A Course in Phonology

IGGY ROCA AND WYN JOHNSON
A COURS.
PHONC
ih
WMrbeed-
Og)
e
Hg
A Course in Phonology is truly comprehensive covering phonetics,
all areas of linear phonology, lexical phonology, and optimality
theory. It is unique in that it can serve as both a text for a basic
introductory course in phonology and for a more advanced
course. The book is pedagogically solid. Phenomena are first
exemplified for English, then other languages are discussed. The
focus on English builds on what students are familiar with. The
integration of phonetics and phonology in the first part of the
book and the introduction to optimality theory are particularly
valuable. The book stands as a remarkable achievement in presenting
the current state of the field in a pedagogically-friendly way.
Stuart Davis, Indiana University
EE
THE INTERNATIONAL PHONETIC ALPHABET
(revised to 1993, corrected 1996)
Consonants (Pulmonic)
Bp vortae nO
feat] al wp a apn
faut -D9'eimg [>Ss jomsagriee orlie Hei Cah a ae
ap cai i|y cifinint[aseres Tanee aI Era
frrcatve |p Blt vie dis z/f 3s alei[x v[x «|n_s]n al
oP pa
yeild
otniobate
ammo ef a
fricative
Lateral
approximant
Where symbols appear in pairs, the one to the right represents a voiced consonant. Shaded areas denote articulations judged impossible.
Consonants (Non-Pulmonic) Vowels
Voiced implosives Front Central Back
© Bilabial 6 Bilabial > Examples: Close 1ay eu weu

1¥ 0}
Dental d Dental/alveolar P Bilabial
Close-mid e€ &B@—9*€&O— ¥e0O
(Post)alveolar f Palatal it Dental/alveolar 3
Palatoalveolar g Velar k° Velar Open-mid [SUG ) \o ——T AWS o)
Alveolar lateral ec Uvular S’ Alveolar fricative Gs 4

Open aeG& aep
Other Symbols Where symbols appear in pairs, the one to the right
M Voiceless labial-velar fricative & Z Alveolo-palatal fricatives represents a rounded vowel.
W Voiced labial-velar approximant J Alveolar lateral flap
U_ Voiced labial-palatal approximant h Simultaneous f and X
H. Voi i
Voiceless epiglottal icati
fricative Affricates and double articula- : :
Saivoiced epiglottal fricative tions can be represented by two uprasegmentals

symbols joined bya tie bar if ;
? Epiglottal plosive necessary. Primary stress
kp t ; Secondary stress
Diacritics ' founa' tifon

Diacritics may be placed above a symbol with a descender, e.g. 0 I Long ex
" Half-long @”
Extra-short é
_ Voiced st ~ Creaky voiced . Apical : | Minor (foot) group
h Aspirated th db _. Linguolabial Laminal t

lI MajorDe Gntonati
nea Saud
Syllable break Ti.2ekt
More rounded 2 W Labialized (he coh? ~ Nasalized Linking (absence of a break)
Less rounded 2 Palatalized di Nasal release
Advanced U ¥ Velarized t® d¥ Lateral release Tones and Word Accents

Level Contour
5 Pharyngealized + cual No audible release rr Extras
€or | high €or / Rising
Centralized ée ~ Velarized or pharyngealized t Caan High é \\Falling
ns = ~
Mid-centralized e Raised € ( T== voiced alveolar fricative) iS 4Mid e€ al High rising
Syllabic n Lowered e (8 = voiced bilabial approximant) e { oa € A OW EES

é | oe é€ 2)Rising-falling
Non-syllabic € Advanced tongue root e
: Lt Downstep 7’ Global rise
© Rhoticity oH a | _ Retracted tongue root e + Upstep \, Global fall
A COURSE IN PHONOLOGY
Iggy Roca and Wyn Johnson

University of Essex
BLACKWELL
feed itishers
Copyright © Iggy Roca and Wyn Johnson 1999
The right of Iggy Roca and Wyn Johnson to be identified as the authors
of this work has been asserted in accordance with the Copyright,
Designs and Patents Act 1988
First published 1999
24681097531
Blackwell Publishers Ltd

108 Cowley Road
Oxford OX4 IJF
UK
Blackwell Publishers Inc.

350 Main Street
Malden, Massachusetts 02148
USA
All rights reserved. Except for the quotation of short passages for the purposes
of criticism and review, 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 the prior permission of the publisher.
Except in the United States of America, this book is sold subject to the condition that it shall
not, by way of trade or otherwise, be lent, resold, hired out, or otherwise circulated without
the publisher’s prior consent in any form of binding or cover other than that in which it is
published and without a similar condition including this condition being imposed on the
subsequent purchaser.
British Library Cataloguing in Publication Data
A CIP catalogue record for this book is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Roca, Iggy.
A course in phonology/Iggy Roca and Wyn Johnson.
aco:
Includes bibliographical references (p. ) and index.
ISBN 0-631-—21345-7 (alk. paper)
ISBN 0-631-21346-5 (pbk. : alk. paper)
1. Grammar, Comparative and general—Phonology. 2. English
language—Phonology. I. Johnson, Wyn. II. Title.
P217.R58 1999
414—dc21 98-51941
Clr
Typeset in 10 on 13 ptPalatino
by Graphicraft Limited, Hong Kong
Printed in Great Britain by T. J. International, Padstow, Cornwall
This book is printed on acid-free paper

Unless it grows out of yourself
no knowledge is really yours,
it is only a borrowed plumage.
D. T. Suzuki, An Introduction to Zen Buddhism
To Morris and Noam,
who created generative phonology.
CONTENTS
Preface XV
Acknowledgements xvii
Reading Logistics XViii
Key to Symbols Xx
Part |: Phonetics and Phonology 1
1 How Are Sounds Made? The Production of Obstruents 8

1 Speech Sounds 3
2 Fricatives: Place and Manner of Articulation 6
3 Phonetic Transcription i
4 A Hissing Fricative 2
5 A Fricative in the Back of the Mouth 11
6 A Laryngeal Fricative 14
7 Voice 16
8 The Stop Gesture 19
9 More Stops 20
10 Still More Fricatives 2s
11 Affricates 25
12 Summing Up 26
Key Questions 29
Further Practice 29
2 Introducing Phonology: Assimilation 31

1 On How Bilabial Stops Become Labiodental ol
2 Total Place Assimilation in Stops 34
3 Voice Assimilation 38
4 The Organization of Language 41
5 Basic and Derived Forms 45
6 The Formalization of Rules 49
7 Derivations Dz
8 Phonetics and Phonology 54
viii Contents
Key Questions 56
Further Practice De
3 Sonorant Consonants 58
General Properties of Sonorants 59

Nasality 60
The Universal Nasal 63
Other Nasal Consonants 65
Liquids 69
Laterals 70
HR
NH
WH
GHW
ND Rhotics 74
8 Summing Up 81
Key Questions 83
Further Practice 83
4 Natural Classes of Sounds: Distinctive Features 85
Descriptive Phonetic Parameters 86

Distinctive Features 87
Naturalness and Formal Economy 90
Place Assimilation in Nasals: Natural Classes 92
The Feature “Coronal”. Active and Passive Articulators 96
Single-Value Features oF
Constraining Rules: Autosegmental Formalism 99
Functional Groupings of Features 102
PR
NH
W
DAK
OON Feature Dependencies 105
Key Questions 112
Further Practice 112
5 Vowel Sounds: Cardinal Vowels 114
1 On What Vowels Are and How They Are Made 143

2 The Two Basic Cardinal Vowels 117
3 The Four Corner Primary Cardinal Vowels:
Two Axial Parameters 119
Four Perceptually Intermediate Primary Cardinal
Vowels: The Roundness Parameter 123
Cardinal Vowels and Real-World Vowels: Diacritic Symbols 126
Some Vowel Typology: The Basic Vowel Triangle 127)
Quantum Vowels 129
Secondary Cardinal Vowels: Front Round Vowels 130
Back Secondary Cardinal Vowels 134
AN Central Vowels
ye DO
SM 137
Contents iIX
Key Questions 140

6 Phonological Processes Involving Vowel Features 143

Distinctive Features for Vowels 144
Feature Dependencies 146
Two More Distinctive Features 147
Back Harmony in Turkish 149
Lexical Underspecification 152
Vowel Disharmony 154
The No-Crossing Constraint 157
German Umlaut 160
W
DAK
OON
HH English Plurals 164
Key Questions 167
7 The Vowels of English 169

Variation in English 169
The Four Corner Vowels 173
Intermediate Primary Vowels 178
More Lax Vowels 181
Central Vowels 186
Homogeneous Diphthongs 190
Heterogeneous Diphthongs 195
CON
HR
NH
W
DTHCentring Diphthongs 199
Key Questions 202
8 The Timing Tier and the Great Vowel Shift 204
A Puzzle with Affricates 205

The Timing Tier 207
A Strange Set of Vowel Alternations in English 210
Short ~ Long Vowel Alternations Ae.
The Great Vowel Shift 214
BM
NH
WO
DABThe Synchronic Reflex of the GVS. Vowel Primes
and Vowel Processes 218
7 The SPE Account 221
8 Further Repercussions of the Vowel Shift 223
9 Multidimensional Phonology: The Skeleton 226
Key Questions 229
x Contents
Part Il: Suprasegmental Structure 239
9 The Syllable 235
The Shape of Children’s Early Utterances 236

Structure of the Core Syllable 238
Sonority and the Syllable 241
The Coda 242
The Rime 244
Basic Syllable Typology 245
The Nature of the Syllable 248
OTK
HR Complex Nuclei
COND
WH 249
© Complex Onsets 252
10 The Sonority Hierarchy 239
11 Sonority Distance 256
Key Questions 258
10 Syllable Complexity: English Phonotactics 261
my Complex Codas 262

Non-Vocalic Nuclei 265
Vowels in Disguise 268
Onset Vowels. The “OCP” 271
Syllabification of [tu] 275
Onset Fulfilment 278
Onset Maximization. English Stop Allophony 279
No Complex Codas in English 284
D
LO
LW
OON
TH The Antics of /s/ 288
Key Questions 294
11 The Phenomenon of Stress: Rhythm 294

Syllable Prominence 295
Word Prominence 296
Metrical Grids 299
Motivating Stress Constrasts 300
The Distribution of Stress in Personal Names 302
Stress Retraction under Clash 304
Word-Internal Stress Retraction 307
Retraction Failures: The Continuous Column Constraint 310
Rhythm 312
ySBS
to
Gon
HS
OT
Gy
NI
Sop
SF Segmental Evidence for Stress: Vowel Reduction 314
Contents Xli
M Stop Allophony 315

Key Questions 318
12 Metrical Principles and Parameters 321
English Phrasal and Compound Stress 321

Extrametricality a0
The Elsewhere Condition 325
Stress Assignment in Words 324
Basic Stress Pattern of English Nouns O27,
The Metrical Foot 331
Main Word Stress in English 3a3
Multiple Stress 335
Stress Typology: Metrical Parameters 338
Word-Level Stress: Line Conflation
ms DAR
WNP
SHAN 342
Key Questions 346
13 Syllable Weight. Further Metrical Machinery 349
— An Additional Pattern of Stress in English 349

Syllable Weight and Metrical Accent voz
The Word-Final Consonant aoe
Long Vowels in the Last Syllable 359
Moras 361
Foot Structure and Universal Rhythm 365
Non-Rhythmic Stress 368
Unbounded Feet 371
LY
W
AHR
DD Idiosyncratic Accent
OWON 373
Key Questions 379
14 Tonal Phonology 382
The Phenomenon of Intonation 383

The Mechanics of Intonation 384
The Primitives of Intonation 385
Autosegmental Intonation 387
Stress and Intonation 389
Non-Lexical Tones 390
Three Types of Intonational Tones 391
Sentence Intonation 392
WH Tone Languages
OWON
BR
DAK 393
XIIii Contents
10 Pitch Accent Languages 395

11 Principles of Autosegmental Association oo9
12 Floating Tones 402
Key Questions 406
Part Ill: Advanced Theory 409
15 Modes of Application: The Cycle 411
Staged Grid Construction 412

Cyclic Tone Association 415
Non-Cyclic Refooting 418
Final Stress Retraction 422
Vowel Shortening 424
Strict Cyclicity 427
Non-Cyclic Accenting 429
Word-Internal Stress Cycle 432
The Structure of the Word-Final Syllable
SOON
WNP
DAK 436
Key Questions 441
Further Practice 44]
16 Domains of Application: Lexical and Prosodic Phonology 444
Three-Mora Feet? tt
Violations of the Three-Syllable Window 446
Cyclic and Non-Cyclic Affixes 449
The Interaction between Morphology and Phonology 451
The Scope of Peripherality 455
Word-Internal Cohesion: The Bracket Erasure Convention 457
Non-Cyclic Processes 459
WN Ordered Affixes
CON
PR
DOB 465
Lexical Phonology: Problematic Orderings 469
The Phonological Phrase 472
The Intonational Phrase 476
The Phonological Utterance 480
Properties of Phonological Domains 481
Subphrasal Phonological Domains 483
mRAR Segmental Affiliation to the Phonological Word
RRR
Oo
SO
WHR 486
16 Small Phonological Words 488
Key Questions 494
Contents xiii
17 Aspects of Lexical Representation: Underspecification,

Markedness and Feature Geometry 498
Effects of Strict Cyclicity 499

Lexical Underspecification 502
Feature Transparency as Underspecification 504
Underspecification and Markedness 507
The Theory of Radical Underspecification 511
Problems for Radical Underspecification 516
Contrast-Restricted Underspecification 521
RP
WHY
DAR
ONFeature Dependencies O22
Feature Geometry 524
Class Nodes 526
Relations between Vowels and Consonants 329
Redundancies between Features Dot
Ts Privative Features 534
Key Questions 539
Further Practice D0.
18 Rules and Derivations 543
Rule Ordering: Feeding and Counterfeeding 543

Bleeding and Counterbleeding 549
Transitivity 552
Palatalization 556
Further Twists 561
Vowel Length Alternations. Tensing 565
Cyclic Rules 570
WN Non-Cyclic Rules
CON
PR
DAB 576
Key Questions 580
19 Constraints: Optimality Theory 584
Naturalness of Phonological Inventory: Markedness 585

Constraint Ranking: Faithfulness 588
Structural Constraints: Syllables 592
The Generator. Tableaux 594
Basic English Syllables 596
Syllable Complexities 600
Basic Metrical Structure 605
Extrametricality 607
Quantity-Sensitivity 609
mS WN
SOW
PR
DAK
AN Secondary Footing 612
XIV Contents
11 Correspondence Constraints 613

12 Cyclic Effects 617
i Word Formation through Truncation 619
14 OT Morphology: English Plurals 620
15 English Possessives and Correspondence Theory 624
Key Questions 627
20 Looking Back and Moving On 630

Phonetics 631
Foundations of Phonology 634
Syllables 637
Stress 639
Tone 641
The Interaction between Morphology and Phonology 643
Phonological Domains . 646
Aspects of Lexical Representation 648
Derivational Theory 651
ym SO
WNP
DAB
AN Optimality Theory 654
Key Questions 660
References 665
Glossary 683
Index of Languages 705
Index of Names 709
Index of Subjects Lo
PREFACE
The field of phonology finds itself in the heat of a revolution. The 1950s
and 1960s saw the triumph of generative phonology over its taxonomic pre-
decessor, and with it the establishment of a new way of doing phonology,
consolidated through the 1970s and 1980s. The 1990s have witnessed the emer-
gence of Optimality Theory.
Writing a textbook in the midst of revolutionary turmoil would seem like
bad timing: aren’t textbooks distillations of what philosophers of science have
referred to as “normal science”? While accepting the truthfulness of this proposi-
tion, we would like to argue that the present state of phonology resembles
more a ripple in the ocean (perhaps a rather deep ripple) than an earthquake.
In particular, we contend that Optimality Theory is still very much generat-
ive phonology. Indeed, the basic constituent elements of Optimality Theory
are identical to those of classical generative phonology: a collection of para-
digmatic data, a modelling of such data into levels of representation (perhaps
only two in OT), and a procedure to effect the mapping between these levels.
It is here, of course, that the two theories appear most strikingly to part
company: classical generative phonology has rules and Optimality Theory
constraints. However, the forms that these constraints evaluate (the “can-
didates”) evidently need to have been composed in the first place, by pro-
cedures which, more likely than not, resemble rules.
In the pages that follow we guide the beginner step by step through the
wonders of phonology, endeavouring at all times to remain non-doctrinaire:
theories provide the mould, but the dough is obviously theory-independent,
a foundation the learner needs to become fully conversant with before diving
into deep theoretical waters. The presentation of facts in a formal vacuum
is, however, ultimately counterproductive, if not simply impracticable.
Accordingly, we have largely relied on the familiar rule format, always keep-
ing sight of the current appetite for constraints. Our goal has been to write
a text which is sufficiently clear for learning to take place, and sufficiently
complete to be worthy of the title “A Course in Phonology”. In order to achieve
this aim, we survey all the major building blocks of phonology, which, by
their very nature, are shared by classical generative phonology and Optim-
ality Theory, and we also supply a preliminary foundation in articulatory
phonetics. We have striven for clarity of style and have left the text free of
Xvi Preface
the usual reference clutter, providing instead the fabric of the underlying
scholarship in chapter 20. We interact with the reader as we go along and
use constant quizzing in the shape of grey boxes. We have a strong belief
that learning grows best out of the learner’s experience, past and present.
As a consequence, we have shied away from potentially mystifying exot-
icism, and structured the presentation around readily accessible facts of
English, while still catering for the exotic through references to foreign data
where appropriate. We dispense advice on how to approach the text with
maximum efficiency in the “Reading Logistics” section that follows this pre-
face and the acknowledgements.
We have deliberately taken sides with the learner, against the expert phono-
logist, whenever we felt a conflict. The expert’s vista of the field, the result
of many years’ hard work, is unavailable to the beginner, whose needs are
necessarily at variance with the needs of the expert. Indeed, the greatest
challenge to teachers and textbook writers alike involves abandoning our
vantage point and trying to sit inside the mind of the average learner. This
is not an easy undertaking, but we have endeavoured to carry it through
to the best of our ability. The result may at times be slightly irritating to the
expert, from whom we beg forbearance. Hopefully, however, it will genuinely
meet the needs of the learner and induce true understanding.
The book contains three interrelated but clearly distinct parts, of approx-
imately equal length. Part I presents the foundations of both phonology and
articulatory phonetics, and explains the differences between these two dis-
ciplines. It also supplies the basic tools of the phonological trade as we have
known it for almost half a century now: features, rules and derivations, all
couched autosegmentally. Part II deals with the familiar suprasegmental
aspects of phonology: syllables, stress, and tone and intonation. Part I is appro-
priate for a short introductory phonology course, by itself or as part of a
phonetics component. Parts I and II together provide a solid foundation in
phonology at an intermediate level. Part III takes the reader to a substan-
tially more advanced level of both phonological awareness and theoretical
sophistication, dealing with such matters as the cycle, multilevel phonology,
prosodic phonology, markedness, underspecification, feature geometry, and
Optimality Theory. Many of the issues we raise are given practical treat-
ment in the companion volume A Workbook in Phonology, which includes
a large section on Optimality Theory to complement the exercises in the
present book.
Having invested much time and energy in the book, we have high hopes
that readers will not need to look back, metaphorically or literally, but will
be empowered to take the leap into full enjoyment, both passive and active,
of the marvels of phonology. If this aim is achieved, we will feel modestly
proud of having provided the soil in which a true and durable knowledge
of this exciting field of inquiry has taken root.
ACKNOWLEDGEMENTS
A book of this kind inevitably owes an incommensurable debt to many more

than memory or paper availability can realistically account for, and there-
fore we shall limit ourselves to mentioning those who have had a direct major
part in its elaboration. First among these are Bernard Tranel, Kevin Varden
and other Blackwell’s anonymous readers, all of whom made an invaluable
contribution to the shaping of the book through the several years of writing.
For phonetic information, we are particularly grateful to Judith Broadbent,
Dick and Katrina Hayward, Paul Kerswill and Linda Shockey. Of our own
students, we must single out Nick Sherrard and, in particular, Paula Reimers,
who carefully read through the whole final draft and offered us most
valuable painstaking comments. Finally, Philip Carpenter skilfully, if not
always painlessly, steered the project to success through its various phases,
and was eventually joined by Simon Eckley, Jack Messenger, Fiona Sewell,
and the remainder of the Blackwell team. To all these, and to the many
others who contributed to the project in a variety of ways and capacities,
our deepest gratitude. It goes without saying that any errors or infelicities
that may have escaped scrutiny are of our sole making.
READING LOGISTICS
This book has been written with you, the reader, in mind. Our aim has been
to present phonology in a way that is both comprehensible and memorable,
and to this end we have resorted to a variety of pedagogical devices. We
have endeavoured to use a clear, honest-to-God style from beginning to
end, giving you the gist of the theory rather than the maze of opinions and
counteropinions. Because we are fully aware that even the clearest of expla-
nations is not sufficient to ensure rapid understanding and memorability,
we make use of a panoply of mnemonics, as follows:
¢ each chapter opens up with a list of key points to be covered, hopefully

in a style that you will understand at once
¢ throughout the chapter, we purposely interact with you by means of grey
boxes, which should ensure that you connect with the text
* margin notes summarize the main points in the text as we go along: you
should definitely endeavour to remember them
e the contents of each chapter are summarized in a dark box at the end
¢ each chapter is followed by a set of key questions for you to answer,
and bya set of exercises for further practice
¢ each of the three parts of the book is preceded by a concise summary of
its substance
¢ chapter 20 further sums up what we say throughout the book
In addition, we give all technical terms in full capitals in the place in the text
where they are defined or become central. Many of these terms, and a number
of others, are brought together in a glossary at the end of the book, to facil-
itate matters further. We also include a copy of the latest official version of
the IPA alphabet, and a key to the special symbols we use throughout the
book. A list of references follows chapter 20 to enable you to take your
reading as far afield as your need or your yearning may drive you. Last,
but not least, there is a subject index, a language index, and an author index.
Our advice to the reader is to make use of all these facilities in the man-
ner that best fits personal needs. For instance, if you are a complete begin-
ner, you will probably gain much benefit from reading the matter in the grey
boxes, more or less systematically. If you already know some phonology,
Reading Logistics xix
however, you may find some of these boxes distracting, particularly in the
preliminary chapters, and therefore you will be best advised simply to skip
them. Similarly, you may wish to ignore the grey boxes, or at least some of
them, in subsequent readings of the text. In a nutshell, we provide you with
a range of materials, grey boxes included, for your benefit, not for your
punishment: it is you who must decide how many of these materials you
use and when. The whole idea is that the text should not be approached as
inert matter, but, rather, as a flexible interactive partner. We obviously wish
you success. We also hope that you will genuinely enjoy the task.
KEY TO SYMBOLS
The following list includes most of the special symbols used in the book.
You must pay attention to the fact that many of them are used for several
purposes, often unrelated or even contradictory.
ANGLED BRACKETS:
indicates a two-way implication between the strings thus enclosed;
marks extrametricality (or, more generally, extraprosodicity) of the mater-
ial thus enclosed
ARROW:
signals that the material on its left ‘rewrites as’ the material on its
right
ARROW HEAD (DOUBLE):
used in OT to mean ‘ranked higher than’
ARROW HEAD (SINGLE):
signals that the material on its left ‘precedes’ the material on its right
ASTERISK:
marks the ungrammaticality of the material it is prefixed to;
signals metrical prominence, thus making up metrical grids;
in OT tableaux, it signals constraint violation;
as a superscript, it indicates optional recursion;
signals “accent” in both stress and tones
BRACES:
enclose several subrules to signal their joint participation in a single
rule, as in the English Vowel Shift;
indicate disjunctivity of features, as in the environment “[+sonorant]
or [-continuant]” in the English rule of Spirantization
COMMA:
in OT, it signals lack of ranking
DASH (SWUNG):
negates the material it is prefixed to;
separates alternants
EXCLAMATION MARK:
in OT, it indicates a fatal constraint violation;
Key to Symbols xxi
in tonal phonology, it is frequently used to signal downstep (the IPA

symbol is J, with ! standing for a (post) alveolar click)
IMPLICATION SYMBOL:
indicates that the material on its left ‘implies’ the material on its right
MU [mju]:
a shorthand for “mora”
) PARENTHESES:
enclose feet;
enclose optional material
POINTING FINGER:
in OT tableaux, it singles out the winning candidate
SIGMA (CAPITAL):
a shorthand for “foot”
SIGMA (LOWER CASE):
a shorthand for “syllable”
SLASH:
separates focus from environment in rules
SLASHES:
enclose underlying representations;
enclose phonemic material
SQUARE BRACKETS:
enclose distinctive features;
delimit distinctive feature matrices;
enclose constituents (phonological, morphological or syntactic);
enclose surface representations;
enclose phonetic material
SUBSCRIPT NOTATION:
an alternative to the class node formalism
TIE BAR:
links two phonetic symbols to indicate the unitary nature of the sound
iy aod? of Ysa.» Lind pe a ee anata
bn oy ee
AT alt) quianvob
fan
alert re
inate
ane “lye a a hh
g ping « eee a
M 11 ly = - 4 [= a eictary ean
suedennosi
. “aieislive” woa & .
; ae 4a? 6
pale tal pny, 40g Ghee sn betoean Oe rah
:. | sdencidicermesregst ee
eget Gel® Go setae on rae
Oo
S”6h
as
ee
= . SPD AAB AS an
ae. re werd remanent of ons
agai ee ete stent
i oper
to p einen
=
-_—-
2.
biel ophap we
anes sane sn laid
‘ye Sey aac balay
i] asY
ee yatts
naparetpason
pile, 9 wm Sor ingle et Get
inulicaio (adm Dihp cd Geet ostn tn oretnemanhtt
or [-uorieusté? tthe Ghali rele af Gokpemileetiony *
COMLIA 7
Ligh <F ep kieg
in (IT gigas
~ DASH (SIVLe
Reyaler te onarkelf eae
mepunanie marie
; Le LAMA TION MAS
iw Ci,W lentes @¢
PHONETICS AND PHONOLOGY
This book is about sound in language. The first distinction we must draw
is between sound as sound (“phonetics”) and sound structure (“phonology”).
By sound as sound we mean the sound we make with our vocal organs when
we speak, an activity equivalent to the playing of the instruments by an orches-
tra: it is as if we were all carrying an orchestra in us! These sounds have
their own physical characteristics, which can be described: here we focus
on the description of the movements we make with our organs in order to
produce such sounds (“articulatory phonetics”). The sounds made by an
orchestra are not random, however, but, rather, the materialization of a sym-
phony. Similarly, the sounds we make when we speak any particular lan-
guage implement the particular symphony that constitutes the phonic
structure of that language (“phonology”). Here we differentiate two levels,
which we can illustrate with the two ps of the English word paper: they are
the same “sound”, but still sound different. They are the same sound in as
much as they are not bs, or ts, or ks, etc. They sound different in as much
as you could conceivably blow out a candle with the first p of paper, but not
with the second one. Therefore, the message is that each language possesses
an inventory of distinctive sounds (“phonemes” or “lexical segments”,
depending on the theoretical framework), but that each of these sounds can
have a number of different realizations (“allophones” or “surface sounds”),
contingent on environmental conditions, a bit like the way we dress up warmly
when we climb up a frozen mountain, but undress when we lie in the
sun on a beach: one important source of sound alternation thus concerns
“contamination” by neighbouring sounds, a process globally known as
“assimilation”. In order best to understand assimilation, and other sound
alternations, we decompose sounds into atomic units, or “features”, to
which we grant a certain degree of mutual autonomy: a p, therefore, is not
a p, but a web of interconnected features, just as the hand with which we
write is basically a constellation of molecules. The distinctive features of sounds
are grounded in the gestures involved in their articulation, and thus in phon-
etics. Formally, features connect to each other by means of “autosegmental”
lines. Sound processes mapping lexical sounds onto surface sounds (equi-
valently, but not identically, phonemes onto allophones) involve readjust-
ments of such lines, a bit like a game of musical chairs. The linearization of
2 Phonetics and Phonology
speech is expressed in a line, or “tier”, of abstract timing units, also known

