the behaviour of sounds.

(c) Write up formally the rule needed to account for the alternations in the final root consonant.

(d) Use these data to illustrate what is meant by neutral- isation in phonology.

2. Luyia (Kenya) Study the b e l o w . Noun class [ m b a k o ] [mbwa] [nduju] fndika] [ndimu] [ n d e m u ] [jrjusi]

[tfu] [jiama] [jiumba]

following

9 ' h o e ' ' d o g ' ' r a b b i t ' 'bicyle' i e m o n ' 'snake' 'jackel' ' h o u s e ' ' m e a t ' ' h o u s e '

data and a n sw e r the questions

Noun class 12 (diminutive) [xaBako] [xaBwa] [xatuju] [xatika] [xatimu] [xaremu] [xajusi] [xaju] [xajiama] [xajumba]

(a) List the allomorphs of the class 9 and class 12 prefixes. (b) What is the class 9 form of the following words:

[xaBusi] 'small goat' and [xateBe] 'small chair'.

152 The abstractness of underlying representations

(c) State verbally the rule governing the alternation in the shape of the class 9 prefix and give it a suitable name. Why is the rule preferable to the list in (a) above?

(d) Write up formally the rule you have stated. (e) State verbally the rules which account for the alter-

nation in the shape of the first consonant of the noun root. Give each rule a suitable name.

(f) Re-write your rules in (e) formally. (g) Must any of the rules you have formulated be extrin-

sically ordered? Justify your answer.

3. Hyman ( 1970 ) reports that in Nupe (Nigeria) conso- nants are palatalised before front vowels (see (a) below) and labialised before round vowels (see (b) below). Where the vowel following a consonant is neither front or round, as in (c), it has no effect on the preceding consonant.

(a) [egji] child (b) [egwu] mud [eg^e] bear [egw°] grass

(c) [ega] stranger [ta] to tell

However, palatalisation and labialisation also occur as in (d) and (e) where there are no front or round vowels following the consonant. Suggest a possible expla- nation for this.

(d) [egja] blood (e) [egwa] hand (Va] to be mild [twa] to trim

Notes

1. The interaction between phonology, morphology and the lexicon is one of the main issues in linguistic theory today. In Chapter 12 another approach, which to me appears more promising, is presented.

2. The vowel [i] causes palatalisation of a preceeding velar so that / k + i / —> [kJ] and / g + i / -* [g*] (but this will be ignored as it is not central to the problem which we are focusing on).

3. These spirantisation processes affect several other morphemes. For instance the perfective and the causa- tive suffixes have the same effects as the agentive nominalising suffix shown here.

CHAPTER 9

The syllable

9.1 The syllable

The syllable is at the heart of phonological representations. It is the unit in terms of which phonological systems are organised. It is a purely phonological entity. It cannot be identified with a grammatical or semantic unit. There are syllables like [An] as in unusual which are co-extensive with the morpheme; there are syllables like [kaet] cat which are co-extensive with the word; there are syllables like [kaets] cats which represent more than one morpheme (the noun root cat and the plural marker -s) and there are syllables like [mAn] and [ki] in [mArjki] monkey which represent only part of a morpheme.

9.2 The representation of syllable structure

The syllable has received a very considerable amount of attention from phonologists, especially in recent years, and a number of alternative models of the syllable have been offered. A serious attempt to compare them is beyond the scope of an introductory textbook of this kind. All that can be presented here is a very brief outline of some of the main trends before focusing on the one that we shall be using.

Many phonologists envisage a BRANCHING, HIER- ARCHICAL syllable structure. For a traditional structuralist statement of this position see Pike (1967) and Pulgram (1970). More recently, writers like like Kiparsky (1979), Halle and Vergnaud (1980), Steriade (1982) and Harris (1983) have presented a revamped version of the hierarchical branching theory in the framework of a MULTI-TIERED

154 The syllable

P H O N O L O G I C A L T H E O R Y . (This is an approach w h e r e phonological representations are viewed as consisting of a n u m b e r of independent levels that are linked to each other. See C h a p t e r 10). O n this view, syllable structure can be represented as in [9.1]:

[ 9 . i ] (a) a / \

O R / \

N M I / \

b ae t s 'bats'

(b)

(c)

a

R C 1 1

N 1

a / \ ) ] R

1

N 1

e v 3

a a / \ / \

O R O R 1 1

N 1

N 1 1

t 1 'little

N o t e : a = syllable, O = onset, R = r h y m e , N = nucleus and M = m a r g i n

Bats consists of o n e syllable. T h a t syllable has t w o constituents, n a m e l y the O N S E T w h i c h comes at the beginning and the R H Y M E w h i c h follows it. In the first syllable of e-ver, the r h y m e is simple. It does n o t B R A N C H . It contains j u s t o n e constituent: the v o w e l . T h e r h y m e of bats, on the other h a n d branches. It contains a v o w e l w h i c h is followed by a consonant. T h e examples a b o v e s h o w that the r h y m e is the H E A D C O N S T I T U E N T (i.e. the only c o m p u l s o r y constituent) of the syllable. T h e onset is the part that branches off on the left of the r h y m e , c o m i n g from the same n o d e . T h u s , in English, it is possible for a w e l l - f o r m e d syllable to contain n o onset, as in the case

The representation of syllable structure 155

of the first syllable of e-ver. But it is not possible for a well- formed syllable to exist without a rhyme.

The binary partition of syllables which we are suggesting is supported by versification practices in English:

[9.2] T w o syllables with the same onset but different rhymes alliterate (e.g. pan, pet, etc.) while two syllables with the same rhyme, but different onsets, are said to be rhymes (e.g. pan, man, etc.). In contrast, the onset and nucleus (e.g. pan, man) do not form a significant grouping for verse.

We noted above that the rhyme is the only essential element of the syllable in English. What is true of English is also true of other languages. The rhyme is always obli- gatorily present in all syllables in all languages. What varies from language to language are the elements that can be part of the rhyme. Typically the nucleus slot in the rhyme is occupied by a vowel but occasionally a consonant may fill that position, as in [9.1c] above where the final [1] of little is syllabic.

Find two more examples of syllabic consonants in English or in some other language. What kind of elements can precede or follow the syllabic consonants? Refer back to section 3.3.1 on page 43.

Another model - that of Hyman (1985), has a different slant. Hyman suggests that the core of phonological repre- sentations consists of rhythmic WEIGHT UNITS rather than onsets and rhymes or C and V slots proposed by other writers (see next section). Segments have weight units associated with them underlyingly. But only associations between weight units and vowels tend to survive to the surface. Normally consonants lose their weight units and get re-associated with the weight unit of an adjacent vowel by the syllabification rules. Only those segments whose association with a weight unit is preserved to the end of a derivation are syllabic.

156 The syllable

9.3 The CV-tier

Most current work in theoretical phonology assumes a model that incorporates a CV-tier (Consonant-Vowel tier) in terms of which the canonical forms of morphemes are stated. Precursors of this approach are Hockett (1947) and Abercrombie (1967). Using a multi-tiered approach in studies of classical Arabic, McCarthy (1979, 1981, 1982) has shown that P R O S O D I C TEMPLATES (see section 10.5 below) are needed to represent sequences of CV elements which function as morphemes. Though McCarthy's work has great phonological interest, his concerns are primarily morphological.

9.3.1 A generative CV-phonology model of syllable structure

It is Clements and Keyser (1983) who have expounded a CV-model of phonology specifically designed to deal with the syllable. Theirs is the model that I shall use in the rest of the book. Clements and Keyser require the theory of the syllable to perform three tasks:1

[9.3] (a) state universal principles governing syllable structure;

(b) state syllable structure TYPOLOGY, i.e. define the range within which syllable struc- ture may vary from language to language;

(c) state language-specific rules governing syllable structure

In order to fulfil function (a), the syllable is assumed to have a THREE-TIERED S T R U C T U R E consisting of a SYLLABLE N O D E 'a'; a CV-TIER whose C and V elements D O M I N A T E (i.e. have below them as constitu- ents in the syllable tree) consonantal and vowel segments; and a SEGMENTAL TIER consisting of bundles of distinc- tive feature matrices which represent consonant and vowel segments (for convenience these features may be abbrevi- ated using letters of the phonetic alphabet):

The CV-tier 157

[ £ M ] Syllable-tier

CV-tier C

segmental-tier

Roughly translated into the terms of the theory in [9.1], a V element of the CV-tier represents a syllable N U C L E U S i.e. peak of sonority (see below) while a C element represents a syllable O N S E T or MARGIN, i.e. an element which is not the peak.

Nowadays a CV-phonology model of some sort is assumed by many phonologists. The version proposed by Clements and Keyser has the advantage of being concep- tually simpler than the alternatives. Intervening between the syllable node and segmental tier there is a 'flat' CV-tier, lacking internal constituent structure. Contrast this with the more complex syllable models with onsets and rhymes which are illustrated in [9.1].

As in syntax, a tree like [9.4] shows IMMEDIATE C O N S T I T U E N T S T R U C T U R E . An element is an immediate constituent of a higher element within which it is contained. This is shown by a constituent being IMMEDIATELY D O M I N A T E D by that higher element. Thus, in pen the elements CVC of the CV-tier are all immediately dominated by o while the elements [p,e,n] of the segmental tier are immediately dominated by C V and C respectively.

In a theory where several tiers are posited it is essential to specify how the tiers are LINKED. In CV-phonology, the linking is done using ASSOCIATION LINES which are subject to a WELL-FORMEDNESS C O N D I T I O N (this principle is discussed in more detail in the next chapter). To relate the CV-tier to the segmental tier, associ- ation lines are drawn following certain universal rules. Normally, these rules link V elements to [-cons] segments

158 The syllable

(vowels) and C elements to [ + cons] segments, as you can see:

[9-5] (a) o a

A A\ C V C C V

m u n t u 'person' (Luganda)

(b) a o o o

I A A A\ v c v c v c v c 3 m e r I k 3 n American

One of the functions of the syllable in all languages is defining syllabicity for segments. Any segment dominated by a C-element of the CV-tier is nonsyllabic while any segment dominated by a V-element is syllabic. An inter- esting consquence of this model is that it obviates the need for the feature [syllabic] (section 3.3.1): the V element of the CV-tier is the constituent of the syllable that contains the S O N O R I T Y PEAK.

The class of segments capable of functioning as syllable peaks is not arbitrary. It has been noted by generations of phoneticians and phonologists that the distribution of segments in syllables follows a clear pattern which can be stated in terms of the S O N O R I T Y HIERARCHY in [9.6] below which was suggested by Hooper (1972, 1976). This hierarchy was introduced earlier in section 6.2.1.

[9.6] Sonority hierarchy least sonority 1 voiceless obstruents 2 voiced obstruents 3 nasals 4 liquids 5 glides 6 7 vowels greatest sonority

The CV-tier 159

The phonological sonority hierarchy has the phonetic correlates of openness and propensity for voicing. The more sonorous a sound is, the more audible it is likely to be. The sonority hierarchy is a mirror image of the strength hier- archy (section 6.2): sonority is in inverse proportion to strength (Hooper 1976).

The element dominated by V (which in other approaches is called the NUCLEUS) is relatively more sonorous than the consonants that surround it. (These are recognised as the O N S E T and MARGIN of the syllable in other models). In a word like bat, the vowel /ae/ is domi- nated by V and it is more sonorant than the consonantal segments / b / and / t / which it is flanked by. The chart in [9.6] correctly predicts that vowels are the most likely and obstruents the least likely segments to be dominated by V, with other sounds occupying intermediate points on the hieararchy.

As the syllabic potential of a sound depends on its propensity to vocalise, it is only to be expected that the more open a vowel is, other things being equal, the more likely it is to be the peak of sonority in its syllable. The algorithm (i.e. formal procedure) that assigns syllabicity works by ranking consonants and vowels on the sonority hierarchy. The most sonorous segment is assigned to the V-element (the nucleus). Less sonorous sounds preceding the nucleus are assigned to the initial C-element (onset) and those following it are assigned to the other C-element (also variously known as the C O D A , MARGIN, or TAIL). This principle predicts that the vowel / u / is dominated by a V- element in [9.5a].

It also predicts that, in [9.7] the vowel / u / is initially assigned to the V-element because it is higher up the sonority hierarchy than the consonant / m / which precedes it. But when the fact that it is followed by / a / is taken into account, association lines need to be re-drawn, making / a / the only [-cons] dominated by V and linking / u / instead with the preceding C-element. This makes it a nonsyllabic glide:

Clements and Keyser's model performs the second task of describing syllable typology by including a range of C O R E SYLLABLES. Linguistic elements which are part of the C O R E GRAMMAR are present in all langauges. At the

160 The syllable

[9.7] Glide formation

a a

A\ A C V V C V

NN I I . . m w a: n a 'child' Luganda

Note: (a) a broken line is an instruction to 'LINK' i.e. insert new association line;

(b) an association line with two lines through it shows DELINKING i.e. termination of an association.

level of the syllable, CV type syllables meet that require- ment. So far, no language has been reported to lack CV type syllables. Other syllable types may be seen as modi- fications of the prototypical CV syllable. Many languages, English included, have syllables containing only V, (see e- ver). Such languages may be assumed to have a rule at the entry to the phonological component which deletes the syllable initial C and thus allows canonical syllables with V only. Languages may also have C V C syllables which are obtained by a rule which adds a C after the V element to form canonical C V C syllables. Languages may have any one of the following canonical syllable types:

[9.8] Type 1: CV e.g. ta Type 2: CV, V e.g. ta, a Type 3: CV, C V C e.g. ta, tat Type 4: CV, V, CVC, VC e.g. ta, a, tat, at

(based on Clements and Keyser 1983:29)

Make up two long words (in a real or imaginary language) which have each one of the four syllable types in [9.8].

Further embellishments of the syllable types in [9.8], peculiar to different languages, do occur. To fulfil the third requirement of syllable theory set out in [9.3 c] a

Glide formation o a

A\ A C V V C V

m u a n a

The CV-tier 161

mechansim is needed to deal with language-specific syllable structure principles. There are languages which allow core syllables to have C* or V* (where C* or V* represents sequences of C or V elements) so that well-formed core syllables may contain combinations like C C C V C C or CCVVC or V elements. Thus, while [9.5] only shows single C and V elements following each other, languages like English allow syllables with C C C V C C sequences as in the word strand. The theory has to provide a mechanism for stating such language-specific facts. We shall explore this in section 9.4.1.

9.3.2 Syllabification

The theory has to provide a way of grouping arrays of CV elements into syllables in situations like this: VCVCCCVC. From the foregoing it is clear that each V-element will be associated with a syllable peak. What is yet to be shown is: to which syllable node are C-elements assigned in ambiguous cases, where they could go with either the preceding or the following vowel? To which syllable, for example, should the middle consonant of panic (CVCVC) be assigned?

The O N S E T FIRST PRINCIPLE (Kahn 1976, Clem- ents and Keyser 1983) has been proposed to deal with such situations. It is stated in [9.9]:

[9.9] (a) 'Syllable-initial consonants are maximised to the extent consistent with the syllable struc- ture conditions of the language in question.

(b) Subsequently, syllable-final consonants are maximised to the extent consistent with the syllable structure of the language in question.' (Clements and Keyser 1983:37)

Principle (a) applies before (b) in any derivation. In potentially ambiguous cases initial consonant clusters take precedence over syllable final ones. This means that unless there is an overriding language-specific reason for doing otherwise, given a string like VCV, the Onset First Prin- ciple requires that the string be divided up as V-CV rather than VC-V : a word like ever [eva] is divided up as [e-va]

162 The syllable

and not *[ev-a]. To take another example, English allows C C sequences like [sp]. They can be initial as in spoon or final as in grasp. In a word like aspire, where the [sp] cluster could be regarded as syllable initial or syllable final, the word can be syllabified as either a-spire or asp-ire. The Onset First Principle predicts that the former is the correct syllabification.

The theory incorporates the following algorithm (formal, step-by-step procedure) for building syllables, with the procedures being applied starting from V outward to successive C-elements in the order specified in [9.10] below, which is based on Clements and Keyser (1983:38):

[9.10] (a) Underlyingly every V of the CV-tier is linked to o; this merely reflects the fact that no syllable exists without a V element (as nucleus).

(b) Link each C element to the nearest V-element to its right provided the resulting sequence of segments does not violate any language- specific rules. This procedure creates syllable onsets.

(c) Repeat the procedure in (b), this time linking the C-elements to the nearest V to its left. This procedure creates syllable margins.

The effect of this algorithm is illustrated in [9. n ]

[9.11] (a)

c v c c c c v c c L I I I I I ! I I c o n s c r i p t

V-elements are pre-linked with o by conven- tion [9.10a]

(b)

/ I ---v- .' I c v c c c c v c c I I I I I I I I I c o n s c r i p t

The CV-tier 163

By convention [9.10b], link C-elements to the V on their right, one at a time, provided the resulting sequence is permissible in the language in question. Thus, in this case the procedure creates sen but stops short of *nscri because in English nasals are not allowed to occur at the beginning of a syllable initial consonant cluster (see section 9.4.1 below).

to

c v c c c c v c c I I I I I I I I I c o n s c r i p t

By convention [9.10c] link C-elements to the V preceding them so long as the resulting sequence is allowed in the langauge.

Write rules and produce derivations similar to [9. n ] above to account for the syllabification of the following words: agony [segani], corner [kona], December [disemba], extinct [ekstirjkt]

You should come up with trees like the following:

(d)

0 0 0 a a a a a a a

I A A AA A A A A ^ 7 K v c v c v e v e v c v c v c c v v c c c v c c c a e g a n i k 0 n 3 d i s e m b a e k s t i r j k t

It must be emphasised that any language-specific restrictions on consonant clustering override universal prin- ciples. For instance, since tl- is not a permissible combi- nation syllable-initially in English, a word like atlas, the Onset First Principle notwithstanding, cannot be syllabified as *a- tlas; rather it must be divided up as at-las. Universal prin- ciples can be viewed as providing the default state of affairs

164 The syllable

which applies, unless specific instructions to the contrary are given by the grammar of a particular language.

9.4 Functions of the syllable

In early generative phonology, although the feature [syllabic] was used, the syllable was not given a place in the theory. It was assumed that segments, boundaries and rules stating permissible combinations of segments in morphemes and words were sufficient to describe the sound systems of languages. That is the stance taken in SPE. But subsequent work (e.g. Hooper 1972; Vennemann 1972; Bell and Hooper 1978; Kiparsky 1979; Selkirk 1980) showed that there are good reasons for rejecting that position. Today the place of the syllable is secure. Below I show the central role it plays in phonology.

9.4.1 The syllable as the basic phonotactic unit

One of the most basic functions of the syllable is to regulate the ways in which lower level units (consonants and vowels) of the phonological hierarchy can combine.

Knowledge of the phonological system which speakers of a language have consists in part of a knowledge of the phonemes of that language and their allophones. But this is not all.

Suggest at least one reason why phonological knowl- edge must go beyond knowledge of phonemes and their allophones. Use the data in [9.12] to support your argument.

[9.12] *f/eg (compare [buf/eg] bootleg) *ndziz (compare [m3tj"3«rfaiz] merchandise) *bmit (compare [sA^mit] submit) *psrrj (compare [kalse/^irj] collapsing)

What [9.12] illustrates is the importance of constraints on the combination of sounds. All the starred pseudo-words

Functions of the syllable 165

contain English phonemes; all the consonant sequences at the beginning of the pseudo-words are permissible in English: [tl] occurs in bootleg, [nd] occurs in merchandise, [bm] occurs in submit and [ps] occurs in collapsing. But, the sequences in [9.12] are not potential English words. Just as at the level of grammar not all sequences of words produce a well-formed grammatical sentence, so it is in phonology: not all combinations of sounds produce possible words. Some non-occurring nonsense words, like those in [9.12] are beyond the pale, while other non-occurring words could easily be turned into real words if a meaning could be found for them. For instance, if you invented a machine which automatically does phonological analysis, you could call it by any one of these names:

[9.13] [fonalaiza] [ausl] [gluka]

As a speaker of English, you know that the non-occurring words in [9.13] are potential words while those in [9.12] are not. The rules which reflect speakers' knowledge of what combinations of sounds are allowed in their language are variously referred to as P H O N O T A C T I C RULES or M O R P H E M E S T R U C T U R E C O N D I T I O N S .

The syllable is the unit in terms of which phonotactic rules are best stated. Thus, for instance, in English the sequence [tl] is allowed so long as the t and the / belong to different syllables as in boot-leg [but-leg], at-las [set-Is], part- ly [pat-li] or litt-le [lit-J] (syllabic / is the nucleus of the second syllable (see [9.1c]). But the sequence tl is not allowed in the same syllable. Hence the impossibility of *tleg as an English word since it has tl as a syllable onset. Likewise, [bm] can occur in English if there is a syllable boundary separating them as in sub-mit but [bm] would not be allowed where those two sounds belonged to the same syllable as in our imaginary word *bmit.

Another English example of a phonotactic constraint is the rule which only allows vowels to follow syllable- initial affricates: while cheap, judge and adjust are English words, *chleep, *jpudge and *adjpudge are neither real nor potential words in English.

166 The syllable

Supply three examples of your own of: (a) non-occurring possible words in English; (b) non-occurring impossible words in English. Explain the grounds for your decisions.

Constraints on syllable structure serve as a filter allowing only certain sound sequences to occur. These constraints are specific to a particular language. What is a well-formed syllable in Swahili may not be in English. In Swahili (and in many other African languages) for instance, N C sequences like [nd] as in [ndugu] 'brother' or [rjg] as in [rjguruwe] 'pig' are allowed in syllable- (and word-) initial position but they are outlawed in that position in English: *[ndu], *[mpig] and *[gget] are not potential words in English. (That is why the names of African leaders like Nkrumah and Nkomo tend to have a short [i] or [a] vowel inserted before them to make them pronounceable by English speakers.)

The nativisation of foreign loanwords and the phenom- enon of'foreign accent' provide interesting evidence of how deeply ingrained syllable structure rules are. For the same reason that native speakers of English insert a vowel before a word-initial (and syllable-initial) velar nasal in words like Nkrumah, speakers of languages like Walpiri (Australia) and Luganda, which only have syllables ending in V in the phonetic representation, will insert a vowel after a syllable- final consonant in borrowed words such as [wAn] 'one'. In Walpiri one is rendered as [wani] and in Luganda as [wanu]. Likewise, English loanwords in Japanese are normally modified to fit in with the predominantly CV syllable struc- ture of the language. So, 'baseball' and 'milk' become [besuboru] and [miruku] respectively.

9.4.2 The syllable as the domain of phonological rules

The relevance of syllable structure constraints is not restricted to loanwords and mother tongue interference. Syllable structure often plays an important role in condi- tioning the application of phonological rules internal to a language.

Functions of the syllable 167

A frequently cited example of the relevance of the syllable in determining whether a phonological rule applies is some form of obstruent devoicing rule which is found in many languages including Russian, German and Turkish. In Turkish it is responsible for alternations like [rengi], (possessive), [renkten] (ablative) and [renk] (nominative) 'colour'; in German for the alternation between [tak] 'day' and [taga] 'days'; in Russian for the alternation between [gorat] 'town' and [garada] 'towns'.

The rule can be written as follows:

[9.14] [-sonorant] —> [-voice]/—

This rule correctly states that obstruents are devoiced word- finally or before another consonant. But it misses the point that the environments 'word-finally' and 'before another consonant' are not accidentally related contexts. The two environments share the property of being syllable final.

The makeshift nature of the solution provided by rule [9.14] becomes even clearer when it is appreciated that the environment 'word-finally or before another consonant' turns up frequently not only with respect to obstruent devoicing, but also with respect to other rules. Take the nasalisation rule of French (which is similar to a nasalisation rule found in numerous languages):

[9.15] V - » [+nasal] / N

Vowels are nasalised when followed by a preconsonantal nasal as in /enfle/ [afle] enfie 'swollen' or word-finally as in / b o n / [bo] bon 'good'.

Account for the nasalisation of the vowel in grand [grfi] 'big' and sentiment [satima] 'feeling'.

The same nasalisation rule applies even where on the surface the nasal is not syllable-final in words like sentiment [satima] (sentimental [satimatal] 'sentimental'); and grand [gra] (grande (feminine) [grad]). A derivation is worked out for grand to show this:

C #

168 The syllable

[9.16] UR: / g r a n d / Rule a

Vowel nasalisation before syllable-final N C : grand —> grand (see [9.15])

Ruleb Consonant deletion: grand —» gra (see [7.4])

PR [gra] (Note: rules a and b may apply simultaneously as the underlying representation satisfies the structural descriptions of both rules.)

If the relevance of the syllable is not recognised, the fact that the same environment turns up in various rules in different languages remains a puzzling mystery. If, however, the syllable is recognised, it becomes obvious that these processes are conditioned by the presence of a syllable boundary.

Furthermore, probably in every language there are phonological processes whose motivation is the preservation or the creation of preferred syllables. The function of such rules can only be understood if they are approached in terms of the syllable. A classic example in the literature comes from the Californian language Yawelmani. In this language consonant clusters are allowed only if they do not exceed two consonants. When a consonant-commencing suffix such as the aorist" suffix - hin is added to verb roots whose last two segments are consonants, a vowel is inserted between the last two root consonants to prevent a tricon- sonantal sequence from occurring:

[9.17] root - aorist /?ilk - hin/ -H> pihkhin] 'sang' / p a ' t - hin/ —> [pa'ithin] 'fought' (t is retroflex)

If a vowel-commencing suffix like the dubitative suffix -al is added, however, no vowel insertion takes place since the requirement of a maximum of two consonants in a cluster is not violated. (The dubitative mood is used to express doubt.)

[9.18] root - dubitative /'ilk - al/ - ^ pilkal] 'might sing' /pa?t - al/ —> [pa?tal] 'might fight'

Functions of the syllable 169

9.4.3 The syllable and the structure of complex segments

The syllable not only regulates the combination of segments, it also controls the combination of features which make up segments. In this, current thinking differs from that of early generative phonology which assumed that phonemes were bundles of features with these two characteristics:

(i) the scope of each feature was one segment and no feature could extend over adjacent segments;

(ii) within a single phoneme the features were unordered which meant that there was no subphonemic structure.

Both these assumptions fail in many cases. Frequently articulatory gestures are not started and completed within a single phonological segment (Firth 1948). Obvious exam- ples are words like guard-room and barbarism in which the feature [+voice] is not restricted to particular segments but extends over the entire word. We need not labour the point that was made in Chapter 4 that since speech organs are not mechanical ratchets, it is only to be expected that articula- tory gestures are going to overlap in the transition from one segment to the next.3

One of the main functions of the syllable is to provide an analysis of the internal structure of segments and to indicate the number of rhythmic units present in a syllable. This depends on the way C and V elements present on the CV-tier are linked with consonant and vowel segments on the segmental tier (Hyman 1985). Three patterns of internal segmental structure are possible:

(a) A one-to-one association of V or C with a segment:

Syllable-tier [9.19]

CV-tier

segmental-tier - c o n t — voice + labial — low

+ syll — back - h i g h

+ nas + ant + cor

170 The syllable

(b) Simultaneous association of one segment with two C or V slots. That is the case when consonants are GEMINATED (i.e. the same consonantal articulation is held for the duration of two consonantal beats) or when a vowel is lengthened (i.e. the same vowel quality is maintained over two V slots). I represent both possibilities with a Luganda example in [9.20]. The word ttaala [t:a:la] 'lamp', begins with a geminate t followed by a geminate vowel.

[9.20]

00 c c (b) v V C

1

(c) The third possibility is the simultaneous association of a single C slot with two segmental distinctive feature matrices. This is what happens when complex segments like affricates occur. Affricates like [pfj, [tf] and [d3] are described using the feature [+delayed release] in SPE (see section 3.3.6). But they can be more revealingly represented in this way:

[9.21]

Affricates

C

[ - c o n t ] [ + cont] e.g. t / = [tf\

This makes the feature [delayed release] superfluous. Diphthongs are treated in analogous fashion:

[9.22]

Diphthong

V

I (as in [en] 'eight')

Functions of the syllable 171

Many languages have prenasalised consonants which, like the complex segments above, do show sequential organization of features at the subsegmental level. An example of a prenasalised consonant is the sound [nd] as in Kikuyu [ndegwal] 'bull'. In the light of [9.21], how should pre-nasalised consonants be represented?

A prenalised consonant can be represented as follows:

[9-23]

£ C / \ A

[+nasal] [-nasal] e.g. n d = [ nd]

9.4.4 Compensatory lengthening

In addition, the approach to the syllable outlined in the last section accounts in a natural way for some traditional prob- lems in phonology such as C O M P E N S A T O R Y LENGTH- ENING. In many languages, if an underlying syllabic segment is deleted or is released as nonsyllabic, an adjacent syllabic gets lengthened 'in compensation'. We shall exem- plify this with the data from Luganda in [9.24].

[9.24] / b a + a + l a b + a/ [ba:laba] 'they saw' (but / b a + k u + l a b + a / [bakulaba] 'they see you') / b a + e + l a b + a / [be:laba] 'they see themselves' / m u + a + l a b + a / [mwa:laba] 'you saw' (but / m u + t u + l a b + a / [mutulaba] 'you see us') /li + ato/ [lja:to] 'boat, canoe' / m a + a t o / [ma:to] 'boats, canoes'

The rule is that a high vowel is realised as a nonsyllabic glide if it appears followed by another vowel; it is delinked from a V slot and re-associated with a C slot (see rule [9.7] above). But a nonhigh vowel is deleted altogether if it appears followed by another vowel. The motivation for the glide formation and vowel deletion rules in Luganda is to

172 The syllable

prevent two dissimilar vowels from being adjacent to each other in the phonetic representation. Interestingly, the number of V-elements on the CV-tier remains stable regardless of the fate of the first vowel. How can this be explained?

Let us assume that in the underlying representation each C and each V element of the CV-tier has a certain amount of potential duration. This can be represented as a BEAT or TIMING U N I T (or WEIGHT UNIT). Normally only those timing units which are associated with vowels survive in the phonetic representation. Timing units tend to reflect syllabic peaks because the syllabification rules in [9.10] ensure that only the timing unit of the more sonorous segment is retained when segments are grouped together in a syllable.

We can now interpret compensatory lengthening as follows: when the first vowel is deleted or realised as a nonsyllabic glide, its V-slot (and timing unit) is inherited by the second vowel which becomes simultaneously asso- ciated with two V slots and hence has virtually the duration of two vowels in the phonetic representation:

[9-25]

a a a

AAA C V V C V C V

IJN I I I I b © e 1 a b a [be:laba]

o a

AA C V V C V

K M I I 1 j a t o [lja:to|

o o u AAA C V V C V C V I I I I I I I b a e 1 a b a

[9.26]

u o

AA C V V C V —

1 i a t o

See if you can find an example of compensatory lengthening in a language that you know.

Functions of the syllable 173

9.4.5 The syllable as indispensable building block for higher phonological domains

In recent years, phonological research has amassed evidence s h o w i n g that the syllable is the h u b of phonological o r g a n - isation. In m a n y languages, higher p r o s o d i c p h e n o m e n a like stress, nasalisation, and quantity (length) can only be insightfully described in t e r m s of the syllable because often, in order to d e t e r m i n e w h e t h e r a given rule is applicable, the n u m b e r of syllables in a w o r d (or part of a w o r d ) has to be counted. T h e r e are rules which require main w o r d stress to fall on a certain syllable of the w o r d w h i c h could be, say, the last syllable or second syllable from the end (penultimate syllable) of a w o r d . W e shall p r e v i e w this kind of rule here and return to it in m o r e detail in C h a p t e r 11.

Which syllable receives stress in the Swahili data in [9.27]?

[9- 27] piga pigana piganija tutawapiga

'hit' 'fight', 'cause to fight'. ' w e shall hit t h e m '

t u t a w a p i g a n i j a ' w e shall m a k e t h e m fight'

T h e answer is that stress falls on the penultimate syllable. T h e relative position of stress remains the same even w h e n the w o r d g r o w s longer w i t h the addition of affixes.

Swahili has another rule which determines w h e t h e r the n o u n class 9 nasal prefix is syllabic. This rule also takes the n u m b e r of syllables in a w o r d into account: if attached to M O N O S Y L L A B I C roots, the nasal prefix is syllabic but if attached to longer r o o t s , it is non-syllabic:

[9.28] (a) nta point (b) m b u z i goat ntji c o u n t r y rjguruwe pig

T h e next e x a m p l e is from Luganda. It involves C L I T I C S . B u t before w e e x a m i n e the Luganda data w e need to digress and clarify the n o t i o n of 'clitic'. Typically a clitic is an unstressed particle which is attached to a H O S T (i.e. main) w o r d and is incapable of standing on its o w n .

174 The syllable

Often clitics affect the stress pattern of the h o s t w o r d . A clitic attached to the b e g i n n i n g of a w o r d is called a P R O C L I T I C and a clitic attached to the end of a w o r d is called an E N C L I T I C . S o m e clitics are derived from self- standing w o r d s . For e x a m p l e , the French first person p r o n o u n form je (as in je le vois 'I see h i m ' ) is an i n d e - p e n d e n t w o r d . B u t in j'ai (from je ai) 'I h a v e ' it is a P R O C L I T I C . In English not is a separate w o r d in she is not b u t it is an E N C L I T I C in she isn't. O t h e r clitics are n o t derived from i n d e p e n d e n t w o r d s e.g. Latin -que ' a n d ' as in mensamque 'and the table' (accusative case). T h e process of adding clitics is called C L I T I C I S A T I O N .

W e can n o w r e t u r n t o Luganda after the digression. In this language there is a v o w e l length rule w h i c h is sensitive to the n u m b e r of syllables. S o m e w o r d s of m o r e than one syllable end in a long v o w e l and others in a short v o w e l . T h e difference b e t w e e n these t w o w o r d types s h o w s u p w h e n an enclitic (a g r a m m a t i c a l particle attached to the end of a w o r d ) like the interrogative m a r k e r -ki is present. A w o r d like mutawama ' t r o u b l e ' w h i c h underlyingly ends in a l o n g v o w e l b e c o m e s [mutawa:na:ki] ' w h a t t r o u b l e ' . But a w o r d like mukozi ' w o r k e r ' w h i c h u n d e r l y i n g l y ends in a short v o w e l , w h e n cliticised (e.g. w h e n the enclitic inter- rogative particle -ki is attached to it), b e c o m e s [mukozi'ki] ' w h i c h w o r k e r ' n o t * [ m u k o z i : k i ] .

Interestingly, in L u g a n d a all m o n o s y l l a b i c r o o t s end in a l o n g v o w e l in the u n d e r l y i n g representation. T h a t long v o w e l is shortened in m o s t contexts in the p h o n e t i c r e p r e - sentation. B u t it is protected and s h o w s up before enclitics like the interrogative m a r k e r as y o u can see:

[9.29] M o n o s y l l a b i c length Underlying Representation

ki-taa 'calabash'

Cliticised Form Phonetic Representation kita: ki ' w h i c h calabash'

ki-loo ' n i g h t '

ma-taa ' m i l k '

ki-lo: ki ' w h i c h m i g h t '

ma-ta: ki ' w h i c h milk'

Non-cliticised Form Phonetic Representation kita 'calabash'

ki-lo ' n i g h t '

ma-ta ' m i l k '

Syllable weight 175

m u - t n ' t r e e '

m u - t i : ki ' w h i c h tree'

m u - t i ' t r e e '

Swahili and L u g a n d a are not e x t r a o r d i n a r y in requiring a count of the n u m b e r of syllables before deciding w h e t h e r or n o t a rule applies. S o m e other languages, for e x a m p l e Spanish, d o the same. Jaeggli (1980) has s h o w n that there are Spanish dialects w h e r e the form that a d i m i n u t i v e suffix takes depends simply on the n u m b e r of syllables in the n o u n to which -the d i m i n u t i v e suffix is added: -sita/sito is added after disyllabic w o r d s and -itaj-ito after trisyllabic ones.