as the “skeleton”: long sounds, like the o in vote or the ee in feet, are associ-
ated with two such units, and short sounds like the o in cot with one. The
introduction of the skeleton allows us naturally to differentiate processes
involving length changes from processes involving changes in the sub-
stance of the segments.
How ArE SOUNDS MADE?
THE PRODUCTION OF OBSTRUENTS
In this chapter you will

m What is characteristic
a variety of sounds.
The primary aim of this book is to present the principles and practice of
current phonology in a manner which is both accessible and stimulating
to the uninitiated reader. Phonology is the study of linguistically signific-
Phonology is the
ant sound patterns, that is, of the organization of the sounds of speech. This study of linguistic-
definition will become clearer as we proceed. In order to study the organ- ally significant
ization of speech sounds, we must first be able to identify the sounds them- sound patterns
selves, and we make a start on this task in the present chapter.
Speech Sounds
We can compare the act of speaking with the act of playing a recorder, with
which many of us are familiar from childhood. You may of course substi- The act of speak-
ing can usefully
tute any similar instrument: the analogy will still hold. To start with, when
be compared with
you want the recorder to produce a noise, you have to blow air from your the act of playing
lungs through the mouthpiece. The sounds that we produce when we speak a recorder
also need to be powered by air from the lungs. In fact, the physical act of
speaking can be likened to “playing” our mouths and larynxes with the air
coming out of the lungs through the windpipe linking the lungs to the mouth,
technically known as the “trachea”. The LARYNX is of course the voice box
4 How Are Sounds Made?
er
a
Se
a
Figure 1.1 Playing a recorder
at the top of the trachea, which in the throat of males protrudes as the Adam’s
apple. A picture is worth a thousand words, and therefore we will adopt
the practice of illustrating many of our statements with drawings (figures
1.1 and 1.2, above and on the next page, respectively).
At this point, a question arises. We take air into our lungs and let it out
again every time we breathe. How come then that we are not continuously
producing speech sounds? An easy experiment will give the answer. Quite
simply, if you pick up a recorder and place the mouthpiece to your lips while
breathing normally, you will find that the recorder only makes a faint
wheezy sound, and that in order to play the recorder you need to discharge
an extra amount of air. In the same way, if you wish to make a speech sound,
you must breathe out more air than usual, and, of course, in order to
breathe out more air, you must have taken in more air in the first place.
How Are Sounds Made? 5
Figure 1.2 Speaking
You have now got the recorder making a noise. In order to play a tune,
however, you need to do more than blow extra air through the recorder.
Specifically, you need to move your fingers over the different holes on the
body of the instrument. Why? Suppose you leave all the holes uncovered.
Air will then come out through all the holes simultaneously, and the
recorder will play a single note. Covering different holes at successive times
will, however, enable you to produce different sounds. By covering (some
of) the holes you are effectively putting up obstacles to the exit of the air
through those holes. Similarly, in order to make the various speech noises,
we have to interfere in various ways with the flow of the air on its way out
through the larynx and, especially, the mouth. We will now look at some
of the ways in which we “play” our vocal organs, starting off with the mouth.
Fricatives: Place and Manner of Articulation
Suppose that you place your lower lip loosely on the lower edge of your
upper teeth and force air out of the mouth, after having filled your lungs
aplenty. The situation is depicted in figure 1.3 (any new technical terminology
in the drawings will be explained directly).
Figure 1.3 Labiodental fricative
What will happen? The obstacle you have created by the loose contact of
the lower lip with the upper teeth does not totally block off the air, and there-
fore air will continue to flow out.
Because the teeth and the lip are touching each other, the air rushing past
will cause friction, and therefore a noise will be produced.
The noise made by the air escaping between the lower lip and the upper
teeth will sound exactly like the f at the beginning of fat, simply because
this is how f is made, or “articulated”. The ARTICULATORS of f are thus
How Are Sounds Made?
the lower lip and the upper teeth, hence the label LABIODENTAL given to
this type of sound: labial is a Latin-based adjective meaning ‘of the lips’, and
dental, related to dentist, means ‘of the teeth’. Because the noise made by the
air escaping through the obstacle being described is caused by friction, we
refer to this type of sound as a FRICATIVE. Accordingly, the sound f at the
beginning of fat can be (partially) described as a “labiodental fricative”.
“Labiodental” defines the place where the sound is made, its PLACE OF
ARTICULATION, and “fricative” defines the manner in which the sound is Where a sound
made, its MANNER OF ARTICULATION. is made is its
PLACE OF
ARTICULATION.
How a sound
Phonetic Transcription is made is its
MANNER OF
ARTICULATION
The labiodental fricative we have just discussed is the first of a range of fricat-
ive sounds which we are going to explore. Before we do this, however, it
will be useful to introduce a method to write sounds down. An analogy with
numbers will make the issue clearer. We call a certain number, say, “one
thousand five hundred and sixty-five”, but we do not usually write out this
lengthy expression; indeed, calculations would be very difficult if we did.
Instead, we use a shorthand version with the four digits one-five-six-five:
1565. In a similar way, we can and shall use a single symbol as a shorthand
for the phrase “labiodental fricative”.
How are we going to write down speech sounds? At first sight, this looks
like a rather silly question — surely we already have a method of writing
down speech sounds: we call it a “spelling system”, the very one we are
using right now to put our thoughts on paper. So, you may think, of course
we are going to write the first sound of fat as f, the second sound as a, and
the third sound as t — what else is the spelling there for? Well, there is actu-
ally something rather unsound about this line of reasoning. One example
will bring out the problem. If indeed we are going to write the first sound
of fat as f because that’s the way it’s spelled, are we also going to write the
last sound of laugh as gh because that’s the way it is spelled, and the first
and the penultimate sounds of philosophy as ph? The point is that the first
sound of fat and the last sound of laugh are identical, but their spellings are
not. This is a bit like writing the number one sometimes as “1”, and other
times as “4”, “23”, etc.: clearly confusing! In fact, of course, there is no
reason not to write “one” with the same symbol every time, say, as “1”. The
reasoning carries over to sounds. The best system of representation is thus
one where the same symbol is always and only used for the same sound.
This way there will be a one-to-one correspondence between each sound and
each symbol, and we will be able to work out what sound we are referring
to just by looking at the symbol - always assuming familiarity with the table
of sound-symbol correspondences, which we will be supplying as we go along.
Clearly, ordinary English spelling is very far from being an adequate sys-
Ordinary English tem for transcribing sound. In fact, no conventional spelling of any living
spelling is very far
language is, but English orthography is notoriously further removed from
from being an
adequate system
the ideal than average, as highlighted by Bernard Shaw’s famous witticism
for transcribing that the word fish could equally well have been written ghoti: gh as in laugh,
sound 0 as in women, and ti as in nation.
few:
n
handful ofE
English spelling is based on the late medieval pronunciation of English, and

is indeed grossly inadequate for representing the way the modern language
sounds. Awareness of spelling shortcomings spurred on the birth of systems
of phonetic transcription in Britain and other European countries from at
least the sixteenth century. At the end of the nineteenth century, a group
of language teachers and phoneticians led by the Frenchman Paul Passy set
up the International Phonetic Association and devised one such trans-
cription system — soon to join was the British phonetician Henry Sweet,
traditionally thought to have inspired the character of Professor Higgins
in Pygmalion/My Fair Lady. The transcription system of the International
Phonetic Association was gradually enriched to make it applicable to all lan-
guages, and it has by now been adopted by most practising phoneticians
the world over. The system is known as the “alphabet of the International
Phonetic Association”, “IPA alphabet” for short. It is called an alphabet because
The goal of the
IPA alphabet is to
it is based on letters — in fact, as we will see, essentially the letters English
have each symbol is spelled with, the Roman alphabet. The chief remit of the International
always stand for Phonetic Association is to ensure the well-being of the international phon-
the same sound, etic alphabet, which it regularly updates and publicizes.
and only for that From now on, when we describe a sound, we will also give its IPA sym-
sound, and, con-
versely, to have
bol. Remember that the goal of the IPA alphabet is to have each symbol always
each sound always stand for the same sound, and only for that sound, and, conversely, to have
represented by the each sound always represented by the same symbol. You will be pleased to
same symbol learn that the IPA symbol for the first sound in fat that we have been describ-
ing is [f]: conventionally, phonetic symbols are enclosed in square brackets.
You should equally not be surprised to hear that [f] also represents the last
sound in laugh, and the first and penultimate sounds in philosophy. Square brackets
are conventionally
used to enclose
phonetic symbols
EY A Hissing Fricative
We now proceed to the description of other fricative sounds. How does the
articulation of the sound at the beginning of sip differ from the articulation
of [f]?
When we pronounce s, as in sip, we place either the “blade” or the “tip” of

the tongue close to the ridge out of which the top teeth grow. The BLADE The BLADE is eas-
ily recognizable as
is easily recognizable as the most mobile and versatile part of the tongue,
the most mobile
located behind the front point, itself known as the TIP (many non- and versatile part
specialist people actually call the whole blade of the tongue the “tip”). The of the tongue,
blade is in fact the part of the tongue that sticks out most easily (figures 1.4 located behind
and 1.5). the TIP
Figure 1.4 The tip and blade of the tongue

Figure 1.5 Alveolar fricative
If you blow air through the narrow gap between the blade of the tongue
and the upper tooth ridge, in the manner you did through the gap between
the lower lip and the upper teeth for [f], a hissing noise will be produced,
which we transcribe as [s].
Pronounce the tw
attention to the dit
Now some terminology in connection with [s]. Because [s] is articulated

on the (upper) tooth ridge, it is given the label ALVEOLAR: alveolus is the
Latin word for ‘socket’, and the teeth obviously grow out of “sockets”, hence
the expression “alveolar ridge”. We can therefore describe [s] more fully (but
still partially) as an “alveolar fricative”.
Note that, strictly speaking, we ought to use the label “linguoalveolar”: both
the alveoli and the tongue, lingua in Latin, contribute to the articulation of
[s]. The prefix linguo- is omitted on the grounds that the tongue is the organ
most commonly involved in sound articulation, its intervention therefore
being taken for granted in the absence of information to the contrary.
A Fricative in the Back of the Mouth
We have now described the articulation of two fricative sounds pro-

nounced in the front region of the mouth: [f] (labiodental) and [s] (alve-
olar). The mouth is quite a big cavity, and other parts of it can also be used
in sound production. In fact, there are so many locations available that typ-
ically languages do not make use of all of them.
The tongue is the most important articulator, and is almost exclusively
composed of muscles, eight in all, some of them quite long — surprisingly
so when you consider we can comfortably tuck the whole lot into the
mouth. In addition, the wealth of innervations and the complex arrangements
of the fibres of its muscles give the tongue a remarkable degree of versatil-
ity, which makes it particularly apt for the articulation of speech sounds.
Consequently, there are many sections of the tongue available for contact
with the structures that lie above it from the teeth to the pharynx. Note There are many
sections of the
that, contrary to popular belief, the PHARYNX is the backmost part of
tongue available
the mouth, above the larynx, rather than the uppermost part of the throat for contact with
(pharyngitis is therefore an inflamed back mouth, rather than a sore throat). | the structures that
See figure 1.6. lie above it from
the teeth to the
pharynx
Figure 1.6 The pharynx, larynx and tongue root
We have explained that [s] is made by positioning the blade of the

tongue in the vicinity of the alveolar ridge. Suppose, instead, that you
make the constriction at the back of the palate, that is, in the soft palate
region. The PALATE is commonly referred to as the “roof of the mouth”,
and the SOFT PALATE is the soft area of the palate at the back. The tech-
nical term for the sounds made at the soft palate is VELAR, from VELUM,
the anatomical word for the soft palate, derived from the Latin word velum
‘veil’ (figure 1.7).
Figure 1.7 The soft palate
Clearly, the quickest and most comfortable way of creating a velar stricture
will not be with the blade of the tongue, as the feeling exercise just suggested
is likely to have revealed. Instead, it will be more convenient to use the part
of the tongue which normally lies under the soft palate: the back part of the
body of the tongue. The BODY is the section of the tongue behind the blade,
more massive and less mobile than the blade (figure 1.8).
Figure 1.8 The body of the tongue

Indeed, the ACTIVE and PASSIVE ARTICULATORS involved in the pro-

duction of a sound, that is, the articulator that moves and the inert articu- ACTIVE and PAS-
lator, more often than not lie directly opposite each other. Suppose now that SIVE ARTICULA-
you move the back of the body of the tongue up towards the soft palate, to TORS more often
than not lie
create the same sort of narrow gap as for [f] and [s] (figure 1.9).
directly opposite
each other
Figure 1.9 Velar fricative
The air rushing through will produce the sound that occurs at the end of
loch as this word is pronounced in Scotland, and also the sound in such
German words as acht (‘eight’) or Bach (the composer).
The phonetic symbol for this sound is [x]. You must of course not con-
fuse this symbol with the letter x, which in English corresponds to two
successive sounds.
On the other hand, [x] is spelled ch in both Scots and German, therefore
with two letters, despite being a simple sound. The spelling of [x] in other
languages confirms the arbitrariness of spelling conventions: as either g or j
in Spanish (depending on the vowel that follows), and as h, as well as ch, in
14. INGVaPArESOUNOS ees
Polish. In the face of this diversity, the usefulness of a truly international stand-
ard for the representation of language sounds should be becoming obvious.
{4 A Laryngeal Fricative
We shall now present another fricative sound which is pronounced even fur-
ther back than [x]. In fact, it is articulated so far back that it is not articu-
lated in the mouth at all, but rather in the larynx — as we said above, the
The larynx is the larynx, or voice box, is the part of the throat that can stick out as the Adam’s
part of the throat
apple: we give cross-sections of the larynx in figure 1.10 below. The sound
that can stick out
as the Adam's we are now introducing occurs at the beginning of the word high.
apple
Clearly, there is no obstacle to the airflow in the mouth here, and yet one
can distinctly hear friction. What is happening, then?
Front
To mouth and nose
Side view
Front
Larynx
eee View from above
To lungs Front view
Figure 1.10 The larynx

When the air comes up the windpipe, it obviously has to pass through the
larynx before reaching the mouth. Anatomically, the larynx is a cylindrical The larynx is a
frame made up of cartilage, across which stretch two folds of muscle, called °Y!indrical frame
the VOCAL FOLDS - also, and perhaps more ?
commonly
i
in a non-linguistic Bee ie ees
age, across which
context, VOCAL CORDS, a somewhat misleading term anatomically. The ctretch two folds
vocal folds are shaped like a pair of small lips and are highly mobile - they of muscle, called
are responsible for the sudden reflex movement that prevents us from _ the VOCAL
choking when a foreign body threatens to make its way into the windpipe FOLDS
(figure 1.11).
f=—Vocal folds
Glottis
Figure 1.11 The vocal folds
In order to create the necessary friction for [h], the sound under observa-
tion, we position the vocal folds near each other to create turbulence in the
airflow, but not so close together that they vibrate, as they do when we hum
(figure 1.12).
Figure 1.12 Glottal friction

The space surrounded by the vocal folds is known as the GLOTTIS, and there-
fore [h] is said to be a GLOTTAL fricative.
a Voice
We have just mentioned humming, and a sort of humming plays a very import-
ant role in the production of many of the sounds of speech.
If you bring the vocal folds together closer than for the fricative sound
[h], and tighten them somewhat, although not so much as simply to stop
the airflow, the air passing through will cause them to vibrate, giving a hum-
ming effect (figure 1.13).
This humming effect is known technically as VOICE. Voice can be super-

When the vocal
imposed on any of the sounds that we have already practised, just as we
cords are brought
close together and could get the combined sounds of a trumpet and a clarinet by playing the
tightened some- two instruments plugged into one another, with the same airstream power-
what, the air ing both instruments: the mouth “instrument”, which gives us the majority
passing through of the fricative sounds we have examined, is in effect permanently plugged
causes them to
into the larynx “instrument”, with which we obtain voice. Clearly, this new
vibrate, giving a
humming effect
set of sounds, composed of oral friction (that is, friction at some place in the
known technically mouth) and voice (that is, vibration of the vocal folds) will sound different
as VOICE from the simple set of voiceless sounds we have been reviewing so far: in
this way, we will effectively double the inventory of speech sounds.
How Are Sounds Made? ae
If you repeat the action by which you produced [f] in fat and super-
impose voicing, the result will be a different sound: instead of fat, you will
get vat. The two articulators, the teeth and the lower lip, are in the same
position for both sounds (see figure 1.3 above), which are only differen-
tiated with regard to voicing. The phonetic symbol for this new sound is,
unexcitingly, [v].
The two other oral fricative sounds we described above also have voiced
analogues. If you add vocal fold vibration to the [s] in sip, you will hear zip.
Therefore, the only difference between [s] (as in sip) and [z] (as in zip) is
again voicing: [s] is voiceless, and [z] voiced.
Adding voicing to the sound at the end of loch does not give us a common
sound of English. This is not terribly surprising, since [x] itself is not a sound
of modern English either — loch is originally a Gaelic word, and most non-
Scottish speakers of English pronounce it like lock. The voiced velar fricat-
ive [y] (a phonetic symbol resembling the Greek letter gamma) is found in
Greek and Arabic, among other languages.
nore familiar
fa
We have now presented four voiceless fricatives ([f], [s], [x] and [h]), and
three voiced ones ([v], [z] and [y]). Voicing of [h] may appear impossible,
since the organ responsible for the friction of [h], the vocal folds, is also
the organ responsible for voice: it would seem out of the question to have
one and the same instrument execute two apparently incompatible actions
simultaneously. Surprising though it may seem, we can indeed perform this
feat, given a bit of vocal fold gymnastics: the vocal folds must be placed
close together at one end whilst heldalittle further apart at the other end.
The closed end vibrates, while at the more open end there is air friction
(figure 1.14).
Figure 1.14 The position of the vocal folds for voiced h
In order to identify the new sound, you can compare the h of head with the
h of ahead. In head, there is no vocal fold vibration on the h, whereas in ahead
voice runs through the entire word.
We must of course differentiate graphically between the two types of h,

and accordingly we use the phonetic symbol [fh] for the voiced h. This sound
completes our initial survey of fricative sounds, which we now tabulate
(table 1.1). Notice that the two relevant criteria for the classification are place
of articulation and voicing.
Table 1.1 Fricatives (first inventory)
Place of articulation
Labiodental Alveolar Velar Glottal
Voiceless f Ss x h
Voiced Vv Z v fh
El The Stop Gesture

We now know that the common denominator for fricatives is the gesture
with which these sounds are produced: the two articulators are held close
together, but not so close that the air is prevented from getting through and
causing friction. Obviously, if the contact is tightened up, the airflow will
be interrupted. Surprising though it may seem, momentarily stopping the
airflow is another common method of producing speech sounds: another
“manner of articulation”. We shall now examine exactly how this is done.
Clearly, if you simply block the air, no sound will be made.
onstratethis
However, if after blocking the airflow for a fraction of a second, you

abruptly release the closure to allow the air to rush out, a sound will be If, after blocking
produced. For instance, if you close both lips tightly and then open them the airflow for a
suddenly, you will hear the sound [p], corresponding to the spelling p in fraction of a sec-
the word spy — do bear in mind, however, that you need to have built up ond, you abruptly
release the closure
sufficient air pressure in the lungs for the “vocal instrument” to “sound” at
and allow the air
all. The type of sound we are now introducing is obviously not a fricative, to rush out, a
because the main phase of its realization does not involve the slow friction STOP sound will
that characterizes fricatives. Instead, the articulation of [p] involves air stop- be produced
page, hence the generic name STOPS given to these sounds (figure 1.15).
Figure 1.15 Bilabial stop

et
If you vibrate the vocal folds during the closure phase of the lips, you will
hear the sound [b] of the word obey.
The sounds [p] and [b] share both manner of articulation and place: they
are stops and BILABIAL, since they are articulated with both lips. There-
fore, [p] and [b] only differ with regard to voice.
EE} More Stops
We shall now describe the stop correlates of the rest of our by now fam-
iliar fricatives. If you place the blade of the tongue roughly in the same
position as for [s], but this time interrupt the airflow by tightening up the
contact, you will get the [t] of the word sty (figure 1.16).
Figure 1.16 Alveolar stop

If you increment [t] with voicing, the result will be the [d] of adorn. What
will happen if you stop the air by placing the back of the tongue against the
soft palate, in approximately the same position as for the fricatives [x] and
[y]? The respective stop sounds will now be [k], without voicing, as in sky,
and [g], with voicing, as in again (figure 1.17).
Figure 1.17. Velar stop
ComparelochwithTock ndnotice
We now have all the information we need in order to give the promised
phonetic interpretation to the English letter x: the x of box corresponds to
the sound sequence [ks] (notice that box rhymes with socks), and the x of
exam to [gz] (compare eggs).
The remaining fricatives in the inventory in table 1.1 are [h] and its
voiced counterpart [h]. They are both glottal, as their production involves
air friction in the glottis - remember: the space surrounded by the vocal folds.
Does either of these fricatives have a stop counterpart? If you say ah (as when
asked to do so by the doctor), then close the vocal folds, and then open them
again with another ah, the result will be the sound that is thought of as a
“dropped t”, heard in words like butter or bottle in many British accents
(in bottle also in some American accents).
The technical name for this “dropped ¢” is, unsurprisingly, GLOTTAL

STOP (figure 1.18).
Figure 1.18 Glottal stop
In order to pronounce a glottal stop, the vocal folds come together to close
the glottis, causing a momentary break in the airstream. This closure is then
released suddenly, exactly as with the remainder of the stops. The gesture
involved in making the glottal stop is in fact similar to the gesture involved
in coughing. Glottal stops do have to be voiceless, since it is not possible
for the vocal folds to vibrate if no air is passing through, just as it is imposs-
ible for a flag to flap in the absence of wind. The phonetic symbol for the
glottal stop looks like a question mark without the dot: [2].
The addition of [p], [b], [t], [d], [k], [g] and [?] notably enlarges our invent-
ory of speech sounds, as we now encapsulate in table 1.2, which obviously
supersedes table 1.1 above.
Table 1.2 Stops and fricatives (second inventory)
Place of articulation
Mouth Larynx
Bilabial Labiodental Alveolar Velar Glottal
M Voiceless p t k 2
a Stops
n Voiced b d | g
n Voiceless f Ss x h
e Fricatives
r Voiced Vv Z Yy A
Still More Fricatives
Fricative and stop sounds with the obstruction to the airstream in the
mouth are known as OBSTRUENTS: the glottal sounds [hl], [A] and [?] are _ Fricatives and
therefore not considered obstruents. In stops, the obstruction takes the Se ee
form of a total blocking of the air, while in fricatives the air forces its. GgcrpyENTs
way through the obstruction. You may well think that the list in table 1.2
exhausts the inventory of obstruents. In fact, there are quite a few more in
store, even if we don’t look beyond English.
The sounds we will now examine also come in voiced—voiceless pairs.
If you place the blade of the tongue on either the inside or the edge of
the upper teeth, allowing the appropriate narrow gap for friction, you
will get the sound at the beginning of thigh if you don’t voice, and the
sound at the beginning of thy if you superimpose vocal fold vibration
(figure 1.19).
_@
Figure 1.19 Dental fricative
Notice that our spelling system cannot distinguish between these two
sounds, but the phonetic alphabet of course must. The respective IPA
symbols are [9] (the Greek letter “theta”), for the voiceless sound in thigh,
and [0] (the Old English letter “eth”), for the voiced sound in thy.
Note that it is also possible to make stop sounds on the inside of the upper
teeth. Indeed, this is the place where speakers of Spanish pronounce their
ts and ds (as also do some speakers of English). This is also the usual ren-
dering by Southern Irish speakers of the common English sounds [6] and
[0], which thus still remain distinct from the alveolar stops [t] and [d].
Another fricative we have not yet discussed is the first sound in ship, also
used extralinguistically to call for silence (shhh!). The phonetic symbol for
this sound is [f]. English spelling has some difficulty in representing [J] -
normally sh; but also, if followed by a vowel, ti, as in ration, or si, as in man-
sion; and even, if followed by u, as a simple s, as in sure. The articulation
of the sound [f] involves drawing the blade of the tongue to the area where
the tooth ridge joins the hard palate — the part of the roof of the mouth which
feels hard, as we mentioned above.
Because their place of articulation straddles the palate and the alveoli,
sounds like [{] are known as PALATOALVEOLARS (figure 1.20 below).
Figure 1.20 Palatoalveolar fricative
If you add vocal cord vibration to [f], you get a sound which, although also
used in English, has no specific representation in the English spelling sys-
tem. This sound is found in the ordinary word measure and in the loanword
rouge, and its phonetic symbol is [3]. This is also the sound which appears
in French at the beginning of genre ‘class, kind’ or Jean, the French equival-
ent of John.
Some authors, particularly in North America, use the symbol [8] for IPA [f],
and [Z] for IPA [3].
Affricates
There is a third and final type of obstruent which is a composite of a stop

and a fricative made in rapid succession, without changing the position of A third type of
obstruent sound is
the articulators. These sounds are known as AFFRICATES — the phonetic
a composite of a
similarity to Africa may be a useful mnemonic for this rather unusual word. stop and africa-
The affricates familiar to speakers of English are the sounds found at the tive, and is known
beginning and the end of church and judge. as an AFFRICATE
In order to produce these sounds, we raise the tongue to the same posi-
tion as for [f] and [3]. Instead of leaving a gap between the tongue and the
roof of the mouth, however, as we do for [f] and [3], we create a total block-
age of the air, as for a stop (figure 1.21).
Stage 1 Stage 2
Figure 1.21 Palatoalveolar affricate
Crucially, at the moment of release the articulators do not come apart

cleanly, as they do for stops, and therefore the outgoing air causes strong
friction.
_ _ Observe how th happens youpr ynounce church or judge.
The phonetic symbols for these affricate sounds reflect their compositional
nature. The symbols [t] and [f] are combined into [ff] to represent the voice-
less obstruent in church — equivalently, [tf], with a TIE BAR linking both sym-
bols to indicate the unitary nature of the affricate. In turn, [d] and [3] are
combined into [d3] (or [d3]) to represent the voiced obstruent in judge. Note
that the transcription [t] and [d] is strictly speaking inaccurate in this con-
text, since the stop element of these affricates is palatoalveolar, as are [J]
and [3], rather than alveolar, like [t] and [d]. An alternative, non-IPA sym-
bol for [ff], particularly popular in North America, is [é]. Its equivalent for
[dg] is [J].
Summing Up
We will now bring the chapter to a close. You need, of course, to famil-
iarize yourself with all the aspects of the sounds we are describing until
you feel totally comfortable in handling them. Before ending the chapter,
we recapitulate our findings in table 1.3.
a9eL¢“T ‘sdoig saayeory
pue sayeonyye
pry) (Aroyuaaut
OO
aov]q
fo uolynoiyav
SSS
YJnowe XUALWT
seseoesesaosaeaeaeaean
joiqviig ]vJuapolquy jojuaq Apjoaajy AUjOaQvoJvIVgAvan 121401
ESSE
A, SS3TSOIO d } } x Z
W sdoyg Pesion q Pp PB
e . 8
u i A, SSITBOIO J Q Ss i x
, SAATILOML oes : : A y
Y
£ R
3
I ey A, SSITSOIO fi
pacto,, &p
So
Two final brief notes about this table. First, as we have hinted, [6] and
[0] can be pronounced interdentally (that is, between the teeth) instead of
dentally — the choice of place of articulation seems to be a matter of indi-
vidual preference here (figure 1.22).
Figure 1.22 Interdental fricative
Second, the dental and alveolar symbols for ¢ and d in the table are ident-
ical except for the extra mark, or DIACRITIC, “.” under the dentals.
Diacritics serve the purpose of increasing the descriptive power of the
Diacritics serve phonetic alphabet without unduly cluttering up the system with new sym-
the purpose of bols. Following on from this, we should transcribe the interdental fricatives
increasing the
with a special interdental diacritic. However, no such diacritic is available
descriptive power
of the phonetic in the regular IPA alphabet, and therefore we resort to the general diacritic
alphabet without “ ”, which simply signals an articulation slightly forward of the point rep-
unduly cluttering read by the symbol: [9], [9].
up the system
with new symbols
Chapter Summ ary”