[9.30] (a) -sita (Jem.)J -sito (masc.)

(b)

Semantically, the addition of a d i m i n u t i v e suffix to a n o u n has the effect of changing its m e a n i n g so that it can be paraphrased in English as 'a little or insignificant s o m e - body or s o m e t h i n g ' .

Noun madre cruz buey -it a comadrc dinosauryo

'mother' 'cross' 'bull'

'midwife' 'dinosaur'

Diminutive madre-sita cruze-sita bueye-sito

comadr-ita dinosaur-ito

9.5 Syllable weight

Traditionally the major distinction d r a w n between syllable types found in languages has been b e t w e e n O P E N SYL- LABLES and C L O S E D S Y L L A B L E S . A n open syllable ends in a v o w e l while a closed syllable ends in a consonant. In some languages syllables typically end in a v o w e l , that is to say, they are open. T h a t is the situation in languages like Japanese and Luganda (if syllabic nasals are disregarded). In other languages, like French and English, syllables can end in a consonant. But, even in those languages which allow closed syllables, there is often a clear preference for open syllables. In French, for instance, syllable final consonants suffer a considerable degree of attrition. T h e r e is a rule which deletes w o r d final consonants in w o r d s like petit [pa

176 The syllable

ti] 'little' and preconsonantal consonants as in enfant [dfa] 'child'. T h e effect of this deletion rule is to to t u r n w h a t w o u l d be a closed syllable i n t o an open syllable.

T h e consensus today is that m o r e i m p o r t a n t than the traditional classification of phonological systems in t e r m s of open and closed syllables is their classification in terms of S Y L L A B L E W E I G H T . In n u m e r o u s languages a factor that determines the applicability of certain phonological rules is the W E I G H T of the r h y m e . Essentially, a syllable is L I G H T if it contains a n o n b r a n c h i n g r h y m e as in [9.31]. B u t a syllable is H E A V Y if it contains a b r a n c h i n g r h y m e as in [9.32]. T h e onset seems never to play any role in the c o m p u t a t i o n of syllable w e i g h t . C o n s e q u e n t l y , its internal structure is irrelevant.

Generally, languages in w h i c h a distinction between light and heavy syllables is d r a w n fall into t w o camps which are s h o w n as type A and type B b e l o w . T o begin w i t h , for convenience, I assume that the syllables have the structure given in [9.1] at the b e g i n n i n g of this chapter. Type A Languages

(a) In a light syllable the r h y m e contains a short v o w e l as in [9-31]:

[9.31]

N

o / \

O R

N

a p a

(b) In a heavy syllable the r h y m e contains e i t h e r (i) a long v o w e l or d i p h t h o n g optionally followed by

one or m o r e consonants; o r (ii) a short v o w e l followed b y at least one consonant

as in (9.32)

[9-32] a

O R

N / \

p a a

a / \

O R / \

N M

Syllable weight 177

Type B Languages (a) In a light syllable the r h y m e contains a short vowel.

As always, the presence or absence of a consonant in the onset is irrelevant. B u t in this case so is the presence of a consonant in the m a r g i n , following the nucleus:

[9-33] o

O R

N

o / \

O R N M

I I p a t

(b) In a heavy syllable the r h y m e contains a l o n g v o w e l or d i p h t h o n g . T h e presence or absence of any consonant in the m a r g i n being again irrelevant.

[9-34] a a

O R O R I

N / \

y \ N M

y\ i P a P a a t

P r o b a b l y Latin is the best k n o w n e x a m p l e of a type A language. Stress placement in Latin was g o v e r n e d by a rule which was sensitive to syllable w e i g h t :

[9.35] Latin Stress Rule (i) In disyllabic w o r d s , stress the first syllable,

as in [re-go:] rego 'I rule' (ii) In a w o r d of m o r e than t w o syllables, stress

a heavy penultimate syllable (the second syllable from the end of the w o r d ) . A syllable is heavy if it contains a long vowel as in [i-ni-mf:-kus] inimkus ' e n e m y ' or, alternatively, if in the m a r g i n , following the v o w e l there is a consonant as in [re:-ksis-tis] rexistis ' y o u (plural) ruled'.

(iii) If the penultimate syllable is short and has n o consonant in the m a r g i n , it is regarded as light and stress in that case is placed on the antepenultimate syllable, as in [l-nsu-la] insula 'island'

178 The syllable

Syllable w e i g h t is at the heart of Latin M E T R E . Traditional Latin p r i m e r s , like K e n n e d y (1948:203) o n w h o m the account here is based, have t a u ght generations of students that a V E R S E (i.e. line of Latin poetry) consists of a certain n u m b e r of F E E T . Each foot is m a d e u p of t w o or m o r e syllables. O n e of the syllables in the foot, usually the heavy one d o m i n a t e s the other(s). Usually the metrically d o m i n a n t part of the foot is called the RISE ( _ ) and the w e a k e r part is called the FALL ( u ) . T h e principal feet in Latin poetry are listed in [9.36]

[9.36] I A M B U S : one light syllable followed by a heavy one e.g. card 'flesh' ( " Fall, ~ Rise)

T R O C H E E : one heavy syllable followed by a light one e.g. mensa 'table' (~ Rise, ~ Fall)

A N A P A E S T : t w o light syllables followed by a heavy one e.g. patulae ' b r o a d ' ( ^w Fall, "Rise)

D A C T Y L : one heavy followed by t w o light ones e.g. litora ' s h o r e ' ("Rise, w w Fall)

S P O N D E E : t w o heavy syllables e.g. herds ' h e r o ' ( wFall - Rise; or " Rise ~ Fall)

T R I B R A C H : three light syllables e.g. temere ' t o fear' ( °FaIl, "Rise; or - Rise, wFall)

T h e spondee and tribrach arc rare. T h e c o m m o n feet all share the characteristic of having s t r o n g , d o m i n a n t syllables which alternate w i t h w e a k ones.

Let us leave Latin and t u r n to an e x a m p l e of a type B language. W e have b o r r o w e d Larsen and Pike's (1949) data from the Mexican language Huasteco for our illustration.

W o r k o u t the placement of stress in this language, s h o w i n g the interaction b e t w e e n stress and syllable w e i g h t . V o w e l length is p h o n e m i c and it is indicated b y a colon in the usual w a y . A n a p o s t r o p h e after a consonant s h o w s that it is globalised. T h e accent m a rk s stress.

Syllable weight 179

[9.37] Disyllabic words / ' a t ' e m / 'salt' / c a l a m / ' s h a d e ' / b u x ' i ? / ' c o w a r d ' / ' e : j a l / ' b o s s ' / c i j o : k / 'chin' / ' a m u : l / ' r u b b i s h ' / ' i i l a : b 'seed' / y a : n r . l / ' m a n y t i m e s '

[9.38] Trisyllabic words / h i l k ' o m a / 'leftovers' / ' a : u l o r n / 'field of garlic' / k w ' a h i : l o m / ' w i n d o w ' / h u : u:k' ik/ 'blisters' /?alabe:l/ ' p r e t t y ' / b i : n o m a c / ' o n e w h o g a v e ' / ? u b a : t ' l a : b / ' g a m e , p l a y t h i n g ' / ' e : l a : w a : j / ' ( t n e y ) surely find each o t h e r '

I h o p e you have w o r k e d o u t a s t a t e m e n t along these lines for predicting stress placement in H u a s t e c o :

[9-39] (i) If a w o r d contains one or m o r e long vowels, stress falls on the syllable w i t h the last long v o w e l ,

(ii) If a w o r d contains n o l o n g v o w e l s , stress falls o n the first syllable.

T h e u p s h o t of this discussion is that while the r h y m e plays a role in d e t e r m i n i n g the applicability of stress rules, the onset does not. As a rule, in order to apply rules, it is necessary to k n o w the constituent s t r u c t u r e of the r h y m e but n o t that of the onset. T h e crucial characteristic of r h y m e s has been formally stated in the literature in t e r m s of the B R A N C H I N G R H Y M E H Y P O T H E S I S : a syllable with a n o n b r a n c h i n g r h y m e , is light while a syllable with a branching r h y m e is heavy. T h e difference between type A languages like Latin and type B languages like Huasteco is accounted for by assuming the tree g e o m e t r y in [9.40] and [9.41] respectively for these languages.

T h e m e t a p h o r of P R O J E C T I O N 4 has been used to express the principle at stake: in quantity-sensitive stress systems, w h e r e syllable w e i g h t plays a key role, we could say that as far as the stress rules are concerned, the syllable onset is n o t relevant, it is n o t 'seen'. W h a t is projected (on

180 The syllable

[9.40] Type A Heavy Syllables

a a

O R O R X \

N M I I

I N

/ \ C V C CV V

[9.41] Type B Heavy Syllables

a

O R

N

C V V

o / \

O R / \

N M / ^ \

C V V C

an imaginary screen, as it were) and 'seen' by the stress rules is the rhyme. In type A languages what is watched out for is branching anywhere in the rhyme, be it at the level of the nucleus and margin, or within the nucleus itself. In type B languages, on the other hand, only the nucleus is projected. The question whether branching occurs is only asked about the nucleus.

Let us now return to the CV-tier model and restate syllable weight within that framework. In CV-tier phon- ology the differences between type A and type B languages can be accounted for by assuming that only V (and the segments it dominates) and any C following the V element is projected. The situation in the two types of language is shown in [9.42] and [9.43] respectively. (I have boxed off the elements that are projected.)

[9.42] Type A Language (a) light syllables

Abstract segments 181

(b) heavy syllables

a o

C

A V C

a t

[9.43] Type B Language (a) light syllables

C V C V C

P I a | t

(b) heavy syllables

c

p

V

A a a

or C

P

V

A a a

9.6 Abstract segments

The theory of the syllable outlined here also succeeds in throwing some light on one of the most recalcitrant prob- lems in phonology: the problem of abstract segments. In many languages phonologists have discovered that 'ghost segments' which do not appear in the phonetic represen- tation may affect the way in which phonological rules apply (see Chapter 8, section 3, pages 145-146).

Consider the facts of French which are shown in [9.44] and [9.45]:

[9- 44][labe] [lane] [leta] [lidantite]

1(e) abbe 1(a) annee 1(e) etat 1(a) identite

'the abbot' 'the year' 'the state' 'the identity'

182 The syllable

A rule deletes the vowel of the definite article w h e n the next w o r d begins w i t h a v o w e l :

[9.45] V o w e l truncation

[-cons] -H> 0 / _ # V

T h e c o n s o n a n t - c o m m e n c i n g n o u n s in [9.46] are unaffected:

[9. 46][b b w a ] le bois 'the forest' [la karafj la carafe 'carafe, w a t e r - b o t t l e ' [ b pa] le pas 'step, pace' [la maladi] la maladie 'illness'

In the light of the solution provided above, h o w should w o r d s below c o m m e n c i n g w i t h the so-called ' H A S P I R E ' be represented in u n d e rl y i n g lexical representations? Justify the rule which y o u p r o p o s e . (You m i g h t find it useful to review first the discussion in section 3 of C h a p t e r 8.)

[9.47] [le ara] *[lez ara] les haras 'the stud f a r m s ' [le arpist] *[lez arpist] les harpistes ' t h e harpists' [le er5] *[lcz er5] les herons 'the h e r o n s ' [le ord] *[lez ord] les hordes 'the h o r d e s ' [le up] *[lez up] les h o u p p e s 'the b u n c h e s '

T h e p r o b l e m is this: a l t h o u g h phonetically they start with a v o w e l , these w o r d s behave phonologically as t h o u g h they started w i t h a c o n s o n a n t . This raises questions about the nature of that consonant as w e saw o n page 146.

T h e solution which the t h e o ry of the syllable outlined here enables us to c o m e u p with is simple. T h e t h e o r y allows C and V elements to exist at the C V - t i e r w i t h o u t being linked to c o n s o n a n t or vowel s e g m e n t s . In that event, they do n o t surface in the phonetic representation — they are n o t p r o n o u n c e d . B u t phonological rules affecting the C V - tier w o u l d have access to t h e m .

In the case of h-aspire w o r d s , there is an unattached C w h i c h inhibits the application of the v o w e l truncation (de- letion) rule although it is not linked to any concrete s o u n d on the segmental tier. T h e situation can be represented in this w a y :

a a [0.48I / \ S^ iv-4°j c VC V

I I I a r a haras 'stud farm'

Extrasyllabicity 183

9.7 Extrasyllabicity

T h e reverse situation also occurs. A s e g m e n t can be phonetically fully specified w i t h o u t being linked by associ- ation lines to a C or V. Such a 'floating' s e g m e n t is n o t part of any syllable: it is E X T R A S Y L L A B I C .

A n o t h e r French e x a m p l e illustrates this. Instead of using the familiar consonant truncation rule (see [7.4] and [9.16]) which deletes syllable-final consonants unless they are followed by a v o w e l , w e can simply attribute the failure of such consonants to surface in the p h o n e t i c representation to their being extrasyllabic. C o m p a r e these data:

[9.49] (a) [pad pres] petit prince 'petty prince' (b) [pati tafa] petit enfant 'little child'

Where the next w o r d begins w i t h a consonant as in petit prince [9.50a], the floating u n d e r l y i n g final / t / of petit remains unattached at the end of the derivation and conse- quently fails to surface. B u t w h e r e the next w o r d begins with a v o w e l as in petit enfant [9.50b], a C - e l e m e n t is inserted at the C V - t i e r and the floating u n d e r l y i n g / t / is attached to it. T h e O n s e t First Principle [9.9] ensures that the inserted C (and the s e g m e n t it dominates) is attached as a syllable onset. T h e / t / is then able to surface phonetically.

[9.50] (a)

A A ^k C V C V C C V C

/ \ I I I I II p 3 t i t p r e s

(b) 0 5. JL c v c v e v e v

/ \ / \ i I I I p a t i t a f a

This analysis implies that French prefers o p e n syllables b o t h in underlying and surface representations.

Syllable-based rules are not e x t r a o r d i n a r y in i g n o r i n g peripheral e l e m e n t s . As w e shall see presently, in m a n y

184 The syllable

languages stress rules also tend to disregard peripheral elements in w o r d s (see section 11.2.3).

9.8 Summary

T o s u m m a r i z e , the syllable has the following functions:

(i) P h o n o t a c t i c regulation: constraining the c o m b i n a t i o n of consonants and v o w e l s of a language.

(ii) Regulation of s u b s e g m e n t a l structure t h r o u g h the C V - tier.

(iii) Serving as the unit of the phonological hierarchy in t e r m s of w h i c h the b e h a v i o u r of h i g h e r units of the prosodic hierarchy such as stress, t o n e , and d u r a t i o n is stated.

Exercises

l . ( a ) M a k e a broad transcription of the data b e l o w . (b) D i v i d e the w o r d s into syllables using the syllabification

c o n v e n t i o n in [9.10]. met fright sphere strict laughed scratched j u x t a p o s e Knesset Gdansk

(c) C o m m e n t on any p r o b l e m a t i c cases. D o e s [9.10] need to be modified to a c c o m m o d a t e them?

2. (a) Study the following data and suggest an informal rule to account for the distribution of lax and tense vowels in English: column A column B column C column D lax bit [bit] get [get] ban [baen] pot [pDt] soot [sut]

tense beat weight barn port suit

tense [bit] bee [weit] w a y [ban] bar [pot] p a w [sut] sue

[bi] [wei] [be]

[pa] [su]

lax sing l o n g b a n g i n g fungus t o n g u e

[sm] [lDQ] [baeriii]] [fAOgas] [tAD]

Hint: T h e l a x v o w e l s a r e / i e ae A D U a / A l l o t h e r vowels are tense. Syllable structure affects the

, distribution o f lax v o w e l s .

3 . ( a ) T h e data b e l o w exemplify the process k n o w n as L I A I S O N in French. U s i n g the approach to the syllable

Summary 185

p r o p o s e d in this chapter, formulate a rule to account for liaison. Specifically, state h o w the segmental tier, the C V - t i e r and the syllable tier are linked,

[le dam] les dames 'the ladies' [le fose] les fosses 'the ditches' [le primat] les primates 'the primates' [le mwa] les mois 'the months' [lez animo] les animaux 'the animals' [lez ebenist] les ebenistes 'the cabinet-makers' [lez idjo] les idiots 'the idiots' [lez ordinatoe:r] les ordinateurs 'the computers' [lez urs] les ours 'the bears' [lez afa] les enfants 'the children'

Notes

1. For a g o o d discussion of the functions of the syllable also see F u d g e (1969).

2. T h e t e r m A O R I S T means simple past tense, e.g. went, or walked as o p p o s e d to had walked or had been walking.

3 . W e shall r e t u r n to this point in the next chapter w h e n w e consider suprasegmental phonological p h e n o m e n a like v o w e l h a r m o n y .

4. T h e account I have presented is n o t universally accepted. It has been argued by H y m a n (1985) that the n o t i o n s of b r a n c h i n g r h y m e s and projection of the r h y m e o n l y offer a partial explanation of the n a t u r e of the syllable. T h i s is because they fail to s h o w that the units w h i c h c o n t r i b u t e to syllable w e i g h t are also the same units w h i c h can carry tone, stress and quantity. T h e r e is n o t h i n g in the projection approach that s h o w s that this is n o t m e r e coincidence. H y m a n argues that a m o r e illuminating account w o u l d be one w h i c h i n c o r - porated the n o t i o n of M O R A (timing unit or w e i g h t unit). N o n b r a n c h i n g r h y m e s contain one m o r a while b r a n c h i n g r h y m e s contain t w o .

CHAPTER 10

Multi-tiered phonology

10.1 Introduction to tone languages

Very many of the world's languages are T O N E LANGUAGES. They have morphemes which are at least in part realised by pitch modulation: pitch differences can be used to make phonemic contrasts. Pitch depends on the rate of vibration of the vocal cords. The more taut the vocal cords are, the faster they vibrate and the higher is the pitch of the perceived sound.

Linguists usually use tone marking diacritics like those in [10. i] to represent pitch in tone languages. The diacritics are to be interpreted as follows:

[10. i] ' = low tone e.g. [awo] 'star' (Igala, Nigeria) ' = high tone e.g. [awo] 'guinea fowl' (Igala,

Nigeria) = mid tone e.g. [awo] 'an increase' (Igala,

Nigeria) = falling tone e.g. [musana] 'sunlight'

(Luganda, Uganda) = rising tone e.g. [okpa] 'length' (Gwari,

Nigeria)

In tone languages pitch can be used to distinguish word meaning or to convey grammatical distinctions. In this respect they differ from STRESS (nontonal) languages like English where pitch does not have those functions. (See section 10.4.i below.)

The Nigerian language Igala is an example of a language which uses pitch differences to contrast word meaning:

Introduction to tone languages 187

[10.2] Igala (Nigeria) W e l m e r s (1973:116) a w o 'guinea fowl' [ ~~] a w o 'a slapv [_ ~] a w o 'an increase' [~ _] a w o 'a combV [_ ~] a w o 'hole (in a tree) [~ _-] a w o 'star' [^_]

In s o m e tone languages tone has a p r e d o m i n a n t l y L E X I C A L F U N C T I O N . It is used almost exclusively to distinguish w o r d m e a n i n g . This is generally the case in oriental tone languages like Chinese. See [10.5] be)low for a w e l l - k n o w n Chinese example of the use of p i t c h to distinguish w o r d m e a n i n g s . H o w e v e r , in s o m ^ other languages the function of tone is primarily G R A M M A T - I C A L . It is used mainly or exclusively for the signalling of grammatical distinctions. This is the case in m a n y African languages. All intermediate possibilities b e t w e e n th^se t w o extremes are also possible.

In [10.3] s o m e data from L o n g u n d a (Nigeria) which illustrate the use of tone to m a k e g r a m m a t i c a l distinctions are presented. In this language, there are three sets of personal p r o n o u n s and in all three sets the difference between first and second person is expressed tonally:

[10.3] L o n g u n d a (John and B o n n i e N e w m a n 1^74:113)

1st person 1st inclusive 2nd person

Neutral sx- na

na

PI. ka ka ka

Future Sg- n

h

in. ka ka ka

Continuous sg- m. nana kaa

kaa nana ka

N o w a t t e m p t [10.4]. W h a t are the functions of tone in Luganda? W h a t is y o u r evidence?

[10.4] L u g a n d a (a) atusomela ' s / h e reads for u s ' (main clause)

atusomela ' s / h e w h o reads for us' (relative clause)

asom a ' s / h e w h o reads' (relative clause) a s o m a ' s / h e reads' (main clause)

(b) kwesela v. refl. ' t o b u b b l e u p in f o i l i n g ' kwesela v. trans, ' t o water cattle' m i i s a n v u 'seven' m u s a n v u ' t w i g '

188 Multi-tiered phonology

I hope in working out the answer to [10.4] you estab- lished the fact that tone need not have an exclusively lexical or grammatical function. Tone has both lexical and gram- matical functions in Luganda. The existence of minimal pairs like those in [10.4b] shows that tone contrasts word meanings while the use of tone to distinguish between main clauses and (subordinate) relative clauses in [10.4a] shows that tone is also used to make grammatical distinctions.

Tone languages can be classified as belonging to one of two categories on the basis of the shape of their 'pitch phonemes': REGISTER T O N E L A N G U A G E S and C O N T O U R T O N E LANGUAGES. In an ideal register tone language the tones have LEVEL high, mid or low pitch. They are almost pure notes. The pitch hardly goes up or down during the production of a particular tone. That is the case in the Igala examples above.

In contrast, in a contour tone language, many tones have fluctuating pitch as can be seen in the following Mandarin Chinese examples:

[10.5] Chinese ma 'mother' (level high tone) [~] ma 'hemp' (high rising tone) [~" ] ma 'horse' (falling rising tone) [v] (also

referred to as dipping rising or low rising)

ma 'scold' (high falling tone) [~]

The distinction between contour tone and register tone languages is not absolute. In reality most systems display some mongrel qualities: a register tone language often has a few contour tones, and vice versa. Above, the Mandarin Chinese word / m a / 'mother' has a level high tone although contour tones are the norm in that language; and the Luganda word / a s o m a / 's/he who reads' ends with a falling contour tone although the norm in this language is level low or high tones.

10.2 The nature of phonological representations

Until the mid-1970s the consensus view among phono-

The nature of phonological representations 189

logists was that phonological representations consist of segmental and suprasegmental representations. The segmental representations were assumed to be made up of consonant and vowel segments, together with empty segments referred to as syllable, morpheme, word and phrase J U N C - TURES or B O U N D A R I E S . Boundaries are included in phonological representations to indicate the domain in which a particular phonological process takes place (see Chapter 12).

Boundaries may have a conditioning or inhibiting effect. Some phonological processes only take when a certain boundary is present, or absent. For instance, in English voiceless stops are aspirated only if two conditions are satisfied: they must be in a stressed syllable and, in addition, they must be immediately preceded by a SYLLABLE B O U N D A R Y . Both conditions are met in [phen] 'pen' and [dithem] 'detain'; a voiceless stop preceded by / s / , e.g. / p / in [spik] 'speak', is unaspirated even when it is followed by a stressed vowel because in that position the stop is not syllable initial.

In SPE it was assumed that super-imposed on the segmental layer were tone and stress, and possibly a few other phenomena such as vowel harmony (i.e. the sharing by vowels within a word of certain phonological features (see [10.6] below)). Both segmental and suprasegmental elements were thought to be arranged in a row one after another. The correctness of the assumption that phono- logical representations consist of linear segmental and supra- segmental levels was taken for granted for a long time. Moreover, the question of how the two levels related to each other was not raised in a serious way.1

In the 1970s a number of studies focused on the relationship between segmental and suprasegmental repre- sentations. The findings arrived at revealed that the assump- tions that the received orthodoxy was based on were questionable. The questioning initially took place in discussions of the representation of tone. A key question that was raised was whether tonal properties such as [high] [low] or [rising] should be regarded as properties of a vowel, much in the same way that features such as [back] or [round] are. Or, rather, were tonal properties to be viewed as distinct from the segmental representation of

190 Multi-tiered phonology

vowels? Should tone be represented using D I A C R I T I C M A R K S like ( / - x ) for high, mid and l o w tone respec- tively, to indicate its peripheral status? T h e answers to these questions have a bearing o n h o w the nature of phonological representations is interpreted. T h e key issue is w h e t h e r SPE (and structuralist p h o n o l o g y before it) is right in assuming that phonological representations are linear, w i t h segments, s o m e of t h e m bearing suprasegmental properties, arranged in a neat sequence.

10.3 The representation of tone

G o l d s m i t h (1976) p r o p o s e d s o m e interesting answers to questions of this kind. H e p r o p o s e d that the division of the speech c o n t i n u u m into ' s e g m e n t s ' m a y proceed in different w a y s in different languages. T h u s , while the parameters of place and m a n n e r of articulation, for instance, are n o r m a l l y treated as segmental properties b e l o n g i n g to individual consonants, there are languages w h e r e these properties m a y extend over several s e g m e n t s . Nasalisation is usually a p r o p e r t y of nasal consonants only, yet in s o m e languages it can be a p r o p e r t y of the syllable or even the w o r d as a w h o l e . T h e rate of vibration of the vocal cords w h i c h determines the pitch of a sound can be a p r o p e r t y of an individual segment, of a syllable or even of an entire w o r d .

A central claim of the n e w t h e o ry inaugurated by G o l d - s m i t h is that in principle the various articulatory parameters, e.g. a s p i r a t i o n , n a s a l i s a t i o n , v o i c i n g , a n d t o n e are A U T O N O M O U S and the articulations that result from t h e m are, in principle, i n d e p e n d e n t . O n e of the main tasks of phonological t h e o r y is to establish the language-specific as well as universal principles w h i c h regulate the linking of these a u t o n o m o u s p a r a m e t e r s . G o l d s m i t h called this m o d e l A U T O S E G M E N T A L P H O N O L O G Y , a n a m e intended to highlight the fact that, in this theory, the potential i n d e - pendence of the various phonological parameters is regarded as crucial, for the reasons outlined b e l o w .

10.3.1 Contour tones

C o n t o u r tones, like rising and falling tones, pose theoretical

The representation of tone 191

difficulties for what was the standard generative approach to phonology until the mid-seventies. This approach incor- porated a principle which Goldsmith (1976) refers to as the ABSOLUTE SLICING HYPOTHESIS. This is the claim that speech can be exhaustively sliced into segments which consist of unordered bundles of features which are linearly ordered. In an SPE inspired model a word like 'mad' would be represented as in [10.6]:

+ cons +nas +lab —cor —cont

m

—cons +nas - l a b —cor +cont

ae

+cons +nas - l a b +cor —cont

d

While the assignment of the various features to discrete segments in [10.6] might look plausible, the same procedure could not be extended to the feature [+voice] in this word since [+voice] is a property of the entire word. Nor could it be extended to the analysis of pitch here because it too could not be vertically sliced and allocated to a single segment, without any leakage into adjacent segments. This evidence undermines the fundamental claim of the 'slicing hypothesis'. As we saw earlier in [4.1], even apparently segmental features can often extend over more than one segment.

In the discussion of complex segments in the last chapter we saw another counter-example to the SPE position of regarding all segments as minimal, unordered bundles of features. Not all segments are unordered feature bundles. Segments can have internal structure involving the linear ordering of some features. That is true of affricates, prenasalised consonants and diphthongs (section 9.4.3).

The same can be demonstrated to be true of contour tones. A contour tone is a combination of two more basic tones. For example, a falling tone is made up of a high tone followed by a low tone, while a rising tone consists of a low tone followed by a high tone.

I shall illustrate this point using data from Gwari

192 Multi-tiered phonology

(Nigeria). The source is Hyman (1973). In Gwari there is a rule which spreads the tone of the first syllable onto the following syllable creating a contour rising tone or falling tone:

[10.7] UR (a) L H

/ o k p a / (b) H L

/ s u k N u /

--» --» --» - • »

PR L LH [okpa] L HL [siiku]

'length'

'bone'

Even where a contour tone cannot, on the face of it, be broken into a sequence of two tones, inspection of its phonological behaviour shows that it acts like two consecu- tive tones. This will be shown by considering the phenom- enon of D O W N D R I F T which is found in many African tone languages. When an underlying high tone is preceded by a low tone it is usually automatically lowered and becomes phonetically indistinguishable from a mid tone. Downdrift is an example of partial assimilation: a high tone is somewhat lowered and becomes more like a low tone which precedes it:

[10.8 UR "H L H"

PR

Thus, in Luganda an underlying high tone preceded by another high tone remains high. However, a high tone preceded by a low tone or a falling tone gets lowered. The later part of a falling tone thus behaves just like a simple low tone.

[10.9] (a) nabadde sigiila 'I was not buying'

[ T (b) twabadde tugula 'we were buying'

(c) kikuukuulu_ 'crowd'

The representation of tone 193

'They will kill themselves'

In a theory which insists on regarding segments as unordered feature bundles the facts which we have surveyed are simply accidental. There is no apparent reason why [+low] and [+falling] tones should both trigger off down- drift. If, however, falling tones are treated in a way anal- ogous to that proposed for complex segments like affricates in section 9.4.3, the reason for the similarity between low tones and falling tones becomes obvious. A falling tone, even when it is borne by a short vowel, is not an unordered bundle of features. It is a sequence of a high tone followed by a low tone and naturally it will cause the same pertur- bations as a simple low tone. In view of this, contour tones should be represented as complex tones which have internal structure:

[10.10] [ +high] [-high]

—cons +syll

The implication of this is that the simple slicing hypothesis should be replaced with a model that recognises that there need not be a one-to-one mapping between elements at various levels. As [10.10] shows, the tonal features [+high] and [—high] are both simultaneously associated with a single vowel.

However, it is not a free-for-all that is being advocated. There are restrictions on the ways in which tones can be associated with segments. The theory incorporates the OBLIGATORY C O N T O U R PRINCIPLE (OCP) which prohibits two identical tones from being adjacent (Leben J973, 1978). Adjacent identical tones are simplified:

[10.11] UR PR L H H -> LH HLL -> HL

A rising tone (LH) is simplified to H before a high tone and a falling tone (HL) is simplified to low before a low tone.

(d) banetta

[":-'

194 Multi-tiered phonology

This can be seen in the Mende examples below which are cited by Leben (1978). The high toned definite singular suffix -1 causes the simplification of the underlying rising tone of mba but has no effect on the falling tone of mbu.

[10.12] UR (a) HLH -*

mbii-i (b) LH H -»

m b a - i

PR HLH m b u i ' o w l ' LH m b e i ' r i c e '

(a)

(b)

H L mbiinga:

LH L m b a - n g a a

HL L m b u - n g a

LHL m b a n g a :

(The change in the vowel is not relevant.)

Likewise, the underlying low tone of the indefinite plural suffix -nga: causes the simplification of a falling tone but has no effect on a preceding rising tone:

[10.13] URPR I L HL L

'owl'

'rice'

It has been proposed by some that the O C P applies not only to tone but also other tiers, including the segmental tier (see McCarthy 1986).

10.3.2 Tone stability

In many languages when an underlying tone-bearing segment (normally a vowel) is either deleted or becomes nonsyllabic and loses its ability to bear tone, the tone still survives and surfaces on an adjacent syllable. Tone shows a kind of STABILITY which cannot be accounted for if it is assumed to be an integral part of the phonological segment on which it appears in the phonetic representation.

Consider the following examples from Margi (Nigeria) and see what happens to tone when a vowel is deleted:

[10.14] Margi (Hoffmann 1963) Indefinite Definite

(a) sal 'man' salari (b) fa 'farm' fari (c) ti 'mourning' tyan

In [10.14a] when the definite suffix -ari, with its HL tones

The representation of tone 195

is added, no phonological process takes place. But in [10.14b] when the same suffix is added, one of the vowels is deleted so that the phonetic representation is firi and not *fidri. Note, however that the underlying LHL tone pattern is preserved although that means the LH being squashed on to a single vowel. Similarly, in [10.14c], when glide forma- tion turns an underlying high vowel into a nonsyllabic glide incapable of bearing tone, the tone does not disappear. It merely gets shunted on to the next syllabic segment. Once again the underlying LHL tone pattern is preserved by having a rising tone on the first available vowel.

Normally, when a phonological rule deletes a conson- ant or vowel it wipes it out in its entirety. The fact that tones can survive either the deletion or loss of syllabicity of the segments that bear them is further evidence of their autonomy.

Note however, that tone stability is not a universal phenomenon. There are many cases where deletion of the tone-bearing element spells doom for the tone which it carries. This is particularly true of languages where only heavy syllables can bear complex contour tones. This is because many such languages have a requirement that there can only be one-to-one association between tones and tone- bearing units. We shall return to this point presently.

10.3.3 Melody levels

A further type of argument for the autosegmental approach is the existence of MELODY LEVELS. There are languages where a given pitch configuration is linked to certain morphemes or words or grammatical constructions regard- less of the number of consonant and vowel segments which they contain. A classic example of this is Mende (Sierra Leone) (Leben 1973, 1978). Most words containing one morpheme in this language have the following tone patterns: H, L, HL, LH and LHL irrespective of the number of syllables they contain:

[10.15] Mende H: L: HL: LH: LHL:

ka kpa mbu mba ba

'war' 'debt' 'owl' 'rice' 'companion'

pcle bile ngila fande nyaha

'house' 'trousers' 'dog' 'cotton' 'woman'

hawama kpakali tela ma ndavula nikili

'waistline' 'tripod chair' junction' 'sling' 'groundnut'

196 Multi-tiered phonology

Leben suggests that if we regard the tone pattern as phon- ologically separate from the consonant and vowel segments in these words, we can capture the fact that a given melody is realised on each class of words regardless of the number of segments present.

To summarise, at the abstract level of lexical represen- tation, tones do have a separate existence. But at some point in a derivation, which differs from one language to the next, tone gets segmentalised so that phonetically it is realised as a feature of a tone-bearing unit in much the same way as segmental properties like [± coronal] or [± back] are.

10.4 The autosegmental model and the representation of tone

In the last section the arguments in favour of the autoseg- mental model were presented. How does autosegmental phonology differ from SPE phonology?

The main concerns of autosegmental phonology differ from those of SPE. In the SPE model (the spirit of which is reflected in the first eight chapters of this book), the main preoccupation was with rules that modify feature specifi- cations in different ways; with the ways in which rules mapping underlying onto surface representations interact and with the degree to which underlying representations may differ from surface representations.

In autosegmental phonology the focus shifts to ways in which phonological rules can change the ORGANIS- A T I O N of phonological representations. In autosegmental phonology phonological representations are no longer seen as simple rows of segments, with all phonological processes taking place at one single level. Rather, they are regarded as complex arrays of (in principle independent) elements arranged on different levels or TIERS.

Various metaphors can be used to express this insight. Phonological representations can be compared to a multi- storied building, with the syllables as the structural pillars and beams which support the weight of different levels of the building. In this building various events can take place at the different levels (e.g. the stress level, the tonal level, etc.) without necessarily having any effect on what goes on

The autosegmental model and the representation of tone 197

at another level. The different levels, though inter- connected, are in principle autonomous.

Alternatively, following Morris Halle, we can use the analogy of a book with many pages to describe the auto- segmental model of the organisation of phonology. Imagine a book with many pages, with each page standing for a different phonological tier. All languages have the conso- nant and vowel segments page (tier). But there is some variation as to what other pages (tiers) are selected by any one language. The possibilities include a page each for stress, tone, nasalisation and vowel harmony. Whatever selection is made, however, the binding or the core that makes the various pages of the book hold together is the syllable. In the last chapter (which was in effect a practical introduction to autosegmental phonology) we saw the central role of the syllable in phonology. We are going to build on that in this chapter.

In this model, establishing the principles that govern the ways in which elements on different tiers are ASSO- CIATED becomes a central task of phonology. ASSOCI- ATION LINES are the apparatus used to LINK elements on different tiers. Phonological rules may not only have the effect of modifying segments, they may also alter the way in which elements on different tiers relate to each other. Phonological theory has to establish universal principles of association and descriptions of particular languages have to show any idiosyncratic aspects of the ways in which associ- ation between different tiers takes place.

The notation used in writing rules is presented below:

[10.16] Autosegmental notation (a) T Vowel linked to a tone

I V

(b) ® Free (floating) tone not linked to a vowel (c) © Free vowel slot (d) T Establish a link between tone and vowel

V (e) T Delink the tone from the vowel

+ V

198 Multi-tiered phonology

As y o u can see in the M a r g i e x a m p l e b e l o w , u n d e r - lying tone association m a y be different from surface associ- ation. Phonological derivations m a y involve deleting or r e - d r a w i n g association lines. Y o u will also observe that b o t h the possibility of a o n e - t o - o n e M A P P I N G of tone o n t o tone-bearing units (vowels in this case) and that of a o n e - t o - m a n y m a p p i n g are b o t h envisaged.

[10.17]

UR Tonal tier L H L

Segmental tier _fa + ari

PR L H L

fa ri

It is n o t expected, given the a u t o n o m y of the different levels, that every tone will be necessarily associated w i t h a t o n e - b e a r i n g unit. T h u s , Leben s h o w s that in M e n d e there exist b o t h tone-bearing suffixes like the definite m a r k e r —i and toneless ones like the post-position -hu 'in'.

E x a m i n e the data in [10.18] and suggest a rule to account for the realisation of tone in these M e n d e suffixes:

[10.18] Citation form k6 mbu mba pcle bele kpakali

(Note:

'war' 'owl' 'rice' 'house' 'trousers'

-hu kohu mbuhii mbahii pekehu belehu

-1

k6i mbu! i mbei pelei belei

'tripod chair' kpakakihu kpakalii_ ([kpakali:]

! indicates a D O W N S T E P P E D HIGH tone. This is ex jlained below •)

In o r d e r t o represent the u n d e r l y i n g l y toneless v o w e l in [10.18], a rule like [10.19] is needed. T h e V slot starts off w i t h o u t being linked to a tone and a rule d r a w s an association line linking it to the tone immediately to its left. B y c o n v e n t i o n such a linking process is s h o w n by using a b r o k e n line.

The autosegmental model and the representation of tone 199

[10.19]

~H

pele + hu [p£lehu]

L

bcle + hu f\\

— • [belehu]

kpakali + hu I • [kpakalihu]

T h e converse of [10.19] is n o t only theoretically p o s s - ible, it is also a practical reality. M a n y languages have tones which are u n d e r l y i n g l y unlinked to any tone-bearing element. Such tones are referred to as F L O A T I N G tones.

T h e w o r d [mbu!i] in [10.18] a b o v e contains an e x a m p l e of a floating t o n e . As y o u can see, the r o o t has an u n d e r - lying falling t o n e . B u t w h e n it is followed b y the h i g h tone of the definite singular suffix, the falling t o n e (HL) is simplified to high (H) (i.e. r l L —» H ) . After the deletion of the l o w part of the falling tone there sh o u l d be t w o high tones next to each o t h e r and they sh o u l d have the same high pitch in the p h o n e t i c representation. T h e surface represen- tation should be * [ m b u i ] ( ). B u t that is n o t the case. This is because before it is deleted by the tone simplification rule, the u n d e r l y i n g l o w tone (the second part of the falling tone) brings a b o u t the l o w e r i n g to m i d of the high t o n e of the suffix. T h e correct surface representation has the high tone of the suffix -1 l o w e r e d so that it is like a mid tone. We assume this derivation:

[10.20] UR / m b i f i /

PR [mbu'i] ( - - )

D O W N S T E P is the n a m e given to this process w h e r e b y a high tone is lowered in the absence o f any preceding l o w t o n e in the phonetic representation. A d o w n - stepped high t o n e is represented by ( o r ! H ) .

It is necessary to distinguish b e t w e e n d o w n d r i f t and d o w n s t e p . D o w n d r i f t is a u t o m a t i c l o w e r i n g induced b y the presence of a l o w tone immediately before a high tone in the phonetic representation. But d o w n s t e p is phonetically

200 Multi-tiered phonology

n o n a u t o m a t i c l o w e r i n g . T h e u n d e r l y i n g l o w tone that causes the l o w e r i n g does n o t occur in the phonetic r e p r e - sentation. T h e nonsurfacing l o w tone w h i c h causes d o w n - step is referred to as a F L O A T I N G tone.

O u r next e x a m p l e of a floating tone comes from M b u i , a language of C a m e r o o n and is b o r r o w e d from H y m a n and Tadadjeu (1976: 61):

[10.21] (a) bakoo 'crabs' b a n d u m ' h u s b a n d s ' (b) bakoo ba sarj 'the crabs of the bird'

b a n d u m !ba sarj 'the h u s b a n d s of the b i r d '

In isolation b o t h 'crabs' and ' h u s b a n d s ' have L - H tone. H o w e v e r , w h e n these w o r d s c o m e t o g e t h e r in the associ- ative construction, w i t h h i g h tone [serj] as the second n o u n , the associative m a r k e r [ba] is realised w i t h its underlying high tone after ' c r a b s ' b u t is d o w n s t e p p e d to H after ' h u s b a n d s ' . Since in M b u i , as in m a n y other languages a d o w n s t e p p e d high ( H ) only arises w h e n a h i g h tone is preceded b y a l o w t o n e , w e can infer that a l t h o u g h n o l o w t o n e is actually phonetically present, there is one at the u n d e r l y i n g level. W e can therefore set u p these t w o different u n d e r l y i n g representations for [10.21]:

[10.22] a. /bak o o + ba '+ sarj/ ' t h e crabs o f the b i r d '

b . / b a n d u m + ba + sarj/ ' t h e h u s b a n d s of the b i r d '

T h e lexical representation o f ' h u s b a n d s ' is given in [10.23] w h e r e the final l o w t o n e of the w o r d is floating (i.e. n o t linked to any tone-bearing s e g m e n t in the u n d e r l y i n g representation):

[10.23] L H ©"

b a n du m

In M b u i w e have recognised a F L O A T I N G L E X I C A L T O N E tone (i.e. a floating t o n e that is part of the u n d e r - lying representation of a n o u n , verb, adjective or adverb). T h i s is a s o m e w h a t rare p h e n o m e n o n . M u c h m o r e c o m m o n are F L O A T I N G G R A M M A T I C A L T O N E S 2 (i.e. floating tones that are part of the representation of g r a m m a t i c a l m o r p h e m e s like tense, n u m b e r , definiteness, possession etc.).

The autosegmental model and the representation of tone 201

Study the following data and work out the evidence in favour of a floating grammatical tone. Having done that show how it could be represented in autosegmental notation.

[10.24] Etsako (Nigeria) (Elimelech 1976: 56) (a) /ame e0a/ —> ameeGa —» [ameGa]

water father 'father's water' (b) /ame oke/ —> ameoke —> [amoke]

water ram (c) /ame bmb/

water child (d) /odzi 5mb/

crab child (e) /bte bmb/

cricket child

—» amebmb -

—* odziomb -

—> bte5mb -

a ram s water •* [amomb]

'a child's water' -» [ojbmo]

'a child's crab' -* [bt5mo]

'a child's cricket'

The associative construction is marked with a floating high tone which occurs between the nouns. Where as in 'a child's crab', the first noun ends in a high tone and the next noun begins with a high tone, the floating high tone cannot be detected. However, where in isolation the two nouns in the construction respectively begin and end in a low tone, the detection of a floating high tone is easy: it is linked to the low tone at the beginning of the second noun. The result of the combination of high and low tone on the same syllable is a falling tone as in [ameGa] 'water of father'.

[10.25]

L L 0-0 L L L L H L L

1 1 am E (

/ a m e eGa/

1 " 1 f N : a a m ©

[ameGa]

i 9a

In autosegmental phonology, given the starting assumption of the autonomy of tone, the existence of such free, unassociated tones is not surprising. But the same facts cause severe embarrassement in a theory of linear phon- ology where tone features are assumed to be an integral part of the segmental representation of tone-bearing units such as vowels because the prediction that tone is always a

202 Multi-tiered phonology

segmental feature is n o t b o r n e o u t by the evidence. T h e existence of floating tones was one of the original a r g u m e n t s used to justify a u t o s e g m e n t a l representations.

T h e E t s a k o data in [10.24] a r e a l s o interesting for a n o t h e r reason. T h e y s h o w that c o n t o u r falling tones arise from u n d e r l y i n g sequences of H L w h i c h h a p p e n to surface o n the same t o n e - b e a r i n g s e g m e n t . We shall n o t labour this point as it has already been m a d e above.

Besides linking tone to tone-bearing units by d r a w i n g n e w association lines, rules m a y also D E L I N K tones. T h a t is to say, the association line between a tone and a t o n e - bearing unit can be severed. T h u s , in Hausa (Nigeria), an u n d e r l y i n g rising L H t o n e is S I M P L I F I E D and becomes a simple high tone w h e n preceded by a high; the L tone of the rise is deleted as y o u can see in the first example, m e a n i n g 'I t o o k (unspecified object)':

[10.26] (a) Hausa (Schuh 1978: 243) / n a a tfaukaa/ [na: tfauka:] 'I t o o k (unspecified object)' / n a a tfaukee t a / [na: cfauke: ta] 'I t o o k it'

(b) / n a a t a m b a y a a / [na: t a m b a y a : ] 'I asked (unspecified object)' / n a a t a m b a y e e t a / [na: t a m b a y e : ta] 'I asked h e r '

T h e delinking o f the association line b e t w e e n the tone and the v o w e l can be formalised as in [10.27] (and as before, the b r o k e n association line indicates tone spreading (see [10.16]) above).

[10.27]

(a)

H

K L j _

1/ cfa

H

K ka a 'I took'

(b)

H L H

na a cfa u X

u

L \ \ \

ke e

L

ta_ 'I took it'

W e have n o w seen the case for a multi-tier approach to phonological organisation. B u t h o w are the tiers related? T h e principles g o v e r n i n g the linking or association of

The autosegmental model and the representation of tone 203

elements o n different tiers have been articulated in w h a t is called the W E L L - F O R M E D N E S S C O N D I T I O N ( W F C ) . This c o n v e n t i o n exists in a n u m b e r of slightly different versions. T h e version b e l o w is based o n G o l d s m i t h (1976):

[10. 28] Well-formedness Condition (WFC) (i) Each v o w e l m u s t be associated with at least

one tone, (ii) Each tone m u s t be associated w i t h at least

one v o w e l , (iii) N o association lines m a y cross.

This c o n v e n t i o n lies at the heart of autosegmental p h o n o l o g y . It has the effect of adding and deleting associ- ation lines as a p p r o p r i a t e at any point d u r i n g a derivation. Its function is n o t to police phonological representations; it is not a filter that only lets t h r o u g h representations w h i c h meet certain pre-set criteria; rather, it is to be seen as a state- ment of the u n m a r k e d , neutral, n o r m a l state of affairs. Aberrations from the u n m a r k e d situation are sorted o u t by tone rules.

T h e discussion a b o v e has implicitly used the W F C . In our earlier M e n d e examples in [10.19], w h i c h is repeated below as [10.29] for convenience, tone spreads from the first syllable of the r o o t to the toneless v o w e l s that follow it so that n o v o w e l surfaces w i t h o u t tone:

[10.29]

H

pe le + hu -*• [pelehu] be le + hu —- [belehu]

kpakali + hu [kpakalihu]

Likewise, the floating (unassociated) t o n e of Etsako in [10.25] (repeated b e l o w as [10.30]) becomes associated by the W F C :

[10.30] "L L (H) L L

am E e 9a

L L H L L

I I N I _a m[cj e 0a

204 Multi-tiered phonology

T h e W F C is assumed to be a universal constraint o n the s t r u c t u r e of phonological representations.

U s e the W F C to p e r f o r m tone m a p p i n g in the following e x a m p l e from an i m a g i n a r y language:

[ 1 0 . 3 1 ]

H L

ta li la lu ku ma zi sa

W h e n d o i n g the tone m a p p i n g w e m u s t r e m e m b e r that association lines are n o t allowed to cross. O b v i o u s l y , in this case that means that the high tone w h i c h is linked to -la- in the u n d e r l y i n g representation has to be m a p p e d o n t o the toneless first t w o syllables of the w o r d and the l o w tone linked to -ma- has to be m a p p e d o n t o the last t w o syllables:

[10.32]

H L

* < ' 1 \ ^» _ta li la lu ku ma zi sa_

T h e p r o b l e m is d e t e r m i n i n g h o w t o n e m a p p i n g should proceed in the case of the syllables / l u - k u / w h i c h are flanked by high and l o w tone-bearing v o w e l s . T h e y could b e b o t h high, w i t h t o n e spreading from left to right; they could be b o t h l o w , w i t h tone spreading from right to left; or could acquire the adjacent l o w and h i g h tones respec- tively. As it stands, the statement of the W F C in [10.28] is n o t explicit e n o u g h to s h o w w h i c h of these three alterna- tives should be chosen. W h a t is needed is a m o r e explicit s t a t e m e n t of t h e m a p p i n g principles.

In order to cope w i t h situations like this w h e r e the r e q u i r e m e n t s of the W F C can in principle be m e t by i m p l e - m e n t i n g a variety of p r o c e d u r e s , the following s u p p l e m e n - tary principles have been a p p e n d e d to the W F C (Clements and Ford 1979; C l e m e n t s and G o l d s m i t h 1984):

Tone and intonation 205

[10.33] (i) Associate free tones to free tone-bearing units going from left to right,

(ii) The association of free (unassociated) segments takes precedence over that of already linked (associated) segments; furthermore, (a) give precedence to segments linked to unaccented elements, if there are any; (b) give precedence to segments on the left,

(iii) Add the minimal number of association lines required to undo the violation of the WFC.

Once we adopt [10.33], there is no more ambiguity as to which tones should be mapped onto -luku- in [10.32]. The second principle of [10.33] states that the association goes from left to right. So the high tone is mapped on to luku- :

[10.34] H L

U li la lu ku ma zi s

In most cases, applying these principles in this order elim- inates any uncertainty about the application of the WFC.

In the remainder of this chapter we shall see how the insights into the nature of phonological representations gained from the autosegmental study of tone have been extended to other aspects of phonology.

10.5 Tone and intonation

Sometimes T O N E LANGUAGES are contrasted with I N T O N A T I O N LANGUAGES. But they should not be. Tone languages also have intonation. One common inton- ation effect found in many register tone languages is T O N E TERRACING. Terracing involves several related phenomena: DOWNDRIFT, D O W N S T E P and UPSTEP.

As we saw in (10.3), downdrift refers to the automatic lowering of pitch in a P H O N O L O G I C A L PHRASE when-

206 Multi-tiered phonology

ever a high tone is preceded by a l o w tone. A phonological phrase is an utterance, often c o t e r m i n o u s w i t h the sentence at the syntactic level, w h i c h can be said in one breath and can naturally be followed by a pause. D o w n d r i f t creates a falling intonation c o n t o u r in a phonological phrase. Because every high t o n e is l o w e r than the preceding high each time there is an intervening l o w , a phonologically high tone can be phonetically l o w e r in pitch than a l o w tone appearing before it in a phonological phrase. T h u s d o w n d r i f t creates a 'terracing' effect:

[ i o . 3 5 ] H

H H

JJ" H J J " U J ^ H

Lkf

D o w n d r i f t only obtains w i t h i n a phonological phrase. Pitch is reset at the b e g i n n i n g of each phonological phrase. A Luganda e x a m p l e o f d o w n d r i f t is given in [10.36].

[10.36]

tetulabye muntu alima mu mmwanyi

J l_r 'we have not seen the/a person cultivating in the coffee plantation'

Recall that in our discussion of floating tones w e noted that that the term D O W N S T E P is used to describe situ- ations w h e r e tone l o w e r i n g occurs which is n o t a u t o m a t i - cally triggered off by the presence of a l o w tone immediately before a high tone. S o m e t i m e s , for n o apparent surface phonetic reason, a high tone m a y be lowered (!H) w h e n it is preceded b y another high tone as in Akan (Ghana) (Schachter & F r o m k i n 1968):

[10.37] / n i l 0 -b'uj ( m r b u ] ' m y s t o n e '

T h e n o u n / 5 - b u / contains the v o w e l / o / which bears a low tone prefix. T h e v o w e l is deleted in the possessive c o n s t r u c - tion following the possessive m o r p h e m e / m i / ; but the tone

Tone and intonation 207

is not. Rather it is left floating. The presence of this floating low tone is responsible for the tonal downstep. While the phonetic motivation of downdrift is present on the surface, that of downstep is not. Yet from the point of view of intonation, downstep is similar to downdrift in that it causes the pitch of the utterance to go down.

UPSTEP is the converse of downstep. It is the raising of the pitch of a tone so that it is phonetically a step higher in pitch than the preceding token of the same tone. Upstep produces rising intonation over part of the utterance.

Cope (1970) reports that in Zulu (South Africa), low tones between highs become high and that in a series of successive high tones, each high tone is higher than its predecessor:

[1O.38J 'the boys who do not want it'

[- -J /abafana abayifuni/ —•- [abafana ibayifuni]

While in Zulu a high tone is raised if preceded by another high tone, creating a crescendo effect, in Engenni (Nigeria) (Thomas 1974), there is automatic upstepping of a high when it is followed by a low in the same phonol- ogical phrase:

[ 1 0 . 3 9 ] 'I saw you' 'I did see you'

L->—: m 1 mom wo

1_ WD bh e

Sometimes intonation contours are associated with particular sentence types. In many languages falling inton- ation (due to downdrift) is associated with declarative sentences while rising intonation resulting from a mitigation or total undoing of the effects of downdrift is associated with interrogative sentences.

Thus in Hausa in questions downdrift is suspended towards the end of the phonological phrase. The last high tone is upstepped and acquires extra-high pitch with a sharp fall (Hombert 1974 cited in Schuh 1978):

208 Multi-tiered phonology

Balaa da Sheehu zaa su zoo da mutaanensu.

'Bala and Shehu will come with their people?'

[ 1 0 . 4 1 ] . - ^

Balaa da Sheehu zaa su zoo da mutaanensu.

'Will Bala and Shehu come with their people?'

10.6 Pitch-accent

The impression might have been given that all languages are either tone languages or stress languages. The reality is somewhat more complex. Some languages are not clearly one thing or the other. A classic example of a language that does not fit neatly into either category is Japanese.

In Japanese pitch is used contrastively as in a tone language. But it is subject to the constraint that there can only be one pitch drop per word. Pitch can go up or down only once in a word. An illustration of this is given in [10.42] below. You will observe that there are only four basic melodies: L, HL, LH and LHL:

[10.42] Standard Japanese (McCawley 1978) (a) L (i.e. all syllables are low)

na 'name' kaki 'persimmon' sakana 'fish'

(b) HL (i.e. the first syllable is high and subse- quent syllables, if any, are low)

na 'vegetable' na ga 'vegetable' (subject) makura ga 'pillow'

Pitch-accent 209

(c) LH (i.e. the pitch starts low and goes up)

kaki 'fence' atama 'head'

(d) LHL (i.e. the pitch starts low, goes up, and goes down again)

"JL kokoro 'heart' atama ga 'head' kaki g a 'fence'

(Note: the particle ga which follows the noun here is the subject marker. It is always low.)

As you can see, all the syllables of the nouns in (a) are assigned low tone; there is no pitch drop in the word. In the rest of the examples, however there is a pitch change. An abstract A C C E N T , which is represented by an asterisk or star (*), marks the syllable where the change in pitch occurs:

[10.43] L (i.e. no asterisk and hence no pitch change)

CV

L na

CVCV

V cvcvcv V-" L

sakana kaki 'persimmon' sakana 'fish'

The well-formedness condition ensures that low tone spreads to all the syllables of this class of word.

[10.44] H or HL *

CV

L H J na

CV

. H na

CV 1 I I I

L_ ga

CVCVCV I /

CV I 1 '*'

makura ga

A HL melody is assigned, with H going to the first syllable and L to the remainder, if any.

210 Multi-tiered phonology

[10.45] L H

vcv 1 1 1 1 1 1 L H_ kaki 'fence'

* vcvcv \ 1 1 \J 1

L H_ atama head'

Conversely, a LH melody is assigned, with H on the last syllable and L on the rest.

[10.46] L H L

CVCVCV CV \ 1 ' '

LHL kokoro ga 'heart'

VCVCV V 1 1

\ ' 1

L H

CV

L_

CVCV CV 1 1

1 1 1 L H L _

atama ga 'head' kaki ga 'fence'

A specified starred high toned syllable is flanked by low toned syllables. Low is the DEFAULT or neutral tone assigned when there is no instruction (in the form of an asterisk) to assign a high tone.

Japanese is called a PITCH A C C E N T language. It shares some of the characteristics of both STRESS languages like English (which are the subject of the next chapter) and those of tone languages like Chinese.

On the one hand, in Japanese pitch functions in much the same way as stress does in English. Words in a stress language are not allowed to have more than one peak of prominence (in English it is not possible to place two main stresses on a word like * disinformation). Similarly, in a pitch accent language there can be at most one peak of promi- nence in a word. A Japanese word can have only one of the following patterns: L, H, HL, LH or LHL [see [10.42]. High tones cannot alternate with low tones.

On the other hand, however, Japanese is like a true tone language in that pitch contrasts are used phonemically. In this it differs from a stress language like English which does not have minimal pairs distinguished by pitch differ- ences alone.

Vowel harmony 211

Study [10.42] and identify words distinguished by pitch differences.

The following are minimal pairs distinguished by pitch:

na (L) 'name' na (H) 'vegetable' kaki (L) 'persimmon' kaki (LH) 'fence'

10.7 Vowel harmony

Vowel harmony is a process whereby within a certain designated domain, usually the word, all vowels are required to share one or more phonological properties. The vowels of a language are divided into two mutually exclusive sets and all vowels within a stipulated domain must be, say, either front or back, high or low, rounded or unrounded, etc. Thus, in classical Mongolian all the vowels in a word are either front as in [kobegyn] 'son, boy' and [kotelbyri] 'instruction' or alternatively, they are all back as in [uyuta] 'bag' (Lightner 1965).

Welmers (1973:34) reports that in Igbo (Nigeria), vowels belong to either set A, which is [ + ATR] or set B, which is [-ATR]. Normally, all the vowels used in a word come from just one of these two sets:

[10.47] Set^A: [i e u o] Set B: [1 a u 0] 6 riri 'he ate' 6 piri 'he carved' 6 mere 'he did' 5 sara 'he washed' 6 gburu 'he killed' 6 zuru 'he bought' 6 zoro 'he did' 5 doro 'he pulled'

Vowel harmony differs in a significant way from arche- typical suprasegmental properties like tone and stress. While these are always located on a separate tier from that where vowels and consonants are found, harmonising phonolog- ical features like [back], [round], [high] and [ATR] are normally part of the segmental representation of individual vowels. But they are extracted from the segmental tier and are placed on a distinct harmony tier and cease functioning

212 Multi-tiered phonology

as properties of individual s e g m e n t s . B u t w h e n v o w e l h a r m o n y occurs, they are extracted from the segmental tier and cease functioning as properties of individual segments: they S P R E A D to all v o w e l s w i t h i n a specified d o m a i n . This d o m a i n is usually the w o r d .

V o w e l h a r m o n y constitutes a theoretically fascinating p h e n o m e n o n w h i c h can t h r o w light o n the n a t u r e of p h o n - ological representations because of the w a y in w h i c h it func- tions partly as a segmental and partly as a suprasegmental p r o p e r t y .

V o w e l h a r m o n y can be described in the f r a m e w o r k of autosegmental p h o n o l o g y by using these principles:

[10.48] (i) identify the set of h a r m o n i s i n g features w h i c h are suprasegmentalised and placed o n a separate tier;

(ii) identify the class of elements (vowels) w h i c h bear the h a r m o n i s i n g feature;

(iii) identify t h e set (possibly e m p t y ) of O P A Q U E S E G M E N T S . O p a q u e segments are v o w e l s w h i c h o u g h t to o b e y the v o w e l h a r m o n y rules but fail to do so because they are specified in the lexicon for the h a r m o n i s i n g feature and are therefore e x e m p t from v o w e l h a r m o n y rules w h i c h fill in blanks for the h a r m o n i s i n g feature d u r i n g a derivation;

(iv) mutatis mutandis, h a r m o n i s i n g features are associated w i t h vowels in accordance w i t h the r e q u i r e m e n t s of the W F C [10.28,10.33]

In [10.49] these principles are used in the m a p p i n g of the feature [ A T R ] in the derivation of the Igbo forms in [10.47]:

[10.49]

(a) Vowel harmony tier [ + ATR]

Segmental tier O zOrO —• [o zoro] 'he did'

(b) Vowel harmony tier [ — ATR]

Segmental tier O d O r O -*• [odoro] 'he pulled'

Vowel harmony 213

Note: Capital letters are used in underlying representations for vowels specified for all features except the harmonising feature, in this case, [ATR].

Attempt writing an autosegmental rule for the Mongo- lian example [kobegyn] 'boy, son' which we referred to earlier.

For Mongolian, the feature back is the one that is placed on the vowel harmony tier. A rule like [10.50] is needed:

[ 1 0 . 5 0 ] Vowel harmony tier:

Segmental tier:

[-back] "

/ 1 \ k O bEgUn_ —•- kobegyn

N o w write a rule to account for the Akan vowel harmony data in [10.51] which are borrowed from Clem- ents (1985):

[10.51] (a) All vowels are [+ ATR] [o-fiti-i] 'he pierced (it)'

(b) All vowels are [ - ATR] [0 - a r e -1] 'he showed (it)'

Your rule should be like [10.52]:

[ 1 0 . 5 2 ]

(a) r H + A T R ] >fr 1 ! * •*

/ 1 » V |_0-fl t l - I — [ofitii]

(b) R - A T R ] "

/ I \ x v O-cI rE - I — [ocirei]

Unfortunately for the phonologist, all Akan forms are not so well behaved. There are words which have a mixture of vowels from the two harmonising sets. That is the case in [o-jiinsejii-i] 'she became pregnant'. This is an instance of O P A Q U E ASSOCIATION, i.e. some of the vowels are linked to harmonising features in the underlying represen- tation. The infinitive form of this verb is:

214 Multi-tiered phonology

[io.53] [+ATR] -ATR]~

'to become pregnant'

T h e derivation of [o-jiinseni-i] proceeds as in [10.54]:

[+ATR] [-ATR] *

/ s e nl - I_

—

[ATR]

A _ o-Jli n

[-ATR]

A\ S EJl I - I P-Jl

V o w e l h a r m o n y is o f theoretical interest because of w h a t it reveals of the 'slicing' p h e n o m e n o n . A phonological p a r a m e t e r m a y be selected for different degrees of supra- segmentalisation: it m a y be p u t o n an entirely separate tier, like stress or t o n e in m a n y languages; or it m a y be only partially suprasegmentalised, like v o w e l h a r m o n y in those languages w h e r e certain features are s o m e t i m e s p u t o n a separate tier and s o m e t i m e s treated as an integral part of v o w e l p h o n e m e s .

10.8 Nasalisation

Like v o w e l h a r m o n y , N A S A L I T Y , is a n o t h e r phonological p a r a m e t e r that can be suprasegmentalised. N o r m a l l y , the specification [+nasal] is an integral part of nasal consonants (and possibly nasal v o w e l s ) . H o w e v e r , in s o m e languages this feature is extracted from the segmental tier and placed o n the suprasegmental tier so that it characterises several syllables or m o r p h e m e s o r even entire w o r d s .

D e s a n o , a C o l o m b i a n language of the A m a z o n basin, is a standard e x a m p l e of a language w h i c h treats nasalisation as a suprasegmental element. N a t i v e D e s a n o w o r d s consist of m o r p h e m e s w h i c h are either all oral or all nasal:

[10.55] D e s a n o (Kaye 1971) nasal [wai] ' n a m e ' [nohso] 'kind of b i r d '

[senanu] [mininul]

'pineapple' 'a small r o u n d t h i n g '

oral [wai] 'fish' [johso] ' k i n d o f lizard' [goru] 'ball' [wyariru] 'a large r o u n d t h i n g '

Nasalisation 215

T h e voiceless s e g m e n t s / p t k s h / are N E U T R A L . T h e y are neither capable of bearing the p r o p e r t y of nasalisation, nor of inhibiting its spread. T h e y are are 'invisible' to the nasalisation rule. But voiced s e g m e n t s are all capable of bearing nasalisation: they have b o t h oral and nasalised versions:

[10.56] Oral Nasal

V

b

d/r

g

J w

V

m

n

0

Ji

w

In D e s a n o , nasalisation is n o t s i m p l y a p r o p e r t y of certain consonants. It is extracted from the segmental tier and placed o n a separate suprasegmental tier. T o represent the D e s a n o situation rules like [10.57] a r e needed:

[10.57] [ +nasal]

/ 1 N

/ ! \ _ w a i _ -*• [wai]

[ — nasal]

I 1 \ / 1 \ / > \

_ w a i __

Where there are neutral voiceless consonants, the m a p p i n g proceeds in the s a m e w a y , only this time being blind to the presence of voiceless c o n s o n a n t s :

[10.58] [ +nasal]

/! \ johso —"-jiohso

In the discussion of b o t h v o w e l h a r m o n y and nasalis- ation I have used the concept of S P R E A D I N G . Originally spreading was seen as a unique p r o p e r t y of tone. In recent years, h o w e v e r , it has been suggested by a n u m b e r of linguists (Steriade 1982; Hayes 1986; Archangcli and Pulley- blank 1986; H y m a n and Pulleyblank 1987) that all assimi- lation should be treated as spreading. T h a t includes

216 Multi-tiered phonology

processes like voice assimilation, palatalisation and place of articulation assimilation which were dealt with by feature copying rules in the past. This is in conformity with the view of speech outlined in Chapter 4.

10.9 Morphemic tier

Students of Semitic morphology have extended the theory of autosegmental phonology in other directions. McCarthy (1979, 1981) has proposed the setting up of a morphemic tier for languages like Arabic and Hebrew. Usually in a Semitic language words are formed by adding vowels to a SKELETON root which consists of three consonants.

Thus, in Egyptian Arabic the root meaning 'write' is ktb, and the root meaning 'understand' is Jhm. Various word-forms are derived by adding vowels (and in some cases consonants) to these skeletal consonantal roots:

[10.59] (a) katab 'he wrote' fihim 'he understood'

(b) katbu 'they wrote' fihmu 'they understood'

(c) kattib 'he caused fahhim 'caused to to write' understand'

(i.e. he explained)

McCarthy's proposal is that the verb is made up of co- existent representations on three separate tiers. There is a CV-SKELETON which is the consonantal root. T w o other tiers are linked to this: a consonantal tier and a vowel tier containing the vowel pattern of the word which is mapped in a manner reminiscent of tone melodies in Mende.

[10.60]

Cons. f h m f h m f h

tier \ \

m

Skeletal C V C V C C VC C V C V C C V C tier V I I I I

i i u a i Vowel tier fihim fihmu fahhim

Morphemic tier 217

N o w provide a derivation similar to [10.60] for katab, katbu and kattib.

Your solution should be essentially the same as that given in [10.52] except for the root consonants which in this case are ktb instead offlim.

We have seen that Arabic regular verbs have a tricon- sonantal root to which various vowel melodies are added, depending on the grammatical form required. Interestingly, wherever there are more slots for C or V segments on the skeletal tier than there are vowels or consonants on the vowel or consonant tier, a phenomenon akin to tone spreading takes place. A single vowel or consonant is linked to two C or V slots as is required by the WFC. Such a segment is realised as geminate in the phonetic representa- tion (section 9.4.3).

Thus notions of nonlinear phonology, originally proposed in order to deal with tone, turn out to make correct predictions for totally unrelated phenomena like vowel harmony, nasalisation and Semitic concatenative (infixing) morphology. This has encouraged linguists to believe that significant principles of linguistic organisation have been unearthed.

In the next chapter we shall consider another non-linear approach to phonology which has been very successful in the analysis of stress and is inspired by principles similar to those of autosegmental phonology.

Exercises

1. Bekwarra (Nigeria) (SIL Introduction to Phonemic Analysis, 1980 page 159) Study the data below where high, mid and low tones are respectively marked by the diacritics ' " and \ pkii 'match' api 'cow' liko 'parrot' ike 'baboon' ige 'water yam' ije 'mother' ubu 'goat' upu 'vulture' awu 'measuring bowl' ide 'father' ine 'gun' ala 'bird' usi 'snake' abi 'charcoal' ubi 'gun powder'

(a) What is the function of tone in Bekwarra?

218 Multi-tiered phonology

(b) Is Bekwarra a register tone or a contour tone language? What is your evidence?

2. Kombe (based on Elimelech 1976) In Kombe, a Bantu language spoken in Rio Muni (Equitorial Guinea) there are only two underlying phonemic tone oppositions, namely high and low. But five phonetic tones occur phonetically. Thus nouns in isolation exhibit the following five tones in the phonetic representation:

H (high) [V] [-] L (low) [V] [ J L° (unreleased low) [V°] [-] F (falling) [y] p ] D downstep high [V] [—]

In the examples in (a) below the tones L, L°, D and F are shown on the final syllable of nouns:

(a) ta6a [ J 'goat'

'tree'

'sand'

'chicken'

'hand'

It is interesting to note that nouns in isolation never end on a high tone,

(b) Kombe shows an interesting distinction between nouns occurring in isolation or prepause position and those nouns which occur within a phrase.

ta6a L L ele° L L° cele H D ku6a F L ebo L F

[--\

[—)

H

N

M

isolation

ta6a

ele°

cele

'goat'

'tree'

'sand'

inside (e-g- ta6a goat ele tree cele sand

a phrase identification)

ndira is that ndira is that ndira is that

'that's a goa

'that's a tree

'that's sand'

Morphemic tier 219

ku6a 'chicken' ku6a ndira 'that's a chicken' chicken is that

ibumu 'belly' ibumu ndijira 'that's a belly' belly is A that (where A = agreement)

8iji5ni° 'bird' Bijioni ndiBira 'that's a bird' bird is A that

ikayi 'leaf ikayi ndijira 'that's a leaf leaf is A that

ilali 'stone' ilali ndijira 'that's a stone' stone is A that

ilo 'ear' ilo ndijira 'that's an ear'

isolation in phrase Jinal position ku6a 'chicken' a toyendi kuba 'he saw a chicken' ilali 'stone' a toyendi ilali 'he saw a stone' ele° 'tree' a toyend_i ele° v 'he saw a tree' Bijidni 'bird' a toyendi Bijioni0 'he saw a bird' cele 'sand' a toyendi cile T 'he saw sand' ikayi 'leaf a toyendi ikayi 'he saw a leaf

Attempt the following problems:

(i) List all the tonal alternations in nouns observed in comparing the left and right columns,

(ii) Suggest rules to predict the occurrence of: (a) F i.e. falling tones (A) (b) L° i.e. unreleased low (L°) (c) D i.e. downstep high tones (')

Notes

Firthian prosodic analysis rejects the structuralist phonemicist's position presented here. Prosodic analysis assumes a more complex view of phonological struc- ture. Phonological representations are said to consist of P H O N E M A T I C U N I T S (roughly equivalent to consonants and vowels) and PROSODIES like tone, stress, vowel harmony, voicing and nasalisation (Palmer 1970).