This chapter has focused on the Bisica properties of spench sounds, with
particular emphasis on obstruents, those sounds produced by setting up an
obstruction in the path of the air from the lungs. Such an obstruction is made
by bringing together two articulators, which either totally stop the airflow,
then releasing it (a stop), or create a partial closure leading to turbulence
of the air particles (a fricative). The _type of obstruent, the affricate,
combines these two types of closure. The number and variety of obstruents
result directly from the fact that these manners of articulation can be used
at various points in the vocal tract. These points, referred ) as ee of

articulation, are generally described in terms of where in the voce
‘ocal tract con-
tact is made (the alveolar ridge, the teeth, etc.), rather than in terms of the
identity of the active articulator involved. Each of the combinations of man-
ner and place of articulation may also be accompanied by vocal cord vibra-
tion or voicing. We showed thatthe set of ah Ee) obstruents discussed
come in voiced and voiceless i
table 1.3. Learning the d
of learning phonology.
Bey titers,
Fi oon,
1. What is the larynx and what is its should we never talk about letters
role in the production of speech? when we are doing phonetics (or
2 How is speaking similar to or dif- phonology)?
ferent from playing the recorder? 7 What do the terms “active” and
3 Discuss the manner of the release of “passive” articulator mean?
air in the production of the three dif- 8 The set of obstruent sounds avail-
ferent types of obstruent. able to a language can be doubled
4 When we talk about “place of ar- by the addition of “voice”. What is
ticulation”, what are we referring to? voice?
Enumerate the distinctive places of 9 What is the purpose of a diacritic in
articulation we have discussed in phonetic transcription? _ Illustrate
the text. your answer.
5 What does a phonetic symbol rep- 10 Define the term “articulation”.
resent? What are the advantages of 11 Enumerate the sounds for which
the IPA alphabet? the soft palate is an articulator.
6 What is the difference between a Which other articulator would you
phonetic symbol and a letter? Why expect to be involved?
Further Practice
Sound to Spelling
We have pointed out that, as phonologists, we are interested in speech sounds

and not in spelling. We also pointed out that there is often, in English as
indeed in other languages, more than one way to represent a single sound
in the spelling system. We list below the phonetic symbols for a number of
obstruent sounds. List as many different words as you can think of containing
each sound with a different spelling (under each symbol is a number indic-
ating the number of different spellings we have been able to think of):
[f] [kl] [s] [z] [f] [3] [Wl] [4]

Bao US! 15 ease aro.
Articulation and Phonetic Symbols
a Write down the IPA symbols representing the following descriptions, and
illustrate each of the sounds with two English words:
A voiced labiodental fricative

A voiceless alveolar fricative
A voiced palatoalveolar affricate
A voiced glottal fricative
A voiceless bilabial stop
A voiceless dental fricative
A voiced velar stop
b_ Provide full descriptions of the sounds represented by the following phon-

etic symbols (you will need to have recourse to the full IPA chart for some
of the symbols):
[vy] [cl [d] [ts] [3] [B) [8]
c Arrange the sounds below into groups according to the following criteria:
(i) Voicing
Gi) Place of articulation
Gii) Manner of articulation
[x] [k] fb] [f] [3] [2] Id] [p] [fl] [gl
Spelling to Sound
Give the phonetic symbols for the sounds represented orthographically by

the emboldened letters in the words below:
rough Thomas think pen phail fact stuff seed cede rise gnome
agnostic Stephen sheep cage jaw gold fission station chocolate
chaos dough kilt knowledge acknowledge question freeze bus
there castle muscle spaghetti fussy busy fuzzy casual causal
sugar Russia rushes cutlass table sign flight bomb vanity
bombard duke of off ascension escape succeed division
INTRODUCING PHONOLOGY
ASSIMILATION
be influenced by the cont

w How language is organize
m= How what we think we
_ say and hear. -.
m How these two levels
@ In partic
In the previous chapter we surveyed the production, or ARTICULAT-

ORY PHONETICS, of a number of consonants, obstruents and glottals in
particular — we naturally assume some pretheoretical familiarity with the
dichotomy consonant—-vowel, from your learning of the alphabet if nothing
else. We will now put the tools we have acquired to good use and make a
start on the study of phonological structure, the primary object of this book.
We will introduce some additional phonetic observations as we go along,
as a necessary background to some of the discussion of phonology. As a pre-
liminary to the delimitation of phonology from phonetics, we shall present
There are three
several cases of sound “alternation” in English, and we shall provide a sketch
basic parameters
of the way language in general is organized. responsible for the
phonetic differen-
tiation of obstru-
On How Bilabial Stops Become Labiodental ent consonants:
place of articula-
tion, manner of
In chapter 1 we saw that there are three basic parameters responsible for articulation, and
the phonetic differentiation of obstruents: place of articulation, manner of voice or its
articulation, and voice or its absence. Pending the introduction of several absence
additional such parameters for the description of other sounds, we shall focus
32 Introducing Phonology
the discussion on the three distinct series of oral stops that result from place
of articulation differences in English.
You will recall that English has a voiceless—voiced pair of stops for each of
the bilabial, alveolar and velar places of articulation.
Thus, for instance, we explained at the time that both [p] and [b] are bilab-
ial sounds, involving the closure of both lips. In our presentation of these
sounds we made use of such words as spy, obey, etc. But now consider other
forms where the sounds corresponding to the p or b spellings immediately
precede the labiodentals [f] or [v], as in the single words cupful and subvert
or in the compounds cup-final or cab-fare. The question is, are p and Bstill
bilabial in this context?
Chances are that you don’t pronounce the sounds corresponding to the
spellings p, b in cup-final or cab-fare as bilabial, as you do when you say the
words cup, cab, etc. in isolation. Instead, in cup-final or cab-fare p and b are
pronounced labiodentally. In the absence of an official IPA candidate, we
will adopt the ad hoc symbols [9], for the labiodental voiceless stop, and
[%], for its voiced counterpart.
The discovery that p is [P] in cup-final, and b [%] in cab-fare, may have
come as a mild surprise. Taking our search one step further, we shall ask
why the situation should be precisely the way it is. Crucially, cup is the same
Introducing Phonology 33
word whether we say it by itself (cup), as an integral part of the longer word
cupful, or as a component of the compound cup-final. In particular, the
meaning of cup is the same in all three cases, and the difference in articu-
lation we are now focusing our attention on is beyond the threshold of
untrained consciousness. Words which clearly have the same meaning but
sound different, like friend, mate, pal, chum and buddy, are technically known
as SYNONYMS: most obviously, though, the cup of cup and the cup of
cupful or cup-final are not synonyms, but simply the same word. Indeed, in
cupful, cup-final, etc. we do not set out deliberately to change the way we
pronunce the p: if we did, we would have been aware of the phenomenon
all along. So, why are we making our lives complicated by pronouncing the
word cup (and, even worse, any other word ending in the same consonant!)
in two different ways, rather than sticking to the single pronunciation [p]?
One possible reason for the two pronunciations could be that bilabial stops
do not sit comfortably in the middle of (simple or complex) words. If this
were indeed so, the p and bin lip-service, laptop, lapdog, sob-story, adoptive,
etc. ought also to be pronounced labiodentally.
In lip-service, etc., p,b are, however, not pronounced labiodentally, but bilab-
ially, in exactly the same way as when they are word-final in lip, etc. This
result inevitably leads to the abandonment of our current hypothesis and
the search for an alternative. The slightly grand term HYPOTHESIS is used
to refer to an idea that explains some body of data. We will gain familiar- The term
3 ; a HYPOTHESIS
cents is
ity with this important concept as we go along.
an idea that
explains some
body of data
In the present case it is not too difficult to come up with an alternative hypo-
thesis. We shall follow our usual strategy and arrive at this alternative
experimentally. If you try to pronounce cupful with a sound sequence

bilabial-labiodental (for the p and f spellings, respectively) you will make
at least two findings.
First, you will probably feel rather awkward, as if the articulatory move-
ments you are making were somewhat unusual, and unduly complex.
Second, if you listen to yourself you will notice a stilted and unnatural dic-
tion. The reason for these two related results is, obviously, that the conscious
pronunciation you have adopted for the sake of the experiment is not the
pronunciation you use in ordinary speech. So, why should it be more
difficult here to maintain the usual bilabial pronunciation of p, b than to
shift it to labiodental? The answer is obvious: the simplest pronunciation
involves only one articulatory gesture for both consonants. However, if you
don’t make the preceding stop labiodental, you will need two sequential ges-
tures, first the closure of both lips, and then the motion of the lower lip towards
the upper teeth. This is thus the answer to our question: in order to sim-
plify the articulation, we allow the usually bilabial articulation of p, b to be
“contaminated” by the labiodentality of its neighbour. Such “contamination”
“Contamination” by (usually adjacent) sounds is a common occurrence in language, and is
of a sound by
technically known as ASSIMILATION. As we shall see directly, assimila-
other (usually
adjacent) sounds
tion abounds, in English and in other languages. Note, however, that assim-
is commonplace in ilation clearly cannot have a free rein, or else we would end up with just
language, and is one sound in every utterance!
technically known
as ASSIMILATION
Total Place Assimilation in Stops

The last sounds in the words that and red are alveolar stops, voiceless and
voiced, respectively.
Consider now the following simple phrases:
(1) that pen

that box |
red pen |
red box
Let us pay close attention to the pronunciation of the sounds spelled t and
d in these phrases. In particular, let us find out whether they can still rightly
be transcribed [t] and [d], respectively.
The answer is that the ts and the ds are indeed [t] and [d] if you pronounce
the phrases in (1) rather slowly, paying special attention to what you are
saying. However, when you utter them casually, as you do in the course of
normal conversation, you are liable to pronounce them as bilabial: tha[p] pen,
thalp] box, re[b] pen, re[b] box.
wee
Consider next the set of phrases in (2):
(2) that can

that gate
red can
red gate
- What is now the place of articulation of the sounds behind the spellings
t, d?
A small amount of attention will reveal that the sounds are now pro-
nounced as velar (tha[k] can, tha[k] gate, re[g] can, relg] gate), rather than
bilabial, as they were in the set in (1) (tha[p] pen), or alveolar (thaltl, re[d]),
as they are pronounced otherwise.
Let us add the final set in (3) to our pool of data:
(3) that table

that door
red table
red door
What is now the pronunciation of the sounds under scrutiny?
You will observe that t, d have now reverted to their original alveolar articu-
lation: tha[t] table, tha[t] door, re[d] table, re[d] door.
Facts like the ones we have gleaned stimulate the researcher to act. The
next step is of course to look for an explanation for why things are precisely
the way they are. And the first stage in this process of explanation is the
formation of a hypothesis, in much the same way as a detective comes up
with a hypothesis about the identity of the criminal. Clearly, in order to get
to a hypothesis, we must pay very close attention to the facts. For the detect-
ive this involves examining all kinds of material evidence, and for the phono-
logist it involves examining the exact shape of sounds. Indeed, we may have
to look at the facts several times over before we reach a conclusion. At some
point along the way, however, we will hopefully come up with “an idea”.
This idea, which often seems to spring up suddenly, is what we are calling
a “hypothesis”.
What is the most reasonable hypothesis to handle the data we are
discussing?
If you look at the data closely you will realize that the place of articulation
of the stop ending the words that and red is consistently the same as the
place of articulation of the stop which begins the following word. Now, why
should this be so?
The answer should be obvious after the discussion in the previous section.
Simply, what under normal circumstances is an alveolar stop (indeed as
reflected in the spelling) systematically assimilates to the place of articu-
lation of the stop that immediately follows it. This is another instance of
REGRESSIVE ASSIMILATION, that is, assimilation to the following seg-
ment — in the opposite type of assimilation, PROGRESSIVE ASSIMILATION,
there is assimilation to the preceding segment, as happens in cubs and cups.
The reason there is no change in the set in (3) is that there the second stop
is also alveolar, and therefore the assimilatory process is vacuous — it has
no material consequences. In front of a bilabial or a velar consonant, how-
ever, the alveolar articulation is replaced by a bilabial or velar articulation.
We summarize the processes in (4):
(4) thalp] pen thalk] can _ thalt] table

thalp] box thalk] gate tha[t] door
re[b] pen relg] can _ re[d] table

re[b] box reg] gate _re[d] door
Notice that, in all the cases of assimilation we have discussed, we are assum-
ing that one of the alternative articulations is more “basic” than the others. amen the cases of
In all
Thus, we have proposed that the labiodental realization of p in cupful is just See
a “deflection” of a basic [p], caused by “contamination” from the adjacent —o14 o¢ the alter-
labiodental [f]. Similarly, the basic articulation of the last sounds in that pative articulations
and red is alveolar, and their alternative bilabial and velar realizations are __ is more “basic”
brought about by assimilation.
After all, if [t] and [d] were not basic, why should these sounds turn up when
no consonant follows, namely, when the words are pronounced in isolation
or when the next word starts with a vowel, as in that apple?
What we are finding, thus, is that sounds can be replaced by other sounds,
Sounds can be under the appropriate contextual circumstances: the changes in question are
replaced by other = Wot random, but occur in particular, well-defined phonetic contexts. Notice
sounds, uncer
Ree pupiae that you also change clothes, makeup or hairstyle ac ording to circumst
at you also change clothes, makeup or airstyle accor ingtocircumstances
contextual — crucially, you remain the same person, although sometimes it may be hard
circumstances to tell just by looking at you.
Voice Assimilation
We shall briefly consider one more case of assimilation before we propose

a formalization for the phenomenon.
In chapter 1 we presented the following voiceless-voiced consonant
pairs, among others:
Oy fr
[s] [z]
[f] [ds]
The sounds in the top two pairs in (5) are fricatives, and those in the bot-
tom pair affricates. The words in (6) illustrate the voiceless—voiced contrasts
in word-final position:
(6) a. Fife b. five

loose lose
etch edge
Now consider the following phrases:
(7) five tons

lose ten-nil
edge trimmer
The first word in each of these phrases comes from the set in (6b), where
we saw that it ends in a voiced fricative or affricate. From a seriously experi-
mental perspective, however, the assumption that in (7) these sounds are
also voiced needs to be tested.
In fact, a minimal amount of observation should reveal that in (7) none of

these sounds is voiced.
A possible hypothesis to account for this fact, HYPOTHESIS A, would
be that the sounds in question are only voiced when the word is said in
isolation. You are becoming familiar with the need to put scientific hypo-
theses to the test before accepting them: an argument such as “because I say
so” does not cut much ice in science. Let us, accordingly, test out HYPO-
THESIS A with an additional list of phrases that include the words under
scrutiny:
(8) five or six

lose eight-nil
edge of the world
Here the words also appear in a phrasal context, and, therefore, HYPO-
THESIS A predicts that the sounds we are investigating ought also to be
voiceless. The empirical question of course is, are they?
Again, a small amount of observation will reveal that the sounds in ques-
tion are voiced in (8), and therefore HYPOTHESIS Afails. Suppose next that
we modify HYPOTHESIS A minimally into HYPOTHESIS B, by attributing
the voicelessness of v, s and dg in (7) above to the presence of a following
consonant — the mind is reluctant to abandon existing hypotheses, and
therefore it is only human to modify hypotheses as little as possible.
Whether the modification is minimal or drastic, HYPOTHESIS B must also

be tested. Consider the following list:
(9) five days

lose data
fridge door
In (9), the sounds we are discussing are followed by a consonant and there-
fore, according to HYPOTHESIS B, they also ought to be voiceless there. The
question is, are they really?
The answer is that in (9) the sounds we are investigating are in fact voiced.
This means that HYPOTHESIS Balso fails.
Having reached this point, you may be tempted to ask yourself whether
there is any way of accounting for the voicelessness of the sounds in the
original set in (7): there is of course no a priori reason why natural phenomena,
whether in the physical world or in the world of language, should be
amenable to explanation. The assumption of the linguist and of fellow
researchers in other branches of science is that there usually is an explana-
tion, for otherwise there would be no more to say, and hence no linguistics
or science in general. Fortunately, in the present case an explanation is read-
ily available.
The answer is that in the set in (7), but crucially not in those in (8) and (9),
the sounds under observation are followed by a voiceless obstruent. Why
should the voicelessness of the following obstruent matter? Because, as in
the cases of assimilation we examined in the previous sections, the voice-
lessness of this obstruent can (and does) “contaminate” the (usual) voiceful-
ness of the preceding fricative or affricate. Therefore, the devoicing of such
sounds in (7) is simply a consequence of assimilation. Note that the para-
meter involved in the present assimilation process is not place of articu-
lation, as was the case in the preceding sections, but, rather, voice. Indeed,
assimilation can involve any of our familiar phonetic parameters, although
assimilation of manner of articulation is considerably rarer. Assirnllabion car
We will wind up this section with an additional set of data which bring Te
ae ae —e : familiar phonetic
out minimally and strikingly the devoicing process we have examined: parameters
(10) five to five

nose to nose
edge to edge
3 The Organization of Language
Having prepared the ground by investigating a number of assimilation phe-

nomena, it is time to turn our attention to the main subject matter of this
chapter: presenting phonology and segregating its concerns from those of
phonetics. As a preliminary, it will be useful to reflect briefly on the way
language is organized.
We can start with the obvious: all languages have words. In fact, it is likely
that for the layman the words of a language are the language. In reality, how-
ever, things are alittle more complicated, although, fortunately, still not hard
to understand.
We can indeed usefully view words as the building blocks of (any) lan-
guage. What we do when we speak is put words together into phrases and
sentences. The way words come together varies (at least partially) from lan-
guage to language. Its study constitutes the subject matter of SYNTAX, one
of the branches of LINGUISTICS, itself the analytic study of language. In

this book we are of course concerned not with syntax, but, rather, with sound.
Besides syntax and sound, language (its words as well as its sentences)
has meaning, studied by another branch of linguistics, known as SEMAN-
TICS. How do we get sentences to have a meaning? For instance, how do
we get the sentence children love sweets to mean precisely what it does? Well,
it is not so difficult once we have the individual meanings of the words
child (of which children is the plural), love and sweet (singular of sweets): in
essence, all we do in order to understand the sentence is combine the mean-
ings of its component words. But why does child mean ‘child’, love ‘love’ and
sweet ‘sweet’? Note that if we were speaking French then we would use enfant
to mean ‘child’, aimer to mean ‘to love’ and bonbon to mean ‘sweet’. Why?
The answer to the question of why words mean what they mean is tauto-
logical: words mean what they mean because they do. There is no other
reason. In particular, the sound shape associated with the meaning of each
The sound shape
associated with the
word is arbitrary, and simply has to be taken on board and memorized. So,
meaning of each the first thing we need to carry in our minds in order to have a language is
word is arbitrary a pretty gigantic list of words, each word essentially consisting of a certain
sound shape and a meaning glued together.
The famous Swiss linguist Ferdinand de Saussure (1857-1913), usually
considered the founder of modern linguistics (linguistics has existed for
millennia in a variety of guises), called the basic elements of language LIN-
GUISTIC SIGNS. Each linguistic sign is thus made up of two integral and
complementary parts, rather as the two sides of a coin make up the coin:
a characteristic meaning, which Saussure called the sign’s SIGNIFIED, and
a characteristic sound, which he called the sign’s SIGNIFIER. Following
Saussure, the linguistic sign for cup can be represented graphically as in fig-
ure 2.1, where, in line with tradition, we represent the signified by means
of a drawing, and the signifier with the italicized conventional spelling.
op Signified
cup Signifier
Figure 2.1 Linguistic sign

Saussure went on to point out that the association between the signifier and
the signified that make up any one linguistic sign is arbitrary: he referred
to this fact as “l’arbitraire du signe”, ‘the sign’s arbitrariness’. From this arbi-
trariness follows the need for the brute-force memorization we alluded to
above.
It is reasonable to assume that each sign only has one signifier and one
signified. In fact, there is a strong logic to this assumption, namely, that it Fach sign has only
is precisely the conjunction of one signifier (that is, a certain sound shape) °"° Benilier are
and one signified (that is, a certain meaning) that makes up a linguistic sign sae iit
(a word, for our present purposes). If two meanings occur with a single form,
as with the English bank, either a financial institution or the side of a river,
we will assume two linguistic signs, rather than one sign with a double
signified. We now display these two alternative construals in figure 2.2. Cups
are of course easier to draw than sides of rivers or financial institutions, and
therefore, following convention, we represent complex signifieds by means
of ordinary language expressions enclosed in single quotation marks.
‘river side’
‘type of ‘type of
financial institution’ ‘river side’ financial institution’
Rejected alternative Proposed alternative

Figure 2.2 Two ways of representing intersecting signifiers
Following the same logic, the signifier of each linguistic sign will also be
unique. For instance, we will assume two linguistic signs for the synonyms
pal and mate, rather than just one sign with a double signifier. We display
these two alternatives in figure 2.3, where we arbitrarily adopt ‘friend’ as a
representation of the meaning shared by the two words:
‘friend’
Be (es ee
Rejected alternative Proposed alternative
Figure 2.3. Two ways of representing intersecting signifiers

Signifier uniqueness appears difficult to reconcile with the dual pronun-
ciation of the p of cup as [p] or [P], with the multiple pronunciation of the
t of that as [t], [p] or [k], or with the dual pronunciation of each of v, s and
dg as [v], [z], [ds] or [f], [s], [ff]. In particular, aren’t all these alternative pro-
nunciations defining independent signifiers, and thus independent signs, by
the logic adopted?
There are at least two reasons why we want to prevent the proliferation of
We want to pre- linguistic signs. First, there is a general principle of economy acting as a back-
vent the prolifera-
drop to the whole enterprise of linguistics, a manifestation of an old philo-
tion of linguistic
signs
sophical dictum known as “Occam’s razor”, after its fourteenth-century
formulator William of Occam: “entities must not be multiplied beyond
necessity”. This of course just means that we must keep things simple
unless there are good reasons not to do so. The second reason to avoid the
proliferation of linguistic signs is more psychological: as speakers of the
language we have a feeling or intuition that cup is the same word in all its
contexts, and similarly for the other forms discussed. Such feelings must be
taken seriously in linguistics (even if not necessarily accepted at face value),
since the very nature of language is psychological, not for example geo-
logical or artistic.
What exacth do )
If we accept that cup, that, five, etc., correspond toa single sign each, what-
ever varied realizations the last consonant of each such form may have, we
must inevitably select one of the competing candidates as the “official” signifier
We select one of of the sign.
the competing
candidates as the
“official” signifier
of the sign
are these competing
The question is, will there be any criterion for the selection of one candid-
ate as the “official” form, or will it be random?
The answer is in fact implicit in our previous discussion, and should be pretty
obvious by now. As you will recall, we have systematically construed one
of the alternatives as basic, because this alternative is more frequent and gen-
eral (for instance, it turns up when the word is said in isolation), because it
is the alternative we constantly think we are pronouncing or hearing (even
when objectively we are not!), because it corresponds to the conventional
spelling of the sound (this is of course a secondary criterion, given the prob-
lems associated with spelling, but still worth bearing in mind as a possible
reflection of the way we “think” about the sound), and so on.
Clearly, this basic alternative will be the one included in the linguistic sign.
All the other variants will be “derived” from their basic counterpart in the
appropriate contexts, in the way we shall explain next.
Basic and Derived Forms
Let us introduce some simple standard terminology. The set of all linguis-
tic signs making up the vocabulary of a language is known as the LEX- The set of all
linguistic signs
ICON of that language. In fact, the label “linguistic sign” is not much used
making up the
in current linguistics — it is usually associated with the type of linguistics vocabulary of a
propounded by Saussure, many details of which have been superseded by language is known
subsequent developments. The expression “linguistic sign” therefore tends as the LEXICON
to be replaced by the expression LEXICAL ITEM or LEXICAL ENTRY, which
simply suggests a unit in the lexicon. Like linguistic signs, lexical items con-
tain all the information relevant to sound and meaning that must be mem-
orized as part and parcel of the process of mastering a language, whether
a second language, the words of which we may memorize consciously, or
a first language, where memorization happens automatically at an early age.