Where a floating tone does surface in some realisations of a morpheme (as is the case in the Mende example

220 Multi-tiered phonology

in [10.19]), the idea of floating tones is not very contro- versial. However, there is considerable disagreement as to whether it is justifiable to posit an abstract floating tone (which never surfaces anywhere phonetically) the existence of which can only be detected from the tonal perturbations which it causes.

The analysis of the Etsako associative marker as a floating high tone is an example of an abstract floating tone. The validity of this analysis is disputed by those who object to underlying representations which are distant from phonetic representations. The essence of the criticism is similar to that made against abstract segments at the level of segmental phonology (see section 8.2). Critics of very abstract underlying repre- sentations would prefer to interpret problematic tonal alternations like that in Etsako in terms of lexical or grammatical conditioning. They would argue, for instance, that instead of a rule like [10.25] the grammar of Etsako should contain a statement to the effect that the associative marker conditions the change of a low tone into a falling tone on the first syllable of the second noun in an associative construction if that noun begins with a low tone.

CHAPTER 11

Stress and intonation

11.1 Introduction: stress

This chapter is in two sections. The first section introduces you to the description of STRESS using a non-linear approach known as METRICAL P H O N O L O G Y . The second part is devoted to intonation.

11.1.1 What is stress?

Linguists have an intuitive understanding of the phonetic properties of stress although they find formulating a precise description of these properties problematic. Stress is primarily a matter of greater A U D I T O R Y P R O M I - N E N C E . It is essentially a perceptual phenomenon, with ill-defined articulatory correlates. An element that is stressed is highlighted so that it becomes auditorily more salient than the rest of the elements in the string of which it is a part. The main phonetic ingredients of stress are PITCH, LENGTH and L O U D N E S S . Stressed syllables tend to have higher pitch and longer duration than their non-stressed counterparts. In addition, they may be somewhat louder than unstressed syllables; but loudness is a much less important parameter than pitch or length. There may be also increased respiratory energy in the production of a stressed syllable although this is by no means essential. Additionally, in some languages, vowels in stressed syl- lables have clear or full vowel quality while vowels in unstressed syllables are reduced and have a somewhat 'muffled' quality like that of schwa (/a/) in English. The phonetic properties of stress will not be explored in this book. Rather it is the phonological properties that will be investigated.

222 Stress and intonation

From a phonological angle, several different kinds of stress can be recognised. One kind is W O R D STRESS. In English, for instance, every lexical item is entered in the dictionary with word stress. A particular syllable of a word is pronounced in a way that makes it more prominent than the rest. If you say words like mother, better, cotton and pity, you will notice that the first syllable in all these words is much more salient than the second. The first syllable is said to be stressed and the second one unstressed.

In longer words, there is often not just one stressed syllable and a host of unstressed ones. Besides the syllable that receives the main or primary stress, there are other syllables which receive secondary stress. Such syllables are more prominent than the weakest syllables of the word.

If you say the word radiator very carefully, you will notice that the first syllable receives primary stress and that the third syllable has secondary stress; it is more prominent than the second and fourth, though not as prominent as the first. Trager and Smith (1951) introduced the convention of recognising four levels of stress in English, marked as ' * ' " to indicate decreasing order of prominence.

The authors of SPE continue this tradition. Their view is that at least five degrees of stress can be easily detected in English. They use the integers 1-4 to mark stress, with 1 as the strongest and 4 the weakest stress. They suggest that unstressed syllables i.e. [-stress] are 'representable as [5 stress] in this case' (SPE, page 116). In practice, for convenience, unstressed syllables are not assigned any integer, as you can see from the examples below which are taken from SPE:

[ I I . I ] I 3 absolute

1 3 kaleidoscope

3 1 intercept

2 1 mentality

2 3 1 relaxation

1 3 survey (noun)

1 3 advocate

3 1 interlock

2 1 gestation

2 3 1 d e p o r t a t i o n

Introduction: stress 223

3 4 1 3 4 i instrumentality complementarity

3 4 1 3 4 1 documentation experimentation

Mark the most prominent syllable in the following words:

[11.2a] market water button little analyse equality antagonise indemnify

The task set above involves assigning the appropriate integer to each stressed syllable (and nothing to any unstressed one). You were expected to work out an answer along these lines:

[11.2b] 1 1 1 1 market water chemistry little 1 3 3 1 1 3 1 3 analyse equality antagonise indemnify

The analysis of English stress presented in [11.2b] might give the impression that stress is a matter of absolute prominence attached to each syllable and that it can be indicated by unambiguously placing the appropriate diacritic on the syllable in question. Disputes, to a large extent sterile, have arisen in the past as to the precise number of degrees of stress which a language like English has. Are all the weak syllables left unmarked (which could be assigned degree 5 of stress) equal in prominence? For instance, is the final syllable of chemistry not somewhat more salient than the middle one? And, furthermore, is the final syllable of chemistry not stronger than the last syllable of

3 1 1 3 water? And are the the vowels of equality and indemnify which are given degree 3 of stress equal in prominence? Reaching agreement on the exact number of degrees of stress has turned out to be almost impossible. The problem is that the syllable that is most heavily stressed in a word (and is therefore assigned degree 1 of stress) is easy to ident- ify. But there is often some uncertainty as to the degree of stress of the less heavily stressed syllables. It is difficult

224 Stress and intonation

to determine whether or not a given non-primary stress is equivalent to some other non-primary stress in prominence.

Fortunately, research has shifted in a more fruitful direction in recent years. Following Liberman and Prince (1977) many linguists today recognise that a theory of stress must take on board the fact that prominence is a RELATIONAL concept. What matters is that a stressed syllable is more salient than its unstressed counterparts.

Examine some familiar noun-verb pairs in [11.3] which are distinguished by stress.

[11.3] verb Noun project project reject reject conduct conduct protest protest refuse refuse

Add three examples of your own to the list in [11.3]. Underline the syllable that receives primary stress in each example in [11.3].

One of the things which these data show is that stress is not an integral part of a vowel. In many disyllabic words in English the location of stress depends on whether the word appears as a noun or as a verb: stress falls on the first syllable when the word is used as a noun and on the second when it is used as a verb. This shows in a simple way that stress is an autosegmental property of the entire word.

The words in [11.3] are in no way extraordinary. There are numerous examples of English words in which stress shows a considerable degree of mobility. Say the following, noting the syllable with the main or primary stress:

[11.4] (a) nation national nationalist nationalise (b) nationality nationalistic nationalisation

Underline the syllable which receives the main stress on each occasion.

Introduction: stress 225

This should be enough to convince you that stress is not an inherent property of any vowel. In [11.4a] the first syllable receives the main stress regardless of the presence or absence of suffixes. But when certain suffixes are present, as in [11.4b] stress shifts from the first syllable to the syllable immediately preceding the suffix. Stress is not an inherent vowel feature. It is an autosegmental property of the word. Its location in the phonetic representation of a word may depend on the presence of certain affixes or grammatical information such as whether the word is real- ised as a noun or as a verb.

11.1.2 Metrical phonology

METRICAL P H O N O L O G Y is an approach developed within the generative phonology framework in recent years to handle stress phenomena. It complements A U T O S E G - M E N T A L P H O N O L O G Y which, as we saw in the last chapter, was primarily designed for the description of tone although it was used later to account for other aspects of phonology.

We have seen that stress is a relational concept: a stressed syllable is more prominent than an unstressed one. This fact is regarded as crucial in metrical phonology. Rela- tive prominence is expressed using BINARY BRANCHING TREES, which are labelled S T R O N G (s) and WEAK (w). The more prominent syllable is dominated by s (i.e. it lies below s on the tree) and the less prominent one is domi- nated by w.

[ I I - 5 ]

s w w s sen der pre tend

/ \ / \ s w w s milk man high priest

Strong and weak syllables are paired together by a procedure called F O O T F O R M A T I O N . What is proposed

226 Stress and intonation

is that stress is assigned to strings which have constituent structure consisting of two elements, one of which is strong (i.e. dominant) and the other weak (i.e. subordinate). The dominant one is the HEAD. The head governs its immediate neighbour to the left or right. In [11.5] a number of metrical feet are exemplified, with the head coming first or second.

The structure of metrical feet plays an important role in English poetry. Find examples of poems where right- headed and left-headed metrical feet are used. What are the literary terms for these feet?

This short passage from Shakespeare illustrates right- headed metrical feet:

He can | not live, | I hope, [ and must | not die Till George | be pack'd | with post-| horse up | to heaven. I'll, in, I to urge | his hat [ r£d more | to Clarence With lies | well steel'd | with weigh | ty ar | guments . . .

(William Shakespeare Richard HI, I.I. 145-8) (Note: V indicates a weak syllable and V a strong syllable.)

The above lines of blank (i.e. non-rhyming) verse typically contain five feet. Each foot in turn tends to contain an unstressed (weak) syllable which is followed by a stressed (strong) one. As you will recall from the discussion of Latin metre in (9.5) this pattern is called IAMBIC METRE. (To be precise, this extract is written in IAMBIC P E N T A - METRE i.e. there are five feet in each line.)

A variety of left-headed feet were introduced in our brief account of Latin metre in (9.5). Examine an English example of left-headed metrical feet where a stressed syllable is followed by an unstressed one. As in Latin verse, this verse pattern is called T R O C H A I C METRE:

How the" I Chimney | sweeper's | cry Every | blackn'ing | Church ap | pals; And the | hapless | soldier's | sigh Runs in | blood down | Palace f walls.

(From William Blake London, Songs of experience)

Introduction: stress 227

The account of relative prominence given in [i 1.5] only deals with disyllabic words. Initially, I was careful to avoid longer words for clarity's sake. But in fact, such words can be comfortably handled by assuming that phonological systems are HIERARCHICAL and that the concept of rela- tive prominence is applicable at various points in the hier- archy. The rank which we considered in [115] is that of the F O O T . Adjacent syllables were grouped together and one of them was identified as being more salient than its neighbour. Where a word contains more than two syllables, the same manoeuvre can be repeated, with adjacent feet being grouped together in order to establish which one of them is more salient. Consider

[11.6]

s

s w s w w sen si ti vi ty

This procedure is known as WORD-TREE FORMATION. We can identify the antepenultimate syllable -ti- as the one receiving primary stress. By inspecting the tree, we can establish that it is dominated by more s's than any other syllable. Degrees of stress are read off the tree by counting the number of s nodes that dominate a particular syllable: the more s nodes there are above a syllable in the tree, the more heavily stressed that syllable is.

Unfortunately, the model just outlined does not always yield the correct results. It fails to capture some prominence distinctions in English. For instance, it would lead us to construct the following trees for radio, rabbi, racer and racy:

[11.7] (a)

(b)

s w s w ra dio ra bbi

s w s w ra cer ra cy

228 Stress and intonation

+ s ra

+ s ra

+ w

dio

\ w

cer

+ s ra

+ s ra

+ w

bbi

\ w

cy

The trees in [11.7] are identical. But we know that the final syllable of the words in [11.7a] is more prominent than the final syllable of the words in [11.7b]. To solve this problem, Liberman and Prince (1977) proposed the retention of the SPE segmental feature [ ± stress] in their metrical theory of stress. If that is done, [11.7] can be re-written as [11.8], bringing out clearly the differences in the relative promi- nence of the final vowel:

[11.8] (a)

(b)

(Here + stands for [+stress] and - stand for [—stress]) There is some understandable unease about a metrical,

nonsegmetal theory of stress which sets out to show the relational nature of stress allowing binary segmental feature [±stress] to play a role in metrical trees. A simpler way of showing the relative salience of syllables which overcomes this problem is proposed by Prince (1983). He advocates the construction of a METRICAL GRID. This is done by representing information of the kind contained in a metrical tree as an array of asterisks. One asterisk is assigned to each syllable at the syllable tier, an additional asterisk is assigned to the stronger syllable in each foot, and if necessary, at word level a further asterisk is assigned to the strongest foot which is nearest the beginning or the end of the word, depending on the preference of a particular language.

Using a grid we can show in a transparent way the relative stress of the different syllables of sensitivity as in [ » - 9 ]

[II-9] •

* * * * * • • • sensitivity

end rule (see section 11.1.3 below) word level

* foot level * syllable level

Introduction: stress 229

Although trees and grids can convey the same infor- mation, grids are preferable because they do so in a more perspicuous manner.

Convert the metrical trees in [i 1.5] into metrical grids.

11.1.3 Metrical trees and grids

If you cast your eye back to [11.5], you will observe that in words like sender the left syllable is stronger than the right one in the same foot while in a word like pretend the right- hand syllable is the stronger of the two. Trees like those in [11.5] where one syllable functions as the HEAD and governs the syllable that immediately precedes or follows it are said to be B O U N D E D .

In addition to bounded trees, there exist U N - B O U N D E D trees where a head governs not just one ad- jacent syllable, but all the syllables that precede or follow it. This is the situation in a language where primary stress always falls either on the initial or final syllable of a word.

Hayes (1981) provides a typological classification of metrical systems. He shows that in some languages, such as Maranungku (Australia), primary stress falls on the first syllable, and secondary stress on alternate syllables to its

tiralk 'saliva' right:

[11.10] merepet 'beard'

yangarmata

langkarateti

welepenemanta

'the Pleiades'

'prawn'

'kind of duck'

An unbounded tree with a dominant left node [11.11] is needed to describe the facts of Maranungku:

[ n - i i ] (a)

s w ti ralk

230 Stress and intonation

Contrasting with a language of this kind, are languages like French which have unbounded trees with a dominant right node. In French stress falls on the last syllable of a word:

[11.12] petit 'little' encoura^er 'encourage' telecommunication 'telecommunication'

Draw a bounded tree to represent stress in French telecommunication.

The tree required is [11.13]

[11.13]

w s

A / \ w s w s

I / \ /K S\ s w s w s w s te le co mmu ni ca tion

The head is on the right and it can have an indefinite number of subordinate syllables preceding it.

Maranungku and French are not unusual. In many languages, primary stress falls at or near the periphery of a word. Metrical grid theory reflects this fact by incorpor- ating the E N D RULE convention. This convention is responsible for the addition of extra salience (shown by an extra asterisk on a metrical grid) to the most prominent left- hand or right-hand column of a word, depending on whether a language favours the beginning or the end of the word as the locus of primary stress.

In Maranungku where the left-hand edge of the word is dominant, the first syllable is assigned an extra asterisk by the end rule:

[11.14] * end rule * * * * * * * * * *

we le pen e man ta

Introduction: stress 231

In Weri (Papua-New Guinea) (Hayes 1981) the situation is reversed. The final syllable is dominant and alternating secondary stress is put on the syllables preceding it jjJka word.

[ I I . 1 5 ]

* • * * • • * • * akunetepal —»- akunetepal 'times'

11.1.4 Extrametricality

In some languages, such as Finnish and Hungarian, p r i r n l j ^ stress normally falls on the initial syllable while in other" langauges, such as Cambodian and Modern Hebrew, it is the final syllable that gets primary stress. In systems of this kind all syllables are taken into account in deciding where to put stress.

However, in many other languages syllables at word fringes are ignored by stress assignment rules. Such syllables are said to be EXTRAMETRICAL. They are completely 'invisible' to the rules which assign stress (see extrasylla- bicity in (9.7)). The concept of extrametricality is useful in the treatment of languages which have their main stress on the P E N U L T I M A T E syllable of the word. In these languages, the last syllable is 'not seen' by the algorithm which proceeds to put stress on the second last syllable. Swahili is a typical example of such a language (the extra- metrical final syllable is put in parentheses):

[ 11.16] pi(ga) 'hit, beat' piga(na) 'fight, hit each other' pigzni(sha) 'cause to fight' tutawapi({><j) 'we shall beat them'

While in Swahili, which has penultimate main stress, it is the final syllable of a word that is extrametrical, in Native American languages like Dakota, Northern Paiute and Southern Paiute, where main stress falls on the second syllable of the word, it is the first syllable that is extrametrical.

232 Stress and intonation

11.1.5 Quantity sensitivity

We have seen above some of the typological parameters used in the study of stress: languages may have bounded as opposed to unbounded trees; the left or the right end of the word may be dominant; either all syllables in a word may be taken into account by stress placement rules or some of them may be extrametrical. Typological considerations of this kind are important because linguistic theory is concerned not only with the description of particular languages, but also with an exploration of the parameters within which languages can differ. For, as I have stressed at various points in the book, differences between languages occur within certain pre-set limits.

In this section we are going to examine a further typo- logical parameter, namely whether stress placement in a given language is sensitive to the internal structure of the syllable that bears it. In particular we shall consider the role of SYLLABLE WEIGHT in stress placement. In the exam- ples discussed above, no mention was made of syllable weight because in the languages which were cited, the internal structure of syllables is not taken into account in determining where to put stress. Syllables are treated in the same way irrespective of whether they are light or heavy.

The languages described in the last few paragraphs have SYLLABLE Q U A N T I T Y INSENSITIVE STRESS RULES. You will recall that we broached the subject of SYLLABLE Q U A N T I T Y SENSITIVE STRESS rules when examining Latin syllable structure in Chapter 9. Like the Swahili rule, the Latin stress rule treats the last syllable as extrametrical, and ignores it. But there the resemblance ends. While the Swahili rule puts primary stress on any penultimate syllable, the Latin rule is choosy. It puts stress on a penultimate syllable only if it has got a branching rhyme (i.e. it has either a long vowel or is closed by a consonant). The Latin stress rule can be re-stated as [11.17]:

[11.17] (i) ignore the final syllable: it is extrametrical; (ii) construct an unbounded right-headed tree;

(iii) put primary stress on the first heavy syllable at the right-hand side of the word if it is in penultimate position (excluding the final syllable which is extrametrical);

Introduction: stress 233

(iv) if the penultimate syllable is light, put primary stress on the ante-penultimate syllable.

Use [11.17] to predict the syllable which gets primary stress in [11.18]:

[11.18] (a) amo (b) amamus (c) amabamus (d) amabimus (e) amaveras (f) amabantur

I love' we love' we loved' we shall love' you had loved' they will be loved'

Your solution should be along these lines (V indicates primary stress):

[11.19] (a) a(mo) Since the final syllable is always extrametri- cal and therefore irrelevant for stress place- ment purposes, in disyllabic words primary stress must be on the first syllable regardless of its structure.

(b) ama(mus) The penultimate syllable contains a long vowel and therefore counts as heavy. It receives the stress.

(c) amaba(mus) The penultimate syllable is stressed again because it is heavy.

(d) amabi(mus) The penultimate syllable only has a short vowel in the rhyme; it is light. Stress skips it and lands on the ante-penultimate.

(e) amave(ras) The statement made for (d) above covers this.

(f) amaban(tur) The penultimate syllable has a branching rhyme consisting of a vowel followed by a consonant. It is heavy. It receives the stress.

234 Stress and intonation

11.1.6 English stress

I shall n o w illustrate the t h e o ry of metrical p h o n o l o g y b y sketching an outline description of aspects of stress in English. T h e stress s y s t e m of English is e x t r e m e l y c o m p l e x . H e r e I can only give y o u a flavour of a metrical analysis of the system. T o c o m e to grips w i t h the intricacies of English stress, y o u will need to study the w o r d s cited in the b i b l i o g r a p h y . 1

As a rule, unless they are l o n g or c o m p l e x , g r a m m a t i - cal w o r d s like prepositions and conjunctions d o n o t receive p r i m a r y w o r d stress. (But long g r a m m a t i c a l w o r d s , e.g. the preposition underneath and the conjunction notwithstanding receive p r i m a r y w o r d stress.) All lexical w o r d s , i.e. n o u n s , adjectives, verbs and adverbs m u s t have one syllable which receives p r i m a r y stress; obviously, in m o n o s y l l a b i c w o r d s like dog, look and bag, the only syllable that there is m u s t be stressed. W o r d stress rules are needed in the g r a m m a r to account for the location of stress in w o r d s of t w o or m o r e syllables. O n l y o n e syllable receives p r i m a r y stress and the rest are s u b o r d i n a t e d to that syllable.

First consider stress placement in a few disyllabic verbs:

[ 11.20] (a) order ferry cover carry dither copy (b) open lollop hiccup lavish covet pocket

[ n . 2 1 ] (a) compare delay endow enjoy survey prefer (b) corrode avoid refine design amuse presume (c) concoct enlist protect attend assent announce

Suggest a rule (or rules) to account for the placement of p r i m a r y stress in these w o r d s .

Y o u r rule needs to be quantity sensitive. It m u s t m a k e crucial reference to syllable w e i g h t . W h a t is needed is an a l g o r i t h m w h i c h constructs a right-headed b o u n d e d tree over the w o r d and places main stress on the first heavy syllable, c o u n t i n g from the r i g h t - h a n d end of the w o r d . W h e r e disyllabic verbs contain n o heavy syllables, stress falls b y default o n the first syllable. B u t o t h e r w i s e stress falls on the r i g h t - h a n d m o s t h e a v y syllable.

Introduction: stress 235

Thus, in [11.20a] there is no heavy syllable. So, primary stress falls by default on the first syllable in words like copy and ferry. In [i 1.20b] the second syllable contains a short vowel and a consonant. O n the face of it, the second syllable should count as heavy. But it does not. This is because the last consonant of a verb is extrametrical, i.e. it is 'not seen' by stress rules. A word like lavish [laevi(J")] is only scanned up to the vowel [1]. Consequently its final syllable is classed as light and primary stress goes on the first syllable.

In [11.21 a], on the other hand, in each case the second syllable contains a long vowel or diphthong. This makes it heavy and capable of bearing primary stress. In [11.21b] too the final consonant is extrametrical but this does not prevent primary stress falling on the second syllable. This is because the part of the syllable that is seen contains a long vowel or diphthong (see avoid [avoi(d)]). Finally, in [11.21c] stress once more falls on the second syllable. In these words the second syllable has two consonants following the vowel. Even though the last consonant is extrametrical, it is the second syllable which is nevertheless stressed. This is due to the fact that the part of that syllable which is 'seen' by the stress rule counts as heavy because it contains a short vowel followed by a consonant (see enlist [enlis(t)]).

Note in passing, however, that the rule stated above is not exceptionless. Words ending in [au] (like follow and sorrow) are stressed on the first syllable although the second syllable contains a diphthong and 'ought to be' regarded as heavy.

Disyllabic adjectives are subject to the same rule as verbs. Adjectives like tender, tiny and holy which contain no heavy syllables are stressed on the initial syllable. Adjectives like evil, timid and livid whose second syllable ends in a consonant and therefore 'ought to be' heavy are, in fact, stressed on the first syllable because the final consonant is extrametrical. The part of the syllable that is 'seen' by the rule is light (a consonant followed by a short vowel). Lastly, adjectives like robust, morose and alive are stressed on the second syllable because their second syllables are heavy.

Next, let us turn to disyllabic nouns.

236 Stress and intonation

Begin by marking the syllable which receives primary stress in the words below:

[11.22] (a) pity picture

(b) magic (c) almond (d) canoe (e) police

baby ferry tulip contact bamboo alert

sugar panda mason subject bazaar debate

river butter foetus dentist settee cartoon

In [ n . 2 2 a - c ] primary stress falls on the first syllable while in [ n . 2 2 d - e ] it falls on the second syllable. As before, a right-headed tree is constructed for each word. Where there is no heavy syllable, a disyllabic word is stressed on the first syllable. But if a word contains a heavy syllable, stress goes on the heavy syllable nearest to the end of the word.

Note, however, this crucial difference between nouns on the one hand and verbs and adjectives on the other: for the purposes of this rule, a heavy syllable in a noun must contain a long vowel or diphthong. While in verbs and adjectives only the final consonant is extrametrical, in nouns, any consonants following the final vowel are extra- metrical. This means that the presence of consonants after a vowel does not contribute to syllable weight in nouns.

Thus, in [11.22a] nouns like pity, whose second syllable is light are stressed on the first syllable. Similarly, the first syllable is stressed in both [11.22b and c] because of the extrametricality of any consonant following the vowel in the second syllable. Consequently, the part of the second syllable that is 'seen' contains a short (lax) vowel and there- fore cannot be stressed.

However, nouns like police [n.22e] whose second syllable contains a long vowel (followed by one or more extrametrical consonants), are stressed on the second syllable just like nouns such as canoe [ n . 2 2 d ] which end in a long vowel or diphthong. All that matters is whether or not the second syllable contains a long vowel or diphthong. In other words, with regard to the disyllabic noun stress rule, only the nucleus of the second syllable is projected. The noun stress rule only 'sees' the syllable nucleus: if the

Introduction: stress 237

nucleus branches (i.e. contains a long vowel or diphthong), stress is on the final syllable. Otherwise, stress is on the initial syllable.

We can now move on to longer words. We shall observe that the stress rule that applies in longer words is remi- niscent of the Latin stress rule.

Study the data in [11.23] and work out the rule that deter- mines location of primary stress:

[11.23] (a) cinema Agatha effigy Malibu jeopardy overture calumny Salisbury

(b) Ga/apagos And'gone America rhinoceros epitome

(c) tornado rhododendron aroma psycnosis bronchitis Theresa brontosa«rus

As in Latin, the stress placement algorithm creates a right-hand headed tree, ignoring the final syllable, which is extrametrical. Stress lands on the penultimate syllable if it is heavy, [see 11.23c] but if the penultimate is light, stress goes on the ante-penultimate syllable (see [11.23a, b]).

Up to this point we have only considered stress in simple words. We have not examined yet the problem of stress placement in complex words which consist of a root and one or more affixes, which may be prefixes or suffixes. Interestingly, in English the addition of a prefix tends not to affect the placement of stress. Usually words are stressed in the same way with or without prefixes:

t * * *

[11.24] write rewrite polite impolite examine re-examine gratitude ingratitude political apolitical moral amoral

The effect of the addition of suffixes is much more fascinating because different suffixes affect stress placement in different ways. First, there are suffixes whose presence has no effect on the primary stress of the root to which they are attached. A good example of this can be seen by observing what happens when the suffix -ment is added to a verb stem like govern to yield the noun government. The

238 Stress and intonation

position of stress remains unchanged. Likewise, the suffixes -ness as in kindness and -Jul as in fdithjul have no effect on stress.

Another class of suffixes attract stress to themselves as though they were magnets. When they are attached to a word they always get the main stress. They include -ette as in maisonette and kitchenette, -esque as in picturesque and grotesque. (These suffixes are borrowed from French and obey the French stress rule.)

Add two more examples to this list.

A third class of suffixes make the primary stress move to the syllable immediately preceding them. That is the case when a suffix like -ic is added to a stem like democrat to derive the adjective democratic or when the suffix -ity is added to the stem public to produce publicity.

(a) Write down three examples of words containing each of the suffixes -some, -hood, -ity and -ic.

(b) Contrast the effect on stress placement of the presence of the suffixes -some and -hood on the one hand and -ity and -ic on the other.

You will have discovered that -some and -hood have no effect on stress. A word is stressed on the same syllable with or without these suffixes. However, the suffixes -ity and -ic do affect stress placement in the base to which they are attached: they attract stress to the syllable immediately before them. Interestingly, the rule, which attracts stress to the syllable before the suffix is a Q U A N T I T Y INSENSI- TIVE rule - it puts stress on that syllable regardless of whether the syllable is light or heavy. Suffixes like -ity and -ic which attract stress to the syllable immediately preceding them are called S T R O N G M O D E SUFFIXES.

There is a fourth category of suffixes which is similar to the strong mode suffixes in that they also affect stress placement in the base form to which they are attached although they are not stressed themselves. But unlike strong mode suffixes, this class contains Q U A N T I T Y SENSI-

Intonation 239

TIVE suffixes which put stress on the immediately preceding syllable only if it is heavy either because it has a long vowel or a diphthong:

[11.25] adjectival arrival homicidal betrayal recusal rental

or because it is closed by a consonant:

[11.26] departmental ornamental detrimental accidental

If the preceding syllable is light, stress misses it out and falls two syllables before it:

f

[ 11.27] problematical practical conditional municipal seasonal professional

Sufffixes of this kind which only manage to attract stress to the syllable immediately before them if it is heavy, and otherwise attract stress to the second syllable to their left are called WEAK M O D E SUFFIXES.

11.2 Intonation

All languages use pitch. This, in part, is an automatic consequence of the fact that any time a voiced sound is produced, the vocal cords vibrate at a certain rate. The rate of vibration corresponds closely to the pitch perceived by the hearer: the higher the rate of vibration the higher is the perceived pitch. The fact that pitch differences can be observed in the utterances of any language is in itself of no great linguistic interest.

What linguists are interested in are the ways in which pitch differences are functionally harnessed. In broad terms, pitch differences can be exploited in two distinct ways. Within the domain of the word, pitch can be used to contrast lexical meaning or to mark grammatical properties, as we saw in the last chapter. In that case we speak of tone. Alternatively, the domain of pitch can be an entire utter- ance, in which case we speak of intonation. It would be wrong to classify languages as either tonal or intonational because all languages have intonation. That includes tone languages, as the discussion of downdrift in Chapter 10

240 Stress and intonation

showed. One of the issues that has rightly received a good deal of attention from phonologists is the way in which intonation meshes together with stress in a stress language, and with tone in a tone language.

In the next section we shall outline some of the main features of English intonation. This sketch is not meant to be a comprehensive description. Rather it is merely intended to give us a glimpse at the workings of intonation in a stress language.2

11.2.1 The form of English intonation

Just as the syntactician focuses on the sentence as the key unit of grammatical analysis, the phonologist focuses on the T O N E U N I T (also called the P H O N O L O G I C A L PHRASE) as the most significant domain in terms of which intonation contours reflecting the pitch of utterances are assigned.

For the purposes of intonation analysis, the English tone unit has the following internal organisation:

tonic [11.28] (pre-head) head or (tai7)

nucleus

The nucleus is the central element in a tone unit. It contains the syllable in an utterance which undergoes significant pitch movement - and is consequently more prominent than the rest. That syllable is the N U C L E U S or T O N I C SYLLABLE. The tonic syllable is normally preceded by a HEAD. The head is the part of the tone unit extending from the first stressed syllable to the syllable immediately preceding the tonic syllable. Optionally, a head can be preceded by a PRE-HEAD. This consists of any unstressed syllables that occur in front of the first stressed syllable of the head. Again, optionally, a tonic syllable can be followed by a TAIL. The tail contains any syllables (which may or may not be stressed) following the tonic syllable.

Say the sentence in [11.29] noting the way pitch fluc- tuates. Produce two analogous examples of your own and state verbally the pitch movements which you observe.

Intonation 241

[11.29] || He will phone you when all the children are back. || Ph H Ts T key: Ph = pre-head; H = head; Ts — tonic syllable; T = tail

Normally pitch is low in the pre-head, more or less level high in the head and falling on the tonic; in the tail the pitch pattern established on the tonic is simply continued.

The most common direction of pitch movement on the tonic is downward. As we saw in section 10.5, linguists talk of D O W N D R I F T in tone languages when they describe the tendency for high tones to be somewhat lowered when preceded by a low tone, a phenomenon which results in falling intonation. In many stress languages like English the same downdrift (also called DECLI- N A T I O N LINE) can be observed. Frequently pitch moves down on the tonic syllable and remains down until the end of the utterance. Pitch tends to be much lower at the end of an utterance than it is at the beginning. FALLING I N T O N A T I O N , which is exemplified by [10.30], is the unmarked intonation pattern in English.

["•30] 'James Thurber was born in O hio.

l_ The widespread tendency to drop pitch as the end of

an utterance approaches might have a physiological expla- nation. Possibly, as the speaker gradually runs out of breath, there is less and less air to cause the vibration of the vocal cords and consequently they vibrate more sluggishly and the pitch of the utterance goes down. (At the same time, in some languages like Luganda, the intensity of the signal also declines, so that the end of the utterance is auditorily less salient than the beginning.)

It is reasonable to ask what the functions of intonation are. Does pitch fluctuation serve any linguistic purpose? Some answers to this question are outlined below.

242 Stress and intonation

11.2.2 Accentuation function

Word stress interacts with intonation. One of the words in a sentence has a syllable which stands out above the rest. This is the syllable which has S E N T E N C E STRESS - also called T O N I C STRESS. Such a syllable is sometimes referred to as the T O N I C SYLLABLE. In unmarked cases, tonic stress goes on the syllable which carries primary word stress in the last lexical item of the tone unit. Such a lexical item is often a noun, an adjective or a verb. The tonic syllable is underlined in the examples below:

[11.31] They are working. Joan has not seen him. The children are in the playground.

In marked cases, the above rule is superseded by a rule which allows CONTRASTIVE (or EMPHATIC) STRESS to fall on grammatical (function) words such as pronouns, prepositions and conjunctions if the speaker wishes, for some communicative reason, to draw attention to such words or alternatively, contrastive stress may occur on non- final lexical items:

[11.32] (a) She is studying linguistics. She is studying linguistics. She is studying linguistics,

(b) She travelled from London. She travelled from London. She travelled fo London. She travelled to London.

How does the location of the tonic syllable affect the ways in which these sentences can be interpreted?

The accentual function is the most basic function of intonation. In unmarked cases tonic stress is often used to make a syllable in the lexical word containing new infor- mation stand out. But in marked cases, the tonic syllable can be the most prominent syllable of virtually any word which the speaker chooses to highlight.

Intonation 243

11.2.3 Intonation and illocutionary force

The choice of intonation pattern is not entirely free. In a language such as English certain ILLOCUTIONARY ACTS (= acts of speaking) such as making statements and asking questions, are typically performed using certain intonation patterns. I use the word 'typically' advisedly. The deployment of intonation contours is not determined by rigid, exceptionless rules. A given illocutionary act may be performed without using the intonation pattern indicated here. But nevertheless, certain intonation patterns are much more likely to be used to perform certain illocutionary acts than others.

The tone unit type with a level head and a falling nucleus (' h " n), which is the only one introduced so far, is unmarked. It is the intonation pattern that is assigned when there is no good reason for doing otherwise. For example, it is normally used in statements, in imperatives and in wh- questions (i.e. questions beginning with the words why, where, when, what, which etc.). You can verify this by reading the sentences below in a neutral way and observing the intonation contour which you use:

[11.33] (a) Statements: ' h ' n 'Mary lives in ' Lancaster. 'The shops are ' dosed.

(b) WH questions: ' h y n 'When did Mary x go there? 'What is your v name"?

(c) Imperatives: ' h " n Just 'eat up your ' dinner\ ySit\

Note: the head and tail are both optional. They can both be omitted from a tone unit, as in the last example. Only the nucleus is obligatory.

Besides falling intonation, another common (but some- what marked) type is the LOW RISE (rh,n) where the head is level and the nucleus rises slightly. This pattern is often found in Y E S / N O Q U E S T I O N S where the speaker queries an item and expects simple confirmation or denial:

[11.34] Are your 'friends going to the ,party~l Will you re'tire next ,year?