The expression The lexical entry for cup will therefore include the information that its
LEXICAL ITEM or
final sound is [p], the lexical entry for that that its final sound is [t], and the
LEXICAL ENTRY
simply suggests a
lexical entry for five that its final sound is [v]. We refer to the information
unit in the lexicon thus contained ina lexical item as its LEXICAL REPRESENTATION.
We refer to the
information con-
tained in a lexical
item as its LEXICAL It should be quite clear by now that the lexical representation of a word is
REPRESENTATION not always identical to its PHONETIC REPRESENTATION, that is, to the
way the word is heard or said. Thus, for cup in cupful, you will not hear (or
pronounce) the [p] that we have just suggested is present in its lexical rep-
resentation. Similarly, in the phrases that pen or that can, you will usually
not be hearing (or pronouncing) the [t] we assume for that in its lexical rep-
resentation. Likewise, in five tons the lexical [v] of five will not occur in the
normal pronunciation of this phrase. You know by now what the motiva-
tion for these divergences is. What we obviously need next is a formalism
that accommodates unity in the face of diversity, or, equivalently, diversity
in spite of essential unity.
Fortunately, the way forward is pretty straight. At the moment, we

already have lexical representations for cup, that and five. In particular, we
have proposed that the last lexical sounds in these words are /p/, /t/ and
/v/, respectively: lexical sounds are enclosed in slashes, to make them
Lexical sounds visually distinct from phonetic sounds, an important convention you must
are customarily
scrupulously adhere to.
enclosed in
slashes, to make
them visually dis-
tinct from pho-
netic sounds
Now, we must assume that, under normal circumstances, phonetic sounds

are identical to lexical sounds. Therefore, circumstances allowing, the last
phonetic sounds of cup, that and five will also be [p], [t] and [v], respectively:
this is indeed the reason we postulate /p/, /t/ and /v/ as the respective
lexical sounds in the first place. Under some restricted set of circumstances,
however, we know that the phonetic sounds of the forms in question are
different from their lexical sounds. What are these circumstances?
The circumstances under which the /p/ of cup becomes [P] can be stated
as follows:
(11) /p/ is realized as [P] when it immediately precedes /f/
Likewise, the observed changes in the realization of /t/ can be summarized

thus:
(12) /t/ is realized in the same place of articulation as the stop that imme-
diately follows it, namely, as [p], [t] or [k]
Similarly, the change from /v/ to [f] in five tons can be accounted for by the
statement in (13):
(13) /v/ is realized as [f] when it is immediately followed by a voiceless

obstruent
Clearly, the three statements we have just given are sufficient to generate
all and only the correct phonetic forms of these and other similar words.
Thus, suppose we intend to say the word cup by itself. In this case, we retrieve
the lexical representation, with /p/ in final position, and simply pronounce
it: [p].
sh we wrote /p/ just
Suppose, instead, that we intend to say cup in cupful. As before, we retrieve

the lexical representation with a final /p/. In this case, however, we will
not realize this /p/ as [p], because the statement in (11) above will “deflect”
/p/ into [P].
At one level, statements such as those in (11) to (13) are simple observa-
tions of reality. In particular, it is indeed the case that /p/ is realized as [P]
when it immediately precedes /f/, and this is what the statement in (11)
says. At another level, however, such statements can be construed as regu-
lators of our phonetic behaviour, which they appear to monitor and guide.
From this perspective, such statements are therefore RULES, and we will
henceforth refer to them as such: rather as a road diversion sign directs traffic
onto a certain route, phonological rules divert the realization of lexical
Phonological rules sounds in certain directions. The state of affairs we are describing can be
divert the realiza- represented schematically as in (14):
tion of lexical
sounds in certain
directions (14) Lexical representations
Phonetic representations
As (14) displays, lexical representations can be interfered with on their

way to phonetic realization by rules, which “divert” or “deflect” them
accordingly — the break in the vertical arrow is intended to convey a visual
impression of this deflection. In psychological terms, you can think of this
process as one of modification of the lexical sound stored in memory before
articulatory implementation: the phonetic realization that will come out of
the mouth will accordingly correspond to the form modified by the rule,
rather than to its lexical base.
It should be clear at this point that rules are as much part of the ma-
chinery of the sound component of language as are lexical representations.
In particular, we must learn the rules of a language if we are ever going to
get its pronunciation right, since the differences in pronunciation between
different languages, or between different accents within the same language,
are (at least in part) the result of the presence of different rules in those lan-
guages. When we are taught a second language, sometimes such rules are
pointed out to us, and we purposely learn them and try to apply them as
well as we can. In the case of our first language, however, we have simply
extracted these rules from the speech around us, in some rather mysterious
way. Such are the wonders of first language learning, which have led lin-
guists, notably Noam Chomsky (the man who has most influenced our ideas
about language since Saussure), to suggest that the human brain is innately
in possession of a LANGUAGE ACQUISITION DEVICE that allows it to
acquire language spontaneously in childhood.
[4 The Formalization of Rules
We shall now propose a formalism to encapsulate the concepts we have intro-

duced. Formalism can be thought of as a shorthand notation that allows the
eye to see the whole picture at a glance, and consequently it must be cher- _ Formalism can be
ished rather than feared. It also forces us to be more rigorous and precise “ought of as a
wna: ’ shorthand nota-
than we usually are when we write in ordinary prose. :
tion that allows
the eye to see the
whole picture at a
glance
system outside linguis
_ simple arithmetic).
We are already in possession of a set of phonetic symbols (the IPA sym-

bols), and of the convention of enclosing them in slashes to refer to lexical
forms, and in square brackets to refer to phonetic forms. Indeed, the pro-
posed split between the lexical and the phonetic levels of representation is
also part of the formal theory we are constructing to introduce order into
the apparent real-world chaos. A word often used to refer to such scientific
idealizations of reality is MODEL. We are accordingly engaged in the build-
ing of a model which hopefully will give us a better understanding of the
mechanics of sound in language.
In the preceding section we stated several rules in prose, and we shall now
trim them down to a leaner format. A number of ways of formalizing rules
suggest themselves. By way of illustration, let us examine the rule chang-
ing /p/ to [P]. Here is one reasonably obvious possibility:
Gb)ib/jesmpiae/
Paes.cul
The first line is flanked by slashes, to indicate lexical status, and the second
line by square brackets, to indicate phonetic status. All we have done is write
the sequence “pf” in the first line, preceded and followed by dots to indic-
ate the irrelevance of additional material on either side, and the sequence
“Sf” also with dots, in the second line. The obvious implication of this form-
alism is, therefore, that a lexical /p/ becomes a phonetic [P] in the given
environment: immediately before /f/.
One reasonable way of bringing out the idea of change graphically is by
means of an arrow between the lines, as follows:
(lo)? (apr y
fake Pic
The arrow suggests the idea “goes to” or “becomes”, and therefore we will
consider this representation an improvement over its predecessor in (15).
Suppose now that we want to save space — we did say earlier, after all,
that formalization amounted to a shorthand notation. One simple way of
achieving this is to merge the lines:
(et ee spleen | eo eeeae|

Let us take our trimming exercise one step further. Assuming asingle for-
mat for all rules, we can dispense with the slashes and the brackets: given
the linear arrangement we are adopting, the material to the left of the
arrow will automatically be interpreted as the input (here the lexical input),
and the material to the right of the arrow as the output (here the phonetic
output):
Celie sapluns ofdnd hire,

Next, we can quite safely omit the dots, since they signal material which is
by definition irrelevant:
(19) pf > Pf
We have almost reached the end of our trimming exercise. Indeed, you may
well feel that we can go no further, since all the material that is now
included in the rule would seem to be substantial. Notice, however, that
“f” appears on both the left- and the right-hand side of the arrow, to the
right of “p” and “SP”, respectively. Moreover, there is a significant differ-
ence between the role of “f” and the role of “p” or “®” in the rule — and,
correspondingly, in the real-world process. Specifically, “p” represents the

segment to be altered and “9” the result of the alteration. By contrast, “f”
represents the constant environment or context in which the alteration takes
place: in the absence of /f/, /p/ remains unaffected and will indeed be
realized as [p]. When we take these facts into account, it becomes obvious
that it is useful to separate the rule’s ENVIRONMENT (“f”) from the rule’s
FOCUS (“p” > “%”), as follows:
G6) Tp 5 gen Geng

This is the final format of the rule. Formalized phonological rules thus con-
tain two main parts, conventionally separated by a slash, not to be confused Formalized rules
contain two main
with the slash pairs that lend lexical status to phonetic symbols. These two
parts: the focus
parts are: (i) the focus, formalizing the change, and (ii) the environment, form- and the environ-
alizing the contextual trigger for the change. We spell this out in (21) for the ment, conven-
case we are discussing: tionally separated
by a slash
(21) Focus: Pi Ff
Environment: _f
In fact, in this formalism the focus is represented twice: once explicitly, with
the actual input and output separated by the arrow (p — ®), and once implic-
itly as a horizontal line adjacent to the environment (__f). Probably a more
perspicuous formalization of the rule being scrutinized would therefore be
as follows:
(22)
|f
9S
Sz
Indeed, this format can be nicely integrated into our graphic model of the
organization of linguistic sound in (14) above, as follows:
(23) Lexical representations
ot (&------------------- Rules
<=
2 Phonetic representations
It is obvious, however, that the gain in perspicuity is more than offset by

the spatial loss, and consequently we will adhere to the slightly more
abstract format in (20), indeed the standard one.
Derivations
Rules of course only represent one aspect of the organization of sound in

Rules only repre- language. The other two aspects we have considered are the lexical repres-
sent one aspect of
entation, which corresponds to the way language sound is inertly stored,
the organization
of sound in lan- and the phonetic representation, which corresponds to the way this sound
guage. The other is dynamically realized.
two aspects are
the lexical repre-
sentation and
the phonetic
representation
We shall now examine how these three aspects of sound structure — lexical
representation, phonetic realization and rules — are formally related.
In fact, we have already given the nub of this relationship: quite simply,
the rules change lexical representations into phonetic representations.
Equivalently, the rules can be thought of as a formal device bridging the
gap between lexical and phonetic representations. This relationship can
be visualized with the help of the diagram in (14) above, which we now
repeat:
(24) Lexical representations
Phonetic representations
Let us see how this schema is implemented in the case of the [p] ~ [P] alterna-
tion in cup. Crucially, whichever sound is chosen it does not affect the
identity of the word, contrary to what happens if we replace [p] with [b] or
[f], as in cub and cuff, respectively. The reason for the difference in outcome
is that in English [p], [b] and [f] are minimal units of sound contrast, or
“phonemes”, but [9] is not. The word PHONEME therefore refers to a unit
of explicit sound contrast: the existence of a MINIMAL PAIR, like the ones
made up by cu[p] and cu[b], or cup] and cu[f], automatically grants phon-
emic status to the sounds /p/, /b/, /f/ responsible for the contrasts: notice
that the symbols of phonemes are enclosed in slashes, just like the symbols
for lexical sounds, an ambiguous use that should not cause confusion, given
the quite different theoretical contexts associated with the two constructs.
Each contextual variant of a phoneme is an ALLOPHONE of that phoneme:
in the case we are discussing, /p/ can be realized as either [p] or [P?], which Fach contextual
are correspondingly allophones of /p/. The words “allophone” and “allo- — Ya"iant of a
phony”, derived from the Greek words allos “other” and phonos “sound”, Pe ae
are, however, connotative of an approach to phonology closely associated that phoneme
with Saussure’s style of linguistics, now superseded. From a more contem-
porary perspective, we refer to the phenomenon as ALTERNATION, and
to the elements partaking in it, such as cu[p] and cu[2] in the case at hand,
as ALTERNANTS. Variant realizations
We know what the lexical representation and the phonetic representations °F lexical form
of cup (or rather, of the part of this form we are focusing on: we ignore the “* Ceci
remainder for the sake of simplicity) are supposed to be: cu/p/, and cu[p]
or cu[P], respectively. These representations thus make up the LEXICAL
or UNDERLYING LEVEL and the PHONETIC or SURFACE LEVEL, also re-
spectively, as we illustrate in (25). Notice that the use of slashes and brackets
is unnecessary in these representations, since we are defining the two levels
explicitly:
(25) Lexical level: cup
Phonetic level: cup cuP
At the lexical level we have entered the hypothesized /p/, and at the phon-
etic level the actual phonetic alternants [p] ~ [9].
jing that
the/p/ ishypothesized?
While this layout conveniently displays the level-bound difference between

the lexical and the phonetic forms, it obviously fails to account for the DIS-
TRIBUTION of the alternants, that is, for the fact that [?] only occurs before
[f]. It is precisely the role of rules to express this.
Suppose then that we include the appropriate rules in the representa-
tion. In the present case, only one rule is relevant, to be referred to by the
transparent label “Labiodentalization”. The result is as follows (“NA” = not

applicable):
(26) Underlying level: - cup cupful
Rules:
“Labiodentalization” NA P
Phonetic level: cup, cu Pf ul
What we are now doing is feeding lexical forms (in context, where neces-
sary) through a given set of rules (here only “Labiodentalization”), writing
in each line the accumulated results of the application of the rules. Finally,
in the last line we enter the result of the operation of all the rules (again,
only one rule is relevant in the present case). Such an output obviously cor-
responds to the phonetic realization, and will consequently constitute the
phonetic transcription of the form in question. Constructs like the one in (26)
are known as DERIVATIONS, because they embody the derivation of the
DERIVATIONS phonetic sounds from the lexical sounds through the mediation of the rules.
embody the
derivation of the
phonetic sounds
from the lexical El Phonetics and Phonology
sounds through
the mediation of
the rules
We are now in a position where we can begin to understand the substance
of phonology, and how it differs from phonetics.
As you now know, phonetics describes sounds: articulatorily (positions
and movements of the speech organs), acoustically (patterns in the air,
detectable with the appropriate technology) and perceptually (impact of the
sound on the ear and subsequent transmission of the signal to the brain).
For instance, in the cases we examined above, phonetic analysis will yield a
description of the bilabial stop in cup as labiodental before [f], of the alveolar
t in that as bilabial before bilabials and as velar before velars, and of the last
fricative in five as voiceless before voiceless obstruents. Moreover, phonetics
will help us to understand the articulatory motivation of these phenomena
in terms of “coarticulation”, the term by which phoneticians express essen-
tially what phonologists refer to as “assimilation”.
In terms of the display in (26) above, the concern of phonetics is, there-
fore, what we have been calling the phonetic (or surface) level: the level The concern of
reflecting the real world of articulatory, acoustic and perceptual events. This PHONETICS is
what we have
limitation in the scope of phonetics obviously leaves both the lexical (or under-
been calling the
lying) level and the rules uncatered for. Consequently we need another dis- phonetic (or sur-
cipline to look after these aspects of the model. This discipline is, of course, face) level.
PHONOLOGY. PHONOLOGY is
Now, simplicity is not a bad companion, in science as in other aspects of concerned with
life (once again, Occam’s razor can be called forth in this connection), and the lexical (or
underlying) level
we can legitimately ask ourselves what the point is of complicating things and with the rules
by introducing “phonology” alongside “phonetics”. The answer is that in
science, again as in life, division of labour often yields rich rewards. Thus,
by splitting the remit of sound investigation between phonetics and phono-
logy, we can allow the former to focus on the description of physical speech
sound, while the latter will be aimed at its more abstract aspects. Specific-
ally, the object of inquiry of phonetics (physical sound) is amenable to experi-
mental investigation, while the object of study of phonology is by its very
nature of a more hypothetical kind.
Thus, it is not difficult to verify that the p of cup in cupful is [P], rather than
[p] — an experiment to this effect was indeed suggested above. By contrast,
our postulation of /p/ as the underlying (or lexical) form of [P] is purely
idealistic — it is contingent on our hypothetical interpretation of the system,
rather than on directly material data, which would indeed be very hard to
come by in this connection: even if we could peep directly into somebody’s
grey matter, it is most unlikely that we would see anything of relevance
to the issue. A similar remark is apposite with regard to our analysis of
the [%] alternant as the product of a rule, clearly only one of many pos-
sible ways of formalizing the situation, if a particularly apt one in the
context of the overall model we are presenting: an alternative to rules
and derivations will be discussed in chapter 19. All in all, therefore, both
the type of work and the tools needed for the investigation of what we
are calling phonetic events differ considerably from the type of work and
the tools needed for the investigation of what we are dubbing phonolo-
gical phenomena. This difference will undoubtedly have practical repercus-
sions, and, more likely than not, different individuals will be drawn to focus
their work on either area, further cementing the conventional separation of
the two fields.
By examining such
before adjacent voiceless seg
one linguistic sign may have
segment in the cases examined. S|

for the sign CUP and are clear!
sequently, speakers do not rm
Key Questions
1 In what way can the articulation 7 What sort of information is con-

of one sound influence that of an tained in the lexicon?
adjacent sound? 8 What are the “focus” and the “en-
2 Assimilation processes are either vironment” of phonological rules?
“progressive” or “regressive”. What 9 Explain the terms “phoneme” and
is the difference? “allophone”.
3 What are the two complementary 10 What is a “minimal pair’? What
parts of a linguistic sign? does it help us to recognize?
4 What are the terms of “Occam's 11. Define “alternants”.
razor"? How does it constrain pho- 12 What is a derivation?
nological analysis? 13 Distinguish the concerns of phono-
5 What is a phonological rule? logy from those of phonetics.
6 What is the connection between
lexical items and their phonological
representations?
FSO Qetchsie?neuPur a Cyteinc.e
Catalan Obstruents
Consider the obstruent alternations (shown in broad phonetic transcription)

in the following masculine/feminine pairs from Catalan, a language spoken
mainly on the eastern side of Spain:
Masc. Fem.
No[p] llo[b]a ‘wolf’
mul[t] mu|[dla ‘dumb’
celk] celgla ‘blind’
francé[s] france[z]a ‘French’
tilp] tilp]a ‘satiated’
petit] petilt]a ‘small’
se[k] se[k]a ‘dry’
gro[s] gro[sla ‘fat’
(i) Describe the differences between the alternants in each pair.

(ii) Describe the environment for each alternant.
(iii) Hypothesize a lexical form for each pair.
(iv) Is there a pattern to the alternations? What is it?
English Past Tense
The following words are all regular past tense forms of English verbs:
jumped, talked, played, toyed, begged, rubbed, washed, dredged, fetched,

keyed, showed, passed, caused, writhed, frothed
(i) How is the past tense formed in English spelling?

(ii) Does this correspond to a uniform sound?
(iii) If not, how many sounds are there?
(iv) Is it possible to postulate one lexical form for the past tense morpheme?
If so what is it?
(v) What is the reason for the surface variation?
(vi) Show by means of a rule how the lexical and phonetic representations
are connected.
In the previous chapter we offered a succint sketch of phonology, backed
up by an analysis of some assimilation phenomena involving obstruents. The
phonetic properties of obstruents and glottals were examined in detail in
chapter 1. Clearly, though, there are many speech sounds besides obstru-
ents and glottals.
In line with our policy of alternating the presentation of phonetics and phono-
logy in these preliminary chapters, we now turn our attention to the phon-
etics of sonorant consonants. We will make some further remarks pertinent
to phonology as we go along, but will of course reserve the bulk of the phono-
logical discussion on sonorants for chapter 4.
Sonorant Consonants 59
General Properties of Sonorants
The common denominator in obstruents is the presence of an obstruc-

tion to the airflow in the mouth; hence the label “obstruent”. The contrasts
between the various obstruents follow from differences in place of articula-
tion (the anatomical point where the obstruction is made), manner of articula-
tion (the degree of obstruction, according to which we classified obstruents
into stops, affricates and fricatives), or vocal fold activity, by which obstru-
ents are divided into voiced (pronounced with vocal fold vibration) and voice-
less (without vocal fold vibration).
ur memory by pr
ing each of the catego
At this point you may be tempted to think that the production of speech
sound necessarily involves some such obstruction — how else could sound be
made?, you may wonder. As we shall see in this and successive chapters,
however, this is in fact far from being the case. Indeed, in all the sounds to
be introduced from now on, the air comes out through a channel wide enough
to avoid friction. Such unobstructed sounds are known as SONORANTS,
because, as follows from the greater openness of the channel, they carry a Unobstructed
sounds are known
greater amount of sound than their obstruent counterparts — cf. Latin sonus
as SONORANTS
‘sound’, sonor ‘resonance’, sonorus ‘sonorous’.
Actually, sonorancy can be construed as a specific setting for manner
of articulation. It should be obvious by now that the articulatory channel
exhibits increasing openness progressively along the scale stops—affricates—
fricatives—sonorants. This means that sonorants are the most open of con-
sonants. In table 3.1 we summarize the settings we have now available for Sonorants are the
most open of
manner of articulation. You will see that the degree of channel opening is
consonants
relative for all categories but stops — “f > a”, for instance, indicates that the
channel for fricatives is more open than the channel for affricates, without
specifying the precise size of the opening:
The degree of
Table 3.1 Manner of articulation settings
channel opening is
relative for all cat-
Category Degree of channel opening egories but stops
Stops O
Affricates a (a>)
Fricatives £ ‘(f>'a)
Sonorants Ss (s> f)
60 Sonorant Consonants
k the following segments with

ad: [ds], (s], (fl, (pl, [3], kl).
A further property of sonorants is that they are voiced, under normal

circumstances, in most languages. This means that, of the three criteria we
introduced in chapter 1 for the classification of consonants, namely, voice,
manner of articulation, and place of articulation, usually only place of articu-
lation implements contrasts between sonorants. However, the articulation
of sonorants involves additional channels or gestures that play no part in
the articulation of obstruents, and divide the class of sonorants into several
subclasses.
Nasality
The blockage or constriction of the air characteristic of obstruents is located

at some point in the mouth. This point, therefore, defines the place of articu-
lation of the obstruent.
However, in all the sounds we have examined so far, the airflow is also
systematically blocked at a second location, which we left unstated in our
descriptions. In particular, we have taken it for granted all along that, as we
attempted to pronounce each of the sounds we presented, the soft palate
would be raised to prevent the exit of air through the nose — you will recall
that the soft palate, or “velum”, is the softish area at the back of the roof of
the mouth (figure 3.1).
Figure 3.1 Raised soft palate
Now, what will happen if the soft palate is not raised during the produc-
tion of speech? See figure 3.2.
Figure 3.2 Lowered soft palate
Let us experiment by deliberately lowering the soft palate as we pronounce

[b].
The resulting sound will obviously be different from [b]: if you utter
stabber while lowering the velum for the sound represented as bb, you will
hear stammer. This can only mean that [m] (the phonetic symbol for the sound
spelled mm in stammer) is identical to [b] in all respects but one: the posi-
tion of the soft palate.
The difference between the two sounds, therefore, involves nasality. On the
one hand, [b] is an ORAL SOUND, since during its production no air comes
out through the nose due to the raised soft palate. By contrast, [m] is a NASAL
SOUND, since the lowered velum allows air to come out through the nose.
Nasality aside, [b] and [m] are identical: bilabial stops (figure 3:0).
C ¢
_.#
,
[m] [b]
Figure 3.3. Articulation of [m] and [b]
They are bilabial because they are articulated with the two lips. They are
stops because they involve total blockage of air in the mouth, even though
in the case of [m] air keeps coming out through the nose.
nt as you pronounce stammer. Now

the same with tabber. oe oe ote 7 ane ces?
What is their cause?
At first sight, the continuous stream of nasal air makes the classification of
[m] as a stop appear contradictory: in chapter 1 we said that sounds with
no air blockage, such as fricatives, are not stops by definition.
: Tryand guess ay [m] and other nasals

_are classified as On —
The answer is that, of the two articulatory actions you now know partake
in the production of [m] and other nasals (the stricture in the mouth and
the lowering of the velum), the stricture in the mouth is more important.
Because sounds are characterized by their PRIMARY ARTICULATOR, [m]
“hanttersed by Will be defined as a (bilabial) stop.
their PRIMARY Turning now to vocal fold activity, you know that both [b] and [m] are
ARTICULATOR voiced, and that [b] without voice is in fact [p]. What will a voiceless [m]
amount to?
In fact, no voiceless [m] exists in English at the level at which the basic
sound contrasts between the words of the language are catalogued, the
lexical level. Thus, while English could have a word stapper forming a min-
imal pair with stabber, it could not have a word sta[mler minimally contrasting
with stammer — the IPA underscripted diacritic “.” indicates voicelessness,
and consequently [m] is the symbol for a voiceless [m].
Indeed, we have already mentioned that sonorants (of which nasals, and there-
fore [ml], are instantiations) are typically voiced in all languages, for reasons
of physics which need not concern us here. The upshot of this is that Eng-
lish lacks a phoneme /m/, and voiceless sonorant phonemes in general.
Some languages of South East Asia do have voiceless nasal phonemes,
however: for example, the words [ma] and [ma] form a minimal pair in
Burmese, the former signifying ‘hard’ and the latter ‘notice’.
The Universal Nasal
The sound [m] does not exhaust the inventory of nasal sounds.
One obvious additional nasal sound in English, and probably in all of the
world’s languages, is present in such words as knit or tin, initially and finally,
respectively (notice the purely orthographic value of k in knit). Having estab-
lished the close correspondence of [m] with [b], you will not be surprised
to hear that the sound symbolized as [n] also has an oral counterpart.
In order to discover which this is, you have to work out the place of articu-
lation of [n], and then identify the English voiced oral stop articulated at
the same place.
Mone can now have a second go akfinding the oral comelite o

[n], after ascertaining the es of articulation of [n] through
your-memory, if i
The articulation of [n] takes place on the alveolar ridge, and consequently
[n] is an alveolar stop. It is also voiced, as we know all nasals (and other
sonorants) are in English and most other languages. On the basis of these
settings, the identification of the oral partner of [n] will offer no special
difficulty: it must be a voiced alveolar oral stop. Familiarity with the informa-
tion presented in the preceding two chapters will enable you to identify this
sound as [d] (figure 3.4).
In] [d]
Figure 3.4 Articulation of [n] and [d]
It should be clear by now, both conceptually and experientially, that

nasal consonants are characterized by the presence of a second resonating
Nasal consonants chamber, the nasal cavity, in addition to the oral chamber that also charac-
are characterized
terizes obstruents — the precise meaning of the term “resonating chamber”
by the presence of
a second resonat-
bears on the physics of sound, or “acoustics”, and need not concern us
ing chamber: the here: we can be satisfied with the idea of a space filled with air in which
nasal cavity sound is produced. The fact that there are two resonating chambers in nasals
confirms our statement at the outset that the articulation of sonorants
involves channels that are not present in the articulation of obstruents.
The articulation of obstruents can, however, also involve two sources.
In particular, you will recall that, in addition to the oral source, the articu-
lation of obstruents can involve a glottal source contributing voice. It fol-
lows that the pronunciation of nasals involves three (rather than just two)
sound sources: the oral source (also present in obstruents), the nasal source
(defining nasals), and the glottal source — remember that nasals are usually
voiced. We can construe these various sound sources as sound-defining para-
meters, alongside manner of articulation. A PARAMETER is therefore a
criterion for classification, akin to one type of building block. Within the A PARAMETER is
a criterion for
musical universe we have been drawing analogies from, a parameter can
classification
be likened to an instrument in the orchestra: the tune played by this instru-
ment is obviously one of the components that make up the symphony.
Z§ Other Nasal Consonants
A third nasal consonant is quite common in English.
The nasal in question never occurs at the start of a word. In fact, we will
see that there are reasons to believe that this nasal is not a lexical segment
of English, that is, that it is not present in the lexical level, from which alterna-
tion is excluded, as you know.
This third English nasal does, however, occur at the phonetic level in non-
word-initial position. It is exemplified in such words as wing, sung or gong,
which contrast minimally with win, sun and gone in most accents. Wing and
win, for instance, make up a minimal pair in these accents. Even if the the
final orthographic g is pronounced, as is typical, for instance, of Birming-

ham or Liverpool, in England, the n of wing will be phonetically different
from the n of win, in spite of their orthographic identity.
The place of articulation of the n in wing is identical to the place of articula-

tion of [g]. Accordingly, such n is the nasal counterpart of [g] (figure 3.5).
Figure 3.5 Articulation of [n]
part in terms of our familiar pa ameter
The phonetic symbol for the velar nasal is [n].

If [n] is absent from the lexical inventory of English, the question arises
of why it occurs at the phonetic level. The answer is that English [n] comes
about as the result of assimilation of the place of articulation of a lexical /n/
to that of a following /g/, which is subsequently deleted in many accents.
As we are seeing repeatedly, assimilation is one of the major factors respons-
ible for the differences between the lexical and the phonetic levels postu-
lated in chapter 2.
In some languages, [n] is a legitimate lexical segment. In those languages

we would not expect the same limited distribution of [n] we find in English,
where it can never occur word-initially. For example, in Malay [n] can occur
in all positions, including word-initially: [nleri ‘terrified’ forms a minimal
pair with [nleri, a type of plant.
Looked at from a more traditional perspective, [n] is therefore phonemic

in languages like Malay, hence formally /n/. Phonemic status has also
been claimed for English [n], on the basis of minimal pairs like those given
above: wing vs. win, etc. You can now see that phonemic status does not
necessarily imply lexical status: phonemic status follows from the existence
of systematic contrastiveness in the surface, whereas lexical status requires
constrastiveness at the lexical level.
A fourth nasal, reasonably common across languages, is very marginal in

English. This nasal can turn up in such words as onion or canyon when said
casually.
The place of articulation of this new sound, [p], standard in French, Spanish
or Italian, among others, is similar, although not identical, to the place
of articulation of the [dg] of judge (figure 3.6 below). In particular, as we
explained in chapter 1, [ds] is a palatoalveolar sound, pronounced in the

area between the palate and the alveolar ridge. By contrast, [n] is purely
palatal: the blockage takes place entirely on the hard palate. Notice that
the term PALATAL is reserved for sounds articulated on the hard palate:
you already know that sounds articulated on the soft palate are referred to
as “velar”, the adjective of “velum”, which we said in chapter 1 is the Latin
word for the soft palate.
Figure 3.6 Articulation of [n]
Define [p] in termsofall the familiar honetic parameters. :
Although many languages have a contrastive palatal nasal, conventional

orthographies do not have a letter for it. Thus, in French and Italian, [yn]
is spelled by means of the digraph gn (French agneau ‘lamb’; Italian ogni
‘every’), in Spanish as n with a tilde (/, as in afo ‘year’), in Dutch as the
digraph nj (Spanje ‘Spain’), in Portuguese as the digraph nh (anho ‘lamb’),
and so on. The historical reason for this spelling diversity is that this sound
did not exist in Latin, and therefore there was no letter for it in the Roman
alphabet, from which the alphabets of these and other Western European
languages were taken.
In Spanish, French and many other languages, there are many minimal
pairs involving [pn] and the other nasals: Spanish ca[pJa ‘rod’ vs. ca[n]a ‘grey _
hair’ and calmla ‘bed’, for instance. In the absence of contrary evidence, /n/
must therefore be analysed as a phoneme in these languages. From the
updated perspective on phonology we are adopting here, it is reasonable to
assume that /y/ also has lexical status in these languages, but obviously
not in English.
Liquids
So far in this chapter we have been seeing that nasal consonants are char-
acterized by a second resonating chamber, the nasal cavity. Nasalization thus
functions in a similar way to voice: it provides an additional source of sound Nasalization func-
which supplements the oral source to give rise to a complex sound. An import- tions in a similar
way to voice: it
ant difference between nasalization and voicing is the manner in which the
provides an addi-
two sound sources are connected, sequentially for voice, but in parallel for tional source of
nasality, as we represent in the following diagrams: sound which sup-
plements the oral
Air — Voice —> Oral articulation ——> Sound source to give rise
to a complex
Sequential coupling of voice sound
Nasalization
Airceig : j thc Sound
Oral articulation
Parallel coupling of nasalization
Resorting once more to a musical analogy, we suggested in chapter 1 that

the addition of voice to an oral articulation is reminiscent of the plugging
of a recorder into a trumpet: the wind will first power one instrument and
then the other. In the case of nasality, however, a more apposite analogy
would be the bagpipes, where the chanter pipe, responsible for the main
melody, and the drone pipe(s), providing the typical background buzz, are
activated simultaneously by the air coming out of the bag.
You also know from the preceding discussion that the primary articulation
Sonorants have a
of nasals, located in the mouth, is of a stop kind, whereas their secondary
simultaneous con-
articulation, responsible for their nasality, is of a continuant kind: the sound tinuant and non-
continues for as long as air is available in the lungs. In nasals, therefore, a continuant
CONTINUANT and a NON-CONTINUANT mode of articulation are effect- articulation
ively superimposed onto one another.
Sonorants can also be articulated exclusively in the mouth, with no nasal

component. These non-nasal sonorants still have a simultaneous continuant
and non-continuant articulation: during their production one part of the oral
channel is blocked, while another part remains unobstructed and allows the
air to escape freely. Such sounds are commonly referred to as LIQUIDS (per-
haps because they sound fluid), and we now turn our attention to them.
fa Laterals
Let us compare the middle consonant in mellow (the doubling of the letter
in the spelling is of course immaterial) with its counterpart in meadow.
The articulation of both these sounds is alveolar, that is, it involves placing
the blade of the tongue on the upper alveolar ridge. Both sounds are also
voiced. Last, they both involve a complete closure at the upper front alve-
olar area. Given these striking similarities, what is it that makes these two
sounds different?
If you pay close attention to the articulation of the two sounds in question,
you will notice that, in the case of [d], the tongue presses firmly against the
upper teeth all around, not just at the front, but also on the sides, to pre-
vent any air from escaping.
By contrast, for the sound found in the middle of mellow, represented
by the phonetic symbol [I], the sides of the tongue (only one side in some
speakers) do not touch the complete set of upper teeth, and air comes out
continuously through the resulting gap. Because the air flows over sides of
the tongue, these sounds are known as LATERALS: Latin lateralis means ‘of
the side(s)’, from latus ‘side’ (compare such English expressions as lateral
thinking, collateral, etc.).
As regards place of articulation, [I] is defined as an alveolar sound,
exactly like [d] (figure 3.7).
Figure 3.7 Articulation of [I] and [d]
The definition of [1] and [d] as alveolar obviously suggests that the gesture
made by the blade of the tongue is regarded as primary, and the gesture
made by the sides of the tongue (closing for [d] and opening for [l]) as sub-
sidiary. Indeed, primary articulation is commonly related to the median line
of the tube through which the air flows out: to the mouth’s median plane, Primary articula-
tion is commonly
in more technical parlance. In addition, [Il] is voiced, also like [d], unsur-
related to the
prisingly so, since we stated above that sonorants (of which class liquids, central area of
and thus [I], are members) are characteristically voiced across languages. the mouth
Finally, [1] is also oral, since the velum remains raised during the whole of
its production, just as it does with [d]. The difference between [I] and [d],
therefore, lies exclusively in the respective status of these two sounds with
regard to LATERALITY, the term referring to the lowering of the sides of
the tongue during articulation: [I] is lateral, while [d] is not.
YM
In the opening paragraph of the section we said that [I], like [d], involves
air stoppage at the front. We now know that during the articulation of [I]
air continues to flow out of the mouth through the gap formed by the sides
of the tongue and the upper teeth on the sides of the mouth. We came across
a similar situation earlier for nasals, characterized by air stoppage in the mouth
but continuous airflow through the nose, and we attributed their conven-
tional description as stops to the fact that the articulation in the mouth is
regarded as primary.
Phonologically, nasals always function as stops, and therefore their clas-

sification as such is uncontroversial. The tendency for [I] is also to function
as a stop, but in a few languages it appears to pattern with the continuants.
Consequently, we must leave the classification of this sound somewhat
flexible. Articulatorily, however, [I] is always considered alveolar: this
confirms the privileged status of the central region of the mouth in the
identification of the primary articulatory gesture.
An additional lateral sound, a patalal lateral, bears a similar relationship
to [1] to that which [p] bears to [n]. The phonetic symbol for the palatal lat-
eral is [A], an inverted “y”. The articulation of this sound, spelled gli in Italian
(zabaglione) and Il in Spanish (paella), is a bit exotic for the English speaker,
and we will accordingly put some extra care into its description. Pronun-
ciation guides tend to suggest the lateral sound in million as a close English
equivalent, but this correspondence is subject to a number of caveats. To pro-
duce [A], the body of the tongue (that is, the area behind the blade) must be
raised to the roof of the mouth to block the exit of air, as was the case for
the nasal [n]. However, for [A] the sides of the tongue must be allowed to
hang free, to let the air flow out through the resulting gap, in the familiar
lateral gesture.
In the British context it is perhaps advisable to caution the reader about

the Welsh sound spelled II, as in such place names as Llandaf, Llandudno,
Llangollen, Llanelli, Llanfairpwllgwyngyllgogerychwyrndrobwillllantysiliogo-
gogoch, etc. (lan simply means ‘church, village’ in Welsh). Although also
an alveolar lateral, this sound involves frication, caused by a considerable
narrowing of the lateral gap(s). Therefore, it is not a sonorant, since the defining
criterion for sonorants is unobstructed exit of air. Indeed, the Welsh sound
in question is an obstruent, which is moreover voiceless. The phonetic sym-
bol for this sound is [4].
The correspondences in place of articulation between nasals and laterals

are completed with the velar laterals that exist in a handful of languages.
In the Mid-Waghi language of New Guinea, for instance, the word [aLate]
‘dizzy’, with the velar lateral [L], contrasts (although not minimally) with
the word [alala] ‘speak incorrectly’, with the ordinary alveolar [1]. The sound
[L] is also reported to occur in some English accents in some contexts (for
instance, before labial or velar consonants), but it is otherwise rare. The
velar lateral [L] must not be confused with the velarized alveolar lateral [1).
The sound [#] occurs allophonically in English in word-final position, and
syllable-finally generally (syllables are dealt with in chapters 9 and 10), as
in pill, mole or cool, although some accents only have plain, or “clear”, Is (gen-
eral Irish), or velarized, or “dark”, Is (general Scottish).
Pronou
in the so
In languages like Russian, [1] functions phonemically: mot ‘pier’ and potka
‘polka’, with [4], contrast with mol ‘moth’ and polka ‘shelf’, without. The
articulatory difference between the velarized I, [1], and its plain counterpart
[1] lies in the additional bunching of the body of the tongue at the back that
characterizes [t] (figure 3.8).
Repeat the last exercise ifyou had difficulties infinding this out.
Plain [1] Velarized [t]
Figure 3.8 Plain and velarized |

In some accents (London Cockney, for instance) the velar lateral [1] loses its
alveolar contact and takes on lip rounding, effectively becoming the sound
represented by w in bow.
We have seen that both laterals and nasals can be realized at a range of
articulatory places. However, all laterals must by our definition involve the
All laterals involve tongue in their articulation; labial nasals, by contrast, are realized exclusively
the tongue in their
with the lips.
articulation
Rhotics
“Rho” is the Greek name for the letter r, and the label RHOTICS therefore
The label RHOTICS refers to a class of sounds that are “r-like”. It will soon become apparent,
refers to a class
however, that the members of this class do not necessarily have much in
of sounds that
are “r-like". The
common with each other phonetically: their common grouping as “rhotics”
members of this is grounded on similarity of phonological behaviour, rather than on shared
class do not phonetic substance.
necessarily have
much in common
with each other
phonetically
The sound represented by the letter r in most accents of English is very

different from its counterpart in many other languages. In turn, the r sound
typical of Scottish English is different from its common English equivalent,
as we will see below. The phonetic symbol for the common Englishris [4],
an inverted r. The articulatory gesture for [1] is almost the opposite of the
articulatory gesture for [I], hence the tongue twister “red lorry, yellow
lorry”.
be
In the previous section we explained that for [1] we blocked the air in the
central part of the mouth by pressing the blade of the tongue firmly against
the alveolar ridge, while letting it flow freely down the sides. By contrast,
for [1] the sides of the tongue touch the back teeth, while a fairly wide gap
is created in the centre of the mouth for the air to pass through without caus-
ing friction (figure 3.9).
[a]
Figure 3.9 Articulation of [1] and [I]
Try comparing ve oe of ihe ce during [i] an -duri

— [i]. Then focus your attention on where exactly you pla
There are two ways in which the blade of the tongue may be positioned
for [1]. The chances are that speakers from Britain will keep the blade flat,
leaving a channel open at the front for the air to escape (figure 3.10).
Figure 3.10 Plain (British) [1]
American speakers, on the other hand, are more likely to curl back the blade
towards the roof of the mouth (without of course touching it or drawing
the tongue too close to it). See figure 3.11.
Figure 3.11 Retroflex (American) [4]
Fortunately, the effect of this articulatory difference on the resulting sound

is not great. The action of curling back the tongue is known as RETRO-
FLECTION, and the sounds thus produced are referred to as RETROFLEXES.
i Ae
4 Try pronounc
Not only rs can be retroflected, but also all other sounds that involve the
tip of the tongue in their articulation. For instance, the d at the end of the
Swedish word smérgdsbord is retroflex. In IPA phonetic transcription,
retroflection is indicated by the addition of a tail to the symbol of the cor-
responding non-retroflex, hence [q] for the English retroflex r, [d] for the
Swedish retroflex d, and so on. In an alternative notation, in common use
for typographical convenience, retroflection is represented diacritically by
means of an underscripted dot: [4], [d], etc. Retroflex sounds are particu-
larly frequent in languages of the Indian subcontinent. For instance, in
Malayalam, the main language of the state of Karala, in Southern India, ku[t]i
‘child’, with a retroflex [t], makes up minimal pairs with ku[t]i ‘stabbed’, with
a dental [t], and ku[t]i ‘peg’, with an alveolar [t].
-Retroflection tends to carry over to

speakers originating in the Indian sube
a range of sounds that are alveola
how much you can retroflect them.
A minimal amount of self-observation will make it obvious that the mech-

anics of tongue positioning for [1] are rather subtle, if not downright fiddly.
Sonorant Consonants Wii.
Not surprisingly, therefore, children often have difficulty in pronouncing

[1] and substitute w, so rock becomes wok. Some adult speakers (including
some well-known personalities) also have a so-called “defective” r, similar,
but not identical, to w: the typical “defective” r is produced by drawing the
upper teeth onto the inside of the lower lip further back than for [f] or [v],
and not quite close enough to cause friction.
The “defective” r is perceptually similar to the more common English r, in

spite of the striking articulatory difference between the two. The “defective”
r is defined phonetically as a (voiced oral) labiodental approximant. Its
IPA symbol is [v]. The term APPROXIMANT, related lexically to approxim-
ate and approximation, in turn connected to approach, is applied to sounds that
are continuant and frictionless.
A very common rhotic, found in Spanish, Russian, Greek and many other
languages, as well as in Scottish English, at least historically, involves
vibrating the tip of the tongue against the upper tooth ridge, hence the label
“alveolar TRILL” (also, informally, “rolled 1’): Spanish rosa, Russian roza,
Greek (arch.) rodo, all ‘rose’. Mechanically, and aerodynamically, the production
of this sound parallels the production of voice, although, obviously, the artic-
ulators are different. You will recall that voice is caused by the vibration of
the vocal folds. Most importantly, such vibration is not created by actively
moving the vocal folds against one another, in the way we actively move
our hands against one another to clap. Rather, it is brought about by posi-
tioning the vocal folds appropriately (not too close, not too far apart; not
too tight, not too loose) and letting the high-pressure air passing through induce
the vibration, like the flapping of a flag in the wind. |
Vocal fold vibration is undoubtedly part and parcel of the sound invent-
ory of all languages, and consequently it is unlikely that readers of this book
will face difficulties with voicing (naturally, pathologies aside). This is not
the case with the alveolar trill, which at least some speakers of English emphat-
ically claim they are unable to produce, even though this sound is used extens-
ively by young boys to mimic the firing of machine guns! Obviously, this
_ inability cannot be physiological (again, true pathologies aside), and must
simply be attributed toa failure to take the articulatory steps needed for the
production of this sound, which we will now make explicit.
In fact, whether or not you think you can pronounce the alveolar trill, it
is most likely that you already have another trill in your non-linguistic reper-
toire, namely, the trill we use more or less unconsciously to indicate cold,
and which involves bilabial vibration.
As was tke case with the production of voice (humming), the production of
a bilabial trill requires the lungs to be well filled with air, in order to
increase the air pressure inside, and concomitantly the force with which the
air comes out. The lips must also be positioned next to each other, without
undue tightening up or slackening. When pressurized air is let out, the lips
vibrate automatically.
Suppose now that, instead of bringing the lips together, you place the tip
of the tongue just above the upper tooth ridge, again neither too tightly nor
too loosely, taking care that the sides of the tongue press against the set of
lateral upper teeth tightly enough to prevent air from escaping through the
sides. If you now let through a substantial amount of high-pressure air, this
air will automatically set the tip of the tongue vibrating, exactly as it sets
the two lips vibrating in the cold gesture. An apposite musical analogy is
the reed of a wind instrument, which obviously the player does not manip-
ulate directly, but rather through the intermediary of the current of air.
The phonetic symbol for the alveolar trill is [r]. This symbol is obviously
identical to the letter r, and it is sometimes used loosely (for typographical
convenience, particularly when there is no likelihood of misinterpretation)
for other phonetic varieties of r also. This includes the English r, which we
saw above is a very different sound, and must strictly speaking be transcribed
as [1]. Such liberal use of phonetic symbols is for better or worse a fact of
transcription life, and must be accepted by the budding phonologist philo-
sophically, if perhaps not always joyfully.
The reader familiar with Spanish will be aware that there is another r in
this language besides the alveolar trill just described, as demonstrated by
the phonetic contrast between such words as pero ‘but’ and perro ‘dog’, with
the digraph rr representing the alveolar trill. This softer r is also found in
other languages, whether or not in contrast with [r]. Indeed, it is nowadays
more typical of Scottish English than the alveolar trill, contrary to popular
stereotype.
We will ease our way into describing the soft r by thinking of the typical
American pronunciation of t in such words as waiting (similarly in many
Irish varieties). Such a t does not of course sound anything like the [t] we
described in chapter 1: it is indeed a different sound, which substitutes for
[t] in various contexts in American English (see chapter 11 for details).
It clearly involves a single flap or tap of the tongue tip, which is essentially
thrown against the alveolar ridge. This means that we are dealing with yet
another alveolar sound.
The sound in question is also oral, since the velum is raised during its pro-
duction, and, being a sonorant, it is voiced.
You may find the classification of this sound as a sonorant somewhat

puzzling, given the fact that its production involves oral closure, and it is
neither a nasal nor a lateral. The reason it is thought of as a sonorant is
that the contact between the tongue tip and the alveoli is fleeting in the ex-
treme, and therefore the airflow remains essentially unaltered. The argument
carries over to the interruptions that make up the trill, also commonly con-
sidered a sonorant.
The American t in waiting we have just described is usually referred to as
a FLAP. The Spanish rin pero ‘but’, or the typical Scottish r, is similar, but
perhaps not absolutely identical, and is usually dubbed a TAP. The differ-
ence between a tap and aflap is subtle, but has been argued in the special-
ized literature. It is not in our interest to go into this level of detail here,
and we will accordingly leave the matter as it stands. The phonetic symbol
for the tap (as in Spanish pero ‘but’) is [r]. This symbol is also proposed for
the flap of the American waiting in the latest version of the IPA symbol chart.
American authors, however, have tended to use a capital d to represent the
flap, hence [D].
We shall now wind up our survey of rhotics — there are still more across
languages, but the present inventory is quite sufficient for our purposes. As
we pointed out at the outset, while they are spelled r in most languages,
rhotics can differ considerably from each other in their articulation. To add
Rhotics can differ to the confusion, the letter r is also used to represent sounds that are not
considerably from
even sonorants. For instance, the standard French and German 1s are uvu-
each other in their
articulation lar fricatives, rather than rhotics as such: [k] (voiced) or [x] (voiceless). The
UVULA is the appendix found at the end of the soft palate, and therefore
the place of articulation of these sounds lies between the place of articu-
lation of such velars as [x] or [y] and the place of articulation of the glottal
[h] (figure 3.12).
Figure 3.12 Uvular fricative
rs tend to function
as sonorants, even
What is interesting, indeed puzzling, is that all the phonetically quite differ-
when they are not ent rs function similarly in the respective phonological systems. In particular,
so phonetically they tend to function as sonorants, even when they are not so phonetically,
undoubtedly the reason they are generally construed as rhotics.
Indeed, in languages like German, Dutch, Swedish and others, r is pronounced

as the alveolar sonorant trill [r] in some dialects, or even individually, and
as the voiced uvular fricative [s], a phonetic obstruent, in other dialects or
idiolects.
E] Summing Up
In this chapter we have examined the articulation of sonorant consonants:

nasals, laterals and rhotics, the last two grouped together as liquids. Fol-
lowing our practice in chapter 1, we now tabulate all of these sounds under
their respective IPA symbols (table 3.2).
Table 3.2 Sonorant consonants
Bilabial Labiodental Alveolar Postalveolar Palatal Velar
Nasals |. m n n 0
Laterals ] A
Rhotics v c I
(in “defective” (flap or tap) (in standard
English) English)
r
(trill)
We have also described the lateral fricative alveolar [#] (as in the Welsh place
name Llandaf). While not a sonorant, this sound is of course closely related
to the lateral sonorant [1]. Also obstruents, rather than sonorants, are the
uvular fricatives [«] and [x], which we have just mentioned are used for r
in several languages. On the other hand, there exists a uvular trill [Rr], which,
as a trill, must be considered a phonetic sonorant. This uvular trill is some-
what reminiscent of gargling to the ear of the English speaker. It occurs in
some older dialects of French, and can be heard in recordings of the singer
Edith Piaf (cf. her classic “Je ne regrette rien”). It is also found in Lisbon
Portuguese, while the usual sound for r in Brazilian varieties of this language
is [x] or [h]. We gather all these heterogeneous sounds in table 3.3, to facil-
itate comparison.
Table 3.3 Additional consonant sounds discussed in the text
Alveolar Palatal Velar Uvular Glottal
K
Plain x x, h
Laterals ¢
Key Questions
How does a sonorant sound differ 7 What does a “continuant" mode of

from an obstruent? articulation refer to?
What role does the soft palate play 8 Define “laterality”. What is lateral
in the articulation of speech sounds airflow?
beyond those listed in chapter 1? 9 Why do lateral sounds always
Define “nasality”. involve the tongue?
What do we mean by a “primary 10 Enumerate the variety of “rhotic”
articulator"? Suggest what a sec- sounds. Why are they generally
ondary articulator is. grouped together?
5 List the parameters relevant to the
description of consonant sounds.
6 How does the relationship of nasal-
ity with oral articulation differ from
that of voice with oral articulation?
Further Practice
Sound to Spelling
We have represented the consonant sounds in the words below by phon-

etic symbols according to our pronunciation. Work out what the spelling
is.
[nJe[vJer [nJeu[m]Jo[nlia [nlow[]e[d&le au[t]u[m]

[nle[mJo[n]ilk] [8]iln] [8]ilyk] filn]ger
[b]o[m] [kJulaly [x]ilt] [taly
(1ilt] [f]ul[t] [f]ullly [kl]ear
[mJe[nfn] [t]a[llialtle[l]i [nJalt] ca[nlon
Odd One Out
- Find the odd one out in the following sets and state reasons.
avcivxk py sl d. [zddfxv] g. [IqxNneo]

b. [ImvB pb] e. [kdbftg] h. [lrA 1]
e {lkur] fe Vk
ro no i. [O0ftpsf]
Articulation and Phonetic Symbols
Give the IPA symbol which corresponds to the descriptions below:
A labiodental approximant
A retroflex nasal stop
A voiceless uvular fricative
A velar nasal stop
An alveolar lateral stop
A voiced lateral fricative
Provide full phonetic descriptions for the sounds represented by the sym-
bols below:
[t] [q) [ct] [mp] [8] [Nn] [A] [py]
NATURAL CLASSES OF SOUNDS
DISTINCTIVE FEATURES
& Interpreting phonetic parameter:

tinctive features”.
In the previous chapters we have described the phonetic characteristics of

both obstruent and sonorant consonants. In addition, in chapter 2 we pres-
ented phonology and disentangled its concerns from those of phonetics, against
the backdrop of the general structure of language. In the present chapter
we introduce a number of formal devices central to the model of phono-
logy we are concerned with. We first introduce the phonological correlate of
the phonetic parameter: the “distinctive feature”. We will see that most dis-
tinctive features are naturally binary, with two complementary values, but
some features are intrinsically unary, and that distinctive features define
natural classes of segments. Most importantly, we will see that the elements
that make up phonological structure, the features in particular, are in prin-
ciple structurally independent of each other, a state of affairs commonly
referred to by the label “autosegmental phonology”.
86 Natural Classes of Sounds
Descriptive Phonetic Parameters
In chapters 1 and 3 we reviewed in some detail the articulatory phonetics

of several groups of consonants relatively familiar to the English speaker.
Each consonant was described on the basis of a range of dimensions, or para-
meters, each representing a relevant aspect of its articulation.
Thus, for instance, [p] was described as a voiceless bilabial stop obstruent,
[z] as a voiced alveolar fricative obstruent, [n] as a voiced velar nasal stop
sonorant, and so on.
Distinct segments must by definition differ in the setting of at least one such
Distinct segments parameter. Segments kept separate by only one parameter are minimally con-
must by definition
trastive. For instance, the pair sip ~ zip, mentioned in chapter 1, is kept sep-
differ in the set-
ting of at least
arate by the fricative obstruents [s] and [z], which differ only in their value
one parameter. for voice: [z] has voice, since it is pronounced with vocal fold vibration, but
[s] does not, since it is pronounced with the vocal folds inactive.
In most cases, however, the contrast between two segments involves more
than one parameter. For instance, in our present terms, [n] is defined as a
voiced alveolar nasal stop sonorant, while [f] is defined as a voiceless labio-
dental oral fricative obstruent: [n] and [f] therefore differ in the setting of
all the parameters mentioned, although of course they still overlap in being
non-lateral consonants.
Natural Classes of Sounds 87
The advantages of a system based on parameters are obvious. First, such

a system enables us to describe each of the sounds of any language in a
reasonably economic and uniform manner: [p], for instance, will be a voice-
less bilabial obstruent stop in any language; indeed, [p] = voiceless bilabial
obstruent stop. A second, related advantage of the parameter system is that
it allows us to see at a glance what the significant differences between any
two sounds are. For example, if we write [p] and [b], in standard IPA sym-
bols, the differences between the sounds thus represented are not obvious to
the eye: “p” is graphically no closer to “b” than it is to “q”, for instance, and
MAT
“b” is at least as similar to “d” as to “p”. By contrast, if we write “voiceless

bilabial obstruent stop” (= [p]) and “voiced bilabial obstruent stop” (= [b]),
respectively, we can immediately see that the difference between the two
sounds hinges on the voice parameter. A third advantage of our parameter-
based system is, of course, its phonetic motivation: we are not defining the
sounds in question with arbitrary labels, but with labels that are directly
grounded on their phonetic realization.
Distinctive Features
One disadvantage of this phonetic grounding concerns the need to prolif-

erate labels to keep pace with the phonetic facts. Indeed, we have come across
quite a number of these labels already, but we are likely to need many more
if we are going to add a label for each phonetic detail. This result runs against
the grain of Occam’s razor, which we mentioned in chapter 2 as favouring
simplicity in scientific modelling. To counter this it could be argued that com-
plex facts impose richness of both concepts and terminology. This is of course
true if we remain close to the phonetic ground. However, the focus of this
book is not on phonetics, but on phonology. Phonology is concerned with
Phonology is con-
structural patterns rather than with phonetic minutiae, and therefore a sim-
cerned with struc-
plification of the repertoire of parameters is in order here. tural patterns
rather than with
phonetic minutiae,
and therefore a
simplification of
the repertoire of
parameters is in
One of the possible strategies we can follow to achieve this simplification order
highlights another disadvantage of the system we have been using: many

of the labels we have supplied are complementary, in as much as the state

of affairs one label refers to is the opposite of the state of affairs designated
by the other label. For instance, “oral” is the precise opposite of “nasal”
(“oral” = “non-nasal”, and “nasal” = “non-oral’”), “sonorant” is the opposite
of “obstruent”, and so on.
Besides being uneconomical, duplication of labels is potentially misleading,