244 Stress and intonation

I will mention just one more tone unit type, HIGH RISE (' ' N). This has a level head and a high rising nucleus. The head may be optionally omitted. This is common in elliptical questions like:

[11.35] ' coming? ('Are you 'coming) ' tea anybody? ' taxi?

Optionally, the pitch of questions may be raised generally so that the upper range of the speaker's voice is employed.

11.2.4 The grammatical function of intonation

The interpretation of the meaning of a sentence in part involves knowing which words should be syntactically bracketed together. In writing, punctuation is used to group together words which grammatically and semantically form coherent units. In the spoken language intonation serves a similar purpose.

Clear evidence of this can be seen by observing the way in which intonation is used to resolve syntactic ambiguity. Many sentences which are potentially ambiguous on paper are not ambiguous in their spoken form. For a long time linguists have been aware that the syntactic bracketing of potentially ambiguous constructions like old cars and buses can be clarified by intonation. The whole phrase is said as one unit if both the cars and buses are old. But it is said with an intonation break after cars if it is only the cars that are old. Effectively, intonation is used to indicate whether the adjective modifies just the first noun or both nouns.

Say the sentences in [11.36] using in each case the intonation contour which reflects the syntactic bracketing. How do the sentences in each pair differ in meaning?

[11.36] (a) Rioting [young men] and [women] were arrested. Rioting [young men and women] were arrested,

(b) They are [cooking apples]. They [are cooking] apples.

Intonation 245

The syntactic functions of intonation can also be considered from the angle of language processing. We have seen that tone groups usually mark off major syntactic constituents like noun phrases and verb phrases. That is what enables them to disambiguate sentences. In addition, the syntactic units marked by intonation seem to have some kind of psychological relevance in language processing. Thus, it is decidedly odd to place a tone group boundary inside a syntactic constituent. Pauses and any hesitations tend to coincide with syntactic boundaries.

(a) Parse the sentences below and show their major syntactic units.

(b) Where could you hesitate in each one of these sentences? Where are the potential intonation breaks?

[11.37] (a) The man to whom John was talking gave him the book.

(b) Frank and David or Helen went to the party.

(c) When she left the dog started barking.

I hope you have discovered that the natural way of marking intonation boundaries and the places where hesi- tation is likely to occur involves identifying major syntactic divisions:

[11.38] (a) subject (noun phrase) || The man to whom John was talking | predicate (verb phrase) gave him the book.||

(b) subject (noun phrase) predicate (verb phrase) || Frank and David or Helen | went to the

party. || (c) subordinate clause main clause

|| When she left | the dog started barking. ||

'GARDEN PATH' sentences like [11.38c], for which one of the most obvious parsings is wrong, are often assigned the wrong intonation before the speaker works out the correct syntactic analysis. You were probably momentarily tempted to parse c. thus:

[11.38] *|| When she left the dog | started barking.||

246 Stress and intonation

with an intonation break after dog, before you realised that the principal division here is between the main clause and the subordinate clause: dog is the subject of the verb started in the main clause and not the object of the verb leji in the subordinate clause.

11.2.5 Attitudinal functions

Intonation also has an attitudinal function. Inferences can be drawn from intonation about the speaker's A T T I T U D E . We all have heard someone say 'It was not what he said, but the way he said it . . .' or 'It is the tone of voice she used that really upset me'.

We can all mean different things with the same words by modulating our intonation. Utterances may perform different illocutionary acts depending on the speaker's atti- tude as it is revealed through intonation. An utterance like sit down or come back can be a command, a polite invitation or a threat.

The meaning of an utterance depends on many factors. Most obviously it depends on the semantic content of the words as well as the grammatical and logical relationships between the words used in an utterance. Less obviously, it depends on the C O N T E X T OF U T T E R A N C E (the linguistic context in which the words are uttered and the background situation known to be relevant by both the speaker and hearer); it depends on the PARALINGUISTIC features, e.g. gestures, facial expression and voice quality employed; and it depends on the intonation used by the speaker. Intonation seems to play a role in conveying atti- tudinal meaning. But the contribution made by intonation per se is not easy to isolate with any certainty. It is therefore not feasible to establish a simple, direct correlation between intonation and attitude.

One area where a clear link has been established is in Q U E S T I O N TAGS. Normally, falling intonation in the tag indicates a quest for confirmation:

[11.39] 'George is a x lawyer, ' isn't he?

But rising intonation indicates a greater degree of uncer- tainty, with the speaker not merely seeking confirmation but rather wanting to have a real doubt cleared:

Intonation 247

[11.40] 'George is a ' lawyer, ,isn't he?

Elsewhere the link between intonation and attitude is obscure. At best what we can do is outline various factors affecting attitudinal judgements. But it would be rash to claim that these factors determine attitudinal judgements.

One important factor affecting attitudinal judgements is VOICE QUALITY. Phoneticians have become increas- ingly aware of the importance of voice quality as a factor influencing attitudinal judgements (Laver 1980; Knowles 1987: 211-14). Some of the properties of a speaker's voice are INDEXICAL in the sense that they convey some infor- mation about the speaker to the hearer. Some indexical features are relatively permanent. A speaker's sex, age, membership of a social group, regional origins or social class membership can often be inferred from voice quality. These indexical features cannot be easily changed by the speaker. Other indexical features of voice quality such as laryngitis or a cold or being drunk are more transient. The addressee can make various favourable or unfavourable judgements about the speaker on the basis of voice quality features of the kind that I have listed.

A speaker can also deliberately attempt to convey atti- tudinal information by combining a voice quality trait with a purposeful use of pitch. We observed above that the commonest intonation contour in English is a level head followed by a falling nucleus. The SLOPE (or gradient) of the fall is variable. In unmarked circumstances the fall is gentle. If a speaker opts for a very steep slope, this is likely to convey a peremptory, abrupt or domineering attitude, especially when accompanied with harsh voice quality.

Try saying 'Come here !' varying the slope and harshness of the voice quality. Describe the circumstances in which each version of the sentence would be appropriate.

A further variable which has to be considered is PITCH RANGE. There is some correlation between pitch range and emotion.

Say this sentence varying pitch range:

[11.41] (a) I am so glad to see you. (b) Hello!

248 Stress and intonation

If a total stranger ran up to you in the street and said

[ii-42]

Hello!

You would probably think it was a case of mistaken iden- tity. If, on the other hand, when a long lost friend turns up at your door you say:

[11.43] I am so glad to see you again.

\

that would be interpreted as a frosty welcome. When people are excited they use the upper part of

their speech range much more than they do normally. This is usually accompanied by an increase in L O U D N E S S and T E M P O . When depressed or sad, on the other hand, people tend to use mostly the lower part of their pitch range and to talk more slowly and more quietly.

There are no cast iron rules which govern the use of intonation to signal attitude. There are only tendencies. The right interpretation of an utterance depends only in part on intonation. As we saw above, there are many other vari- ables which need to be considered. It would be naive to attempt to establish a simple, one-to-one correlation between attitude and intonation.

11.2.6 Discourse function

Talk rarely consists of isolated utterances. The norm is for utterances to be strung together, and in conversation for the roles of speaker and hearer to be swopped frequently. Inton- ation plays a crucial role in the structuring and gluing together of discourse.

Intonation is normally used to indicate NEW INFOR- M A T I O N (which the addressee is assumed by the speaker

Intonation 249

not to already possess) as opposed to GIVEN INFOR- M A T I O N (which is known or assumed to be already known to the addressee). Given information is kept in the background. It is not in F O C U S . But new information is foregrounded. It is in focus. One of the words conveying new information is highlighted by making one of its syl- lables the tonic syllable.

Imagine an exchange like [ n . 4 4 ] in a butcher's shop:

[11.44] John: Do you 'like ' tripe? Mary: I ' loathe tripe, (or / ' loathe it)

When John asks the question, tripe is new information and receives tonic stress (the word 'information' is being used in a very broad sense). But when Mary replies, tripe is already established as the topic of the conversation and needs no highlighting. So, tonic stress falls on loathe. It would be very odd if in Mary's reply tonic stress was on tripe (* / 'loathe * tripe).

Furthermore, intonation plays an important role in turn-taking in dialogue. A fall usually indicates a completed turn, without the expectation of one's interlocutor replying. But a rise gives a strong indication that the speaker intends to continue, or that a response from one's interlocutor is expected. A simple example of this can be seen in V O C A - TIVES and in YES-NO Q U E S T I O N S :

[11.45] (a) Vocative ' ' Janet (in fact the fall rise is often levelled to —Ja~net)

(b) Yes-No question Have you 'lived here for a long ,time?

It is possible to use falling intonation when calling a person's name e.g. when a school teacher shouts at a naughty boy: "Smith rather than ' ' Smith. What is the difference between these two utterances?

A fall-rise calling nucleus (coupled with generally raised pitch) is the intonation pattern normally used to attract the addressee's attention. However, a falling nucleus can also be used for this purpose - with somewhat different prag-

250 Stress and intonation

matic force. In our example, (especially if the teacher has a stern look and uses loud and harsh voice quality) this 'call' will in fact probably be interpreted as meaning:

[11.46] ' pay attention * Smith.

One of the things that have emerged from this chapter is the fact that phonology has to be placed in a wider context. The discussion of the attitudinal functions of inton- ation showed the way in which phonology overlaps with verbal paralinguistic properties such as voice quality, pitch- range and tempo. These paralinguistic properties are not part of the linguistic system proper although they impinge on it.

The relationship between language and paralanguage (which includes not only verbal non-linguistic properties such as voice quality and pitch-range but also non-verbal features like eye-contact and posture) is an interesting topic and has attracted considerable interest in the literature (see Argyle (1974); Laver and Hutcheson (1972); Laver (1980) and Lyons (1977)). Unfortunately, we do not have the space to pursue it here.

What we shall explore is the relationship between phonology and the other core components of the linguistic system. At several points in the book we have seen that phonology impinges on grammar and meaning. In the concluding chapter we shall focus on this interaction.

Exercises

1.(a) Study the data below and state verbally the placement of stress in Arabic,

(b) Draw metrical trees to show where stress falls in the following words: kdtab, kdtabu, kdtabit, katdbt, katdbti, lamuuna and lamundat. katab he wrote katabit she wrote katabt you (m.) wrote katabti you (f.) wrote katabna we wrote katabtu you (pi.) wrote katabu they wrote

Intonation 251

lamuuna lemon lamunaat lemons mooza banana mozaat bananas xooxa peach xoxaat peaches

2. Ngiyamba (Australia) (Based on Donaldson 1980) Study the data below.

(a) Referring to syllable weight, state in general terms the constraints on stress placement in roots and suffixes in Ngiyamba. Which syllables receive primary stress? Which syllables receive secondary stress? And which syllables are unstressed?

(b) Illustrating your answer with examples show how (i) primary stress is assigned in roots;

(ii) secondary stress is assigned in suffixes; (iii) secondary stress is assigned in roots; (iv) secondary stress is assigned in monosyllabic

suffixes with short vowels. Note: In the data below, 1 indicates primary stress, and

2, 3 and 4 indicate secondary stress. Unmarked syllables are unstressed.

1 girala 'star'

1 3 giralarj-ga 'on star'

I 2 girlam-bidi 'big star'

3 1 gabada:-ga 'on moon'

3 1 gabada:-bidi 'big moon'

1 girbadja-gu '(grey) kangaroos'

1 2 girbadja-bidi 'big kangaroo'

1 4 2 yana-wa-y-gara:-dha 'go along all day'

252 Stress and intonation

I 2 yana-wa-y-ga:-dha

I 2

yana-wa-y-ga:-giri I 2 4

yana-buna-wa-dha

3 i binjdju-binjdju:ri-nji

I 2

girbadja-gu-baga:-dhi:

budha:ni-nji I 2

rjiyam-ba: i

gadawu-ga

bayirga-gu I

maliyan

I 2

mayim-buwan

go along a bit'

(Let's) go along a bit'

go along back'

(I) half-recovered (my) balance'

But the kangaroo scented me'

scent-past tense'

Ngiyamba'

on large-leaved sandalwood'

leech -dative case'

eagle hawk'

with person'

Notes

There is a very rich literature on stress in English. SPE is the classic generative work on the subject and any serious student of English stress ought to be familiar with its contents. For an introductory textbook which explores English stress in an accessible way from a non- metrical, non-generative point of view see Knowles (1987). Fudge 1984 is another recent non-metrical, well-exemplified general survey of English stress. Hogg and McCully (1987) is an advanced introductory textbook on metrical phonology. It examines English stress in considerable detail.

Intonation 253

The account given here is based on the work of metrical phonologists like Hayes (1982, 1983, 1984) Liberman and Prince (1977) and Prince (1983).

2. Knowles (1987) is an introductory book which deals with English intonation extensively. There are numerous advanced studies of English intonation e.g. Halliday (1967, 1970); Ladd (1980); Pierrehumbert (1980); Cruttenden (1986) which you can consult for more detailed discussion.

CHAPTER 12

Phonology in the Wider Context

12.1 The role of the lexicon

Linguistic theory is interested in sound not just for its own sake, but because it is the medium in which speech is real- ised. Phonology has to be put in a wider context so that its interaction with the other components of the grammar can be understood. At the beginning of the book, we saw how sounds are used phonemically to contrast word meaning; in the last two chapters we have considered the grammatical and semantic/pragmatic and discourse functions of tone, stress and intonation. In this final chapter we are going to continue with the theme of the relationship between phon- ology and the rest of the grammar, but this time from a different perspective. We shall highlight the ways in which phonology interfaces with the lexicon and morphology on the one hand, and with syntax on the other.

I think the most promising analysis of the relation between phonology, morphology and the lexicon is the m o d e l of L E X I C A L P H O N O L O G Y A N D M O R P H O L O G Y (normally referred to as LEXICAL P H O N O L O G Y for short). Although it is a recent devel- opment, this trend already boasts a substantial amount of literature (see Kiparsky 1982a, 1982b, 1985; Halle and Mohanan 1985; Mohanan 1985; Kaisse and Shaw 1985; Rubach 1985; Pulleyblank 1986).

As its name suggests, lexical phonology gives the LEXICON a key role. This represents a significant depar- ture from previous models. Traditionally, all regular processes were dealt with by rules of the grammar. The lexicon was seen as being no more than an appendix to the grammar, containing unpredictable idiosyncratic phono-

The role of the lexicon 255

logical, grammatical, semantic and lexical information about morphemes and lexical items. These different kinds of information would be included in the lexicon because of their relevance to the application of semantic, syntactic, morphological and phonological rules.

Phonological information is relevant because speakers need to know how words are pronounced. A lexical repre- sentation must contain a distinctive feature matrix showing the representation of the 'segments' which make up the word. In addition, it must contain any idiosyncratic infor- mation concerning the association of phonological elements at different phonological tiers. For instance, in a tone language, it must show any syllables which are underlyingly linked with a tone in the lexicon and which do not get assigned tone by regular association principles. In a stress language like English, the lexicon may have to show the special effects which a particular suffix has on the stress pattern of the base to which it is attached. As we saw in the last chapter, some suffixes like -ee (in detainee) attract stress to themselves while others, like -ity (in electricity) attract stress to the immediately preceding syllable.

Furthermore, the lexicon must contain a list of forms which are exceptions to particular rules as well as a state- ment of sub-regularities in the language. For example, it needs to show that sheep is exceptional in having no overt marking of plural while words like memoranda, memorabilia, agenda and data, which are borrowed from Latin, form a minor subsystem, where the plural ending -a is used.

The lexicon needs to show the various subclasses to which words belong because some morphological and phonological rules only apply to certain subclasses of words. This is most obvious in the differential treatment of loanwords in many languages. In English, for instance, word-formation rules are sensitive to the distinction between native forms and forms borrowed from Latin or French. Thus, except for the word oddity, only foreign words of Latin or French origin take the noun-forming suffix -ity as in banality, ferocity and community. The same is true at the phonological level. To take one example, the phoneme /$/ only occurs in English words of French origin. Some of these words are fairly recent borrowings such as genre, rouge and beige while others, such as vision,

256 Phonology in the wider context

usual, occasion and pleasure were borrowed from Old French during the Middle Ages.

It is also necessary to include in the lexicon grammat- ical properties of words like [+noun], [+adjective] or [+verb], not only because they are, obviously, relevant to the application of syntactic rules but also because they may be important for certain word-formation and phonological rules. For instance, in English -ly can only be attached to adjective bases to form adverbs (e.g. the adverb wisely from the adjective wise). To take another example, /f/ is realised as [v] when an -s suffix is added after it only where the -s in question represents the noun plural ending (as in wife (singular) and wives (plural). Where the -s suffix represents the genitive ending as in wife's, the final /f/ of the noun is unaffected by the suffix.

Finally, the reason for including semantic information in the lexical representation of words hardly needs explaining. To the lay person, listing word meanings is the function of the dictionary. But semantic information must be included for an additional reason: namely, because it can have a role in both morphological and phonological processes. It has been noted by students of English word- formation, for example, that the negative prefix un-, which occurs in words like uninteresting, is not freely attachable to adjectives at the negative end of an implied evaluative scale. We can form words like un-happy, un-well and un-exciting but not *un-sad, *un-ill or *un-boring.

Our last illustration of a semantic constraint on the applicability of a phonological rule comes from Luganda. Although Luganda allows nasal plus glide clusters e.g. [mwa:na] 'child', [mja:lo] 'ports', [jiwa] 'drink', the cluster [nj] with an alveolar nasal followed by a palatal glide, though not outlawed, is severely restricted in its use. This is because it occurs almost exclusively in taboo words which belong to the semantic area of defecation - words such as [kunja] 'defecate', [kinjo] 'anus' and [munjoBoBo] 'dysen- tery'. Because [nj] has taboo connotations, nowadays speakers tend increasingly to optionally palatalise under- lying / n / wherever it is followed by a palatal glide so that words like [kanja:la] 'immature banana' or [munjongo] 'miserable foreigner' are realised as [kajia:la] and [mujio:jigo] respectively, with a palatal nasal. Thus the offending [nj] sequence is avoided.

Lexical phonology 257

12.2 Lexical phonology

We noted in the last section that traditionally, the lexicon has been regarded as nothing more than an appendix to the grammar which contains the idiosyncratic properties of lexical items and morphemes.

Lexical phonology takes a different stance: the lexicon is recognised as a central component of the grammar which contains not only idiosyncratic properties of words and morphemes, but also regular word-formation and phono- logical rules. It is assumed that word-formation rules of the morphology are directly paired with phonological rules grouped together at various levels. The output of each morphological rule is cycled through the phonology so that the relevant phonological rules of that level are applied to it. One of the main claims of lexical phonology is that both inflectional and derivational word-formation processes can be displayed on a series of linked LEVELS (also called STRATA). This is shown in the diagram in [12.1] which is adapted from Kiparsky 1982a:

[12.1]

LEXICON

MORPHOLOGY

Underived lexical items

< '

Level 1 Morphology

• > Level 2

Morphology

PHONOLOGY

Level 1 Phonology

Level 2 Phonology

SYNTAX POST-LEXICAL PHONOLOGY

258 Phonology in the wider context

The rules of morphology and phonology applying within the lexicon are essentially CYCLICAL. This is because, as we shall see presently, rules are made to apply in a CYCLE first to the root, then outward to the affixes nearest to the root and then again outward to the outer layer of affixes. In other words, in this theory the word can be likened to an onion with the root of the word as the core and LEVEL i as the inner layer, LEVEL 2 as the outer layer and POST-LEXICAL P H O N O L O G Y as the skin on the outside.

During a derivation there is a constant cycling of data through the interlocking phonological and morphological rules at each level:

[12.2]

' ' Word-formation rules

level x

i >

Phonlogical rules

The rules themselves are assumed to have the structure shown in [12.3]:

[12.3] morphology phonology insert A / [Y Z] x —» apply relevant

phonological rule(s) e.g. insert [past]/[fid ] v - » /fid +past/ v - » [fed]

'feed+past'-» 'fed'

12.2.1 Level 1 of lexical phonology

In lexical phonology it is assumed that the lexicon has internal structure. It is not merely a list. Furthermore, it is assumed that the structure of the lexicon is HIERAR- CHICAL. There has been some dispute as to how many structural levels need to be recognised. Halle and Mohanan (1985) argue for a four-level morphology. This is contro- versial. The consensus is that the theory should be as simple as possible. Nothing would be gained by putting each pair of phonological and morphological rules found in a

Lexical phonology 259

language on a level of their own. But determining the minimal number of strata that is sufficient to account for all the wrinkles in the data is not easy. It is a problem which is still being intensively researched. My preference for the simple model with two lexical strata and one post-lexical stratum is not meant to prejudge the outcome of these investigations.

The onion metaphor captures succinctly the essentials of lexical phonology. At the centre of the word there may be an U N D E R I V E D lexical item [12.1]. Underived lexical items consist of a single morpheme. They are forms such as big, boy, girl, talk and soon. No word-formation rule of any kind is used to produce such words. They appear in the lexicon with the phonological, grammatical and semantic properties with which they surface.

Level 1 contains B O U N D M O R P H E M E S like ah- and -duct in abduct or con- and -ate in conjugate which cannot occur independently but must always be attached to some other form.

Some more words containing level 1 bound morphemes are given in [12.4] below. They show one of the important characteristics of level 1 affixes: they tend to be semantically opaque (i.e. their meaning is difficult to pin down):

[12.4] (a) pertain contain detain (b) perceive conceive deceive

(i) Identify the bound morphemes occurring in the words above.

(ii) Identify the base and the affix in each example.

(iii) Does each morpheme you identify have a constant meaning in all the words in which it occurs?

Note: You are not expected to find the answer. This is an open ended question. For further discussion see Aronoff 1976:10 -17-

It is easy to identify the bound morphemes in [12.4], which are all borrowed from Latin. They are the prefixes

260 Phonology in the wider context

per-, con- and de- which were prepositions, and the bases -tain and -ceiue. None of these forms can occur in isolation.

The difficulty is in establishing their exact meaning, independent of some other morpheme with which they combine. For instance, in Latin per- was a preposition meaning 'by, through, by means o f but it is doubtful whether any one of these meanings is always relevant to contemporary English speakers' use of words like pertain and perceive. Com- (of which con- is a variant) meant 'with, together with' but it is doubtful whether modern English speakers using the words contain and conceive are normally in any sense aware of these meanings. Finally, de- meant 'from' or 'concerning' in Latin. Here again it is debatable if these meanings are salient in the minds of English speakers who have not had a classical education.

Most of the affixes put on level i are the ones which, in studies inspired by SPE, were associated with the + boundary. These affixes have a much more intimate relation with the root to which they are attached than level 2 affixes, as we shall see below.

Let us begin by considering inflectional morphology (i.e. grammatically determined alternation in the shape of a word involving categories like number and tense). We shall consider past tense formation in English for our first example. There are several ways of forming the past tense. In the second exercise at the end of Chapter 5 we saw the regular rule which suffixes [d], [t] or [id] to the verb stem. Verb morphology is not altogether regular. Instead of adding that ending, so-called STRONG VERBS undergo changes in the root itself of the kind shown below (ignore the consonants):

[12.5] Present

W [1] sing ring

(b) [ai] fight

(c) [a.] drive write

Past

sang rang

fought

M drove wrote

Perfective ~ [A]

(has) sung (has) rung

~ [0] (has) fought

~ M (has) driven (has) written

Lexical phonology 261

(d) [i] ~ [e] ~ [e] read read (has) read meet met (has) met

The forms in [12.5] show different sub-regularities which a grammar of English needs to capture. Level 1 rules would be used to state the vowel changes which are informally shown above. This illustrates another important property of level 1 rules: they tend to cause drastic changes in the root to which they are attached.

By allowing the verb affixation rules responsible for the alternation in the root itself to apply at level 1, we are able to clear the way for the regular affixation processes which apply to verbs like walk ~ walked ~ (has) walked. Having received their past tense and perfective affixes at level 1, the verbs in [12.5] cannot undergo the more regular verbal affixation processes which apply later at level 2. The theory stipulates that all level 1 rules must precede all level 2 rules which, in turn, precede all post-lexical rules. If a particular rule applies at level 1, it will always have precedence over those rules which are at level 2; if a particular rule is at level 2, it will always precede any rules which apply post-lexi- cally. Thus the ordering of levels has serious implications for the way in which rules interact.

Let us now take another batch of examples:

[12.6] Noun Verb blood bleed food feed tooth teethe bath bathe house house

In [12.6] to derive verbs from nouns, two level 1 rules are applied. One rule, which is applicable to all the exam- ples, changes the vowel. In addition, where the final consonant is a voiceless fricative, another level 1 rule oper- ates, changing it into a voiced fricative.

Interestingly, the voicing of a final fricative is a phono- logical rule which applies elsewhere in level 1 morphology: besides figuring in the derivation of verbs from nouns, it also plays a role in the formation of plural forms of certain nouns such as the following:

262 Phonology in the wider context

[12.7] Singular Plural [f] ~ [v]

hoof hooves wife wives leaf leaves loaf loaves

First, state the level 1 word formation process going on in the examples below. Next state the phonological processes that are accompany it.

[12.8] long length strong broad wide

strength breadth width

The morphological process involved is the derivation of nouns from adjectives. Phonologically the -th suffix is added, triggering off a change in vowel quality from / D / to / e / , / o / to / e / and / a i / to / i / . '

All level 1 affixes are not restricted to affecting the segmental phonology of the forms to which they are attached. They can also affect stress. The discussion of some English stress rules in section 11.1.6 highlighted on the one hand strong mode suffixes like -ity which attract stress to the immediately preceding syllable and on the other hand weak mode suffixes like -al which attract stress to the immediately preceding syllable, only if it is heavy; if the syllable preceding -al is light, stress is placed on the second syllable to the left of -al:

[12.9] (a) Strong mode suffix -ity e/ertric electricity timid timidity account accountable accountability impartial impartiality

(b) Weak mode suffix -al medicine medicinal congress congressional adjective adjectival accident accidental

Lexical phonology 263

In the light of what has been said above, describe the effect of the following suffixes on stress placement:

[12.10] (a) -ic as in phonemic -tide as in patricide -ious as in acrimonious

(b) -ate as in accommodate -an as in American

and academic and insecticide and fallacious and arrogate and Franciscan

The suffixes in [12. ioa] are STRONG M O D E ones: they place stress on any syllable preceding them while those in [12.10b] are WEAK M O D E suffixes and only put stress on the preceding syllable if it is heavy; otherwise they shunt it onto the second syllable to their left.

Certain level 1 suffixes can both affect stress and lead to the modification of the segments in a word. A classic example of this is the suffix -ity:

[l2.ll] Adjective (a) sane

profane (b) serene

obscene (c) virile

senile (d) profound

pronounce

H N H /ao/

Noun sanity profanity serenity obscenity virility senility profoundity pronounciation

/ * /

/ e /

N N

Find one more pair exemplifying the same vowel alternation as that displayed by the pairs in each group above.

Observe that not only does the presence of the strong mode suffix -ity make stress move to the immediately preceding syllable (if it is not already in that position), it also causes the shortening (or laxing) of the diphthong or long (tense) vowel of the root which, as a result, is then realised as the corresponding short (lax) vowel. This vowel shortening rule is called TRISYLLABIC LAXING (SHORTENING) since it only applies to forms with at least three syllables (see section 8.2 page 139).

264 Phonology in the wider context

Trisyllabic laxing w h i c h is a level i rule, only applies to w o r d s formed at level i. A n o n - d e r i v e d form, to w h i c h n o m o r p h o l o g i c a l processes of affixation have applied, is e x e m p t from trisyllabic laxing even w h e n it has three or m o r e syllables and o n the face of it appears to satisfy the structural description of this rule. A w o r d like nightingale [naitirjgeil] w o u l d be p r o n o u n c e d as *[nitirjgeil] if it w e r e subject to trisyllabic s h o r t e n i n g . B u t it is n o t . Nightingale / n a i t m g e i l / is an u n d e r i v e d lexical item entered as such in the lexicon before any level i rules can apply. It is n o t eligible to u n d e r g o trisyllabic laxing w h i c h comes later at level i .

This e x a m p l e is w o r t h dwelling on for it illustrates an i m p o r t a n t principle, n a m e l y S T R I C T C Y C L I C I T Y . In this m o d e l , a phonological rule can only affect those strings of s o u n d s that are p u t t o g e t h e r b y a w o r d - f o r m a t i o n rule applying at the same level. T h u s level i rules m a y only modify structure created b y level i m o r p h o l o g i c a l processes and level 2 rules m a y only change structure created b y level 2 m o r p h o l o g i c a l processes. T h e strict cyclicity principle ensures that rules only apply to one layer at a time. T h e o u t p u t of a m o r p h o l o g i c a l rule at level 1, for instance, cannot u n d e r g o a level 2 phonological rule, and vice versa. A rule is n o t allowed to peep back at a state of affairs preceding arrival at the s t r a t u m w h e r e it applies, n o r is it allowed to glance f o r w a r d to w h a t m i g h t h a p p e n at a later stage in a derivation. In a w o r d , strict cyclicity results in a constrained t h e o r y w h i c h prohibits phonological rules from applying globally. T h e p u r p o s e of i m p o s i n g restric- tions of this k i n d o n the t h e o r y is to ensure that w e do n o t allow the g r a m m a r to h a v e m o r e p o w e r than is absolutely needed to describe the facts of natural language.

Strict cyclicity ensures that phonological rules only have access to m o r p h o l o g i c a l information at the same level. W h e n rules referring to g r a m m a t i c a l i n f o r m a t i o n (which b y c o n v e n t i o n is associated w i t h brackets as in [deny/]verb|all]n0un) apply in the lexicon, the brackets get automatically erased so that subsequent rules at later levels are debarred from referring to that g r a m m a t i c a l information. This principle, w h i c h is k n o w n as the B R A C K E T E R A S U R E C O N V E N - T I O N , means that in o r d e r for a rule to apply, it is n o t necessary to delve i n t o derivational h i s t o r y . For instance,

Lexical phonology 265

the regular plural formation rule in English applies in the same way to an underived noun like bird as it does to a derived noun like denial.

In lexical phonology, the ordering of levels serves a n o t h e r f u n c t i o n . It reflects D E G R E E S O F P R O D U C T I V I T Y . 3 Level i contains the most idiosyncratic morphological and phonological elements. Many level i rules are phonologically idiosyncratic: they tend to have many exceptions; they are rarely used in the formation of new words; the class of items they affect and the ways in which they affect them are usually messy.

We have already noted that strong mode suffixes like -ity and weak mode suffixes like -al which cause stress shift when they are attached to a form all belong to level I. Level i affixes tend to cause more drastic modifications of the base to which they are attached than level 2 affixes. Thus, unlike the level i suffixes -al, -ic and -ity, level 2 suffixes like -er and -ness do not have any effect on stress: words are stressed in the same way regardless of the presence or absence of these suffixes. Thus, in words such as entertain ~ entertainer and aware ~ awareness no stress shift is caused by the addition of the level 2 suffixes -er and -ness.

At the segmental level too, level i rules tend to show little regard for the integrity of the base form to which they are added. As we have seen, the suffix -ity frequently causes a modification of the segments of the base by triggering off trisyllabic laxing [12. n ] . In fact, the disruption caused may be even more drastic. It may include the deletion of part of the root:

[12.12] frivolous frivolity (*frivolousity) anonymous anonymity (*anonymousity) credulous credulity (*credulousity)

In [12.12] when -ity is attached to a form ending in -ous in the lexicon, deletion -ous takes place so that -ity surfaces in the phonetic representation next to the segment which preceded -ous.

But this deletion process is sporadic. It fails in [12.13]:

[12.13] religions religiosity (*religiousity) curious curiosity (*curiousity) precious preciosity (*preci'ousity)

266 Phonology in the wider context

Here the addition of -ity changes the pronunciation of -ous from [as] to [DS]. Clearly, the phonological effects of the addition of this suffix are messy (Aronoff 1976).

The -ity suffix is not unique in having idiosyncratic phonological properties. Look back at [12.5], [12.6], [12.7] and [12.8] and see the changes in the base form due to the presence of various level 1 suffixes. Describe the phono- logical effects of each suffix.

Not only do level 1 affixes tend to be phonologically irregular, they also often tend to have idiosycratic semantic properties. Many level 1 affixes are semantically difficult to pin down. For instance, it is difficult to be specific about the meaning of the suffix -ity per se. True, in [12.11] it seems to derive from an adjective a noun which has the general meaning 'having the quality/state of being X', e.g. serenity is the state of being serene; divinity is the quality/state of being divine, etc. However, in many other cases it is not poss- ible to identify a common thread of meaning which is shared by words containing that suffix:

[12.14] amity 'friendship' annuity 'yearly income' community 'group of people living in one

locality' credulity 'readiness to believe'

Perhaps because of their phonological and semantic irregularity, many level 1 affixes are nothing more than historical relics. They are not productive. And others, though still productive, are only used infrequently to form new words. Thus, the suffix -th in length, width, depth is not used to form new nouns from adjectives in contemporary English, while the suffix -ity discussed above, though still used in word-formation is less productive than the level 2 suffix -ness which it partially overlaps in meaning. The explanation, according to Aronoff (1976), might be that speakers are more confident to use -ness rather than -ity because they are unlikely to get things wrong (and make a fool of themselves) as -ness does not cause any phono- logical changes in the base to which it is attached. More-

Lexical phonology 267

over, unlike the semantically elusive -ity, the meaning of -ness is unproblematic. Its presence consistently produces a noun meaning 'the quality/state X specified by the adjec- tive' (e.g. kindness, illness, etc.).

Likewise, another level 2 suffix, the nominalising suffix -er that derives nouns from verbs (as in writer or maker) causes no phonological complications and has an obvious and uniform meaning of 'person who does whatever X is specified by the verb'.

Study the words in [12.15]. Divide them into morphemes and state the meaning of each morpheme. Comment on any semantically problematic cases which you encounter.

adultery (b) artillery machinery refinery finery nunnery greenery treachery cannery fishery mockery

employ embalm embrace embody enlarge ennoble endure engrave engender encash encase

(c) denial arrival betrayal recital trial *teachal *writal *arrestal *announcal *appointal *sendal

If all morphemes are by definition 'meaningful units', it should be possible to state unequivocally the meaning of morphemes like -ery, -en and -al in [12.15]. But I suspect that you did not find that task altogether straightforward. The suffix -ery has no consistent meaning which recurs in all its occurrences in [12.15]. m cannery, fishery and nunnery it seems to have a locative meaning, but that meaning is absent from treachery, adultery, etc. Although the prefix en- often has the meaning 'to put into or enclose within' as in to encase, attempting to extend that analysis to a word like employ would be questionable. (Note in passing the homorganic nasal assimilation process affecting the prefix in [12.15b]). Finally, the nominalising suffix -al typically has the meaning 'an act of doing X': an act of arriving consti-

268 Phonology in the wider context

tutes an arrival, an act of denying constitutes a denial, etc. But this interpretation is not always valid - an act of appointing does not constitute an *appointal any more than an act of writing constitutes a *writal. There seems to be no good reason for the fact that certain verbs will take a given derivational suffix while others will not.