Duplication of or at least confusing. In particular, it is easy to lose track of the fact that
labels is potentially “obstruent” means exactly the same as “non-sonorant” (and “sonorant” the
misleading
same as “non-obstruent”), and similarly for a number of other label pairs.
The obvious solution to this problem involves the adoption of one of the two
complementary labels as the only “official” label, its counterpart being entirely
disposed of, or at least relegated to informal prose. The terminological gap
left behind will of course be filled with the negated term of the surviving
label: the dichotomy “sonorant” vs. “obstruent” now becomes “sonorant” vs.
“non-sonorant”, and so on. You must, however, be aware of the fact that the
selection of labels has sadly not been carried out uniformly by all practitioners.
For instance, the opposition “sonorant” vs. “non-sonorant” is expressed as
“non-obstruent” vs. “obstruent” by some, and likewise for other oppositions,
a minor inconvenience we will have no choice but to tolerate. This is one
reason you still need to gain familiarity with the less common labels. Another
reason is that many such labels are in current use in the phonetic literature,
to which the phonologist must of course have ready access.
As a further small, but still important, step in the process of formal rational-
ization we are engaged in, we shall express the negation “non-” by the neg-
ative algebraic symbol “—”, so that, for instance, we will write “non-sonorant”
as [-sonorant]. For reasons of symmetry, we will write “sonorant” with the
opposite algebraic symbol, “+”: [+sonorant].
We now have the kernel of the formalization we will be adopting hence-

forth: a restricted set of DISTINCTIVE FEATURES (“distinctive” because they
keep sounds distinct; “features” because they express properties of the sounds:
“parameters” would of course have been just as good alabel), which are
endowed with an alternative binary value, positive if the property named
by the label is present in the sound being defined, and negative if it is not. DISTINCTIVE
Some such distinctive features simply translate the phonetic parameters FEATURES are
endowed with an
you are already familiar with in a self-explanatory manner: [+sonorant],
alternative binary
[+voice], [tnasal], [+ateral] and [+continuant] ([+continuant] was men- value
tioned in chapter 3, and refers to continuous airflow through the central area
of the mouth).
Note that, by convention, distinctive feature labels are always enclosed in

square brackets. Such labels are preceded by the operators + or — expressing
the precise value of the feature: [+sonorant] stands for either [+sonorant] or
[—sonorant], and so on.
The list of distinctive features is provided once and for all for all languages: The list of dis-
it is assumed to be part of UNIVERSAL GRAMMAR, the set of principles tinctive features
for language all humans are endowed with innately. We will be presenting is provided once
this list gradually, then providing a unified picture in chapter 17. Note that, and for all for
all languages:
while the list of features is assumed to be universal by pretty well all phono-
it is assumed
logists, its actual contents vary slightly from proposal to proposal: you must to be part of
not lose sight of the fact that, like any other aspect of phonology, this list is UNIVERSAL
hypothetical, given the conjectural, rather than mechanistic, nature of the GRAMMAR
~ enterprise.
The system we are proposing — with a fixed list of distinctive features, each
assigned one of two values (+ or —) — is ideal to implement classification,
which is after all what we have been doing so far with the sounds of speech,
the set of consonants, to be more precise. It is rather like classifying people
into [+female] or [—female] (equivalently, [-male] or [+male]), bicycles into
[+racing] or [—racing], or animals into [+pet] or [-pet]: the obvious advant-
age of such a system is that it immediately brings out both the criteria for
classification and the exact position of any given element in the system. Notice
that the system is maximally simple (it should only contain the features neces-
sary to implement classification), clear (each value is immediately transparent:
[-sonorant], for instance, refers to non-sonorants, that is, obstruents), and
unambiguous: there is no chance of missing the complementarity between
[+sonorant] and [-sonorant], as there would be if we used the labels “sonor-
ant” and “obstruent”. A further, even more important advantage will be
discussed in the following sections.
Naturalness and Formal Economy
In the preceding chapters we saw that phonetic symbols are very useful when
Phonetic symbols we want a written record of language sounds. In this chapter, however, we
are very useful
are seeing that such symbols are not at all helpful when we want to find or
when we want a
written record of explain sound patterns.
language sounds. Consider for instance the assimilation of /p/ to [P] examined in chapter
However, such 2 ([p] is of course bilabial, while [P] is our ad hoc symbol for a voiceless
symbols are not at labiodental stop). We saw then that the context for such assimilation is the
all helpful when presence of a labiodental sound, such as [f], in the following position.
we want to find
or explain sound
patterns
The rule we proposed for the description of the change from /p/ to [?] reads
as follows:
Now, you already know that /b/ changes to [%] (remember, our ad hoc sym-
bol for a labiodental voiced stop, in the absence of an official IPA proposal)
in the same context, which also triggers a change of /m/ into [m] (the IPA
symbol for the labiodental nasal), as in the phrase come for tea.
The addition of [%] and [m] to [Pf] means that we need two new rules in
our rule repertoire: /b/ > [RB] and /m/ => [mn].
In actual fact, we will need to double our present set of three rules, since
the assimilation process also occurs in front of /v/, not only before /f/, as
in rule (1) above: home video, top value, subvert, etc. So, we will need six indi-
vidual rules in total, not a very economical outcome. Moreover, both com-
mon sense and linguistic intuition are telling us that underlying the six rules
is one single phonological process, by which lexically bilabial sounds are
pronounced as labiodental before a labiodental sound.
If we used phonetic symbols, we would miss an important linguistic gener-

alization. At first blush, our reinterpretation of phonetic symbols as distinctive
features does not take us any further: if anything, the string of phonological
labels that replaces the string of phonetic symbols in each of the six rules
seems an even more complex formula.
The problem extends to all instances of assimilation, of which we examined
a few in chapter 2. We will be looking at a handful more in the remainder
of this chapter, with a view to finding a solution to the problem.
Z¥ Place Assimilation in Nasals: Natural Classes
Consider the following data:
(2) a. intolerant b. indefinite

interminable indistinct
intractable indiscreet
All these forms start with the negative prefix in- — a PREFIX is an ante-
posed affix; an AFFIX is a morpheme that needs to be attached to a base; a
MORPHEME is a minimal unit of grammatical function.
Compare now the forms in (2) with the ones in (3):
(3) a. impossible b. imbalance

imperfect
impure
The examples in (3) contain the prefix im-, also negative. Is im- simply a
different prefix from in-? If we are really confronted with two negative pre-
fixes, what are the principles governing their distribution? Why don’t we get
imtolerant and inpossible, for instance? Linguistic intuition tells us that in- and
im- are in fact one and the same prefix. In particular, both forms convey the
same meaning, and are remarkably similar in phonetic composition, even
in the diverging segments [n] and [ml], only differentiated by their place of
articulation: they are both [+sonorant], [—continuant], [+voice] and [+nasal].
It does not require a great deal of observation and thinking to realize that the
alternation between [m] and [n] in the prefix in question can also be attributed
to assimilation: the labial [m] shows up before alabial ([p] in impossible), and
the alveolar [n] shows up before an alveolar ([t] in intolerant). In order to
formulate the rule responsible for this alternation, we will have to decide
between /m/ and /n/ as the lexical consonant.
sonant ha’
How are we going to select the lexical consonant? The general idea is that
the sound that has the wider distribution (that is, turns up in most contexts) The sound that
has the wider dis-
is lexical, whereas the sound or sounds with a more limited distribution is
tribution is basic,
or are created by contextually restricted rule(s). Particularly important is the that is, lexical,
form that occurs in as neutral a context as we can find, where by “neutral” whereas the
we mean unable or unlikely to induce assimilation. One such neutral context sound or sounds
in the present case would be the word end: when a word is said in isolation with the more
limited distribution
there is no following consonant to trigger assimilation. Unfortunately, pre-
is or are created
fixes by definition cannot be word-final, and so this particular test is not avail- by contextually
able for the set of data we are discussing. restricted rule(s)
At this point we have two choices: we can look for the next best envir-
onment in the same set of forms, or we can extend the data set. We will
now see that in the present case both strategies lead to the adoption of /n/
as the lexical segment.
Let us first look for the next best environment to the word’s end. Clearly,
the context we are seeking involves a vowel-initial base, because there is
little reason to believe that the prefix-final nasal will assimilate to a follow-
ing vowel:
(4) inability
inevitable
inoperable
inimitable
We have purposely supplied forms with an assortment of initial vowels, to

allay any suspicions that the identity of the vowel may after all affect the
place of articulation of the nasal. You can see that the nasal in the prefix
turns up as [n] in all cases. Crucially, no parallel forms starting with im- can
be found.
rying to fir
nat can you say about immature, for example? —
The facts instantiated in (4) therefore point to /n/ as the lexical form of the
prefix.
This conclusion is confirmed when the data set is expanded in other

directions, the second strategy for selecting the lexical form. Consider the
pronunciation of word-final nasals:
(5) ten pens

ten boxes
ten tables
ten doors
The normal pronunciation of these sequences is, in fact, te[m] pens, te[m]
boxes, te[n] tables and te[n] doors. These data therefore parallel the data with
the prefix im- ~ in- in (3) and (2) above. Crucially, however, we can extricate
ten from any possible assimilation context by pronouncing it in isolation.
When we do so, we of course get te[n], confirming /n/, rather than /m/,
as the lexical representation. Indeed, forms that must be assumed to have
lexical /m/, like some, manifestly fail to undergo assimilation under the
circumstances:
(6) some tables

some doors
some pens
some boxes
The data in (6) suggest that a putative lexical prefix im- would never become
in-, as it would need to have done in intolerable, etc., at least other things
being equal:
By contrast, we have encountered abundant evidence supporting the
assimilation of /n/ into [m]. Therefore, we postulate /1n/ as the lexical form
of the prefix we have been discussing, although of course this is a hypo-
thesis, not a factual conclusion.
At this point in the exposition, we can propose the following (streamlined)

rule of labial assimilation for nasals:
(7) Labial assimilation in nasals:

[coronal] —> [labial] / [labial]
+nasal
Notice that we are now replacing phonetic symbols with distinctive features.
The feature [+nasal] is self-explanatory, while [labial] implies involvement
of the lips in the articulation, as in labials and labiodentals. As for [coronal],
all we need to say for now is that it formalizes alveolar articulation: in the
next section we discuss this feature and give the rationale for the particu-
lar terminology.
As formulated, rule (7) simply states that a coronal nasal immediately pre-
ceding alabial itself becomes labial. Notice that we are writing the features
[labial] and [coronal] without the “+” sign: we give the reason for this prac-
tice in section 6 below.
The superiority of distinctive features over phonetic symbols in the formula-

tion of rules should now be apparent. Even for the limited data set we have
been considering, the use of phonetic symbols would entail four distinct nasal
assimilation rules, triggered by__ p,__ b, __ f and __ v, respectively: coronal
nasals also become labial in front of [f] and [v].
Once more, this four-rule outcome would come up against Occam’s razor.
Moreover, postulating four rules misses the generalization that their four con-
texts constitute one single natural class: the class of [labial] consonants.
Notice that we have not included [+sonorant] and several other feature
specifications in rule (7). Indeed, features are omitted from rules when they
are irrelevant to the process described in the rule (for instance, [—voice] is
irrelevant to the operation of (7)) or when they are predictable from the fea-
tures that are present (for example, we know from chapter 3 that [+sonorant]
is predictable from [+nasal]). We will return to this important matter in
chapter 17, simply bearing in mind for the moment that rules should be
Rules should be formulated in as economical a manner as possible.
formulated in
as economical
a manner as
possible The Feature “Coronal”. Active and Passive Articulators
We must now explain and justify the feature [coronal] that we included in
the nasal assimilation rule in (7).
We said at the time that [coronal] encodes alveolar place of articulation.
The feature This is true, but not the whole truth: it is more accurate to say that [coronal]
[coronal] refers
refers to a movement of the blade of the tongue. You will recall that the blade
to activity of the
blade of tongue
is the flexible portion at the front of the tongue that can be curled back or
stuck out unproblematically. The blade enjoys considerable mobility, and
therefore it can articulate in an area larger than just the tooth ridge.
Thus, for instance, the blade can go forward beyond the alveoli to produce
such a dental or interdental sound as [8] in thigh or thistle. It can also posi-
tion itself further back than the alveoli, as it does for the sounds [Jf] or [#]
of sherry and cherry, respectively. Now, as will become apparent later in the
chapter, all these sounds, while different, behave in a similar way with respect
to a number of phenomena. The assumption behind the distinctive feature
model is, of course, that similarity of behaviour follows directly from mem-
bership of a common class: each feature defines a class. The feature relevant
in the present case is “coronal”. The sounds [t], [d], [s], [z], [6], [5], [f1, (31,
[f{] and [dg], among others, are defined as [coronal], since they all involve a
gesture of the tongue blade.
Turning briefly to the terminology, the expression “coronal” is defined
as ‘pertaining to the blade of the tongue’. A more transparent label would
obviously have been “bladal”, except that this word does not exist in
Natural Classes of Sounds Dh
English. An alternative strategy to preserve terminological transparency would

have involved the reconstruction of the noun “crown” from the adjective
“coronal”, therefore talking about the “crown of the tongue”, rather than the
“blade of the tongue”. As it happens, however, phonologists have settled
for “blade” for the area of the tongue in question, and for “coronal” as its
related adjective.
A more substantial point in connection with the feature [coronal] concerns
the reference it makes to the active, rather than the passive, articulator: notice
that there are no features [alveolar], [dental], etc.
articulators.
The grounding of distinctive features in the active articulator is not arbitr-

ary. In particular, while both articulators are equally important phonetically
(obviously, no sound would be possible without the passive articulator), only
the active articulator is believed to be endowed with cognitive, or, more Only the active
articulator is
strongly, neural, substance. Using a computer analogy, phonetics is concerned
believed to
with the hardware, while phonology is concerned with the software: the pass- be endowed
ive articulator obviously falls outside the scope of the software, since it is with cognitive
motionless. substance
[4 Single-Value Features
We pointed out above that the features [labial] and [coronal] appear without
an algebraic operator: they are not given as [+labial], [+coronal], respectively.
What is the reason for this?
There is a crucial difference between features like [labial] or [coronal], on
the one hand, and [+voice] or [tnasal], on the other. Thus, as we said above,
a binary formalization +, — is ideal to capture a situation of complementar-
A binary formal-
ity: when one value is present, the other value must of necessity be absent.
isation +, — is
From this perspective, [voice] or [nasal] indeed are binary features, since ideal to capture
any particular sound will be voiced (= [+voice]) or voiceless (= [—voice]), a situation of
but not both or neither. The same remarks apply to [tnasal] and many other complementarity
features.
The obvious consequence of binarism is, of course, that the negation of
one value implies its opposite.
For instance, if we know that a segment is not [+nasal], we will automatic-

ally know that it is [-nasal] (= oral), and therefore that we must articulate
it with a raised velum. Consider now a feature such as [labial]. Knowing
that a given sound is not labial still does not tell us what it is: there are more
than two places of articulation, of which the feature [labial] only defines one.
Another such place of articulation is, of course, [coronal], the other feature
included in rule (7). A third place of articulation, [dorsal], defines sounds
articulated with the body of the tongue against the velum, or soft palate. The
rationale for the label “dorsal”, rather than “velar”, parallels the rationale for
the label “coronal”: “velar” refers to the passive articulator, while “dorsal”
(from the Latin dorsum ‘back’, here referring to the body of the tongue) denotes
the active articulator, that is, the articulator that moves. As we have already
said, only the active articulator is thought to have cognitive (and neural)
reality, hence its selection to underpin the distinctive features.
An important difference between one-valued features like [labial], [coronal]
or [dorsal] and such obviously binary features as [+voice] is that unary place
The unary place of articulation features can co-occur, since the gesture each such feature
of articulation
features [labial],
represents is not incompatible with the gesture represented by the others
[coronal] and — by contrast, the two values of a binary feature are by definition mutually
({dorsal] can exclusive. For instance, the English sound spelled w in wet or war is articu-
co-occur lated with narrowing both at the lips and at the velum, thus containing the
features [labial] and [dorsal].
More spectacularly perhaps for most of us, many West African languages have
stops that are not simply labial or simply dorsal, but rather a combination
of the two, that is, labiodorsal: [kp] and [gb], where the tie bar indicates this
unity. In Yoruba, a language of Nigeria, for instance, [akpa] ‘bridge’ con-
trasts with both [aka] ‘wheel’ and [apa] ‘lizard’, and [agbal] ‘jaw’ contrasts
with both [aga] ‘axe’ and [aba] ‘palm nut’
Have a g
-[gblah, e
difficult).
All the facts we have considered confirm that at least [labial], [coronal] and
[dorsal] are unary, that is, features with only one value, in this way differing
from their binary counterparts like [+voice]. Note that the unary approach to
place of articulation features is relatively recent: in the older formalism, asso-

ciated with the monumental work The Sound Pattern of English, authored by
Noam Chomsky and Morris Halle, all distinctive features were binary. In
fact, some recent trends attempt to extend the unary approach beyond the
set of place of articulation features (you can refer to chapters 8 and 17 for
more information).
Constraining Rules: Autosegmental Formalism
The assimilation rule in (7) above obviously yields the desired result: it replaces
the nasal coronal with a nasal labial before a labial consonant. There is a
very serious shortcoming in this formalism, however. In order to understand
this shortcoming, you must bear in mind that a rule like (7) is found not
only in English, but also in many of the world’s languages: there is clearly
something very natural about this type of process, which speakers are
extremely reluctant to suppress. Consider now the following formally pos-
sible rules:
(8) [coronal] —> [labial] / eee

+nasal
(9) [coronal] —> [labial] / [dorsal]

+nasal
The difference between these rules and the rule in (7) lies in the relation-
ship between the output and the context: in rule (7) the output feature is
identical to the contextual feature, but in (8) and (9) it is not. In (8) the nasal
becomes labial immediately before a coronal, and in (9) immediately before
a dorsal, and the sequences [mt], [md], and [mk], [mg], respectively, are
thereby created. The fact that rules of this kind do not exist, in English or
(almost for certain) in any other language, clearly cannot be coincidental.
Rather, the process expressed by the attested rule in (7) must be more nat-
ural than its logically possible but unattested counterparts in (8) and (9), which
are so unnatural as to fail to occur.
The reason is, of course, that the rule in (7) is a genuine assimilation rule, but
the rules in (8) and (9) are not. In particular, the rule in (7) brings the sub-
stance of the input segment (the coronal nasal) closer to the substance of the
contextual segment (the labial); indeed it makes it identical to it in place of
articulation. Clearly, nothing of the kind happens in rules (8) and (9).
The bottom line is that the natural relationship between the output of assim-
ilation rules and their context is not easily expressible in the rule formalism
we have been using up to now. This formalism is essentially that of the early
generative phonology literature, as compiled in The Sound Pattern of English,
“SPE”, to which we have already referred. Accordingly, we will now intro-
duce a more restricted and up-to-date formalism.
We shall proceed in small steps, to ensure that we leave no room for
uncertainty. First, you know that the segment that is input to the rule must
contain the features [coronal] and [+nasal], since only coronal nasals are
affected by this assimilation. In the type of theory propounded in SPE, this
information would be represented in a unified feature matrix, as in (10):
(10) et
+nasal
e.g. [n]
We obviously need other features for a complete description or definition

of [n], the incumbent segment, but we have already said that features that
are not relevant to rules or are predictable from the features present in the
rule are customarily left out, for the sake of simplicity.
The second ingredient in our assimilation rule is obviously the contextual
feature [labial], since the change only takes place when the nasal is immedi-
ately followed by a labial. Accordingly, we need to add the context [labial]
to our representation:
(11) |coronal] [labial]

+nasal
e.g. [n pl
We now have astring of two feature matrices, which define the sound
sequences [np] and [nb], among others.
Upon application of the rule, [labial] replaces [coronal] in the first matrix,
as follows:
(12) [| labial [labial]

+nasal
eg.[m p]
In the formalization in (7) above, the input and output feature matrices are
related by means of an arrow, indicating the transition. The equivalent form-
alization for (11) and (12) is as in (13) (rule (7) was of course further stream-
lined by factoring out the environment):
(13) Pe [labial] —> bec [labial]

+nasal +nasal
e.g. [n pl => 5[on pl
What we are now seeing is that this type of formalism is also consistent with
other, non-occurring processes, illustrated in rules (8) and (9) above.
Suppose therefore that we reformulate rule (7) as in (14):
(14) Labial assimilation in nasals:

[coronal] [labial] [coronal] [labial]
[+nasal] [+nasal]
e.g. [n pl SP [iat pl
All we have done, in effect, is break up the unified multifeature matrix and
grant functional (and, correspondingly, graphic) autonomy to each of its com-
ponent features, [coronal] and [+nasal] here. The association line linking these
two features indicates that they occur simultaneously in the input. In the
output, the association line has been transferred from [coronal] to [labial],
to encapsulate the assimilation process.
The simple innovation of breaking up multifeature matrices into as many

The simple innova- single-feature matrices as features they contain captures the essence of the
tion of breaking
AUTOSEGMENTAL approach to phonology, which grants autonomy of action
up multifeature
matrices into as
to each feature.
many single-
feature matrices
as features they
contain captures
the essence of
the AUTO-
The label “autosegmental” is a perhaps not quite felicitous blend of “auto-
SEGMENTAL
approach to
nomous” and “segmental”, where “segmental” should really be paraphrased
phonology as “featural”, since segments have been autonomous all along.
The advantage of writing the assimilation process in this novel way should
be clear. In particular, the autosegmental rule in (14) includes the following
information:
1 there is a sequence of two segments, the first containing the auto-

segmental features [coronal] and [+nasal], and the second containing
[labial];
2 the link between [coronal] and [+nasal] in the first segment gets severed:
notice the absence of an association line between these two features in
the rule’s output;
3 the [+nasal] feature associates instead to the adjacent [labial] autosegment.
The crucial difference between this autosegmental formalism and its pre-
decessor is that in the autosegmental formalism the assimilation process is
The assimilation interpreted as a simple change in the association of one feature: [+nasal]
process is is connected to [+coronal] in the input, and to the adjacent autosegment
autosegmentally
[labial] in the output. The result we were seeking is thus achieved in a
interpreted as a
simple change in way which expresses the assimilation process directly. By contrast, the two
the association of pseudoassimilation rules in (8) and (9) above simply cannot be formulated
one feature autosegmentally — or, if they could be formulated in such a way, their
unnaturalness would immediately be evident.
Ei Functional Groupings of Features
In its present formulation, the assimilation rule in (14) is confined to labials.

We shall now show that this restriction is unwarranted.
Consider the data in (15):
(15) a. inconceivable b. ingratitude

incorrect inglorious
incapable
These words contain our familiar in- prefix, spelled precisely in this Way.
Spelling is of course no sure guide to pronunciation, and therefore you should
not be unduly surprised to hear that in these forms the final nasal in the
prefix is not [n].
Indeed, it would be rather strange for a nasal to assimilate to a following

labial (/p/ in impossible and /b/ in imbalance), but not to a following velar.
Now, in inconceivable and ingratitude the segments following the nasal are
indeed velar (/k/ and /g/, respectively), and therefore we expect the nasal
to come out as [ny].
id you : getthis result

repeat the experiment trying rT
ee
Assimilation to a velar is, however, not predicted by the rule in (14), which
is specifically contextualized to labials. Therefore, in order to account for the
data in (15), we need an additional assimilation rule. We formulate this new
rule autosegmentally in (16), where we have collapsed the input and the out-
put into one single schema - specifically, the crossing out of the relevant
association line indicates dis(as)sociation, and the dotted line association:
(16) Dorsal assimilation in nasals:

[coronal] [dorsal]
i.‘in~ your ov7

Rule (16) says that a coronal nasal immediately followed by a dorsal conson-
ant loses its coronality and becomes dorsal (the feature [dorsal], as you know,
expresses velarity).
While the rule of nasal labialization and the rule of nasal dorsalization are
not identical, they are functionally similar, as they both implement assimila-
tion of nasals. Therefore, economy as well as intuition call for the reduction
of the two rules to one.
The reason that the assimilation process affects the unary features [labial]
and [dorsal] in the same way is, of course, that both these features express
place of articulation, the object of the assimilation in question. One simple way
of formalizing this type of functional unity involves adding a common mark
to all the features thus related, say, a subscript “p”, for “place of articulation”,
A subscript “,”, in the case we are discussing: [labial]p, [dorsal]p. This formalism allows us
for “place of
to refer to all places of articulation simultaneously simply by replacing the
articulation",
allows us to specific feature labels with a variable ranging over them: [X]p, or, perhaps
refer to all places more perspicuously, [ . . . ]p. A more complex, but essentially equivalent,
of articulation formalism will be presented in chapter 17.
simultaneously The common process of place assimilation in nasals can now be ex-
by the simple pressed straightforwardly, as in (17):
strategy of repla-
cing the specific
feature labels by (17) Place assimilation in nasals:
a variable ranging [coronal], [.. lp
over them, thus
[X]p, or, perhaps
more perspicu- [+nasal]
ously, [... ]>
The interpretation of this rule is as follows:
1 a [coronal, +nasal] consonant is followed by another consonant, of irrel-

evant place of articulation, as indicated by the emptiness of the second
P-subscripted matrix;
2 the autosegmental feature [+nasal] of the first consonant loses its asso-
ciation to the P-subscripted coronal;
3 [+nasal] reassociates to the place autosegment of the consonant following
[coronal]p, whatever the specification of such a place autosegment may
happen to be.
un
a lainhow (17) acco
A question that arises at this point is whether, in its present more general
formalization, rule (17) is not too unrestricted.
In particular, it is possible to interpret the variable “”...” of (17) as [coronal]p,

and thus to predict that coronal nasals become coronal, an apparent tautology:
(18) Coronalization of nasals:

[coronal], [coronal],
[nasal]
Even if it isn’t outright wrong, such an extension of the rule’s scope may
seem otiose and wasteful. Note importantly, however, that the application
of the rule in this context will at worst be vacuous: while it will have no
positive effects, it will do no harm either, since we saw in (2) above that — Vacuous applica-
lexical coronal nasals are indeed pronounced as coronal in front of coronals. tion ofa rule has
This means that we can leave things as they are at no cost: the rule in (17) oe PS
but does no harm
es aca "
is in any event needed in its present general formulation. either
Indeed, it would be more costly to attempt to exclude coronals from the con-
text of rule (17), in the same way as it takes a greater effort to prevent the
alarm clock from going off at the daily preset time on the occasion the sun
or a noise happen to awaken us earlier.
Actually, we will see in the next section that, contrary to initial appearances,
the rule in (18) has specific empirical consequences.
[J Feature Dependencies
Consider the forms in (19):
C19) ae tent b. tenth c. trench

hint plinth finch
punt month bunch
The nasal in the a. forms is alveolar, as would be expected from the fact that
the following segment is alveolar: /t/. The question is, are the ns in the forms
in the other two columns also alveolar?
Pronounce these fo
— alveol
It should be obvious that they are not alveolar. In the b. forms (tenth, etc.)
the passive articulator is the inside of the upper teeth, or the tooth edges,
depending on whether [6] is pronounced as a dental or as an interdental.
In turn, for the forms in c. (trench, etc.) the blade of the tongue is retracted
relative to the tooth edge, onto the palatoalveolar area where [Jf] and [tf] are
pronounced. All the consonants in question are, however, still [coronal],
because they all have the blade of the tongue as their active articulator.
Therefore, the differences between these various segments must hinge on
features other than [coronal].
One such feature is [/_DISTRIBUTED], which refers to the “distribution”
The feature of the tongue over the passive articulator. “Distribution” is perhaps not the
[+distributed]
most transparent of terms to refer to the length of tongue area involved in
refers to the
“distribution”
the articulation, but once more we will have to make do with standard usage.
of the tongue The substantive point is that the noted versatility of the tongue blade makes
over the passive it possible for a substantial portion of it to be engaged in the articulation:
articulator [+distributed]. Alternatively, the portion of the blade carrying out the con-
striction may be minimal, essentially the tip: [—distributed]. With regard to the
precise part of the tongue involved, the contrast corresponds to a LAMINAL
vs. an APICAL gesture, that is, a gesture with the full blade of the tongue vs.
a gesture with just the tip. The binary contrast [+distributed] vs. [-distributed]
thus captures the difference in place of articulation between dental [6] or
[t], on the one hand, and alveolar [t], on the other.
[+distributed]
comes by The crucial point in the present context is that the feature [+distributed]
definition autoseg-
comes by definition autosegmentally attached to [coronal]: only coronal sounds
mentally attached
to [coronal], of can be [+distributed], and therefore the feature is only relevant to coro-
which it consti- nals, at least on our current conception of distinctive features — in older
tutes a subdivision models, SPE for instance, the scope of [tdistributed] included other places
of articulation.
The geometric consequence of the dependency of [+distributed] on [coronal]

is that of the structures in (20) below; only a. is well formed. To make the
dependency relation visually clear, we will systematically misalign depend-
ents from their superordinates, and reduce the font size of their labels:
(20) Yes No No
a. [+nasal] b. [coronal], c. [+nasal]
[coronal], [+nasal] [coronal],
[+distributed] [+distributed] [+distributed]
If [+distributed] must by definition be a dependent of [coronal], the fact that

in b. and c. it is not makes these configurations illegitimate.
The graphic misalignment of the dependent feature is also aimed at sug-

gesting that it occupies a different plane from that of its superordinate. Indeed,
phonological representations are multiplanar, not just multitiered. In (20a),
the nasal and coronal tiers share the same plane, but [+distributed] occupies _ Phonological rep-
a different plane, which branches off the [coronal] tier, in a manner remin- thee we
iscent of the leaf of a drop-leaf table.
The multiplanar mode of representation has far-reaching repercussions for

the theory of phonology, as we will have plenty of occasion to observe.
Let us now explore the consequences for coronals of our rule of place assim-
ilation in nasals in (17) above. We have suggested that the application of
this rule to coronals is not vacuous, contrary to first appearances. Consider,

for instance, the effects of the rule on the relevant string in te/n@/ (we are
of course assuming at this point that [+nasal] and [coronal], occupy the same
plane, since neither is a dependent of the other):
(21) Distributed assimilation in nasals:
[+nasal]
[distributed] [+distributed]
Although the rule has no visible effect at the [coronal], level (NB both the
input and the output segments are coronal), it does have consequences at the
level of the dependent [+distributed], which is automatically dragged along
by [coronal]. This result matches the facts in (19b) above, thus providing
strong backup for our approach, which includes the general assimilation rule
(17) and the built-in dependency of [tdistributed] on [coronallp.
The argument carries over to the set of data in (19c) above (trench, etc.).
Here the relevant feature is [tanterior]. This feature divides the hard palate
into two regions, the forward or anterior region, with the alveolar and dental
areas, and the posterior region, which includes the palatoalveolar and palatal
areas (figure 4.1).
Posterior
Anterior
Figure 4.1 The anterior and posterior areas
“Mention “ couple of [anterio r]segment sand a. of

1a couple¢
‘Eanterior] Se : : -.
The feature [+anterior] is the only distinctive feature that refers to the pass-
ive articulator — indirectly, however, it still refers to the active articulator,
which must retract or otherwise to meet the passive articulator.
The feature [tanterior] was first proposed in SPE. Like [+distributed], its §=—-————
scope used to extend over all places of articulation, but it is now also gener- _ [anterior] is re-
ally restricted to coronal sounds. Formally, therefore, [tanterior] is a further __‘tricted to coronal
dependent of [coronal]:
aS ae
further dependent
(22) [coronal], of [coronal]
[+distributed]
[tanterior]
The assimilation observed in tre/ni{/ will consequently be expressed as

in (23):
(23) Anterior assimilation in nasals:
[+nasal]
[+anterior] [-anterior]
As is the case with [+distributed] in (21) above, linking to the second [cor-
onal], has a side effect on the dependent, here the replacement of [+anterior]
with [—anterior]. Therefore, the process is not vacuous.
We finish the chapter with the tabulation of all the distinctive features we
have introduced, specifying their values in the segments we are familiar with
(we have simply ticked unary features where present). We have arranged
the segments by place of articulation from labial to velar, indenting depen-
dent features under their superordinate (table 4.1).
Table 4.1 Distinctive features
po to dk. ee UO Gea ee en
Sonorant target nt nl mm at nacre

Continuant a hen a = Se Se DG Se te ge ear
Voice Sep ee pe Send a Se ee ee ct ee eo
Nasal Se ee ee ee er ee
Lateral ee ee ee ee
Labial, Slo 4 vo
Coronal, Yo SONS S SV eee
Anterior +o + Lh YEO Io Gt Eee
Distributed a OG HOOD Rpt aet
Dorsal, Sis ory
fiovds vaneen= qt cpr). A oies Dove Cone ee
Sonorant S rosyedt Se ocit w. Shoe Sactstig <OSSF TPiesat fie Pe

Continuant ee a ee ee ee MR ee SO
Voice ely Te OF ewe ea cths ane e atta tees ec chee haere tr
Nasal ha ta aoe
Lateral eH HH a aS aS
Labial, J vy
Coronal, 4 Vv v v v Yo Vv
Anterior - - + mee Ei cee ORE
Distributed + + — = = Soha
Dorsal, A v Jf
Palatals and velars are undifferentiated in this table: compare the values
of [pn] and [n]. The features which distinguish palatal from velar sounds
will be introduced in chapter 6, with additional discussion in chapter 17.
Notice also that we have purposely left the glottal stop [?] and the glottal
fricatives [h] and [h] out of the place of articulation count: in phonology,
“place of articulation” specifically refers to place of articulation in the oral
cavity.
nitive units Seis to theactive arti

that tthepepanehs BARRE used
ina ace.
thal
nn a c
complementary pair of labels (such a
Key Questions
What is the relationship between Why do some distinctive features

distinctive features and the phon- only have a single value, as opposed
etic parameters discussed in earlier to a binary value?
chapters? Define the term “coronal”.
Explain “binarity” with respect to What do we mean by “autoseg-
distinctive features. mental formalism"? Describe the
In what way is it more revealing to advantage of this type of formalism
break sounds down into thei com- over that involving feature matrices.
ponent features than to represent them Why is it more revealing?
in terms of their phonetic symbol? What is the advantage of the sub-
The place of articulation of a sound script system in the grouping of
is the point at which the active and features?
passive articulator come into contact. In what respect are some features
Why do we refer only to the active dependent on others? What does
articulator when defining place of a setting for the feature [+anterior]
articulation in phonology? necessarily imply?
Further Practice
Natural Classes
a. In each of the groups (i-vi) below there is one odd member, the rest
belonging to a natural class which can be identified by means of one or more
common feature(s). Identify the odd one out and say which feature(s) is
or are common to the remainder. (There may be more than one possible
answer in some cases.)
[v, n, m, v, B]
ll. [9,f7t, s/¢]
ili. [y, 3 0) -v 7B]
IV. [n, 1,4, d,.n]
ides 2) £0)
vi. x, y, 0,k,pl
Consider the following putative processes:
/m, b, p/ => ind)

/n, d, t/ => ins ki
(3,2) => Ars]
iv. /b,d,g/ [B, 0, yJ

V. /®, 0, f, S, S/ [B, é, V, Z, 3]
a /lyn/ {l, nJ
vii. /b,d,d,}4,8/ [m,n, n,n, 9]
viii. /g,x,k, y/ LtLitddt
[b, ®, Pp, B]
(i) | State the common features of the groups involved.