The problem of erratic semantic behaviour is not restricted to affixes. Many roots to which affixes are attached do not have their own specific and consistent semantic content. It is not possible to say for instance what the roots -trib-, -duce and -tend mean below (unless you happen to know their original Latin meanings: tribut- 'allot, grant', due- 'lead' and tend- 'stretch'). As English morphemes, they are bound forms which never occur in isolation. Their meaning on any one occasion seems to depend on the meanings of the other morphemes with which they co-occur in the word:

[i2.i6](a) contribute (b) conduce (c) contend attribute adduce attend distribute deduce distend retribution reduction retention tributary produce pretend tribute reduce intend

The upshot of this discussioin is that we must abandon the view that the morpheme per se is the minimal mean- ingful unit. The meaning of a root or affix is often unclear until it is put together with other morphemes in a word. The morpheme may be meaningful, as in the case of -er or in the case of a mono-morphemic word like dog, but it need not be. There are many bound roots like -trib- ,-duce and -tend which have no clear, identifiable meaning that remains constant when different affixes are attached to them. Simi- larly, there are many affixes, as we have seen already, to which no steady meaning can be attributed. It is prudent to regard the morpheme primarily as a DISTRI- B U T I O N A L U N I T . What occupies centre stage, as far as signalling word meaning is concerned, is the word rather than the morpheme. Every word found in the lexicon of a language must be meaningful but every morpheme need not be (Aronoff 1976).

Lexical phonology 269

12.2.2 Level 2 of lexical phonology

In the last section we established that level i rules are normally more idiosyncratic than level 2 rules. We saw that often the meaning of level i affixes is unclear, their phono- logical effects unsystematic and their very applicability erratic, with many forms which on the face of it ought to undergo a particular level i rule failing to do so for no apparent reason. Level 2 rules on the other hand have fewer exceptions and their phonological effects and semantic prop- erties are more predictable. Consider again -er, a typical level 2 affix. By suffixing -er to it, we can turn virtually any verb base into an agentive nominal meaning 'doer of activity X designated by the verb':

[12.17] read-er mind-er paint-er speak-er teach-er join-er lead-er cook-er

I say 'virtually' because even very regular level 2 rules can have occasional exceptions. For instance, there are agen- tive nominals which are not formed by suffixing -er. Some agentive nominals are formed by a process known as C O N V E R S I O N or Z E R O SUFFIXATION whereby morphological derivation is achieved without the overt addition of an affix, as in the case of judge („) which is derived from judge („) or bore („) from bore („) or cook („) from cook („).

Yet other agentive nominals are formed by adding a suffix other than -er. That is the case in the formation of the nouns applicant and defendant from the verb apply and defend. These irregular nominalisations take place at level 1 and block the regular level 2 process of adding -er to form agentive nominals.

Another kind of irregularity is exemplified by the word cooker. It takes the regular ending -er, but its meaning is not the expected one of 'a person who cooks', rather it refers to an appliance used to perform the activity of cooking. Joiner is another problematic example. This word normally refers to a craftsman who constructs things by joining pieces of wood but it can also refer to a machine for doing various

270 Phonology in the wider context

kinds of work in wood. This second meaning is less common.

Find more examples of semantic irregularity in word formation. Suggest a way in which the irregularity could be dealt with by the grammar. We shall come back to this problem in the next section.

From a phonological point of view, the presence of the -ex ending has no repercussions: it does not affect stress placement nor does it affect the segments of a word in the way that a level i suffix like -th [12.8] or -ity [12.12] and [12.13] c a n do.

The same applies to other level 2 derivational suffixes such as -jul, -ly, and -ness in [12.18]. There are almost no arbitrary restrictions on the class of forms to which they can be suffixed and they have no quirky phonological or semantic properties:

[12.18] (a) handful (b) kindly (c) kindness cupful sadly illness plateful happily tiredness spoonful badly loudness

12.2.3 The elsewhere condition

The ordering of levels in lexical phonology implicitly deals with the problem of rule interaction which was discussed in Chapter 7 by incorporating the ELSEWHERE C O N - D I T I O N (Kiparsky 1973, 1982b; Koutsoudas et al. 1974). This principle can be stated informally thus:

If two rules compete for the same territory, the more specific rule applies first, blocking the more general rule.

Normally, a more specific rule applies first and later the general rule applies elsewhere. So, putting an inflectional or derivational process at level 1 means that it is subject to a restricted rule which can pre-empt the application of a more general level 2 rule. In inflectional morphology, for instance, the elsewhere condition ensures that the formation

Lexical phonology 271

of the past tense forms of strong (irregular) verbs like those in [12.5] belongs to level 1 while the m o r e general rules deriving the past tense forms of weak (regular) verbs belong to level 2:

[12.19] (a) [sing + past] —> (saerj] (level 1)

(pre-empts regular past tense formation at level 2)

(b) [love + past] (i) (no access to level 1 m o r p h o l o g y : so

miss any level 1 w o r d - f o r m a t i o n rules) (ii) [IAV + d] —* [1Avd] (level 2 regular past)

( l o v e + e d —> loved)

T h e effect of level o r d e r i n g can be seen m o r e clearly by e x a m i n i n g the p h e n o m e n o n of C O N V E R S I O N (or Z E R O S U F F O C A T I O N ) w h e r e b y a w o r d changes its category w i t h o u t the overt addition of an affix, as in a chair („) and to chair („) or a man („) and to man („).

Begin b y n o t i n g that m a n y s t r o n g verbs ending in -ing or -ink follow the pattern of sing ~ sang ~ sung (viz. ring ~ rang ~ rung; sink ~ sank ~ sunk, etc.). Find t w o similar examples). H o w e v e r , w h e n a verb is derived from a n o u n ending in -ink or -ing, the expected irregularity is lost. T h u s , whereas w e say 'ring ~ rang ~ rung' of ringing a bell, w e say the t o w n is 'ringed by mountains' n o t * ' rung by mountains'.

Lexical p h o n o l o g y offers a simple explanation for these facts. T h e s t r o n g verb past tense formation takes place at level 1. T h e derived verb ring, w h i c h comes from a n o u n and has the m e a n i n g 'encircle' does n o t exist at the point w h e n level 1 m o r p h o l o g y takes place because the derivation of verbs from n o u n s occurs at level 2. Verbs derived from n o u n s can only u n d e r g o level 2 past tense inflection in the form of the regular -ed ending.

T h e elsewhere condition is also useful in describing the formation of the plural of n o u n s . T h e regular plural formation rule w h i c h adds -s to form the plural of n o u n s is a level 2 rule while irregular plural formation w h e r e b y n o u n s take endings like -ren as in children or -/ as in cacti or -a as in addenda is a level 1 process. T h e addition of these irregular plural endings at level 1 blocks the application of

272 Phonology in the wider context

the regular plural formation rule. When forms such as [cact+ ipi] or [addend+ api] are produced at level i, it is not possible for the regular level 2 plural formation to apply later, yielding *[cact+i p i+s p i] or *[addend+ ap|. + spl.]

How can we extend our analysis to cover nouns such as equipment, accommodation, news and mumps which are inherently singular, or nouns like people, police, cattle, scis- sors and trousers which are inherently plural?

The case of inherently singular nouns is not proble- matic. They have to be marked explicitly with a diacritic or a [-rule n] feature which shows that they are invisible to the rules that assign plural number. So, a word like equip- ment would have a mark on it saying that it is not available to any plural formation rule.

More interesting is the treatment of the inherently plural nouns. What we want to say is that certain words are unavailable for plural formation rules of any kind because they occur as plural underived lexical items in the lexicon. Kiparsky (1982b) suggests that each lexical entry be deemed to be a 'rule' - a special kind of lexical identity rule where the non-derived lexical item is the input and that same non- derived lexical item emerges as the output. Then the deri- vation of cattle or police, etc. could be handled as in [12.20]:

[12.20] Structural description Structural change [[police] N + pi.] -> [[police]N + P J [[cattle]N + pi] -> [[cattle]N + pi]

When the output of [12.20] reaches levels 1 and 2, no plural formation rule can affect it since it already bears the mark of plural. The elsewhere condition correctly predicts that being more specific, the idiosyncratic plural rules peculiar to certain nonderived lexical items like police take precedence over any more general level 1 or level 2 plural formation rule.

We have seen how level ordering and the elsewhere condition offer a natural way of dealing with rule inter- action in situations where one rule blocks another. Now we are going to see how the theory also provides a mechanism for dealing with rule interaction in cases where rules are in

Lexical phonology 273

a feeding relationship (i.e. w h e r e one rule creates the input to another rule (Chapter 7, page 125)).

T h e t h e o r y predicts the irregular inflection derived at level 1 can form the input to m o r e regular w o r d - f o r m a t i o n processes taking place at level 2. T h i s is b o r n e out by [12.21]:

[12.21](a) p a w - m a r k s (*paws-marks) (b) lice-infested (*louse-infested)

(Find t w o similar examples.)

T h e possibility of level 1 irregular plural formation appearing in c o m p o u n d s suggests that level 1 rules can feed level 2 rules. H a v i n g u n d e r g o n e plural formation at level 1, lice is still available for c o m p o u n d i n g later at level 2 in the formation of lice-infested. U n f o r t u n a t e l y , the facts d o n o t always give unequivocal s u p p o r t to this analysis. W e w o u l d expect other irregular plurals formed at level 1 such as feet and teeth to appear in c o m p o u n d s as *feet-steps and *teeth- brushes b u t instead w e find footsteps and tooth-brushes. S o m e m o r e investigations are needed to deal w i t h the p r o b l e m .

12.2.4 The order of affixes

In (12.2) w e used the o n i o n m e t a p h o r to describe w o r d structure in lexical p h o n o l o g y . W e return to that m e t a p h o r n o w and see h o w it sheds light on rule interaction in w o r d formation by capturing the generalisation that level 1 affixes are placed nearer the r o o t than level 2 affixes:

[12.22] [[level 2 affi] [level 1 aff.] r o o t [level 1 a f f ] [level 2 aff.]]

C o n s i d e r the nominalising suffix -ian as in politician, electri- cian, grammarian and librarian, (and also the adjective- forming suffix -ian as in Olympian, Aeolian, Bavarian, Edwardian, Dickensian and sesquipedalian) as well as the nominalizing suffix -ant as in applicant, inhabitant, attendant and communicant. T h e s e are all level 1 suffixes. Each one of t h e m attracts stress to the syllable i m m e d i a t e l y preceding it in the same w a y that the level 1 suffixes -ic and -ity described earlier d o . O n the other hand, the suffix -ism as in baptism and communism, is a level 2 suffix. It does n o t modify the stress of the base to which it is attached.

274 Phonology in the wider context

The theory predicts that a level I affix will be nearer to the root than a level 2 suffix in cases where both occur in the same word:

[12.23] Shakespear-w«-ww *Shakespear-»5m-w« antiquar-ujn-iim *antiquar-tsm-ian protest-ant-ism *protest-ism-ant

A suffix at a given level can be followed by other suffixes at the same level as in [12.24a] where all the affixes in question are at level 1, or [12.24b] where all the suffixes are at level 2 :

[12.24] (a) pub\-ic-ity pm-if-ic-at-ion

(b) re-re-make czre-jul-ness

Determine which of the suffixes in [12.25] are at level 1 and which are at level 2. Does this have a bearing on the ordering of suffixes in a word?

[12.25] electrification leaderless

In electrification the suffixes -if(y) -ic-(at)ion are all level 1 suffixes. This can be verified by seeing the way in which, like numerous level 1 suffixes, they all attract stress: e^lectrijy; electric and electrification. But in leaderless both suffixes are at level 2. They have no phonological effect on the base. Moreover, their meaning is reasonably clear and consistent as is the norm for level 2 affixes.4

12.2.5 Post-lexical rules

After all lexical rules have applied, words can be inserted in syntactic representations to form sentences. Once sentences have been formed, level 1 and level 2 rules of lexical phonology cannot apply any more. But that does not necessarily render words secure from all phonological modification. The sounds of individual words can still be modified by rules which apply in C O N N E C T E D

Lexical phonology 275

SPEECH, affecting phrases or longer chunks of utterances. In this model such rules are called POST-LEXICAL RULES. They are late rules which apply after all lexical rules (see figure [12.i]). Unlike lexical rules, post-lexical rules are not intrinsically cyclic. The output of a lexical phonological rule is not submitted to the morphology or vice versa. Nor do they respect the strict cyclicity principle (section 12.2.1). Unlike lexical rules, post-lexical rules are not restricted to forms derived at the same stage in a deri- vation. They can affect forms derived at an earlier stage.

For our first illustration of a post-lexical rule we shall consider the floating tone in associative constructions in Mbui described by Hyman and Tadadjeu (1976: 61) which was introduced in [10.21]. The data are repeated here for convenience as [12.25].

[12.26] M b u i : (a) bakoa 'crabs' bandum 'husbands' (b) bakaa ba sag 'the crabs of the bird'

bandiim ba sag 'the husbands of the bird'

In isolation both 'crabs' and 'husbands' have L-H tone. When, however, in the associative construction these words come together with a high tone word like [sag] as the second noun, the associative marker [ba] is realised with its underlying high tone after 'crabs' but is downstepped to 'H after 'husbands'. Obviously, this downstepping is a post- lexical rule since it only occurs in an associative phrase and is not present in words in isolation.

For our next example of a post-cyclical rule we shall turn to Chi-mwini, a Bantu language of the Horn of Africa. Kenstowicz and Kisseberth (1977: 86) report that in this language all words have a short final vowel when they occur in isolation:

[12.27] xpala 'to scale' rjguwo 'clothes' kuja 'to eat' chisu 'knife' navaile 'that he dress' mashu:rjgi 'hair' hujo 'one who eats' husoimo 'one who reads'

But all word final short vowels are lengthened if they occur in phrase medial position in a sentence:

276 Phonology in the wider context

[12.28] xpalainsi kuja: nama navaleirjguwo hujo:zijo husomo:chuwo rjguwo:mphiya chisMichile mashurjgiimale

to scale fish' to eat meat' that he put on clothes' one who eats zip one who reads' new clothes' knife' long hair'

Suggest a rule to account for the alternation in vowel length in the words: navaile, husoimo and mashuir/gi when they occur phrase medially.

A long vowel is shortened when it is followed by another long vowel within the same phonological phrase (marked by a tall bracket) as in navaleiQguwo:

V [12.29] V -> [-long]/— C0[long]_

In addition to rules which apply only either lexically (within lexical items), or only post-lexically (across word boundaries), there exist other rules which apply anywhere their structural description is satisfied. Such rules are purely phonetically motivated. They get automatically triggered off wherever auspicious phonetic circumstances occur. An example of such a rule is the flapping rule in American English:

V [12.30] / t / —• [r] / V [- stress]

/ t / is realised as [r] (a voiced consonant which sounds like a very short [d]) whenever it occurs intervocalically at the beginning of an unstressed syllable.

This automatically happens within a single word like [pin] 'pity' or across a word boundary as in [gent] 'get it'.

The glottalisation rule that applies in broadly similar environments in non-standard British English, producing a gjottal shop allophone of / t / , is equally unrestricted. The phoneme / t / has allophone p ] when it occurs either in intervocalic position followed by an unstressed vowel or in

Lexical phonology 277

word-final position. Glottalisation applies lexically (within a single lexical item) as in [pi?i] ' p i t y ' , and post-lexically (i.e. across a w o r d b o u n d a r y ) as in [ge'i?] 'get it'.

T h e typical fast casual speech p h e n o m e n a of elision and assimilation w h i c h occur in m a n y languages are m o s t l y post-lexical. For instance, English consonants, in particular alveolar ones, tend to assimilate to the place of articulation of the following consonant, even across a w o r d b o u n d a r y :

[12.31] (a) bad b o y (b) [baeb boi] g o o d girl [gud gal] nice shoe [naij J"u]

Elision of w o r d final consonants is also c o m m o n p o s t - lexically, especially w h e r e a w o r d which ends in an alveolar consonant is i m m e d i a t e l y followed by a another w o r d beginning w i t h a c o n s o n a n t . This can be seen by contrasting the lexical and post-lexical forms in [12.32]:

[12.32] (a) lexical representation best m a n / b e s t maen/ r o u n d peg / r a u n d p e g / h a n d grenade /haend g r a n e i d /

(b) post-lexical (connected speech) representation [bes maen] [raon peg] (—» [raum peg]

if elision optionally feeds assimilation)

[haen graneid] (—> [[risen graneid] if elision optionally feeds assimilation)

An i m p o r t a n t difference b e t w e e n lexical and p o s t - lexical rules is that the former can take into account lexical bracketing: they are sensitive to w o r d boundaries. W e saw in the last chapter that lexical rules which assign stress to w o r d s m a y refer to the first or last or penultimate syllable as the case m a y be. Lexical rules can also look at w o r d internal m o r p h o l o g i c a l bracketing i n t r o d u c e d at the same pass t h r o u g h the lexicon - see the trisyllabic laxing rule in section 12.2.1 and D a h l ' s law in K i r u n d i w h i c h only applies when a prefix is i m m e d i a t e l y followed by a r o o t (section 5-4).

But post-lexical rules cannot do that. T h e y cannot refer

278 Phonology in the wider context

to word internal bracketing: they cannot refer to elements like 'stem' or 'affix' because each time lexical rules apply morphological bracketing relevant to their application is erased by the bracket erasure convention. Consequently, once we leave the lexicon we cannot have any access to the internal organisation of words.

Another interesting difference between lexical and post- lexical rules is that while the former always preserve the canonical morpheme structure of a language, post-lexical rules need not always do so. For instance, in a language where in the underlying representation all syllables and words end in vowels the output of a rule of lexical phon- ology must be an open syllable. However, a post-lexical rule may create syllables which violate that condition. Syllable structure found in casual fast speech in many languages differs substantially from canonical syllable structure.

Thus in English 'prohibited' consonant sequences may occur in the phonetic representation in these circumstances: underlying voiced consonants which occur in unstressed syllables may assimilate to an adjacent voiceless consonant and become voiceless when a weak [a] or [i] vowel following them is deleted. This process, which is especially common where grammatical words like the and of are present, may combine with consonant elision where appro- priate, to create syllable structure which is radically different from the normal syllable structure of English. Consider these examples from Brown (1977: 69):

[12.33] UR PR /'sauld ta Sa ['sau/f0 'pAblik] 'sold to the pAbhk/ public' /'bsegk av 'irjgbnd/ ['baeflkf 'iijgland] 'Bank of

England' /5a 'f3st 'raund/ ['0f3s 'raund] 'the first round' /'Saets 8anjuz/ ['5sts6 'njuz] 'that's the news'

The consonant sequences U6, gfef, 0f and ts6 found in [12.33] are not typical of English. They are not allowed to follow a vowel in the same syllable in the underlying representation.

N o w take Luganda, a language where in lexical repre- sentations all canonical syllables are open and where the only consonant sequences allowed are those where a nasal is

Lexical phonology 279

followed by another consonant, as in mbwa 'dog' or nte 'cow'. Notwithstanding the fact that all canonical syllables are open, in fast, casual connected speech vowels can be optionally dropped, giving rise to a number of consonant clusters disallowed at the underlying level:

[12.34] (a) lexical representation situka 'stand up' fukamila 'kneel down' kasikonda 'hiccough' mafuta 'oil'

(b) post-lexical (connected speech) representation [stuka] 'stand up' [ffeamira] 'kneel down' [kajfeonda] 'hiccough' [mafta] 'oil'

The rule required states that vowels are optionally dropped between a voiceless fricative and a following voice- less consonant.

[12.35] V ^ 0 +cont —son —voice

—cont —voice

Finally, note that while lexical rules are exception ridden, post-lexical rules are free from exceptions. Post- lexical rules apply across the board wherever their structural description is met, without being subject to various restric- tions. You can witness this by considering glottalisation of / t / in non-standard British English or flapping of / t / in American English (see [12.30] above).

12.2.6. Summary

I shall end ths part of the chapter with a summary based on Pulleyblank (1986):

[12.36] (i) At each level (stratum), morphological rules are paired with phonological rules,

(ii) Lexical rules are cyclic: the output of each set of word-formation rules is submitted to the phonological rules of that level. Furthermore, lexical rules only apply to

280 Phonology in the wider context

words formed at the pass through the lexicon at which they apply.

(iii) The ordering of strata (levels) determines the sequencing of morphological processes in word-formation: level one rules precede level two rules; lexical rules precede post-lexical rules.

(iv) The output of each lexical stratum of derivation is a word.

(v) The inventory of lexical items of a language is the output of the morpho- logical and phonological rules of the different levels put together.

(vi) Post-lexical rules are not cyclic. (vii) A sharp distinction is drawn between

LEXICAL and POST-LEXICAL rules. The former have access to word-internal structure, are structure preserving, cannot apply across word boundaries, apply cyclically and have exceptions; the latter have none of these properties.

12.3 Prosodic domains

We said above that post-lexical rules apply to the output of the syntax after words have been put together to form sentences. A key theoretical issue that arises concerns the precise nature of the syntactic information which post- lexical phonological rules need to have access to. In SPE the assumption was that phonological rules apply to the output of the syntactic component without being sensitive to syntactic information; these two components of the grammar were thought to be independent of each other and to function without referring to each other closely. In recent years, however, that position has been called into question.

Evidence has been amassed showing that there exists an intimate relationship between phonology and syntax. Post-lexical phonological rules need to know many things about syntactic structure since, in languages as diverse as Ewe, Italian, Spanish and Luganda, certain rules only apply

Prosodic domains 281

in specific synactic environments (Clements 1978; Nespor and Vogel 1982, 1986; Hyman et al. 1987).

The two key issues are:

(a) What syntactic properties are post-lexical phonological rules sensitive to?

(b) In what form does phonology have access to those syntactic properties?

For different languages different syntactic properties have been shown to play a role in phonology . Having recognised the relevance of syntactic information we could proceed to incorporate that information directly into the phonology using features like [+relative clause], [+impera- tive], etc. But as the same syntactic features tend to recur in rule after rule in a particular language such a solution would fail to capture any important generalisations - it would be as unrevealing as it would be repetitive.

That pitfall is avoided if we incorporate into phono- logical theory P R O S O D I C D O M A I N S within which phonological rules apply. Post-lexical rules would then refer to prosodic domains and not directly to syntactic features which determine them.

In the case of English, Selkirk (1984: 26) suggests that the prosodic hierarchy includes at least the following categories:

[ 1 2 . 3 7 ] Utterance

(IP) (IP = Intonational Phrase)

(PhP = Phonological Phrase)

(Wd = Prosodic Word)

(Ft = Foot)

Syl Syl. (Syl = Syllable)

Prosodic domain theory makes the prediction that the domains are arranged in a hierarchy which is strictly

282 Phonology in the wider context

adhered to. Domains never overlap. Lower domains are not allowed to include within them elements which belong to higher domains: lower domains form constituents of the higher ones, e.g. phonological phrases contain words; words contain feet; feet contain syllables. The reverse is not envisaged: syllables cannot contain feet, which in turn contain words, which contain phonological phrases.

An introductory book of this kind is not the place to explore the complex topic of prosodic domains in depth. All I shall do is to briefly air some of the issues at stake. I shall do so by outlining the role of syntactically determined phonological phrase domains in the application of liaison in French.

12.3.1 Prosodic domains and French liaison

The essential principles of liaison in French were introduced in Chapter 9 (page 184). Recall that liaison occurs when a word ending in a consonant is followed by another word which begins with a vowel. Re-syllabification takes place so that the consonant becomes the onset and the vowel the nucleus of the new syllable. The effect of this is to preserve open, CV syllable structure as in [12.38a] where the article is followed by a vowel-commencing noun. But there is no liaison when the article is followed by a noun beginning with a consonant as in [12.38b]:

[12.38] (a) le5 amis 'the friends' le5 endroits 'the places, spots'

(b) les copains 'the pals' les places 'the places'

Liaison (here marked by underlining the appropriate consonant and vowel) can take place in many other gram- matical contexts as you can see below:

[12.39] mow ami (Poss. + N) 'my friend' en bort anglais (Prep. + Adj. 4- 'in good

N) English' cet oMvrier (Dem. + N) 'this worker' il es£ alle (Aux + V) 'he is gone'

It would be possible to formulate the rule in such a way that it makes direct reference to the relevant grammatical infor-

Conclusion 283

mation, e.g. plural definite article followed by a noun, possessive followed by a noun, adjective followed by a noun etc. But that would miss the important generalisation that in all cases of liaison, the forms involved belong to the same phonological phrase (Selkirk 1978).

Interestingly, the domain of the phonological phrase is syntactically determined. It is governed by X-bar syntax.6

Liaison only applies where the appropriate consonant and vowel belong to the same major syntactic constituent, such as a noun phrase or verb group, whose elements share the same head. In [12.39] where the constructions in question belong to the same phonological phrase liaison occurs but in [12.40] where the potential candidates for liaison (marked below by = under the consonant) belong to different phono- logical phrases, liaison fails to apply:

[12.40] II part a six heures. 'He leaves at six o'clock' Est-ce le matin ou le 'Is it m o r n i n g or soir? evening?' M. Dupont a deux 'Mr Dupont has two fils. sons' Jean etait ainsi occupe 'John was thus occupied'

French liaison is not a rule that applies blindly anywhere. In addition to requiring the right segmental phonological input, it also requires that the sounds belong to the same phonological phrase domain. Membership of a phonological domain is in turn determined by X-bar syntax. Clearly, the relationship between phonology and syntax is intimate.

12.4 Conclusion

This introduction to modern phonological theory has not been completely impartial. My selection of topics has been influenced to some extent by my own interests and to a greater degree by the emerging consensus among generative phonologists over the last twenty years. I have made no attempt to compare the merits of alternative phonological theories. Others have done that. There are already several excellent surveys of twentieth century phonology available

284 Phonology in the wider context

(cf. Fischer-Jorgensen 1975; Sommerstein 1977; Anderson 1985).

My aim has not been to provide an encyclopedia of established phonological facts. The state of flux in which contemporary phonology finds itself precludes that. What I have attempted to do is to introduce you through the examination of a considerable amount of data from a wide range of languages, to one coherent model in which important phonological questions are being raised - and in a significant number of cases answered satisfactorily.

Exercises

1 .

(a)

(b)

(c)

(b)

2 .

(a)

U s i n g the conventions of lexical p h o n o l o g y , s h o w h o w the following inflectional processes could be dealt with: Singular plural addendum addenda agendum stimulus cactus mouse louse d o g cat bitch

S t u d y the language 0 long vowel

kikoloomb

mikaate

l y k a a m b a

kituumbili

m b o o p o

agenda stimuli cacti mice h'ce d o g s cats bitches

following data from f Tanzania ( O d d e n 1 ( s h o w n by VV)

» 'cleaning shell'

' l o a f

' s t r i n g '

' m o n k e y '

' m a c h e t e '

K i m a t u m b i , a Bantu 987):

short vowel ( s h o w n by V) k i k o l o m b e changu ' m y cleaning shell'

m i k a t e m i k u l u 'large loaves'

l u k a m b a l w a l u p u w a a n i j k e 'string w h i c h b r o k e '

kitumbili ywaawiile ' m o n k e y w h o died'

m b o p o ye ' t h e m a c h e t e '

Conclusion 285

(b) [kikoloombelNP [chapuwaaniike]Vp shell broke 'The shell broke'

[naampei [kik61oombe]NP [Mamboondo]Ni>]vi' I-him-gave shell Mamboondo 'I gave Mamboondo the shell'

[naakibweni [kikoloombe]NP liilj] VP I-it-saw shell NEG 'I didn't see the shell'

(c) [naampei [ kikolombe kikulu]Np]vp I-him-gave shell large 'I gave him a large shell' [naampei [kik61ombe]NP [ 0 kikulu]Np]vp I-him-gave shell large 'I gave the large (thing) a shell'

[naampei [kikoloombe]NP [cha [0[ywaangu]Np]pp]Np]vp I-him-gave shell of mine 'I gave him the shell of my (human)'

(d) [kikolombe kikeele chaagu]NP 'my red shell' shell red my *kikolombe kikele chaangu [ikolombe yadngu yanaanchima]NP 'my many shells' shells mine many *ikolombe ydngu yanaanchimaJNP

Note: N P — noun phrase, VP = verb phrase, PP = prepositional phrase

(a) At what point in a derivation does shortening apply? (b) Referring to syntactic structure describe in detail the

circumstances in which long vowels are: (i) shortened and (ii) not shortened.

3- Referring to the appropriate prosodic domain, state the rule that regulates glottalisation of / t / so that it is realised as a glottal stop in nonstandard British English in the following data. Glottalisation is marked by underlining; Glottalisation of intervocalic ftj

(i) Tom and his caf a(e \t. (ii) I let it in.

(iii) Thaf is the water bofrie in which the bu«on fell.

286 Phonology in the wider context

Notes

1. The vowel alternation between e ~ o is a historical relic which goes back all the way to Proto-Indo-European. The name for it is ABLAUT. It is found in other languages of this family e.g. Greek lego 'I read' but logos 'word' and Latin tegere 'to cover' but toga 'toga i.e. outer garment'

2. Halle and Mohanan (1985) have relaxed this restriction and allowed 'looping back' from a later stratum to an earlier one in a derivation in certain circumstances.

3. If you can, now read Aronoff (1976: 7-17) for an elucidation of the concepts of 'word', 'morpheme' and 'productivity'.

4. The claim that level 1 affixes are always closer to the root than level 2 suffixes is probably too strong. There are cases where a level 2 affix is nearer to the root than a level 1 affix. For instance, in a word like desirability, the suffix -abil- which is a level 2 suffix (witness its semantic and phonological predictability) is closer to the root than -ity which is a level 1 suffix. We need to adopt a weaker position: in unmarked cases level 1 affixes are nearer the root than level 2 affixes.

5. The syntactic properties considered relevant to phon- ology have included X-bar categories, traces, c- command and labelled brackets carrying any syntactic features, e.g. definiteness, relative clause, past tense, which play a role in phonology (see Selkirk 1984, Hyman et al. 1987, Kaisse 1985, Nespor and Vogel 1982, 1986). Labelled brackets offer the most flexible and at the same time least constrained way of incor- porating syntactic information. Ideally their use should be curtailed. The concept of X-bar which is the one we shall be using in this book is explained in the note below.

6. The phrasal categories within which liaison applies are known in syntax as X-bar domains. The head of any syntactic phrase (e.g. a noun phrase) is referred to as X. The phrasal category containing X (e.g. a noun) is called X (or X ' ) , the phrasal category containing X is

Conclusion 287

in turn referred to as X (or X " ) . An example will clarify this. A phrase like a very clever idea can be represented thus:

Determiner N

Adj. Phrase N

Degree Adj

very clever idea

As you can see X and X are 'projections' of X. By definition the head of a noun phrase must be N (a noun). There is an^intermediate category N which is contained in the N which corresponds to the noun phrase. The fact that features of X are projected throughout the phrasal domain of which it is head means that a language may not have a noun phrase which has a verb or preposition as its head. (See Radford 1981: 79-117 or Riemsdijk and Williams 1986: 39-54-)

Bibliography

Abercrombie, D. (1967) Elements of general phonetics. Edinburgh University Press.

Anderson, S. R. (1974) The organization of phonology. Academic Press, New York.

Anderson, S. R. (1985) Phonology in the twentieth century: theories of rules and theories of representations. University of Chicago Press, Chicago.

Archangeli, D. and Pulleyblank, D. (1986) The content and structure of phonological representations. MS, University of Arizona and University of Southern California.

Argyle, M. (1974) Bodily communication. Methuen, London. Aronoff, M. (1976) Word formation in generative grammar. MIT

Press, Cambridge, Mass. Bell, A. and Hooper, J. B. (eds) (1978) Syllables and segments.

North-Holland, Amsterdam. Bendor-Samuel, J. (ed.) (1974) Studies in Nigerian languages. No. 4:

Ten Nigerian tone systems. Institute of Linguistics, Jos and Centre for the Study of Nigerian Languages, Abdullahi Bayero College, Ahmadu Bello University, Kano, Nigeria.

Bloch, B. (1941) 'Phonemic overlapping'. American speech 16, 278-84

Bloomfield, L. (1926) 'A set of postulates for the science of language'. Language 2, 154-64

Bloomfield, L. (1933) Language. Holt, New York. Bolinger, D. (1965a) Forms of English: accent, morpheme, order.

Harvard University Press, Cambridge, Mass. Bolinger, D. (1965b) 'Pitch accent and sentence rhythm'. In

Bolinger 1965a. Brown, G. (1972) Phonological rules and dialect variation: a study of

the phonology of Lumasaaba. Cambridge University Press. Brown, G. (1977) Listening to spoken English. Longman, London. Chapman, W. H. (1966) Introduction to practical phonetics. 3rd edn

1978, Summer Institute of Linguistics. Horsleys Green, High Wycombe.

Bibliography 289

Chomsky, N. and Halle, M. (1968) The Sound Pattern of English. Harper and Row, New York. (This is normally cited as SPE.)

Clements, G. N . (1976) 'Vowel harmony in nonlinear generative phonology: an autosegmental model'. Indiana University Linguistics Club, Bloomington.

Clements, G. N . (1978) 'Tone and syntax in Ewe'. In Napoli D. J. (ed) Elements of tone, stress and intonation. Georgetown University Press, Washington, D C .

Clements, G. N . (1985) 'Akan vowel harmony: a nonlinear analysis'. In Goyvaerts, D. (ed) African Linguistics: essays in memory of M. W. K. Semikenke. John Benjamins. Amsterdam.

Clements, G. N . and Ford, K. C. (1979) 'Kikuyu tone shift and its synchronic consequences'. Linguistic inquiry 10, 179-210

Clements, G. N . and Goldsmith, J. (eds) (1984) Autosegmental studies in Bantu tone. Foris Publications, Dordrecht.

Clements, G. N . and Keyser S. J. (1983) CV phonology. MIT Press, Cambridge, Mass.

Cope, A. T. (1970) 'Zulu tonal phonology'. Journal of African Languages 9, 111-152

Crothers, J. (1971) 'On the abstractness controversy'. Project on linguistic analysis n - 1 2 , C R 1 - C R 2 - 2 9 . University of Cali- fornia, Berkeley.

Cruttenden, A. (1986) Intonation. Cambridge University Press. De Courtenay, B. (1894) Versuch einer Theorie phonetischer Alter-

nationen. (German translation) Strassburg, 1895. Dimmendaal, G. J. (1983) The Turkana language. Foris Publi-

cations, Dordrecht. Dinnsen, D. A. (ed) (1979) Current approaches to phonological theory.

Indiana University Press, Bloomington. Dodds, R. W. (1977) Malay. Teach Yourself Books, Hodder and

Stoughton. Donaldson, T. (1980) Ngiyambaa : Language of the Wangaaybuwan.

Studies in Linguistics, Cambridge University Press. Doncgan, P. J. and Stampe, D. (1979) 'The study of natural

phonology'. In Dinnsen 1979. Elimelech, B. (1976) 'A tonal grammar of Etsako'. UCLA working

papers in phonetics 35 Firth, J. R. (1948) 'Sounds and Prosodies'. Transactions of the Phil-

ological Society. Reprinted in Palmer 1970. Fischer-Jorgensen, E. (1975) Trends in phonological theory: a histori-

cal introduction. Akademisk Forlag, Copenhagen. Fromkin, V. (1970) 'The concept of "naturalness" in a universal

phonetic theory'. Glossa 4, 29-45 Fromkin, V. (ed) (1978) Tone a linguistic survey. Academic Press,

New York. Fudge, E. C. (1969) 'Syllables'. Journal of linguistics 5, 253-86

290 Bibliography

Fudge, E. C. (1984) English word stress. George Allen & Unwin. Goldsmith, J. (1976) Autosegmental phonology. MIT doctoral disser-

tation. (Published by Garland, New York, I979-) Goldsmith, J. (1979) 'The aims of autosegmental phonology'. In

Dinnsen 1979. Grammont, M. (1961) Traite pratique de prononciation francaise.