(ii) |Which feature changes are represented by the arrows?
(iii) Write the rules out using distinctive features and autosegmental
representations.
Selayarese Reduplication
Reduplication is a process in which all or part of the phonological material

of the base is repeated. In the case of Selayarese, an Austronesian language
spoken in Indonesia, a reduplicated form conveys the meaning of what might
roughly be translated as ‘sort of’:
Basic form Reduplicated form

[pekan] ‘hook’ [pekampekan] ‘hook-like object’
[tunruy] ‘hit’ [tunruntunrun] ‘hit lightly’
[kelon] ‘sing’ [kelonkelon] ‘sort of sing’
[janan] ‘chicken’ [jananjanan] ‘bird’
[hukkun] ‘punish’ [hukkuynhukkun] ‘punish lightly’
[mannan] ‘tired’ [mannammanyay] ‘sort of tired’
[gintan] = ‘chili’ [gintangintan] ‘chili-like object’
[rongan] ‘loose’ [ronganrongan] ‘rather loose’
(i) Is the reduplicated form identical to the base form in all these data?
(ii) Why?
(ii) | Write a rule accounting for the facts.
(iii) Give the reduplicated forms for the following:
[dodon] ‘sick’
[nungayn] ‘hit’
[bamban] ‘hot’
[soron] ‘push’
VOWEL SOUNDS
CARDINAL VOWELS
In chapters 1 and 3 we deliberately relied on English for the presentation

of the articulation of consonants, introducing them where possible with the
help of one or more English words containing the sound in question. The
rationale behind this strategy is twofold. First, it is our belief that real under-
standing of each sound can only be achieved by relating abstract descrip-
tion to intuition and experience, hence our selection of English as the main
exemplificatory language. Second, the basic mechanics of consonant articu-
lation are readily accessible through guided self-exploration. We trust that
our goal has been met, and that by now you feel reasonably in command
of the panoply of consonants we have been discussing. Unfortunately, the pres-
entation of vowels cannot follow the same pattern, for the simple reason that
the mechanics of vowel articulation are far less accessible to observa-
tion. In order to see why this is so, we must first gain some understanding
of the differences between vowel and consonant sounds.
Vowel Sounds 115
On What Vowels Are and How They Are Made
The key articulatory difference between vowels and consonants resides in

the fate of the airflow coming out of the lungs as it passes through the mouth.
In consonantal sounds the airstream finds a radical constriction or even total
blockage at some point along the central passage in the oral cavity. By con-
trast, when a vowel sound is pronounced, no such obstacle is present.
Not surprisingly, a musical analogy will help us understand the nature When a vowel
of vowels. At the beginning of chapter 1 we commented on the fact that if sound is pro-
nounced, no
you blow air through a recorder without covering up some of the holes in
obstacle is present
turn, there will be no melody: just one single, invariant note. We now want
to focus on a different aspect of the music. In particular, whatever note we
play, the sound quality of a recorder is readily distinguishable from the sound
quality of a trumpet or a clarinet. You may think that the reason for this is
simply that the recorder is a different instrument. This is tautologically true,
of course. Consider, however, the fact that recorders come in different sizes:
recorder fans will be able to name the sopranino, descant, treble, tenor, etc.
These different types of recorder are all the “same” instrument, and yet
they also produce different sound qualities, according to recorder size. This
clearly suggests that the instrument’s size (that is, not just its shape, mater-
ial, etc.) plays a crucial role in the determination of the instrument’s sound
quality. The sound quality of the instrument is of course crucially different
from the notes we play on it: all notes can be played on all instruments,
although they still sound different.
From now on, we shall simply assume the veracity of our finding that sound
quality is a function of the size of the instrument — this is indeed a fact of Sound quality is a
function of the
physical life, the technical reasons for which do not concern us here. Now,
size of the instru-
in a recorder concert recorders of different sizes will play at different time ment
intervals, giving the effect of a symphony of recorder sound qualities. Sup-
pose, however, that we only have one recorder to play the symphony with:
how can we produce the same variety of sound quality? The answer is, obvi-
ously, that we cannot if our recorder is made of rigid wood or plastic, as
recorders conventionally are. Imagine, though, that new technologies allow
our single recorder to be flexible, so that we can vary its size by stretching
or compressing it as we play along, as effectively happens with electronic
gadgets. In this case, we will indeed be able to play the symphony with just
the one recorder — at least if only one of the different-sized recorders is required
to play at a time: in real-life symphonies, several different-sized recorders
can of course find themselves playing simultaneously. Vowels differ from
How does all this relate to vowels? Each vowel sound is like the sound each other only in
made by one of the different-sized recorders of the analogy. Specifically, sound quality
vowels differ from each other only in sound quality, in the same way as
116 Vowel Sounds
the sounds given out by the different-sized recorders. In the case of

vowels, however, we clearly have only one instrument to make them with:
the mouth. The question therefore is: how can we vary the size of the mouth
to produce the different vowels present in language? After all, the mouth
is made of rigid materials, namely, the different bones that give it its basic
shape and structure, and therefore you are likely to wonder how the vari-
ation in the size of the mouth that we need to produce different vowels can
be attained.
If you think carefully about the anatomy of the mouth, you will realize that
not all of it is rigid. For instance, the mouth can be opened and closed. More
to the point here, we can and do form a tube inside the mouth, with the
roof of the mouth, the position of which, of course, we cannot alter, and the
tongue, which we can move. By drawing the tongue closer or less close to
By drawing the the roof of the mouth, we can vary the size of this tube. As you just learnt
tongue closer or from our discussion on recorders, each such tube size will produce a dif-
less close to the
roof of the
ferent sound quality, hence a different vowel.
mouth, we can You can verify this prediction on yourself. Suppose you utter the sound
vary the size of ah, as when asked to do so by the doctor.
the tube. Each
such tube size will
produce a differ- fe
ent sound quality,
hence a different
vowel
Clearly, ah is a real sound of language, similar in fact to the sound spelled

a in such English words as spa or father in most accents (see the caveat on
p. 119). How is it made? You can guess the answer from the function of the
doctor’s request. Obviously, the doctor wants to look down your throat, and
consequently needs your mouth to be as open as possible, with the tongue
at its lowest. What is interesting is that, instead of instructing you to carry
out these actions, the doctor asks you to say ah, a sound with which you
are familiar since an early age and which you can therefore reproduce auto-
matically. It follows logically that the pronunciation of this sound must involve
an open mouth and a tongue away from the roof, so as not to interfere with
the doctor’s field of vision down the throat.
We now have one vowel in our repertoire, just like one recorder in the
range of recorders. Remember that there is no constriction of any kind in
the mouth when we say ah. Let us contrast this sound with another one which
Vowel Sounds 117
is, in many ways, its polar opposite. In particular, push the body of your
tongue as high and forward as you possibly can without making contact or
creating a constriction with the roof of the mouth — if you do block the pas-
sage or create a constriction, the sound will of course not be a vowel, but a
consonant.
When you feel that you have got the tongue to the desired position, let out
the air: the sound that will come out of your mouth will be similar to the
vowel in the English words bee or sea. This sound thus constitutes a second
vowel, which we can add to our list. Again, it is important to notice that
the air flows out of the mouth unimpeded, precisely as is characteristic of
vowels. The reason this second vowel sounds different from the first one is,
of course, that the size of the tube in which it is produced is different. In
fact, it is considerably different: the doctor would have quite a job examin-
ing your throat while you pronounce the vowel ee.
The Two Basic Cardinal Vowels
The number of vowels in the world’s languages can vary dramatically, from
one or two to ten or twenty. We said above that our awareness of exactly
how each vowel is pronounced is less than with consonants. You now know
why this is so. In consonants we can find out easily where we are making
the constriction or blockage. The pronunciation of vowels, however, involves
creating a makeshift tube with the tongue and the roof of the mouth, and The pronunci-
ation of vowels
our awareness of such a tube and its properties is usually very poor, un-
involves creating
fortunately for the student of phonetics. In addition, the tube is well inside a makeshift tube
the mouth, and therefore we cannot observe it directly, even if we enlist with the tongue
the aid of a mirror. We could of course take still X-rays, or even motion and the roof of
pictures, as we articulate the vowels, and indeed these and other such tech- the mouth
niques have led to important advances in our understanding of how vowels
are produced. Clearly, though, this technology is well beyond the reach of
the average reader — and must anyway, for medical reasons, be used with
great care, under proper supervision only.
118 Vowel Sounds
Naturally, phoneticians have been aware of the problem for a long time.
In the early 1900s, the English phonetician Daniel Jones developed a chart
on which to plot vowels in a way reminiscent of the cardinal points in nature.
For the cardinal points we look for the point in the horizon where the sun
rises, and we define such apoint as the east. If we now face east (which we
have just established), we automatically define west, south and north, as the
points behind our back, to our right and to our left, respectively. Once we
have these four cardinal points, we can define any intermediate points
(south-west, north-east, etc.) simply by dividing the space accordingly.
We will now do something similar with vowels. We already have two
CARDINAL VOWELS: ah and ee. As we have explained, ee, roughly as in
bee or sea, is articulated with the body of the tongue as high and forward as
is compatible with a vowel sound — in particular, there must be no blockage
or constriction of the airstream. If you observe yourself in a mirror as you
do this, you will notice that your lips spread automatically. The articulatory
instructions for the ah of father or spa, or at the doctor’s surgery, are the oppos-
ite: the tongue is pulled back and lowered, with the concomitant lowering
of the jaw (unless the speaker happens to be chewing a pipe!).
If you observe yourself in a mirror as you articulate ah, you will notice
that the lips are in a neutral position: neither spread nor rounded. At the
moment, such lip movements are quite automatic: our articulatory instruc-
tions concern exclusively the tongue. However, it will become clear later that
lip position also plays an important role in the articulation of vowels.
The interest of the two vowels we have just considered is that they can
be produced simply by following the given articulatory instructions, in a
manner similar to what we saw in previous chapters is the case with con-
sonants. In this respect, these two vowels are exceptional, as we will see.
Their usefulness lies in the fact that they provide us with the two basic car-
dinal points to use as a reference to define the rest of the vowels.
The IPA symbols for our two current cardinal vowels are [a], for ah, and
[i], for ee. We have been saying all along that these vowels are similar to the
Cardinal vowels vowels in the English words father or spa, and bee or sea, respectively.
are by definition
However, although they are similar, they are not identical, for two reasons.
artificial, made by
following a set of First, the cardinal vowels are by definition artificial, made by following a
articulatory set of articulatory instructions, and there is no guarantee that the resulting
instructions articulatory gesture will be picked up literally by any natural language. Second,
we will see below and in successive chapters that vowels have a tendency to
Vowel Sounds 119
move about in the articulatory space much more than consonants. There-
fore, vowels are likely to acquire their own nuances and idiosyncrasies in each
language, or even in each dialect or idiolect (the word IDIOLECT refers to
the specific manner of speaking of one speaker). By contrast, the cardinal
vowels are idealized reference points, analogous to the metre standard kept
in the Pavillon de Breteuil in Paris: the metre standard is stored under ideal
conditions of temperature, humidity, etc., quite unlike the circumstances found
in the outside world, and therefore no other metre stick will have exactly
the same length as the standard, although it will of course be equivalent to
it for all practical purposes. So, when we talk about the cardinal vowels [a]
and [i] henceforth, you must bear in mind that, although they are similar to When we talk
the vowels in father and bee, they are bound not to be absolutely identical, about specific car-
dinal vowels, you
and likewise for the other cardinal vowels we will introduce.
must bear in mind
that they are
bound not to be
absolutely identical
to the vowels of
any language
The Four Corner Primary Cardinal Vowels:

Two Axial Parameters
Before we proceed, we must issue an important caveat. The vowel system

of English is remarkably complex in two ways. First, English has a much The vowel system
of English is re-
larger number of vowels than is usually the norm: something in the region
markably complex
of twenty. Second, as English speakers will be well aware, there is striking
variation in the way these vowels are pronounced throughout the English-
speaking world, and these pronunciations are still changing. The difficult-
ies in illustrating the pronunciation of the cardinal vowels with English are
therefore considerable, and, like other writers, we are bound to rely on the
best-known or “standard” accents, namely, RP for British English and GA
for North American English. The term RP (“Received Pronunciation”)
refers to a socially, rather than geographically, determined accent, traditionally
associated with the English upper middle classes and their “preparatory”
and “public” schools. This accent has, also traditionally, been favoured by
the national media, hence the alternative expression “BBC English”. The
expression “General American” (GA) refers to a pronunciation of English
common in North America, characterized by lacking any obvious regional
traits - “Network English” is an alternative label. Hopefully, the wide-
spread access to these standard accents in the respective communities will
120 Vowel Sounds
help most readers grasp without much trouble the points we are end-
eavouring to make. We will also appeal to more localized accents when
required by the facts being discussed — we will indeed find Scottish English
particularly useful to identify some of the primary cardinal vowels. In the
final analysis, however, readers need to refer our descriptions to their own
specific accents if true understanding is to emerge.
We saw above that, once we have the four cardinal points, we can define
an infinite number of intermediate points in the intervening space. We will
now show that something similar happens with cardinal vowels.
The cardinal points are of course plotted in an open space. The space avail-
able to vowels is, however, confined to the volume delimited by the roof
of the mouth and the (moving) tongue. Geometrically, the vertical section
of this space can be construed as a trapezoid, as we represent schematic-
ally in figure 5.1. Note that, by convention, the vowel space systematically
faces left.
Figure 5.1 The vowel space
The space thus shaped is imposed by the anatomical structures of the

mouth and the tongue. Such space limits the scope of the tongue’s move-
ments inside the mouth, and therefore it defines the maximum size of the
tube responsible for the vowel sounds. Smaller tube sizes will be formed by
Vowel Sounds 121
the appropriate tongue movements inside the space, aS we now represent

in figure 5.2, again schematically (the strings of arrows are intended to sug-
gest the airflow).
Figure 5.2 Tongue positions and tube sizes
In figure 5.2 it is visually apparent that different positionings of the tongue

result in tubes of different sizes and shapes. Each such size and shape will Vowel quality is a
function of the
of course produce a different vowel sound, as we know from our earlier
size of the tube
discussion — you can usefully refer back to the analogy of the range of recorders through which the
in this connection. airstream exits the
In the same way as the set of cardinal points conventionally contains mouth
four primary points, the set of PRIMARY CARDINAL VOWELS is con-
ventionally made up of eight vowels. We already have two primary cardinal
vowels. in our repertoire, which we will now plot in the vowel space
(figure 5.3). The set of PRI-
MARY CARDINAL
VOWELS is con-
ventionally made
up of eight vowels
Figure 5.3 Points of the two basic cardinal vowels
The two blobs in figure 5.3 stand for the two points where our first two car-
dinal vowels are plotted. Clearly, given the convention that the vowel space
faces left, the blob in the top left-hand corner identifies the position where
[i] is articulated, and the blob in the low corner on the right of the figure
the position for [a] (figure 5.4).
122 Vowel Sounds
Figure 5.4 The two basic cardinal vowels
Having plotted these two vowels in the top left and the bottom right
angles, respectively, two other vowels naturally suggest themselves in the
two remaining corners. The IPA symbols for these vowels are [a] and [ul]
(we will, however, propose another symbol for [a] in section 6 below). See
figure 5.5.
a a
Figure 5.5 The four corner cardinal vowels
You can obtain these two vowels articulatorily from the two more basic
vowels by sliding the tongue forwards, for [a], or backwards, for [u], while
maintaining its height constant in each case.
The vowel [u] is instantiated (again, approximately) in the words shoe and
clue. Its articulation thus involves simultaneous retraction and raising of the
tongue. Additionally, the lips purse out rather heavily: the technical term
for this lip gesture is “rounding”.
Vowel Sounds 123
For [a], by contrast, the tongue needs to go down (as it does for [a]) and
come forward (in contrast to [a]). A sound similar to cardinal vowel [a] is
found in such words as heart in eastern New England, and also in hat and
tap in some accents of Northern England, in Yorkshire for instance. For [a] oe
as for [i] and [a], there is no lip rounding.
. onecann practi bystartingwith lal an pushing¢the
A glance at figure 5.5 above will reveal that the two basic parameters for
vowels are height, on the vertical axis, and backness-frontness, on the The two basic
horizontal axis. According to these parameters, [i] and [u] are both high parameters
defining vowels
vowels, whereas [a] and [a] are low vowels. Simultaneously, [i] and [a] are
are height, on
front vowels, and [u] and [a] back vowels. We summarize this distribution the vertical axis,
in the table in (1): and backness—
frontness, on the
horizontal axis
(1) High Back

[i] + ae
Pade age =
Foal Ec! os
[u] - ~
EJ Four Perceptually Intermediate Primary Cardinal Vowels:

The Roundness Parameter Once we have [i]
and [a] at the
front, we can pro-
nounce a vowel
We said above that there are eight cardinal vowels, and we must now describe which sounds
the remaining four. Importantly, no particular articulatory instructions can one-third of the
be issued for these four vowels, which instead fill in the perceptual space way closer to [i]
than to [a], and
at regular intervals at the front and back. The idea is that, once we have [i]
another vowel
and [a] at the front, we can pronounce a vowel which sounds one-third of which sounds
the way closer to [i] than to [a], and another vowel which sounds one-third one-third closer to
closer to [a] than to [i]. These two vowels are represented in the IPA alpha- [a] than to [i]
bet as [e] and [e], respectively (figure 5.6).
124 Vowel Sounds
a
Figure 5.6 The front cardinal vowels
In order to pronounce both [e] and [e], we must position our tongue some-
where in between [i] and [a]. The size of the tube must therefore increase
along the scale [i], [el, [e], [a] - exactly how this is achieved is too complex
a matter to go into here: what is relevant to us is that [e] sounds half-way
between [i] and [e], and [e] sounds half-way between [e] and [a].
The cardinal vowels [e] and [e] approximate to the Scottish English vowels
in words like late or raid, and let or red, respectively. In more common English,
the vowel in late is diphthongal, as we shall explain in chapter 7, while the
vowel in let is higher than [e]. The cardinal vowels [e] and [e] also occur in
more or less their pure form in French (bébé [bebe] ‘baby’; béte [bet] ‘beast’)
and in German (See [ze] ‘lake’; Bett [bet] ‘bed’).
The situation we have described carries over to the back vowels [u], [a]. In
The perceptual particular, the perceptual space between these vowels can be filled in at equal
space between the intervals with the vowels [o], one-third close to [u], and [5], one-third close
back vowels [ul], :
[al can be filled in © Lal (figure 5.7).
at equal intervals
with the vowels u
[o], one-third
close to [u], and
O
[9], one-third close
to [a]
oO
a
Figure 5.7 The back cardinal vowels
Vowel Sounds 125
Lip rounding is restricted to the primary cardinal vowels of the back set,
and increases with height. We already know that [a] is unround, whereas
[u] is round. The intermediate vowels [o] and [5] are also round, [o] more
heavily so than [5], and less so than the high vowel [u]. Lip rounding makes
up for the shallowness of the back area of the mouth, as compared with the
area at the front. Thus, you know by now that vowel quality is a function
of the size of the tube through which the airstream exits the mouth. The lips
are pursed to varying degrees to compensate for the reduced volume at the
back of the mouth, so as to achieve differences in tube size equivalent to
those in the front vowels.
Our two new vowels [o] and [5] again occur in Scottish English, respect-
ively in coat or road, and cot or rod. They also occur in French (beau [bo]
‘beautiful’; botte [bot] ‘boot’) and in German (Sohn [zon] ‘son’; sonst [zonst]
‘otherwise’).
We now summarize the complete set of primary cardinal vowels, plotted

on the vowel space. To highlight their a priori nature and make reference
easier, each cardinal vowel is standardly provided with a number, which
we include in figure 5.8.
ie u 8
ore a6
4a Cao
Figure 5.8 Set of primary cardinal vowels (numbered)
As we have already explained, the two main vowel parameters are height
and frontness—backness. The vowels [i], [e], [e] and [a] are front, while [ul],
[ol, [>] and [a] are back. With regard to height, [i] and [u] are high, while [a] A ROUNDNESS
and [a] are low. The intermediate vowels [el], [o], [e] and [5] are considered mid: parameter has
[e] and [o] mid-high, and [e] and [5] mid-low. An additional ROUNDNESS been implicit in
parameter has been implicit in the discussion. With regard to this third para- the discussion
meter, [ul], [o] and [9] are round, and all the other vowels unround.
126 Vowel Sounds
Cardinal Vowels and Real-World Vowels: Diacritic Symbols
We have been pointing out that the cardinal vowels are idealized vowel sounds
The cardinal evenly distributed around the vowel space. The cardinal vowels do not there-
vowels are ideal-
fore necessarily correspond to the real vowels of any natural language.
ized vowels
We have, of course, offered some approximate illustrations to help readers
understand the sound quality of each such vowel, but none of these strictly
corresponds to the pristine cardinal vowels — recordings of these vowels by
Daniel Jones, and others after him, are available commercially.
This said, some vowels in some languages approximate more to the car-
dinal vowels than some vowels in other languages. Unfortunately for most
of us, the common English vowels are far from being good representatives
of the cardinal vowels, even in cases where there is a reasonable corre-
spondence: we shall see this in detail when we carry out a survey of English
vowels in chapter 7. For instance, we have provided words like sea and shoe
to illustrate the high cardinal vowels [i] and [u], respectively. Two similar-
sounding words exist in French, although their meanings are quite differ-
ent: si ‘yes’ and chou ‘cabbage’. In IPA symbols, both English sea and French
si will be transcribed [si], and both English shoe and French chou will be
transcribed [fu]. In spite of this, there is a perceptible shade of difference
between the pronunciation of the two English words and the pronunciation
of their French correlates, which a speaker of one language will readily
attribute to the presence of a “foreign” accent in the speaker of the other
language. One of the reasons the two high vowels sound slightly different
in the two languages is precisely that the French vowels are closer to the
corresponding cardinal vowels than are the English vowels, which deviate
from the cardinal specimens in several ways we will examine in chapter 7.
Of course, similar considerations apply to the differences between English
accents the world over. These differences are in fact far from trivial, in
particular with respect to vowels, again as we will see in some detail in
chapter 7.
The cardinal vowels are therefore equivalent to the straight lines we
draw in real life, which can never be absolutely straight — the gap between
conceptual objects and real-world objects is well known to philosophers, and
is quite beyond dispute. Naturally, this caveat also applies to the other sounds
represented in the IPA alphabet, except that for consonants no claim has
ever been made of absolute uniformity across languages: [s], for instance, is
defined as a voiceless alveolar fricative, and so any voiceless alveolar fricat-
Vowel Sounds 127
ive can and must be transcribed as [s], whatever additional nuances it