Nouvelle edition, Delagrave, Paris. Halle, M. and Clements, G. N . (1983) Problem book in phonology.

MIT Press, Cambridge, Mass. Halle, M. and Mohanan, K. P. (1985) 'Segmental phonology of

modern English'. Linguistic inquiry 16, 57—116 Halle, M. and Vergnaud, J.-R. (1980) 'Three dimensional phon-

ology'. Journal of linguistic research 1, 83-105 Halliday, M. A. K. (1967) Intonation and grammar in British English.

Mouton, The Hague. Halliday, M. A. K. (1970) A course in spoken English: intonation.

Oxford University Press, Oxford. Harms, R. T. (1968) Introduction to Phonological Theory. Prentice-

Hall, Englewood Cliffs, New Jersey. Harris, J. W. (1969) Spanish phonology. MIT Press, Cambridge,

Mass. Harris J. W. (1983) Syllable structure and stress in Spanish. MIT

Press, Cambridge, Mass. Hayes, B. (1981) A metrical theory of stress rules. MIT doctoral

dissertation. (Available from Indiana University Linguistics Club, Bloomington.)

Hayes, B. (1982) 'Extrametricality and English stress'. Linguistic inquiry 13, 227-76

Hayes, B. (1983) 'A grid-based theory of English meter'. Linguistic inquiry 14, 357-93

Hayes, B. (1984) 'The phonology of rhythm in English'. Linguistic Inquiry 15, 33-74

Hayes, B. (1986) 'Assimilation as spreading in Toba Batak'. Linguistic inquiry 17, 467-99

Herbert, R. K. (1974) 'Seven will get you. five: Luganda vowels'. Paper presented at the fifth conference on African linguistics, Stanford University.

Hockett, C. F. (1947) 'Problems of morphemic analysis'. Language 23, 321-47

Hockett, C. F. (1955) Manual of phonology. Indiana University Press, Bloomington.

Hoffmann, C. (1963) A grammar of the Margi language. Oxford University Press.

Hogg, R. and McCully, C. B. (1987) Metrical phonology: a course- book. Cambridge University Press.

Hombert, J.-M. (1974) 'Universals of downdrift: their phonetic

Bibliography 291

basis and significance for a theory of tone'. In Schuh, R. (1978) 'Tone rules'. In Fromkin (1978).

Hooper, J. B. (1972) 'The syllable in phonological theory'. Language 48, 525-40

Hooper, J. B. (1976) An introduction to natural generative phonology. Academic Press, New York.

Hyman, L. M. (1970) 'How concrete is phonology?' Language 46, 58-76

Hyman, L. M. (1973) 'The role of consonant types in natural tonal assimilations'. In Hyman, L. M. (ed) Consonant types and tone. Southern California Occasional Papers in Linguistics 1. Department of Linguistics, University of Southern California, Los Angeles.

Hyman, L. M. (1975) Phonology: theory and analysis. Holt Rinehart Winston, New York.

Hyman, L. M. (1985) A theory of phonological weight. Foris Publi- cations, Dordrecht.

Hyman, L. M., Katamba, F. and Walusimbi, L. (1987) 'Luganda and the strict layer hypothesis'. Phonology yearbook 4, 87-108

Hyman, L. M. and Pulleyblank, D. (1987) ' O n feature copying: parameters of tone rules'. In Hyman, L. M. and Li, C. N . (eds) (1987) Language Speech and Mind: studies in honor of Victoria A Fromkin. Croom Helm.

Hyman, L. M. and Tadadjeu, M. (1976) 'Floating tones in Mbam-Nkam'. In Hyman, L. M. (ed) Studies in Bantu tonology. Southern California Occasional Papers in Linguistics 3. Depart- ment of Linguistics, University of Southern California, Los Angeles.

Introduction to Phonemic Analysis. 1980 Summer Institute of Linguistics. Horsleys Green, High Wycombe.

Jaeggli, O . (1980) 'Spanish diminutives'. In Nuessel, F. H., Jr (ed) Contemporary studies in Romance languages. Indiana University Linguistics Club, Bloomington.

Jakobson, R., Fant, G. and Halle, M. (1952) Preliminaries to speech analysis: the distinctive features and their correlates. MIT Acoustics Laboratory. 5th printing 1963, MIT Press, Cambridge Mass.

Jakobson, R. and Halle, M. (1956) Fundamentals of language. Mouton, The Hague.

Jones, D. (1931) ' O n phonemes'. Travaux du cercle linguistique de Prague 4, 74-9.

Joos, M. (1957) Readings in Linguistics 1. University of Chicago Press, Chicago.

Kahn, D. (1976) Syllable-based generalizations in English phonology. MIT doctoral dissertation. (Published by Garland, New York 1980.)

292 Bibliography

Kaisse, E. M. (1985) Connected speech: the interaction of syntax and phonology. Academic Press, New York.

Kaisse, E. M. and Shaw, P. A. (1985) 'On the theory of lexical phonology'. Phonology Yearbook 2, 1-30

Kaye, J. D. (1971) 'Nasal harmony in Desano'. Linguistic inquiry 2, 37-56

Kennedy, B. H. (1948) The revised Latin primer. Longmans, Green and Company, London.

Kenstowicz, M. and Kisseberth, C. (1977) Topics in Phonological Theory. Academic Press, New York.

Kenstowicz, M. and Kisseberth, C. (1979) Generative Phonology. Academic Press, New York.

Kiparsky, P. (1968) 'How abstract is phonology?' Indiana Univer- sity Linguistic Club, Bloomington.

Kiparsky, P. (1973) 'Elsewhere in phonology'. In Anderson, S. R. and Kiparsky, P. (eds) A festschriji for Morris Halle. Holt Rine- hart Winston, New York.

Kiparsky, P. (1979) 'Metrical structure assignment is cyclic'. Linguistic inquiry 10, 421—41

Kiparsky, P. (1982a) 'From cyclic phonology to lexical phon- ology'. In van der Hulst H. and Smith, N. (eds) (1982a).

Kiparsky, P. (1982b) 'Lexical morphology and phonology'. In Yang, I-S (ed) Linguistics in the morning calm. Hanshin, Seoul.

Kiparsky, P. (1985) 'Some consequences of lexical phonology'. Phonology Yearbook 2, 85-138

Knowles, G. (1987) Patterns of spoken English. Longman, London. Koutsoudas, A., Sanders, G. and Noll, C. (1974) ' O n the appli-

cation of phonological rules'. Language 50, 1-28 Ladd, D. R. (1980) The Structure of intonational meaning. Indiana

University Press, Bloomington. Ladefoged, P. (1971) Preliminaries to linguistic phonetics. University

of Chicago Press, Chicago. Ladefoged, P. (1975) A course in phonetics. 2nd edni982. Harcourt

Brace Jovanovich, New York. Larsen, R. S. and Pike, E. V. (1949) 'Huasteco intonations and

phonemes'. Language 2, 268-277 Lass, R. (1984) Phonology: an introduction to basic concepts.

Cambridge University Press. Laver, J. (1980) The phonetic description of voice quality. Cambridge

University Press. Laver, J. and Hutcheson, S. (eds) (1972) Face to face communication.

Penguin, Harmondsworth. Leakey, L. S. B. (1959) First Lessons in Kikuyu. East African

Literature Bureau, Nairobi. Leben, W. (1973) Suprasegmental phonology. MIT doctoral disser-

tation. (Published by Garland, New York.)

Bibliography 293

Leben, W. (1978) 'The representation of tone'. In Fromkin 1978. Liberman, M. and Prince, A. (1977) 'On stress and linguistic

rhythm'. Linguistic inquiry 8, 249-336 Lightfoot, D. (1982) The language lottery. MIT Press, Cambridge,

Mass. Lightner, T. M. (1965) 'On the description of vowel and conson-

ant harmony'. Word 19, 376-87 Lyons, J. (1977) Semantics 1. Cambridge University Press. McCarthy, J. (1979) Formal problems in Semitic phonology and

morphology. MIT doctoral dissertation. (Available from Indiana University Linguistic Club, Bloomington.)

McCarthy, J. (1981) 'A prosodic theory of nonconcatenative morphology'. Linguistic inquiry 12, 373-418

McCarthy, J (1982) 'Prosodic templates'. In van der Hulst and Smith (eds) (1982a).

McCarthy, J. (1986) ' O C P effects: gemination and anti-gemi- nation'. Linguistic inquiry 17, 207-63

McCawley, J. D. (1978) 'What is a tone language'. In Fromkin (1978).

Mohanan, K. P. (1982) Lexical phonology. MIT doctoral disser- tation. (Available from Indiana University Linguistic Club, Bloomington.)

Mohanan, K. P. (1985) 'Syllable structure and lexical strata in English'. Phonology Yearbook 2, 139-155

Mohanan, K. P. (1986) The theory of lexical phonology. Reidel, Dordrecht.

Nespor, M. and Vogel, I. (1982) 'Prosodic domains'. In van der Hulst and Smith (1982a).

Nespor, M. and Vogel, I. (1986) Prosodic phonology. Foris Publi- cations, Dordrecht.

Newman, J, and Newman, B. (1974) 'Longunda'. In Bendor- Samuel (1974).

Odden, D. (1987) 'Kimatumbi phrasal phonology'. Phonology yearbook 4 37—59

Okoth, D. O. (1979) Dholuo morphophonemics in a generative frame- work. Unpublished MA dissertation University of Nairobi, Kenya.

Palmer, F. R. (ed) (1970) Prosodic analysis. Oxford University Press. Pierrehumbert, J. (1980) The phonology and phonetics of English

intonation. Unpublished doctoral dissertation. MIT Press Cambridge, Mass.

Pike, K. (1947) Phonemics: a technique for reducing language to writing. University of Michigan Publications in Linguistics 3, Ann Arbor.

Pike, K. (1967) Language in relation to a unified theory of human behaviour. Mouton, The Hague.

294 Bibliography

Prince, A. (1983) 'Relating to the grid'. Linguistic inquiry I I , 19-100

Pulgram, E. (1970) Syllable, word, nexus, cursus. Mouton, The Hague.

Pulleyblank, D. (1986) Tone in lexical phonology. Reidel, Dordrecht.

Pullum, G. K. (1978) Rule interaction and the organization of a grammar. Garland, New York.

Radford, A. (1981) Transformational syntax. Cambridge University Press.

Riemsdijk, H. C. van and Williams, E. (1986) Introduction to the theory of grammar. MIT Press, Cambridge, Mass.

Ringen, C. O. (1972) ' O n arguments for rule ordering'. Founda- tions of language 8, 266-73

Rubach, J. (1985) 'Lexical phonology: lexical and postlexical derivations'. Phonology Yearbook 2, 157-172

Sapir, E. (1925) 'Sound patterns in language'. Language 1, 37-51 Sapir, E. (1933) 'La realite psychologique des phonemes'. Journal

de psychologie normale et pathologique 30, 247-65 Schachter, P. and Fromkin, V. A. (1968) 'A phonology of Akan:

Akuapem, Asante, and Fante'. UCLA working papers in phonetics 9

Schane, S. A. (1968) French phonology and morphology. MIT Press, Cambridge, Mass.

Schane, S. A. (1973) Generative phonology. Prentice-Hall, Engle- wood Cliffs, New Jersey.

Schuh, R. (1978) 'Tone rules'. In Fromkin (1978). Selkirk, E. O. (1978) ' O n prosodic structure and its relation to

syntactic structure'. Paper presented to the Conference on Mental Representation in Phonology. (Available from Indiana University Linguistics Club, Bloomington.)

Selkirk, E. O. (1980) 'The role of prosodic categories in English word stress'. Linguistic inquiry 11, 563-605

Selkirk, E. O . (1984) Phonology and syntax: the relationship between sound and structure. MIT Press, Cambridge, Mass.

Sommerstein, A. (1977) Modern Phonology. Arnold. Stampe, D. (1973) A dissertation on natural phonology. Unpublished

doctoral dissertation, University of Chicago. Steriade, D. (1982) Greek prosodies and the nature of syllabification.

Unpublished doctoral dissertation, MIT, Cambridge, Mass. Thomas, E. (1974) 'Engenni'. In Bendor-Samuel (1974). Toweett, T. (1975) Kalenjin nouns and their classification. Unpub-

lished MA dissertation. University of Nairobi, Kenya. Trager, G. L. and Smith H. L. (1951) An outline of English struc-

ture. Studies in linguistics occasional paper 3. Battenburg Press, Norman, Oklahama.

Bibliography 295

Trubetzkoy, N . S. (1939) Grundziige der phonologie, trans, by Baltaxe, C. (1969) Principles of phonology. University of Cali- fornia Press, Berkeley.

Twaddell, W. F. (1935) 'On defining the phoneme'. Language monographs 16. Reprinted in Joos (1957).

van der Hulst, H. and Smith, N . (1982a) The structure of phono- logical representations (Part I). Foris Publications, Dordrecht.

van der Hulst, H. and Smith, N . (1982b) The structure of phono- logical representations (Part I). Foris Publications, Dordrecht.

Vennemann, T. (1972) ' O n the theory of syllabic phonology'. Linguistiche Berichte 18, 1 —18

Vennemann, T. (1974a) 'Phonological concreteness in natural generative grammar'. In Shuy, R. and Bailey, C. J. (eds) (1974) Toward tomorrow's linguistics. Georgetown University Press, Washington, D . C .

Vennemann, T. (1974b) 'Words and syllables in natural generative grammar'. Natural phonology parassession. Chicago Linguistic Society, Chicago Linguistic Society, Chicago.

Welmers, W. (1973) African language structures. University of California Press, Berkeley and Los Angeles.

Whiteley, W. H. and Muli, M. G. (1962) Practical introduction to Kamba. Oxford University Press.

Suggested answers to exercises

Some of the questions have more than one 'correct' answer. A solution different from the one suggested below may also be valid. The answer I suggest need not always be regarded as the definitive solution.

Chapter 1

I.(a)Phonetics is the study of SPEECH S O U N D S which humans can produce.

(b) The three main branches of phonetics are acoustic phonetics, auditory phonetics and articulatory phonetics.

(c) (i) (a) airstream mechanisms (b) production of voicing (technically called

P H O N A T I O N ) (c) place of articulation (d) manner of articulation

(ii) (a) lip rounding (b) tongue height (c) location of the highest point of the tongue at

the front, centre or back of the mouth. (iii) (a) monophthongs like [e] in bet

(b) diphthongs like [ei] in bait

2. (i) [1] /aw (ii) [s] 5aw (iii) [n] new (iv) [h] he (v) [b] bee (vi) [p] />ea (vii) [n] longer (viii) [8] then (ix) [J] she (x) [d3] jet

3. (i) a high front vowel f@t fell pat wet full p@t

(ii) a low front vowel what b@d a3t saw these eggs

Suggested answers to exercises 297

(iii) a high back vowel women sQt p@l fijx^l bi@d fl@d

(In popular North of England speech [blud] blood and [dud] flood also have a high back vowel.)

(iv) a front vowel w@d word whgn hat card hit

(v) a back vowel hut aail g © r d sell s@t mist

(vi) a central vowel skin hgr wintgr pertain doctor sir

(In American, Scottish and South-West of England accents there is considerable R- C O L O U R I N G of the central vowel.)

(vii) a rounded vowel her g@d dumb hall pod cart

(In North of England accents [dum] dumb would also have a rounded vowel.)

(viii) a high vowel w© d o s@ ten pan hjri

(ix) a mid vowel s£nd card keys school hall you

(x) a low vowel man moon @rt cup knot teeth

(xi) a diphthong why) he m@) b@) tar h@se bird

Chapter 2

word bet tell zoo pail debt call see chest char lad

initially met sell sue tail net gall she zest jar dad

word final cub hit his harp maid back bass watch batch feel

iy come hiss hiss heart main bag bash was badge feed

Voiceless [ m \ r] occur word finally following a voice- less consonant. Voiced [m 1 r] occur elsewhere.

298 Suggested answers to exercises

Chapter 3

I . The segments with the specified feature are circled: (i) [+syll]

(ii) [+ant] (iii) [—cons] (iv) [+cor] (v) [+round]

(vi) [—voice]

Original segment

( i ) i

(iii) b (iv) e ( v ) d

(vi) g

w p(i)©m h @ g v © @XD9®(§>@N(B)<5)<D © f v 1 r h<gxj)x@ v®©@0(DD x(3)k © t i @ © ® @ e 1 ae o w©(g)l(Dg m i ®

Feature value changed [+syll] j [+back] y [—nasal] m [—round] 0 [ + voice] t [—nasal] n

New segment

[-syll] [-back] [+nasal] [+round] [—voice] [ + n a s a l l

3.(a)The phonetic symbols for the initial segment of each one of the words below are: that cat band wet [6] [k] lb] [w]

philosophy m

shy June knee [d3] [n]

write

[r] tea Itl

(b)

d3

cons son syll voice cont nas ant cor lab back high

+ —

+ +

+ + - - -

+ —

— -

- - - + +

+ -

+ -

+ - + - -

- 4

+ +

- - + + +

+ +

+ +

+ + - — -

+ -

- +

+ - + — -

+ -

- +

- + - - +

+ -

+ -

- + — — +

+ +

+ - + + + — — -

+ —

— - - + + — — -

Suggested answers to exercises 299

Chapter 4

1. The words which you were asked to read aloud are: (i) den [den] (vi) linguistics [lirjgwistiks]

(ii) pan [paen] (vii) house-breakers [hausbreikaz] (iii) science [saians] (viii) individual [mdividjual] (iv) sixteen [sikstin] (ix) knighthood [naithud] (v) angle [aerjgj] (x) profusion [prafju3n]

2. (a) full fool foot coot cut /ful/ /fill/ /fut/ / k u t / /kAt/ but boot tuck took /bAt/ / b u t / /tAk/ /tuk/ (In the north of England usually standard English / A / corresponds to /u/. Hence cut, but and tuck are realised as /kut/, / b u t / and /tuk/ respectively.)

(b) glass path mast plastic bath last laugh /glas/ / p a 0 / /mast//plaestik/ / b a 0 / /last/ /laf/ (In northern English accents standard English / a / usually corresponds to /as/ when it is followed by a fricative. The above words are rendered as /glaes/, /paeB/, /maest/, /plaestik/, /baeG/, /laest/ and /laef/.

(c) stair stare fare rare fair /stea/ /stea/ /fea/ /rea/ /fea/ fur where wear were / f j / /wea/ / w e a / / w 3 /

(d) philosophy finish enough fish caution /fibsafi/ /fmif/ /inAf/ / f i j / /kojn/

(e) ringing wringing bringing longer long /nrjirj/ /rirjin/ /bngirj/ / b n g a / /lorj/

3.(a)apart attention atmosphere [aphat°] [athenj"n] [astmDsfia]

(b) button cotton bottom bacon baking [bAtn] [kmn] [bDtm] [beikn] [beikirj]

(c) kettle little medal metal [khett] [lit*] [medt] [mett]

(d) drain train strain play [drem] [threm] [strein] [phlei] splay sweep scream [splei] [swip] [skrim]

300 Suggested answers to exercises

Chapter 5

I . ( a ) A broad p h o n e t i c transcription of the English data in e x a m p l e [5.1] a b o v e is given b e l o w :

A B Singular Plural Singular Plural dog dogs dock docks / d o g / /dDgz / d o k / /drjks/ bid bids bit bits / b i d / / b i d z / / b i t / / b i t s / rib ribs tip tips / r i b / / n b z / / t i p / / t i p s / love loves giraffe giraffes / I A V / / I A V Z / /d3iraf/ / d 3 i r a f s / / b a e d 3 / /baed3iz/ sea sed5 mof/j wof/j.*; bus fen5f5 / s i / / s i z / /moG/ /mrjOs/ / b A s / /bAsiz/

(b) T h e genitive e n d i n g has the same shape as the plural e n d i n g . T h e rule w h i c h predicts the shape of the plural ending also applies to the genitive; It is realised as:

(i) [—z] after w o r d s ending in a voiced s e g m e n t (see c o l u m n A),

(ii) [—s] after w o r d s e n d i n g in a voiceless s e g m e n t (see c o l u m n B).

(iii) [— iz] after w o r d s e n d i n g in a sibilant (i.e. a shrill fricative like / s z J" 3 / or a shrill affricative like /tf/ or / d 3 / ) (see c o l u m n C ) .

Singular witch / w i t / / nose / n a u z / marsh / m o f / badge

C Plural witches / w i t f i z / noses / n a u z i z / marshes / m a / i z / badges

2. (a) A b r o a d present walk / w o k / look / l u k / trap

/ t r a p / wish / w i / / laugh

M watch / w D t j /

phonetic transcription of the data: past walked / w o k t / looked / l u k t / trapped /traept/ wished / w i j t / laughed / l e f t / watched / w D t J t /

present paint / p e i n t / U'dMf

/ w o n t / part / p a t /

fit / f i t / court /kDt/ land / l a n d /

past painted / p e i n t i d / wanted / w o n t i d / parted / p a t i d / fitted / f i t i d / courted / k o t i d / landed /laendid/

Suggested answers to exercises 301

launder / b n d a / arm / a m / warn / w o n / sue

H

laundered / b n d a d / armed / a m d / warned /wond/ sued /sud/

souna sounded /saundid/ faded /feidid/ padded /paedid/ rou^d /raud/

Jade /feid/ pad /paed/ row /rao/

The regular past tense is realised as: (i) /—id/ after a verb ending in / t / or / d /

(ii) /—t/ after a verb ending in any voiceless segment except ft/

(iii) /—d/ elsewhere (i.e. after any voiced segment except /d/) (Note that this rule also applies to regular past participles e.g. a fitted ([(\tid\) carpet

a padded ([p&did]) anorak armed ([amd]) forces a trapped ([trxpt]) animal).

Both the past tense and the plural endings are subject to the same basic phonological principle of voice assimilation. In the case of the plural, an underlying alveolar fricative agrees in voicing with the last segment of the noun. In the case of the past tense suffix, the underlying alveolar stop agrees in voicing with the last segment of the verb.

However, where the suffix is too similar to the root to which it is attached, a vowel ([i] or [a]) is inserted to separate the consonant of the suffix from that of the root. This happens when an alveolar sibilant ([z] or [s]) representing the plural suffix (or the genitive suffix or the third person singular present tense marker) would be adjacent to another sibilant as in buses [bAS- iz] (not *[6AS-.?]) or noses [naoz-iz] (not *[n3u^-z]). It also happens where an alveolar stop of the past tense (or past participle) ending would be next to another alveolar stop as in painted [peintid/ (not *[pein/-/]) or padded [paed-id/ (not *[pxd-d]).

302 Suggested answers to exercises

Chapter 6

1. (a) r i m © example (J) is an obstruent but r 1 m are sonorants

(b) p t x(g)k s f (g) is voiced, unlike the other consonants

(c) l r n j i d w j © © is the only voiceless sound in the group

(d) f J t p (g) b 8 (g) is the only nonanterior consonant (e) s z f v x J"(t) 0 is a stop; the rest are fricatives (f) pf(3)ts tj bv d3 @ is a stop; the rest are affricates (g) a 6 © u y © is oral; the rest are nasal (h) p @ t s b d © is nasal; the other consonants

are oral (i) b @ g d @ is implosive; the other stops are

plosives (j) e i e(3) @ is the only front rounded vowel

2. Tairora (Papua New Guinea) (SIL: 1980: 39) (a) The sounds [1] and [r] are not distinct phonemes. They

occur in complementary distribution and are allophones of the same phoneme: (i) [1] occurs before the back rounded vowels [u] and

[o] (as in [bi'/o], ['bu/o], [a/Hke'/oma] and ['biri 'timi/o]).

(ii) [r] occurs elsewhere i.e. before any other vowel (as in [birdie], ['iria] [bi'reraBe ti'lo]).

(b) [b] occurs word-initially and [B] between vowels within a word (as in [bi'Ba] and [bu'aaBa]).

(c) Intervocalic position tends to be a weak position. In many languages consonants occurring in that environ- ment are moved lower down the strength hierarchy. When flanked by vowels, stops often become fricatives. The phonetic basis of this phonomenon is simple: as vowels are made without obstructing the airstream in the oral tract, in order to minimise articulatory effort in the articulation of a consonant in inter-vocalic position (like the stop [b] in our example), it makes sense for a speaker not to form a strong obstruction of the airstream (which would have to be undone a few milliseconds later). Thus Tairora speakers will produce the bilabial fricative [B] in between vowels instead of the bilabial stop [b].

Suggested answers to exercises 303

3. The addition of the genitive suffix has no effect on the final consonant of the noun. As here the genitive ending is attached to a root ending in a voiceless consonant, the allomorph of genitive morpheme selected as /—s/. This ensures agreement in voiceless- ness between the suffix and the final root consonant. (The rule accounting for regular plural endings which was worked out on page 300, in the answer to question 1 in Chapter 5, applies here as well.)

In the plural, however, something unusual happens. The final labiodental fricative which is realised as voiceless [f] elsewhere in words like wife is realised as voiced [v] in wives and other words of this ilk. The form of the plural suffix selected in this case is predict- ably /—z/. This ensures agreement with the voiced final consonant of the noun root.

The process described above is not purely phon- ological. T w o suffixes which appear to be phonologi- cally identical /—s/ trigger off different phonological alternations. In order to predict the behaviour of these suffixes it is necessary to take into account morphol- ogical information as I have done above. The voicing of final fricatives before /—s/ is an idiosyncratic prop- erty of the plural morpheme. It is not an automatic, natural phonological alternation.

Chapter 7

1.(a)The rules needed are: (1) / d / DELETION: d -> 0 / — C

(In these compounds the final / d / of the first word is deleted when the second part of the compound word begins with a consonant.)

(ii) H O M O R G A N I C NASAL ASSIMILATION The nasal acquires the place of articulation of the following consonant: alveolar / n / is realised as bilabial [m] to match the bilabial [b] that comes next. (The homorganic nasal assimilation rule is written formally in the answer to the next question.)

304 Suggested answers to exercises

(b) T h e first rule feeds the second one. If w e assume extrinsic linear o r d e r i n g , the deletion of / d / m u s t apply first, creating the input to h o m o r g a n i c nasal assimi- lation rule.

H o w e v e r , it is possible to handle this rule inter- action w i t h o u t insisting o n extrinsic linear o r d e r i n g . O b s e r v e that until rule (i) has applied deleting / d / , the nasal is n o t adjacent to the initial c o n s o n a n t of the second w o r d and cannot therefore assimilate to it. For b o t h / d / deletion and h o m o r g a n i c nasal assimilation to apply, universal principles require that / d / deletion precedes h o m o r g a n i c nasal assimilation since a rule cannot apply until its structural description is met. A feeding rule m u s t apply before the rule w h o s e appli- cation it facilitates. N o explicit linear o r d e r i n g state- m e n t is required.

2 . (a) Generative phonologists claim that differences between dialects of the s a m e language can be attributed either, (i) to innovations in the rule system of o n e dialect

w h i c h are n o t shared b y other dialects or to the loss of a particular rule in s o m e dialects,

or (ii) to a change in the o r d e r in w h i c h the same rules

are applied to the same u n d e r l y i n g representations in different dialects.

A n y o n e of these factors will result in differ- ences in the phonetic representation (Kiparsky 1968).

T h e Lumasaaba data exemplify innovation in the rule system. W e can set up / i N - / as the u n d e r - lying representation of the n o u n class prefix. T h e same form w o u l d also serve as the u n d e r l y i n g r e p - resentation of the first person singular personal p r o - n o u n . T h e differences in the phonetic manifestation of the prefixes are due to the rules stated below:

(b) (i) H O M O R G A N I C N A S A L A S S I M I L A T I O N aant Bcor yback

/

aant Bcor _yback_

(This rule was introduced in [7.16] and [7.17] o n page 125.)

Suggested answers to exercises 305

In dialect A the underlying nasal shares the place of articulation of the following consonant (see impiso, intemu, ijicese and igkafu). Dialect B as well has the homorganic nasal assimi- lation rule. Its effects can be seen where the root begins with a voiced consonant (see [imbeba, iggwe, indimal]). However, in dialect B, if the root begins with a voiceless consonant (as in [iitemu, iijula, iikuba]) two additional rules come into play. They are nasal deletion and compensatory lengthening of the vowel preceding the deleted nasal,

(ii) C O M P E N S A T O R Y L E N G T H E N I N G

V *- [+long] / _ N

(iii) NASAL DELETION

+cons —voice

N ^ 0 /_ +cons —voice.

Rules (ii) and (iii) are an innovation in dialect B. They are absent from dialect A.

(c) First, observe that rule (iii) bleeds rule (i) in dialect B. Nasals are deleted before voiceless consonants. This deprives the homorganic nasal assimilation rule of some of its input. As we have already seen, in this dialect homorganic nasal assimilation is restricted to roots commencing with voiced consonants

Secondly, if extrinsic ordering is assumed, rule (iii) cannot apply before rule (ii) because, if it did, it would remove the nasal whose presence is required by the compensatory lengthening rule.

But extrinsic ordering of these last two rules is not essential. If we stipulate that rules apply at the earliest opportunity when their structural description is satis- fied, both (ii) and (iii) can apply directly and simul- taneously to the underlying representation since at that point the structural description of both rules is satisfied.

306 Suggested answers to exercises

Chapter 8

(c)

I.(a)The plural morpheme has the shapes [i], [e] and [e]. (b) The underlying representations of the noun roots are:

(a) abd 'vegetable' luo 'stick' god 'hill' ko8 'rain' guog 'dog'

(b) lep 'tongue' lak 'tooth' adit 'basket'

I have posited as underlying the consonants that appear in the plural form of the noun rather than those which occur in the singular form because this allows us greater generality. The solution I am suggesting requires an obstruent devoicing rule which changes underlying voiced obstruents into voiceless obstruents in the data in (a). Obviously, the devoicing rule has no effect on the underlying voiceless obstruents in (b).

If, instead, we assumed that the words in 1. (a) end in a voiceless obstruent which becomes voiced when it occurs between vowels as a result of the addition of the plural ending, we would be unable to explain why our voicing rule fails to apply to the words in (b). They would have to be marked as exceptions. This solution is to be rejected in favour of the more general first alternative. The devoicing rule which I propose can be written thus:

+cons —son +voice

[" -voice • ] / _ #

The voicing rule which I reject would be written thus:

-> [+voice] /V _ V +cons —son +voice

(d) In the first solution I suggest that underlying voiced obstruents are devoiced word finally (in the singular form). They become indistinguishable from underlying

Suggested answers to exercises 307

voiceless obstruents like those in (b). In other words, the opposition between voiced and voiceless obstruents is NEUTRALISED (i.e. SUSPENDED) in word final position. For a better known example of the neutral- isation of the voicing opposition see the Turkish data in [6.7] on page 103.

2.(a)The only allomorph of the class 12 prefix is /ka-/. The class 9 prefix has these allomorphs:

[m] before roots commencing with labials [n] before roots commencing with alveolars [ji] before roots commencing with palatals [n] before roots commencing with velars

(b) [mbusi], [ndeBe] (c) The rule in question is homorganic nasal assimilation.

Proposing a rule is preferable to making a list because it allows us to capture the generalisation that in Luyia nasals assimilate to the place of articulation of the following consonant.

(d) H O M O R G A N I C NASAL ASSIMILATION

'aant Bcor yback

/ -

' aant Bcor yback

(e) C O N T I N U A N T S T R E N G T H E N I N G The continuants [B r j] become the stops [b d J\ respec- tively when a nasal prefix is present:

'hoe' [xaBako] ~ [mbako] 'snake' [xaremu] ~ [ndemu] 'jackal' [xajusi] ~ [jijusi]

VOICE ASSIMILATION: [t] becomes [d] after a nasal as in 'lemon' e.g. [xatimu] ~ [ndimu]

PALATAL S O N O R A N T DELETION The palatal glide [j] is deleted following a nasal when the next syllable also begins with a nasal as in 'house' / N + j u m b a / —» /ji+jumba/ —* [jiumba]. The same happens to the palatal nasal [ji] when the next syllable also contains a nasal as in 'meat' / N + j i a m a / - * /.P + jiama/ —» [jiama].

(f) C O N T I N U A N T S T R E N G T H E N I N G [ + cont] > [-cont] / [+nasal]

308 Suggested answers to exercises

VOICE ASSIMILATION:

— voice — cont

[+voice] / [ + nasal]

PALATAL S O N O R A N T DELETION

-* 0 / N _ V N + cor +high +sonorant

(g) All the rules above except one are mutually non- affecting. The order in which they apply makes no difference. The only case where the order is relevant is the last two examples.

Since the nasal prefix surfaces as a palatal nasal although the initial nasal of the root is absent from the phonetic representation in class 9, we can assume that the rule of homorganic assimilation applies when the palatal sonorants / j / and /ji/ are still present. The dele- tion of /']/ and / j i / must apply after homorganic nasal assimilation. If we tried to apply palatal sonorant dele- tion before homorganic nasal assimilation, the deriva- tion would abort because the palatal nasal which conditions the homorganic nasal assimilation rule would be missing.

3. Hyman (1970) suggests this solution: (a) A rule palatalising consonants before front vowels

applies to the data in (a). V

[+cons] -+ [+high] / _ [-back] (b) A rule labialising consonants before round vowels

applies to the data in (b). V

[ + cons] —» [ +round] / [ +round] (c) Nothing happens to consonants followed by / a / which

is neither front nor round. (d) The surface [a] in these words is derived from an

underlying / e / which never occurs phonetically because it is absolutely neutralised with [a]. This derivation is assumed: (i) UR / e g e /

Suggested answers to exercises 309

(ii) Palatalisation before the underlying front vowel (by rule a) /ege/ —> /eg>c/

(iii) Lowering: underlying jzj —> [a] /egi e / -H> [eg

Ja] (e) The surface fa] in these words is derived from an

underlying [0/ which never occurs phonetically because it too is absolutely neutralised with [a]. This derivation is assumed:

(i) U R / e g o / (ii) Labialisation before the underlying round vowel

(by rule b) /ego/ —> /eg w o/ (iii) Lowering: underlying / o / —» [a]

/eg w o/ - [egwa]

Chapter 9

I.(a)Make a broad transcription of the data below. met / m e t / strict / s t n k t / juxtapose /d3Akst3pauz/

(b) D i v i d e the w o r d s c o n v e n t i o n in [9-J

fright / f r a i t / laughed /loft/ Kne5ie( / k n e s e t /

in syllables [Of,

s»/iere /sfia/ scratched /skrsetjt/ Gdansk /gdaejisk/

using the syllabification

c v c c c v c c c v

I I I M M I I K m e t f r a i t s f i a

met fright sphere

C C C V C C C V C C C C C V C C

s t r i k t l a f t s k r a e t / t

strict laughed scratched

310 Suggested answers to exercises

a a

A A\ C C V C V C

k n e s e t

knesset

Gdansk

Remarks: (a) Diphthongs and affricates are complex elements. In a

diphthong (e.g. / a i / as in fright) two segments with different phonetic properties occupy a single V slot. Likewise, with affricates like / d 3 / in juxtapose a single C slot is occupied by two consonantal segments.

(b) Knesset and Gdansk are foreign words which have an initial consonant sequence that does not fit in the English syllable template: / k n / and / g d / are not permissible syllable onsets in English. The consonants / k / and / g / are EXTRASYLLABIC. In speech the transition from the extrasyllabic consonant to the next consonant may involve the insertion of a short voiceless [a]. (See Clements and Keyser 1983: 39-40) Knesset and Gdansk have syllable trees in which the initial [k] and [g] respectively are unlinked to any syllable node.

The syllabification principles in [9.10] need to be amended to allow underlying consonants, in excep- tional circumstances, to be unlinked to any syllable node.

2. Syllable formation rules may need to take into account segment type. There may be restrictions on the conso- nant or vowel segments which can fill a particular C or V slot. In our data we have this pattern of restrictions:

c v c c c v c v c A A

d 3 A k s t a p a u z

juxtapose

c c v c c c

g d se ji s k

Suggested answers to exercises 311

Column A: lax vowels may occur in STRESSED CLOSED syllables

Column B: tense vowels may occur in STRESSED CLOSED syllables

Column C: only tense vowels occur in STRESSED O P E N syllables in monosyllabic words

Column D: only lax vowels can precede /rj/.

3. The plural definite article /lez/ contains an underlying final jzj which only surfaces when it is followed by a vowel (as in les animaux [lez animo]) and is deleted in preconsonantal position (as in les dames [le dam]). We can assume that this consonant is underlyingly EXTRAMETRICAL. It cannot be pronounced unless it gets associated to a syllable node during a derivation. French being a language which generally prefers open syllables, in cases where the extrasyllabic consonant is followed by a consonant it is deleted:

UR PR a o a 0

/ \ A\ / \ A\ c v c c v c c v c v c

I I ! J ! J - l ! J ! i But when the extrasyllabic consonant is followed by a vowel it is preserved. It is linked by the liaison rule to the syllable node to its right and functions as a syllable onset. This is shown by the broken line drawn between / z / and the syllable node dominating /a/:

a 0 0 0

A I A A c v c v c v c v 1 e z a n i m o

PR a o a o

A A A A c v c v c v c v 1 e z a n i m o

les animux

312 Suggested answers to exercises

Chapter 10

I.(a)In the Bekwarra data tone has a lexical function. There are no minimal pairs distinguished soley by pitch but at the same time it would not be possible to predict the pitch that occurs in a given position in a word. The examples contain many near minimal pairs. We can recognise high ('), mid (—) and low (N) tone as phonemic using the principle of contrast in analogous environments on the basis of pairs like: [6ku] ~ [uko], [Ige] ~ [Ije] [ide] ~ [ine] [ubii] ~ [upu] [api] ~ [abi] [ubi/ ~ [abi]

(b) The data shows that Bekwarra is a register tone language. All the tones have level high, mid or low pitch. There are no rising or falling contour tones.

2. (i) The following tonal alternations are observed (based on Elimelech 1976):

L-L° vs L-H (cf. ele° 'tree' vs ele ndira 'that's a tree') H-D vs H-H (cf. cele 'sand' vs cele ndira 'that's sand') F-L vs H-L (cf. ku6a 'chicken' vs ku6andira 'that's a chicken') H-F-L vs L-L-L (cf. ibumu 'belly' vs ibumu ndijira 'that's a belly') L-L-L° vs L-L-H (cf. 6iji5ni° 'bird' Bijioni ndiBira 'that's a bird') L-H-D vs L-H-H ^ (cf. lkayi 'leaf vs lkayi ndijira 'that's a leaf) L-F-L vs L-H-L (cf. ilali 'stone' vs ilali ndijira 'that's a stone') The L-L of ta6a 'goat' and the L-F of ih 'ear' do not alternate. (ii) In underlying representations only high and low

tones contrast in Kombe. The rules below derive falling tone (F), unreleased low tone (L°) and downstep high tone (D).

Falling tone The F occurs in penultimate position before a L tone followed by a pause as in:

Suggested answers to exercises 313

kuba 'chicken' a toyendi 'he saw a chicken' kuba

ilali 'stone' a toyendi 'he saw a stone' ilali

Where these nouns occur phrase medially the F tone is simplified to H. This is the case in the identification construction: ku6a 'chicken' ku6andira 'that's a chicken' ilali 'stone' ilali ndijira 'that's a stone' Falling tones are derived from underlying high tones by a rule which spreads the L tone in pre- pausal position to the preceding vowel:

UR

"H L

_V V # # _

PR

"H L

N V V # #

As a result of this rule, the first vowel is simul- taneously associated with a high and a low tone. This makes it a falling tone. Unreleased low tone (L°) L° occurs word finally after a L and followed by a pause: ele° 'tree' a toyendi ele° 'he saw a tree' Bijioni0 'bird' a toyendi Birjoni0 'he saw a bird' But when the relevant nouns occur in phrase medial position, instead of L° we find a H tone: ele° 'tree' ele ndira 'that's a tree' Bijioni0 'bird' Bijioni ndiBira 'that's a bird' We shall therefore assume that L° tone is derived from an underlying H. A high tone is lowered to L° when preceded by L and followed by a pause.

L H

V V # #

_^ L H

N _V V # #

The L tone spreads from the first vowel to the second. Interestingly, in the phonetic represen- tation the language distinguishes these derived

314 Suggested answers to exercises

unreleased (non-falling) l o w tones from falling l o w tones w h i c h c o m e from u n d e r l y i n g l o w tones (see taba). Downstep tone D o w n s t e p is restricted to final syllables of n o u n s : cele r 'sand' a t o y e n d i cele T 'he saw sand' ikayi 'leaf a toyendi ikayi 'he saw a leaf A l w a y s D is preceded by H and followed by a pause. D does n o t occur w h e n a n o u n is inside a phrase: file 'sand' cele ndira 'that's sand' ikayi 'leaf ikayi ndijira 'that's a leaf T h e rule stating that u n d e r l y i n g H b e c o m e s D w h e n it is preceded b y H and followed by a pause is given b e l o w (with # # m a r k i n g a phrase b o u n d a r y ) :

H H

V V # #

H H D

V V # #

T h e s e data foreshadow the p r o b l e m of P H O N - O L O G I C A L D O M A I N S w h i c h is explored in the final chapter: certain rules only apply within a syllable, other rules only apply w i t h i n a w o r d or even only w i t h i n w o r d s of a particular category and other rules only apply w i t h i n a phonological phrase.

Chapter 11

I.(a) Stress placement in Arabic is sensitive to syllable w e i g h t . A syllable w i t h a short v o w e l (CV) followed b y n o t h i n g is light. B u t a syllable containing a l o n g v o w e l (CV:) or a s h o r t v o w e l followed b y at least one consonant ( C V C ) is heavy. F u r t h e r m o r e , a w o r d final consonant is extrametrical and does n o t contr ibute a n y t h i n g to the w e i g h t of a syllable. So, C V C syllables in w o r d final position are light. In that position only C V C C syllables and syllables w i t h l o n g vowels count as heavy.

T h e Arabic stress rule is reminiscent of the Latin

Suggested answers to exercises 315

w

stress rule: stress falls on the first heavy syllable, counting from the right-hand end of the word. In a disyllabic word with no heavy syllables stress goes on the initial (penultimate) syllable. In longer words with no heavy syllable stress goes on the antepenultimate syllable. The operation of this rule is shown below in (b). Metrical trees to show where stress falls in the words: kdtab, kdtabu, kdtabit, katdbt, katdbti, lamuuna and lamundat are shown below. (Note that it makes no difference whether trees are drawn above the words as I did in the text of Chapter 9 or below the word as I have done here.) (a) ka - ta(b) The final consonant is extrametrical.

Therefore the final syllable is light although it ends in a consonant. It is skipped over. Stress goes on the first syllable.

ta - bi(t) Final / t / is extrametrical. In a word ta - bu of more than two syllables, where

as here, none of the syllables is heavy, stress goes on the antepenul- timate syllable.

Stress is on the heavy CVC or CV: syllable in penultimate position.

(d) la - mil - naa(t) Even though the final / t / is extra- metrical, the last syllable is still heavy (CV:). As it contains a long vowel it receives the stress.

Ngiyamba (Australia) (Based on Donaldson 1980) The most salient characteristics of Ngiyamba stress are summarised in this section.

(i) This language has a quantity sensitive metrical structure, with the left-hand dominant. Primary stress falls on the initial syllable if a word contains no long vowels or on the heavy syllable (with a

316 Suggested answers to exercises

long vowel) nearest to the beginning of a word, if there is one.

(ii) The assignment of secondary stress is subject to a rhythm rule requiring an alternating stress pattern. On the one hand, adjacent syllables cannot receive stress even where they otherwise qualify to bear it, and on the other hand, no more than two unstressed syllables can occur in a sequence.

(iii) Any word-final syllable is extrametrical. It is 'invisible' to stress rules - unless it contains a long vowel,

(b) The following analysis, based on Donaldson (1980: 42) informally exemplifies Ngiyamba stress rules: STRESS RULE 1: PRIMARY STRESS IN R O O T S

(i) If a root contains no heavy syllable (with a long vowel), primary word stress falls on the first

1

syllable (as in girala). (ii) If a word contains a heavy syllable, primary word

stress falls on the first heavy syllable, i.e. the 3 1

syllable containing a long vowel (as in gabadai-ga, 3 1 3 1

gabadai-bidi and binjdju-binjdjuiri-nji). (iii) In a trisyllabic root without long vowels, if the

final syllable is open and a monosyllabic final suffix with a short vowel follows, primary stress

1

falls on the second syllable, as in girbadja-gu and 1

bayirga-gu. STRESS RULE 2: SECONDARY STRESS IN SUFFIXES (i) Secondary stress falls on each (heavy) syllable con-

taining a long vowel in each suffix so long as no adjacent syllables are stressed.

(ii) If a suffix has no long vowel, and has more than one syllable, stress falls on the first syllable. This is an ITERATIVE rule. It applies to each suffix in turn, starting with the first to be affixed to the root, provided that none of the general constraints mentioned above is violated. For instance, the rule

Suggested answers to exercises 317

does not apply in cases where it would otherwise place a secondary stress adjacent to a root-final syll-

i able with primary stress as in gabadai-bidi, or to a suffix-final syllable with secondary stress, as in

i 2 yana-wa-ga\-giri.

STRESS RULE 3: S E C O N D A R Y STRESS IN R O O T (i) Secondary stress falls on the first syllable of a root

if primary stress falls on some other syllable than 3 1

the first or second (adjacent) syllable, as in gabadai- 3 1 3 1

ga, gabadai-bidi and binjdju-binjdjw.ri-nji. (ii) Secondary stress falls on alternate syllables of the

root to the right of the syllable carrying primary 1 3

stress, as in giralan-ga unless this would result in two adjacent syllables being stressed. The effect of

1 2 this constraint can be seen in giralam-bidi and

1 2 girbadja-bidi..

STRESS RULE 4: S E C O N D A R Y STRESS IN MONOSYLLABIC SUFFIXES WITH S H O R T VOWELS

Secondary stress falls on monosyllabic suffixes with short vowels which are adjacent to unstressed syllables,

1 4 2 1 2 4 as in yana-wa-y-garai-dha and yana-buna-wa-dha.

Chapter 12

1. We assume that the irregular plural formation processes in (a)-(c) take place at level 1. In order to qualify for any one of these ways of forming the plural, a noun must be expressly marked [+ rule a] to show that it undergoes the rule which 'replaces' the singular ending -um with the plural ending -a, or [+ rule b] which 'replaces' the singular ending -us with the plural ending -1 or [+ rule c] which causes the internal vowel change

318 Suggested answers to exercises

/ a o / —> /ai/. Any of these level I rules BLOCKS the application of the later level 2 regular plural formation rule exemplified in (d). Having been already inflected for plural at level 1, the forms in (a)-(c) cannot subse- quently undergo the regular level 2 rule.

2. Kimatumbi (Based on Odden 1987) (a) Shortening is a POST-LEXICAL rule. It applies after

words have been put together by the syntax to form phrases.

(b) Shortening in Kimatumbi only applies within a single X-bar domain. (i) A long vowel in a noun stem is shortened if the

stem is the head of the noun phrase and occurs in phrase medial position. This is illustrated by the data in (a),

(ii) Shortening fails to apply if the noun is not immedi- ately followed by a modifier within the noun phrase of which it is the head.

Thus in (b) shortening fails because the noun is followed by a verb phrase (chaapuwaaniike) or by a noun which is not in the same N P (Mamboondo).

Likewise, in (c) the long vowel is shortened in the first example where [kjkolombe kikulu]N P is a single noun phrase whose head is [kikoloombe] but shortening is blocked in the second example where the same words are adjacent to each other but do not belong to the same noun phrase: [kjkoloombe]Ni> [0kikulu] N P . This shows that it is constituent structure rather than lexical category or grammatical agreement that determines the applicability of the short- ening rule.

Finally, in (d) we can see that shortening only applies to a noun which is the head of a noun phrase. Thus, the long vowels of the adjective kikeele and the possessive yaangu fail to shorten when they occur in phrase medial posi- tion.

3. The English glottaHsation rule applies to / t / when it is initial in an unstressed syllable and occurs either

Suggested answers to exercises 319

between vowels or between a vowel and a sonorant like N or / n / :

—stress —consonant +sonorant

This rule applies both lexically as in water, bottle and button and post-lexically as in let it in.

t ^ -> J V

Language Index

Akan, 206-7 Aleut Eskimo, 27 American English, 5, 36-37, 72,

276 Arabic, 27

morphemic tier, 216-7 stress, 250-1

Azerbaijani, 27

Bantu languages, 95, 106 Bekwarra, 217 British Columbian French, 27 British English, 17

Cockney, 5 northern English, 78, 299 west country accents, 72

Burmese, 27

Cairo Arabic, 92 Cambodian, 231 Cherokee, 16-17 Chi-mwini, 275-6 Chinese, 16, 188 Cocopa, 28 Czech, 27, 28-30

Dakota, 231 Danish, 106 Desano, 214-6 Dyirbal, 27

Intonation, 239-50 Latin borrowing, 92 LEXICAL

P H O N O L O G Y , 254-274 non-standard British

English, 276-7 obstruents, 102-3 plural, 81-84 potential English words, xi,

164-6 present tense suffix, 82-3 stress, 53, 222-9, 234-9 syllable structure, 154-5,

161-6, 169 underlying representation,

83-4, 134-40, 141-2 voice assimilation, 83-4 vowels, 78, 184

Etsako, 200-2 Ewe, 23

Finnish, 231 French, 5, 29-30, 56, 99, 105

FINAL C O N S O N A N T DELETION, 119-20

'ghost segments', 145-6, 181-4

LIAISON, 184-5, 282-3 stress, 230 voice assimilation, 89 VOWEL

NASALISATION, 120-23, 167-8

German, 5, 99 obstruent devoicing, 167

Greenlandic Eskimo, 109 Gwari, 186, 192

Efik, 27 Engenni, 54 English, 6-12, 19-23, 25, 27, 41,

56-7, 60-74, 9 0 - 1 , 9 4 - 5 , 110-11, 117-9, 133

GREAT VOWEL SHIFT, 140

Language Index 321

Hausa, 202, 207-8 Huasteco, 179 Hungarian, 211, 231

Igbo, 211-4 Igala, 54, 186, 187 Italian, 27, 105 Ivatan, 29

Japanese, 88 English loanwords in

Japanese, 166 Pitch-accent, 208-11

Kalenjin, 89 Kikamba, 52, 106 Kimatumbi, 284 Kikuyu, 27, 93-4, 101 Kirundi, 95-6, 277

R U N D I , 106 Kombe, 218-9

Latin, 88, 105-6 METRE, 178 Latin Stress Rule, 177, 232-4

Longunda, 167 Luganda, 52, 84-5, 101, 106, 107,

109, 111-3, 166 CLITICS, 173-4 C O M P E N S A T O R Y

LENGTHENING, 171-2 GEMINATED, 170 Glide formation, 123-4, 160 SPIRANTISATION, 146-9

Lumasaaba, 133 Luo, 150-1 Luyia, 151-2

Malay, 89-90 Maranungku, 229, 230 Margi, 194-5 Marshalese, 28 Mbui, 200, 275 Medieval Spanish, 106 Mende, 194, 195, 198-9, 203 Middle English, 140 Modern Hebrew, 231 Mongolian, 211, 213

Ngiyamba, 251-2 Northern Paiute, 231 Nupe, 152

Old English, 137 Old French, 120

Palauan, 32 Pare, 107 Persian, 27 Pitta-Pitta, 109 Pokomo, 107 Proto-Bantu, 106, 149

Rimi, 107 Romance languages, 108 Runyankole, 114 Russian, 142-4, 167

Shona, 49 Siriono, 28 Southern Paiute, 231 Spanish, 27, 88, 105-6

diminutive suffix, 175 underlying representation,

140-41 Sukuma, 107 Swahili, 27, 100, 125-7, 129

English loan words, 114 Swahili stress, 173, 231 syllable, 166

Swedish, 29, 99

Tairora, 115 Turkana, 100 Turkish obstruent devoicing,

103, 167

Vietnamese, 29-53

Walpiri, 166 Wangkatja, 5 Weri, 231

Yawelmani, 168 Yoruba, 5, 94

Zulu, 47, 207

Subject Index

Abercrombie, 15, 156 ABLAUT, 286 Absolute neutralisation, 145-50 ABSOLUTE SLICING

HYPOTHESIS, 191 ABSTRACT SEGMENTS,

145-6, 181-2 Abstractness, 134-5 A C C E N T , 209 Accentuation function, 242 A C O U S T I C P H O N E T I C S , 2,

77 A D V A N C E D T O N G U E

R O O T , 47 affixation, 261 affixes, 237

Order of Affixes, 273 AFFRICATE {see also

DELAYED RELEASE), 6 AIR STREAM MECHA N I SM,

2 ALLOMORPH, 81 ALLOPHONES (or

VARIANTS), 18 ALPHABETIC WRITING

SYSTEM, 16-18, 60-66 Anderson, xiv ALVEOLAR, 5, 6 ANTERIOR, 44 aphasia, 111 APICAL sounds, 45 A P P R O X I M A N T , 7 Aronoff, 266 ARTICULATORS, 4 ARTICULATORY

P H O N E T I C S , 2 ASPIRATED, 48, 71

ASPIRATION, 70 ASSIMILATION, 80-4

BIDIRECTIONAL, 85 PROGRESSIVE

(ANTICIPATORY), 85-5 REGRESSIVE, 84 ASSIMILATION

PROCESSES, 36-38, 86-94

{see also spreading) ASSOCIATION LINES, 157,

197 Attitudinal functions, 246-50 A U D I T O R Y P H O N E T I C S , 2 AUDITORY P R O M I N E N C E ,

221 A U T O S E G M E N T A L

P H O N O L O G Y , 190-2, 225 {see also MULTI-TIERED

P H O N O L O G I C A L T H E O R Y , 153-61)

BACK, 10, 46 BASE FORM, 83 {see also

underlying representation) Bell, Alexander Graham, 76-77 BILABIAL, 5 BINARISM, 40 BINARY B R A N C H I N G

TREES, 225 BLEEDING ORDER, 125-7 BLEEDING RELATIONSHIP,

128 Bloomfield, 35 borrowed, 256, 259 B O U N D MORPHEMES, 259 B O U N D A R I E S , 189

Subject Index 323

B O U N D E D trees, 229-31 BRACKET ERASURE

C O N V E N T I O N , 264-5 BRANCHING RHYME

HYPOTHESIS, 179-81 HIERARCHICAL syllable

structure, 153-4 branching rhyme, 176-7

BROAD TRANSCRIPTION, 69-70

canonical syllable types, 160-1 CENTRAL vowel, 10 citation forms, 87 Chomsky, x-xi Chomsky and Halle (see also SPE),

ix, 123 Clements, 281 Clements and Goldsmith, 204 Clements and Keyser, 156, 161-3 CLICK S O U N D S , 46-47 CLITICS, 173-75

CLITICISATION, 174 ENCLITICS, 174 PROCLITIC, 173

CLOSE vowel, 11 CLOSED SYLLABLES, 175 C O M P E N S A T O R Y

LENGTHENING, 171-2 C O M P E T E N C E , x COMPLEMENTARY

DISTRIBUTION, 19, 33-34, 70 complex segments (see also

internal structure), 169-71 Concrete phonology, 135-7 C O N N E C T E D SPEECH, 274-9 C O N S O N A N T A L , 43

production of consonants, 2-8

C O N S T R I C T E D GLOTTIS, 48-49

C O N T I N U A N T , 50 C O N T R A S T , 21-22

C O N T R A S T IN A N A L O G O U S E N V I R O N M E N T S , 23

C O N T R A S T IN IDENTICAL E N V I R O N M E N T S , 22-23

C O N V E N T I O N , 109 C O N V E R S I O N , 269, 271 CORE GRAMMAR, 159 CORE SYLLABLES, 159-61 C O R O N A L , 44 C O U N T E R - B L E E D I N G

relationship, 127 Courtenay, 34, 67-9 CV-phonology, 156-61

CV-SKELETON, 216-7 CV-tier, 156

CV syllables (see also O P E N SYLLABLES), 114-5

creaky voice, 49 CYCLE, 258

Dahl's law, 96, 277 DECLINATION LINE, 241 DEEP S T R U C T U R E , xii DEFAULT, 210 DEGREES O F

P R O D U C T I V I T Y , 265 degrees of stress, 222-5 DELAYED RELEASE, 51, 170 DELINK, 197

delinking, 203 DENTAL, 5 derivation, 258 D I A C H R O N I C , 105 DIACRITIC FEATURE, 148 dialects, 304-5 DICTIONARY (see also

LEXICON), 83 D I P H T H O N G S , 12, 170 DISSIMILATION, 94-96 DISTINCTIVE FEATURES,

35-38, 80 DISTINCTIVENESS, 21 DISTRIBUTED, 44-45 Disyllabic, 235-7 Donaldson, 124 Doncgan and Stampc, 110 DOWNSTEP, 199-200, 205-7,

314 D O W N S T E P P E D HIGH, 198 DOWNDRIFT, 192, 205-6

EGRESSIVE, 49 EJECTIVES, 49 elision, 277

324 Subject Index

ELSEWHERE C O N D I T I O N , 131, 270-3

E N D RULE convention, 230 EXTRASYLLABIC, 183-4 EXTRAMETRICAL, 231 EXTRINSIC linear ordering,

128

fast casual speech, 277-79 FEEDING ORDER, 125-7 Firthian prosodic analysis, 219 Fischer-J(|>rgensen, xiv FLOATING, 199-202

FLOATING GRAMMATICAL T O N E S , 200

FLOATING LEXICAL T O N E , 200

Foot, 281 F O O T F O R M A T I O N , 225-6 'foreign accent', 166 formal notation, 118 FORTITION, 103-4 FRICATIVE, 7 F R O N T vowel, 10 F U N C T I O N , 66-69 F U N C T I O N A L P H O N E T I C S ,

69 F U N C T I O N A L , 114 Functions of the syllable, 164-75

GEMINATE consonant, 170 GEMINATE vowel, 170 GENERATIVE GRAMMAR,

ix-xiii GENERATIVE

P H O N O L O G Y , ix, xiii-xiv geographical dialect, 76 'ghost segments', 145 Glide formation, 160 GLOTTAL, 5 GLOTTALIC AIRSTREAM

MECHANISM, 48-9 GLOTTALISED

C O N S O N A N T S , 48-49 glottalisation, 70, 318-9

GLOTTIS, 2, 8 Goldsmith, 190-1, 203 GRADUAL OPPOSITIONS,

41

'H ASPIRE', 182 HALF-CLOSE vowel, 11 HALF-OPEN VOWEL, 11 Harris, 141 Hayes, 229-31 HIERARCHICAL, 227 hierarchy, 281-2 HIGH, 9, 45 H O M O R G A N I C , 90

H O M O R G A N I C NASAL ASSIMILATION, 90-91, 112-4, 303-5

Hooper, 105, 140, 158-9 HOST, 173 Hyman, 155

illocutionary force, 243 IMMEDIATE

C O N S T I T U E N T S T R U C T U R E , 157

IMPLOSIVES, 48-9 INDEXICAL, 247 inflectional morphology, 260 I N F O R M A T I O N , 248-50 INGRESSIVE, 47, 49 innovation, 304 I N S T A N T A N E O U S

RELEASE, 61 I N T E R A C T I O N BETWEEN

RULES, 117 internal structure (see also

complex segments), 37 International Phonetic

Alphabet, 13 Intonation, 239-40 Intonational Phrase, 281 INTRINSIC linear ordering,

128 ITERATIVE, 316

Jakobsonian features, 38-41 J U N C T U R E S , 189

Kenstowicz and Kisseberth, 142-3

Kiparsky, 149, 254-58, 270 Koutsoudas et al, 128

LABIAL, 44 Labialisation, 87-88

Subject Index 325

LABIO-DENTAL, 5 LABIO-VELAR, 5 Ladefoged, 41 LAMINAL, 45 language acquisition (see also

LEARNABILITY), 111 Laryngealised, 48 LARYNX, 2 Lass, xiv LATERAL, 7, 50 Latin Stress Rule, 177, 232-3 LAX, 48 LEARNABILITY (see also

language acquisition), 149 Leben, 195 LENITION, 103-4 Lexical phonology, 257-8

LEVEL 1, 258-68 LEVEL 2, 269-70

LEXICAL P H O N O L O G Y A N D M O R P H O L O G Y , 254-6

LEXICAL VIA RULES, 140-1 LEXICON, 254-6 LIAISON in French, 184, 282-3 Libcrman and Prince, 224, 228,

229 Lightfoot, xii Linear rule ordering, 122-25 linguistic knowledge, x LINK, 160, 197 LIQUID, 50 loanwords, 166 LONG (see also

C O M P E N S A T O R Y LENGTHENING), 52

LOUDNESS, 221 LOW, 9, 45-46

MANNER OF ARTICULATION, 6, 8

MAPPING of tone, 198 MARKEDNESS, 39

MARKED, 39, 99, 101-2 MAXIMUM PERCEPTUAL

DIFFERENTIATION, 108 McCarthy, 216 MELODY LEVELS, 195-6 mcntalist approach, 68 METRE, 178, 226

METRICAL GRID, 228-9 METRICAL P H O N O L O G Y ,

225 Multi-tiered phonology, 186

Metrical trees and grids, 229 B O U N D E D TREES, 229 U N B O U N D E D TREES,

229 WORD-TREE

F O R M A T I O N , 227 MID, 9, 53 MINIMAL PAIR, 22-3 M O N O P H T H O N G S , 12 MONOSYLLABIC, 173, 234 M O R P H E M E , 81, 153, 267-8 M O R P H E M E S T R U C T U R E

C O N D I T I O N S , 165 Morphemic tier, 216-7 M O R P H O L O G Y , 112

Rule morphologisation, 122 M O R P H O P H O N E M I C

ALTERNATIONS, 140 MULTI-TIERED

P H O N O L O G I C A L T H E O R Y , 153

(see also A U T O S E G M E N T A L P H O N O L O G Y , 190-2, 225

MULTIVALUED O P P O S I T I O N S , 41

NARROW TRANSCRIPTION, 69-70

NASAL, 7, 50 NASAL T R A C T , 4 NASALISATION, 93-4, 167-8,

214-6 NATURAL CLASSES, 38, 101 Natural Generative Phonology,

140 NATURAL P H O N O L O G Y ,

109-11 NATURALNESS, 98 Ncspor and Vogcl, 281 NEUTRAL, 215 NEUTRALISATION (see also

vowel neutralisation), 39 neutralised, 149 non-alphabetic writing systems, 16

326 Subject Index

N U C L E U S , 154, 157, 240

OBLIGATORY C O N T O U R PRINCIPLE (OCP), 193-4

O B S T R U E N T , 43 obstruent devoicing rule, 142-3 one pitch drop, 208 O N S E T FIRST PRINCIPLE,

161 O P A Q U E SEGMENTS, 212 O P A Q U E ASSOCIATION, 213 O P E N SYLLABLES, 175, 183 O P E N vowel, 11 OPPOSITIONS, 39 ORAL, 7, 50 ORAL TRACT, 4 orchestral score, 63-4 Organs of speech, 4

PALATAL, 5 PALATALISATION, 86-7 PALATO-ALVEOLAR, 5 PARALINGUISTIC features,

246 parameters, x PENULTIMATE syllable, 231 PERFORMANCE, x PHARYNX, 7 P H O N A T I O N , 296 P H O N E , 18-22 P H O N E M E , 36, 38

Phoneme inventory, 100 P H O N E M I C LEVEL, 69 TARGET S O U N D , 67-9

P H O N E M A T I C UNITS, 219 P H O N E M I C

TRANSCRIPTION, 69 P H O N E T I C LEVEL, 69 P H O N E T I C

REPRESENTATION (PR), xiii, 117, 121

P H O N E T I C TRANSCRIPTION, 69

P H O N E T I C S , 1, 60-1 P H O N O L O G I C A L

C O M P E T E N C E , xi P H O N O L O G I C A L

D O M A I N S , 281 P H O N O L O G I C A L LEVEL, 69 P H O N O L O G I C A L PHRASE,

205-8, 240, 281

phonological processes, 80 P H O N O L O G I C A L RULES, xiii

formalisation and ordering, 117-22

phonological similarities, 80 P H O N O L O G I C A L

SYMMETRY, 25-31, 99-102 PHONOLOGICALLY

C O N D I T I O N E D alternation, 82, 137

P H O N O L O G Y , xiv, 1, 60-9 P H O N O T A C T I C RULES,

165-66 Pike, 75 PITCH, 186, 221 PITCH A C C E N T , 208-11 PITCH RANGE, 247-8 PLACE OF ARTICULATION,

4, 8, 106 PLOSIVES, 6 POST-LEXICAL

P H O N O L O G Y , 258 Post-lexical rules, 274-80 Prague, 38 PREDICTABILITY, 144 prefix (see also level 1 and level

2), 237, 256 pre-nasalised consonants, 171 primary stress, 222 Prince, 228 PROCESSES, 110 production of speech, 2 Prosodic Domains, 281 PROSODIC FEATURES, 51-54 PROSODIC TEMPLATES, 156 Prosodic Word 281 PROSODIES, 219 PROJECTION, 179-81 Pulleyblank, 279-80 P U L M O N I C EGRESSIVE, 2

QUALITY of a vowel, 10 Q U A N T I T Y SENSITIVITY

(see also syllable weight), 232-3

R E D U N D A N C Y , 56-7 SEGMENT

S T R U C T U R E C O N S T R A I N T S , 56

SEQUENTIAL R E D U N D A N C Y , 56

Subject Index 327

'r-colouring', 72 RHOTACISED, 72 rhyme, 155 RELATIONAL, 224 Re-syllabification, 282 RETROFLEX, 5 R O U N D , 44 R O U N D E D , 10 RULE, 110 Rule morphologisation, 112 rule ordering, 122

counter-feeding, 130 MUTUALLY

N O N A F F E C T I N G , 129 POTENTIALLY FEED,

130 R A N D O M

SEQUENTIAL RULE APPLICATION, 131

rule formalisation, 118-9 S I M U L T A N E O U S RULE

APPLICATION, 129

Schane, 146 SEGMENTS (see also

UNITARY S O U N D S ) , 17-8 SEGMENTAL TIER, 156 semantic constraint, 256 S E N T E N C E STRESS, 242 SIBILANTS, 51 SLOPE, 247 S O N O R A N T , 43 S O N O R I T Y , 104-6, 158-9 SONORITY HIERARCHY,

104, 158 SPE, IX, 123, 145, 190, 196, 222,

260 SPE system of distinctive

features, 42 SPEECH ERRORS, 17 SPEECH S O U N D S , 1 SPIRANTISATION, 146 spoonerisms, 17 SPREAD GLOTTIS, 46 SPREADING, 36-7, 62, 203-5,

211, 215-6 Stampe, 109-11 STOP, 6 STRATA, 257 S T R E N G T H E N I N G (also

called FORTITION), 103-4

Strength hierarchies, 103-5 STRESS, 53, 221-5

CONTRASTIVE (or EMPHATIC) STRESS, 242

Discourse function, 248-50 phonetic properties of

stress, 221 STRESS (nontonal) languages

186 STRICT CYCLICITY, 264 STRIDENT, 50-51 STRONG, 281 STRONG (s), 225 STRONG M O D E , 61 STRONG VERBS, 57 STRONG M O D E SUFFIXES,

238 suffixes, 237-9 (see also level 1

and level 2) SUPPRESS, 110 surface (phonetic)

representation, 117 SURFACE S T R U C T U R E , xiii SUSPENDED, 39 SYLLABIC, 43, 159-60 Syllabification, 161-4 Syllable, 153

C O D A , 159 IMMEDIATE

C O N S T I T U E N T S T R U C T U R E , 157

MARGIN, 157 N U C L E U S , 154-55, 159

O N S E T , 157 SYLLABLE Q U A N T I T Y

INSENSITIVE STRESS RULES, 232

SYLLABLE Q U A N T I T Y SENSITIVE RULES, 232

SYLLABLE WEIGHT, 175-77, 232

SYMMETRICAL, 26-7, 99-101 (see also P H O N O L O G I C A L SYMMETRY)

S Y N C H R O N I C , 105 SYNTACTIC C O M P O N E N T ,

xii-xiii SUSPICIOUS PAIRS, 24

TELESCOPING, 113

328 Subject Index

T E M P O , 248 TENSE, 48 TIERS, 196 TIMING U N I T (see weight

units), 256 T O N E , 53-54, 186, 205 •

C O N T O U R T O N E LANGUAGES* 188

DIACRITIC MARKS, 190 diacritics, 186 falling tone, 188, 192 GRAMMATICAL

F U N C T I O N , 187 LEXICAL F U N C T I O N ,

187 REGISTER T O N E

LANGUAGES, 188 rising tone, 188, 192 simplified, 193, 202

Tone and Intonation, 205 tone features, 53 tone spreads, 203-5 Tone stability, 194 T O N E TERRACING, 205 T O N E U N I T , 240 T O N I C STRESS, 242 T O N I C SYLLABLE, 240, 242 TRACHEA, 2 TRISYLLABIC LAXING, 139,

263 Trubetzkoy, 38-39 TYPOLOGY, 156 TYPOLOGICAL approach, 39

UMLAUT, 137-8 UNDERIVED lexical item, 259 UNDERLYING

REPRESENTATION (UR), 83, 117, 121, 132, 135-6, 142-44, 150

UNDERSPECIFIED consonant, 146

UNITARY S O U N D S , 61 UNIVERSAL

P H O N O L O G I C A L PROCESS, 109

Universal Grammar, x U N M A R K E D , 39, 99, 102 U N N A T U R A L , 110 U N R O U N D E D , 10 UPSTEP, 28, 30

VELAR, 5 VELARIC INGRESSIVE, 46-7 VELAR S U C T I O N , 46-7 VELIC CLOSURE, 46 VELUM, 7 Vennemann, 140 VOCAL T R A C T , 4 VOICE ASSIMILATION,

81-84, 88-89 VOICE QUALITY, 247 VOICED, 3, 49 VOICELESS, 3, 49 VOICING, 3 VOWEL, 24, 25, 27 (see also CV

phonology) production of vowels, 8-12

vowel deletion, 145-6 Vowel harmony, 211-4 Vowel neutralisation, 142-3 Vowel truncation, 182

WEAK (w), 225 WELLFORMEDNESS

C O N D I T I O N (WFC). 203-5 WEAK M O D E SUFFIXES, 239 WEAKENING, 103-4 WEIGHT UNITS, 155, 172 WIND PIPE, 2 W O R D STRESS, 222-5

X-bar, 286-7 X-bar syntax, 283