may possess. In the case of the cardinal vowels, however, each vowel is
defined to a level of precision that leaves no room for latitude, and there-
fore most, if not all, real-world vowels will technically be at variance with
the cardinal vowels.
The IPA alphabet does provide a set of diacritic symbols to express at
least some of these deviations. Such diacritics are added to the basic vowel The IPA alphabet
symbol to indicate the nature of the divergence. For instance, greater provides a set of
diacritic symbols
tongue lowering is indicated by underscripting “.” to the vowel symbol,
to express subtle
and a small degree of increased tongue height by underscripting “,”. deviations
Similarly, a slightly greater backing is represented by underscripting “”,
and a slightly greater fronting by underscripting “,”. We illustrate these
symbols in (2):
(2) .& 29
While use of diacritics enhances the precision of the IPA alphabet, it also
quite clearly encumbers the transcription. Consequently, phoneticians tend
to be sparing in the use of such symbols, reserving them for cases where
confusion may arise between two otherwise similar sounds: in all other
cases the IPA symbol is used with no accessories. For instance, the English
vowels in sea and shoe will normally appear transcribed as [i], [ul], exactly
as the vowels in the French words si and chou, despite the fact that there
is a perceptible difference between them - as we said, the French vowels
are reasonably close to the cardinal specimens 1 and 8; the common English
vowels are, among other things, slightly lower.
[al Some Vowel Typology: The Basic Vowel Triangle
We have seen that the set of primary cardinal vowels is made up of eight
vowels. Four of these are basic, two of which are fundamental to the pro-
cedure, as they can be defined on an articulatory basis quite precisely.
128 Vowel Sounds
Not surprisingly, not all eight primary cardinal vowels show up in all the
world’s languages, whether in their pure form or even approximately. A vowel
system with one or two members is obviously not very functional. Some
North Caucasian languages are claimed to have only two vowels, although
these possess numerous allophones (we have talked about allophones and
phonemes in chapter 2). The vowel system commonly considered basic to
language in fact contains three vowels, the two high vowels [i], [u] and a
central low vowel articulated between [a] and [a], that is, between cardinal
vowels nos. 4 and 5, in fact the common Scottish pronunciation of the vowel
in hat. Quite unexpectedly and most inconveniently, there is no special IPA
symbol for this vowel, which is usually transcribed with the same symbol
as cardinal vowel no. 4, that is, [a]. This practice can only induce confusion
in the reader, who will need extensive textual support to be sure whether
the vowel referred to at any particular point is the front vowel or the
central vowel. This vagueness runs against the very heart of the philosophy
of the IPA. In order to keep to the standard of formal precision we are pur-
posely adopting for this book, we shall reluctantly take the bold step of depart-
We shall reluc- ing from IPA doctrine and substitute [ze] (a symbol known as “the ash”) for
tantly take the [a] to designate cardinal vowel no. 4, the vowel of Yorkshire hat or eastern
bold step of
departing from
New England heart. This obviously frees the symbol [a] for the central low
IPA doctrine and vowel, the vowel of Scottish hat. We wish to emphasize that we are not mak-
substitute [ze] for ing this move light-heartedly, but on balance it seems to us that grasping
[a] to designate this particular nettle will in the long run be more beneficial than trying to
cardinal vowel bury our heads in the sand. We return to this matter in chapter 7 in con-
no. 4
nection with the pronunciation of the English vowels.
The three basic vowels [i], [ul], [a] pattern in the manner displayed in (3):
(Besa u
Notice that these three vowels are as far apart in the space as any vowels
The most common can be. With regard to the vowel parameters, [i] and [u] are high, and [al]
vowel sys-
low; [i] is.front, [u] back and [a] central; and [i] has spread lips, [u] rounded
tem world-wide is
the five-member lips and [a] neutral lips.
vowel triangle Systems with more than three vowels simply tend to add other vowels
i, €, a, O, U to this basic three-vowel system. Indeed, the most common vowel system
world-wide is the five-member vowel triangle schematized in (4):
Vowel Sounds 129
(4) i u
This system obviously results from the enrichment of the basic vowel tri-
angle with the intermediate vowels [e] and [o]. Note that in these systems
[e] and [o] are genuinely half-way between [i]-[a] and [u]-[a], respectively,
and therefore the symbols [e] and [o] are strictly speaking misleading.
The reason for this lax use of phonetic symbols has to do with typograph-
ical economy, always an important consideration in transcription practice,
sometimes perhaps excessively so. The practical motivation that inspired the
IPA alphabet historically also often lends transcription practice a phonemic
orientation that is not always consistent with the goal of phonetic faithful-
ness. Phonetically oriented transcription is known as NARROW TRANSCRIP-
TION, and phonemically oriented transcription as BROAD TRANSCRIPTION. —Phonetically
oriented transcrip-
tion is known
as NARROW
TRANSCRIPTION,
and phonemically
oriented transcrip-
tion as BROAD
TRANSCRIPTION
Note that the adoption of [e] and [o], rather than [e] and [5], as representat-
ives of the intermediate sounds is also down to typographical considera-
tions: the characters “e” and “o”, but not their counterparts, are part of the
Roman alphabet, and thus included in all standard typing and printing sets.
Quantum Vowels
An interesting question arises at this point concerning the basic three- and
five-member vowel sets given in the previous section. In particular, it is per-
haps not obvious why precisely the vowels [a], [i] and [u] are included in
those sets, rather than, for instance, [a], [o] and [5].
130 Vowel Sounds
One plausible reason for the prevalence of the three basic vowels [al, [i],
[u] could be that they are easier to articulate. If we interpret ease of articu-
lation as minimal movement, however, [a], [i], [u] don’t seem easier to
articulate, since they require extreme positions of the tongue and lips, with
concomitant greater articulatory effort. On the other hand, minimizing
articulatory effort cannot be the only consideration for the selection of
sounds in language, since, if it was, only one sound (the sound easiest to
pronounce) would be in existence universally.
What is special about the vowels [a], [i] and [u] is that each can be articu-
The vowels [a], [i], lated over a reasonably broad space with minimal effect on perception. For
[u] are called
instance, while we said that cardinal vowel no. 1 [i] corresponds to an extreme
QUANTUM
VOWELS because
forward and upward movement of the tongue, the essential sound quality
each of them can of the vowel will still be obtained with a more relaxed movement: the dif-
be articulated over ference in articulation may of course be perceptible, as we saw above in
a reasonably connection with English sea and French si, but the vowels in question will
broad space with still be identified as, broadly, [i]. For this reason, [a], [i] and [u] are given
minimal effect on
the label QUANTUM VOWELS, suggesting a perceptual quantum leap
perception
between them. By contrast, variation within each such vowel will at best be
perceived as a matter of nuance, or “accent”. This state of affairs arises from
the fact that the relationship between the size of the tubes and the sound
they produce is not a direct one — this matter falls within the domain of acous-
tic physics, and therefore its details do not concern us here.
El Secondary Cardinal Vowels: Front Round Vowels
We have seen that the eight primary cardinal vowels [il, [el], [e], el, [al,
[5], [o], [u] are basically definable by means of the parameters of height and
frontness—backness. In particular, the first four such vowels ([i] down to
[z]) are front, while the second four ([a] up to [u]) are back. Concomitantly,
[i] and [u] are high, [ez] and [a] low, and [e] and [o] mid. All these descript-
ive expressions refer of course to the positioning of the tongue in the mouth
cavity, which creates differences of size and shape in the tube it forms with
the roof of the mouth.
We have also seen that the space is somewhat more limited at the back
of the mouth, the lips accordingly undergoing rounding in all back vowels
but [a], to compensate. Rounding of the lips for the primary cardinal vowels
nos. 6, 7 and 8 comes therefore quite naturally.
Suppose now that we reverse the normal setting and round the lips for
all the primary cardinal vowels but precisely these three. What will hap-
pen? Clearly the sizes and shapes of the resulting eight new tubes (one tube
for each cardinal vowel) will be different from the sizes and shapes of the
Vowel Sounds 131
tubes that correspond to the original eight primary cardinal vowels. If the
tubes are different (considerably different, in fact, since the contribution of
the lips is not negligible), the sound they will produce will obviously be dif-
ferent. So, in effect, we will have doubled our vowel inventory, from eight
to sixteen members.
The vowels produced by deliberately reversing the normal action of
the lips are known as SECONDARY CARDINAL VOWELS. We will now The vowels pro-
duced by deliber-
review these vowels, bearing in mind that they are not well represented in
ately inverting the
the most common accents of English, and therefore we will be forced to refer normal action of
to other languages for description and exemplification more often than we the lips are known
would have wished. as SECONDARY
Fortunately, most of the front secondary cardinal vowels exist in languages CARDINAL
VOWELS
with which readers are likely to be reasonably familiar, French or German,
for instance. The round vowel in the front high position corresponding to
[i] is transcribed as [y] ([ti] in an alternative transcription not uncommon in
America), and is present in the French word rue [by] ‘street’ and in the German
word friih [fy] ‘early’.
Interestingly, it is not uncommon among English learners of French and

German to replace [y] with the vowel sound in the English word cue. While
this is obviously wrong, the rationale for the process is both clear and log-
ical, since the word cue contains precisely the vowel sounds [i] and [u] in
succession: [kiu]. These learners are therefore “aware” that the target vowel
[y] overlaps with [i] in frontness and with [u] in rounding, but find it diffi-
cult to combine frontness and roundness in the same vowel. Consequently,
they resort to the strategy of articulating frontness and roundness in suc-
cession, precisely as in the English sound represented in cue: [y] is in fact
one of the historical sources of the modern English sound [iu]. In chap-
ter 8 we briefly discuss some formal machinery that makes direct sense of
this behaviour.
We now incorporate [y] (a front high round vowel) into our familiar vowel
chart, retaining its primary cardinal vowel counterpart to facilitate comparison.
The two vowels are articulated in exactly the same point (figure 5.9).
132 Vowel Sounds
—e <
Figure 5.9 The secondary cardinal vowel ly]
The next secondary cardinal vowel down the chart is [@] ([6] sometimes
in America), the round counterpart of [e]. Again, this vowel does not occur
in English, but it does in French and in German: feu [fo] ‘fire’ and schon [fon]
‘beautiful’, respectively. Monolingual English speakers are likely to have dif-
ficulties with this sound, and simply replace it with an unround central vowel
which we shall discuss in chapter 7. In particular, there is no sequence [eo]
or [oe] in English to fall back on, the way that there is [iu] for [y].
The updated vowel chart looks as shown in figure 5.10.
Figure 5.10 The secondary cardinal vowel [g]

Vowel Sounds 133
The next round front vowel is [ce] (also [5] in America), corresponding to
the primary vowel [e]. As with the other front round vowels, this vowel does
not exist in English, although again it does in French and German: coeur [koex]
‘heart’ and zwilf [tsvcelf] ‘twelve’, respectively.
\\
We now incorporate [ce] in the vowel chart (figure 5.11).
—T
y
Figure 5.11 The secondary cardinal vowel [ce] added
The last front round vowel is the low vowel [CE], corresponding to the
primary cardinal vowel [z]. This sound is truly rare among the world’s lan-
guages. It is, however, reported to exist in a Bavarian dialect of Austria. The
extreme rarity of this sound is evidence of its unnaturalness, easily traced
back to the physiological difficulties in lip rounding inherent to low vowels,
and to the weak perceptual salience of the sound that follows from the
tenuousness of the rounding. All this makes the practice of [GE] unneces-
sary, although we will still add it to our chart for completeness, as an addi-
tional reference point (figure 5.12).
pio
¥
& CE
Figure 5.12 The front cardinal vowels (primary and secondary)
134 Vowel Sounds
E) Back Secondary Cardinal Vowels
You are well aware by now of the tendency of back vowels to be spontane-
ously round, and of the fact that this tendency does not extend to the low
back primary cardinal vowel [a].
If [a] is unround, its corresponding secondary cardinal vowel will have to

be round, if only slightly so on account of the difficulty in rounding low
vowels referred to at the end of the previous section. The IPA symbol for
the back low round vowel is [p], and for once this vowel exists in most accents
of England, Wales and the southern hemisphere, in such words as hot or
loss. A remarkably precise version of [bp] is found in Yorkshire. In North
America and Southern Ireland, these and similar words tend to exhibit the
vowel [a], despite the identity of the spelling. However, in some of these
accents the sound [np] can still be found to some degree of approximation in
words like cloth, caught, claw, watch, horrid, etc.
The vowel [p] completes the inventory of round secondary cardinal vowels,
which we now display in the familiar vowel chart (figure 5.13).
we CE D a
Figure 5.13 The set of round secondary cardinal vowels
Vowel Sounds 135
The remainder of the back secondary cardinal vowels are unround, a re-
versal of their primary condition. The unround counterpart of the primary
cardinal vowel no. 6, [95], is [A]. This vowel (or something quite similar to this
vowel) existed in pre-Second World War RP English in such words as up
or cuff, but its position has now shifted forward. In spite of this change, this
vowel sound is still usually transcribed with the IPA symbol [a], confus-
ingly for RP English and similar accents: in some North American and south-
ern Irish accents the vowel of up or cuff does approximate to the cardinal
vowel [a] (the accents of northern England typically have a much higher
vowel in these words).
We now incorporate [a] into our vowel chart (figure 5.14).
Pay
eS OG
S ce A a)
a CE D a
Figure 5.14 The back secondary cardinal vowel [a] added
The next secondary cardinal back vowel on our way up the chart is [¥],
the unround version of [o]. This vowel also does not exist in common vari-
eties of English. It can, however, be heard in words like up and cuff in the
speech of (normally south-dwelling) speakers of northern English origin in
their attempt to approximate RP - as mentioned above, the vowel of these
words in genuine northern English accents is higher, and also round: we
will describe this vowel in detail in chapter 7. We now add [x] to the vowel
136 Vowel Sounds
chart (figure 5.15) — if you wish to practise this vowel you simply have to
unround your pronunciation of [o], as expected.
wy yi
c Oo ¥ O
o ce A Oo
re CE D a
Figure 5.15 The back secondary cardinal vowel [x]
The last secondary cardinal vowel is [w]. This is a high back unround vowel,
the unround counterpart of [u]. This vowel occurs in Vietnamese, where [ku]
‘continue’ contrasts with [ku] ‘owl’ and with [ki] ‘note’. A bit closer to home,
it also turns up in Turkish: [dw] ‘exterior’.
We give the complete set of cardinal vowels in figure 5.16.
11 y9 16W U8
ze 15 ¥| O07
14A] 96
Figure 5.16
angie
Complete set of cardinal vowels
Vowel Sounds 137
As you can see, there are sixteen cardinal vowels in all, of which eight (nos.
1 to 8) are primary, and the rest secondary (from 9 to 16). As one would There are in all
expect, secondary cardinal vowels, being less natural than their primary coun- sixteen cardinal
vowels, of which
terparts, are less widespread, but they are all still found in some languages
eight (nos. 1 to 8)
— some of them in quite a few languages. We have mentioned an alternat- are primary, and
ive, but less common, transcription system for the front rounded vowels: the rest secondary
placing a dieresis on the symbol of the corresponding round back primary (from 9 to 16)
vowels ([ii], [6], [5]). This system, particularly associated with North
America, is potentially confusing, since the dieresis indicates centralization
(not fronting) in the more widely used IPA system.
Central Vowels
We have already observed that vowels can be advanced or retracted —- as

noted earlier, the corresponding IPA underscripted diacritics are “,” and
““", respectively. Suppose now that we simultaneously retract a front vowel
and advance a back vowel of the same height until the two vowels meet
in the middle. When we get to this position, it will obviously be difficult
to say whether we are dealing with a retracted front vowel or with an
advanced back vowel, since the intermediate sound is both and neither. As
a consequence, it will be useful to set up a third set of vowels which are
neither front nor back. This set of CENTRAL VOWELS is not part of the A third set of
vowels are neither
official inventory of cardinal vowels, for historical reasons, but is in com-
front nor back,
mon use: for instance, central vowels are included in the IPA chart. but “central”.
We have already noted that the low vowel of three- or five-member Central vowels are
vowel systems, [a], is central: [i], [a], [u], and [il], [el], [a], [o], [ul], respectively. not part of the
A slightly higher variety of this vowel, represented in IPA transcription by official inventory
of cardinal vowels,
means of an inverted “a”, [ev], occurs in European Portuguese: [ene] Ana ‘Ann’.
but they are still
It is also the most common pronunciation of the vowel in words like up or important
cuff in current RP English and similar accents, despite its usual mistranscription
as [a], which we have seen is a legacy of history.
Moving up the central vowel space we come across the most neutral of
all vowels: the vowel that, so to speak, comes out of the mouth without any
138 Vowel Sounds
tongue or lip movement. In English, many stressless vowels are realized thus.
The IPA symbol for this central mid vowel is an inverted “e”: [a]. This vowel
is endowed with its own label, in recognition of its importance: schwa (pro-
nounced [fwa]), the old Hebrew term for a diacritic indicating a missing
vowel (Hebrew writing usually only includes consonants). Schwa is a very
common sound in the world’s languages, although it is important to real-
ize that not all the sounds thus labelled (or indeed transcribed) are in fact
phonetic schwas. For instance, schwa is supposed to occur abundantly in
French (in correspondence with non-silent orthographic es), but it has been
argued that such alleged schwas are phonetically [ce], rather than [a].
Similar situations obtain in other languages, mutatis mutandis.
We shall now examine briefly the two central high vowels [i] and [zl],
unround and round, respectively. The vowel [3] occurs in northern Welsh:
un ‘one’. A similar vowel can be heard in words like bit or wish in some
accents of English, notably Scottish and Northern Irish, as well as in New
Zealand. Scottish and Northern Irish English (but not New Zealand
English) often have [] in words like choose or foot. This sound occurs more
generally in other languages, such as Norwegian or Swedish: Swedish [hws]
‘house’, [ful] ‘ugly’, [nw] ‘now’.
We now enter in the chart the position of the central vowels we have dis-
cussed (figure 5.17).
Figure 5.17 Central vowels

Vowel Sounds 139
We wind up the chapter with a chart of all the vowels we have presented,
and with a tabulation of these vowels in terms of the basic parameters
“front” /“back”, “high”/“low”, and “round” (figure 5.18 and table 5.1).
TRY, + U uw u
e\. O
ee So
ee a
a
Figure 5.18 Complete vowel chart
Table 5.1 Parametric vowel table
Front Central Back
Unround Round Unround Round Unround Round
me, i y i a w u
Mid High e @ 3 ¥ oO
Mid Low € ce A 3
Low ee G& a a D
The speech sounds known as v

consonants we have been disc
are articulated with no co
‘ith sonorant onsonants, mak-
sounds. The fact that no -con-
140 Vowel Sounds
for ease of reference. The relationship c

is, in the majority of cases, only approxi
able with the use of diacritics. Not all of tl
approximations of them, occur in all of the v
imal three-vowel system, conse of the q
other in all languages. Then

vowels, with [e] and [o] ade
ting a five-oe figure ont
«
Key Questions
1 What distinguishes a vowel from 7 What are the minimal and the most
a consonant? common vowel systems in the lan-
2 How do we modulate vowel quality? guages of the world?
3 What are “cardinal vowels”? How 8 Define the term “quantum vowels”.
are they identified? List the primary What are the particular attributes
set. of these vowels?
4 What do the initials RP and GA 9 Which of the settings for primary
stand for? cardinal vowels is reversed in the
5 How-do cardinal vowels differ from secondary set?
real-world vowels? What is their 10 List the parameters used for the
use for descriptive purposes? characterization of vowels.
6 Define the terms “broad” and “nar-
row” phonetic transcription.
Vowel Sounds 141
Further. Pra ¢t ice
Vowel Sets
What properties do the following sets of vowels share?
a. ly ce o &]) d. [faean]
b. liiwy uw) e In30v ul
caf [ur x a al f. «le ve el
Vowel Systems
Study the following vowel systems. Explain how each of the sets can be
described as either symmetrical or asymmetrical. What would be required
to make the asymmetrical sets symmetrical?
Italian Albanian Sundanese Papago

i u ery u {te OU ye 4
e Oo
€ 3 € or ca. 0 a
a a a
Itonama Persian Hungarian
mi U a ty U iz y!
e oO ~we to eT cet O85 el Ot
a ea D at
Azerbaijani Chuckchi Chuvash

a Se i u 1, yy ee
@ Oo a €
€ a a
ze a
Vowel Descriptions and Phonetic Symbols
a. Provide phonetic symbols for the following descriptions:

A high-mid unrounded front vowel
A high-mid unrounded back vowel
A high rounded central vowel
A low rounded front vowel
142 Vowel Sounds
A low-mid rounded back vowel

A mid unrounded central vowel
A low unrounded back vowel
A high rounded back vowel
b. Provide descriptions to match the following phonetic symbols:

fe] [a] [eo] fi] [v] [le] fol Ly]
PHONOLOGICAL PROCESSES INVOLVING
VOWEL FEATURES
m The distinctive features u

w More feature dependencies.
m Features that are not specifi
by some facts of Turkish.
= A mechanism for filling 1
spreading of lexical feat
In chapter 4 we saw that the primitive units of phonological sound are the
distinctive features, and we presented and discussed a substantial number
of them. In this chapter we will augment the universal list with a handful
of additional features specifically relevant to vowels.
t eend of chapter§
ight berelevanti
We will put these features to work in Turkish “vowel harmony” and in

German and English vowel “umlaut”. The Turkish phenomenon will show
the need for feature “underspecification”, that is, for assuming that certain
distinctive features are altogether absent in some lexical forms. Turkish vowel
harmony will also demonstrate the prohibition against crossing association
lines, one of the pillars of autosegmental phonology.
144 Phonological Processes Involving Vowel Features
Distinctive Features for Vowels
Table 5.1 at the end of chapter 5, classifying vowels phonetically by the para-
meters height, frontness—backness and roundness, can be reinterpreted
almost verbatim in terms of distinctive features.
We must first formally differentiate between consonants and vowels, by
means of the feature [tconsonantal], which we define as follows:
[+consonantal] sounds involve a drastic constriction in the central oral

[+consonantal] passage
sounds involve a
drastic constriction
in the central area By “drastic constriction” we mean a constriction that results in total block-
of the mouth age of the airflow, as for stops, or in restricted air exit causing friction, as
in fricatives. The condition that the constriction must involve the central
oral passage implies that obstruents, nasals, laterals and most rhotics will
be [+consonantal], whereas all vowels (possibly with some constriction on
the sides of the mouth) will be [—consonantal].
We can now turn to the features that distinguish the various vowels. If we
first consider the height parameter, you will remember that some vowels
are high, in that they involve the raising of the body of the tongue ([i], [u]),
some are low, involving the lowering of the body of the tongue ([e], [al),
and some are neither ([el], [o]).
of formalizing the hi
_ of distinctive featu
This tripartite classification of vowels with respect to height can be imple-

A tripartite classi- mented by means of the binary features [thigh] and [tlow]. We illustrate
fication of vowels
this in (1) for the familiar, and highly natural, five-member system:
vis-a-vis height
can be imple-
mented by means
(1) fi] fe] [a] [o] [ul
of the binary fea-
[high] + - - - +
tures [thigh] and
[tlow] [low] - - + —- =
Mathematically, there is of course another possible combination, namely,

[+high, +low], but this is ruled out on the reasoning that the body of the
tongue cannot be simultaneously raised and lowered.
Phonological Processes Involving Vowel Features 145
The major difference between [i], [e], on the one hand, and [ul], [o], on the
other, is one of backness. We captured this difference by means of a feature
[+back] bearing on body of the tongue retraction, as we now illustrate for a The difference
six-vowel system: in backness is
captured by
means of the
(2) li] [fe] fe] [a] [ol] [ul feature [+back]
[high] + - - - - +
[low] - - + + - =
[back] - - - + + +
In systems with an odd number of vowels the unpaired central low vowel
[a] is usually considered [+back] phonologically.
Secondary cardinal vowels are differentiated from the primary cardinal

vowels exclusively on the basis of inverted roundness. The phonological inter-
pretation of this parameter is the distinctive feature [tround], as we now The roundness
illustrate, first for the primary cardinal vowels, and then for their secondary parameter involves
the distinctive
counterparts:
feature [+round]
(3) fi] [e] [fe] [e] [a] [5] fo] fu]

ishiiaf tobe sa bie ay Rink
[low] —y) 4d) 8) feb otttroet oti et
[back loins eure fbi i) oF etasnodimn +
sound ot Shregen ate cokerodt Bae cect ovek
(4) [y] to] [ce] [2] [vo] [a] [vy] fw]

Lie ia a cit Ee il
[low] -~.,-.--, + - - =
[back] ~ + + +
[round] x +| +| +| +
The four features proposed distinguish all the segments reviewed but the
two mid vowels in each of the front and back sets. We obviously need a
further feature to differentiate the vowels in these pairs. Before we go into
this matter, it will be useful to notice the existence of dependencies between
There are depend- the features we have just presented and some of the features we introduced
encies between
in chapter 4.
features
tures are typically |

features we have mentio
Feature Dependencies
Consider first [tround]. Clearly, the rounding captured by this feature is

implemented by the lips. In chapter 4 we introduced a unary feature [labial]
to capture involvement of the lips in articulation, all labial sounds being phono-
logically classified as [labial]. We are now seeing that [+round] sounds are
necessarily [labial], suggesting that [+round] is a dependent of [labial]. If
[+round] is a [+round] is a dependent of [labial], [-round] will also have to be, for the
dependent of
simple reason that [—round] is not a different feature from [+round], but the
[labial]
same feature with the opposite value:
(5) [labial]
[+round]
Notice that the only reason to specify a sound as [-round] is that it could
conceivably have been [+round]. If so, the sound in question must of neces-
sity be [labial]. For instance, the consonants [p], [f], [b], etc., are all [labial]
and [-round]. By contrast, consonants like [t], [d], etc., are articulated with
the blade of the tongue, as we know, and therefore [tround] is simply irrel-
evant to them.
Let us now examine [thigh], [+low] and [+back]. We already know that
these features capture movements of the body of the tongue upwards, down-
wards and backwards — notice that the blade can simultaneously be tucked
in behind the lower teeth.
It follows from this that [thigh], [tlow] and [+back] are dependents of
[thigh], [tlow]
[dorsal] (see, however, chapter 17 for a different opinion):
and [+back] are
dependents of
[dorsal]
(6) [dorsal]
[thigh]
[tlow]
[+back]
Notice that each of the dependent features is placed on a different plane, in

line with the multiplanar geometry we introduced in chapter 4. Once more,
if a segment is not [dorsal], the dependents of this feature will simply be
irrelevant to it.
In the feature tables in (3) and (4) above, the mid vowel pairs [e]/[e], [0] /[o],
etc., are not differentiated by the four features discussed so far. We will now
explain how these vowels are distinguished.
Two More Distinctive Features
As follows from our remarks in chapter 1, the tongue is an organ of con-

siderable length. Consequently, you will not be surprised to hear that a fur-
ther section hides behind the body of the tongue, the tongue ROOT, which
we now illustrate (figure 6.1).
The vowel [o] with advanced The vowel [9] with retracted
tongue root ([+ATR]) tongue root ([-ATR])
Figure 6.1 Advanced and retracted tongue root
The tongue root has both phonetic and phonological relevance. In particular,
if the tongue root is brought forward during the articulation of vowels, the
pharyngeal cavity gets enlarged. Concomitantly, the size of our familiar tube
responsible for vowel quality is modified: vowels with such a pharyngeal
enlargement will automatically sound somewhat different from vowels which
are identical apart from the enlargement. The forward movement of the root
of the tongue is captured by the distinctive feature [- ATR] (ADVANCED
TONGUE ROOT), a dependent of the unary feature [radical], a word meaning
‘of the root’:
(7) [radical]
Wa
ae
[+ATR]
The incorporation of the feature [+ATR] into our feature inventory allows
[+ATR] accounts
us to specify mid-high vowels positively as [+ATR], and mid-low vowels
for the difference
as [-ATR]:
between the two
shades of mid
vowels (8) [e] fo] [e] [od]
[ATR] + + = =
We have now defined the sixteen cardinal vowels (eight primary cardinal
vowels and eight secondary cardinal vowels) in terms of a set of only five
phonological distinctive features.
The features we have available cannot differentiate the central vowels we

added to the vowel repertoire at the end of the previous chapter. This defici-
ency could be overcome by enriching the distinctive feature inventory with
a feature [+front] to supplement [+back], paralleling the contrast between
[thigh] and [+low]. A [+front] feature, however, does not form part of most
current theories of distinctive features: central vowels are analysed phono-
logically as unround back vowels. There are two reasons for the absence of Central vowels
are analysed
[+front]. The first is that central vowels are thought never to contrast phono-
phonologically
logically with unround back vowels. The second is that the articulatory gesture as unround
that would justify a feature [+front], a deliberate forward movement of the back vowels
body of the tongue, is claimed not to be viable.
Before finishing the section, we will draw your attention to the existence
of implicational relationships between features over and above the rela-
tionships implicit in the dependency relations we have been examining. Thus,
for instance, if a sound is specified as [-consonantall], it is automatically also
[+sonorant] and [+continuant]: these two values are therefore redundant and
can be left out of the representation, then said to be “underspecified” for
these features.
“Underspecification” will be found to be particularly useful in the next sec-

tion. In-depth discussion of this construct can be found in chapter 17.
Z¥ Back Harmony in Turkish

There are
languages where
There are languages where the class of vowels which occur in a given domain,
the class of vowels
typically the word, is restricted in some way. This phenomenon is commonly which occur in a
known as “harmony”, on the grounds that the vowels harmonize for some given domain is
feature or features in the domain in question. The vowel harmony of Turkish restricted in some
has been thoroughly investigated in the literature, and provides a conveni- way
ent point of entry into this area.
Consider the following data:
(9) Nom.sg. Gen.sg. Nom.pl. Gen.pl.

‘rope’ ip iping, _ ipler iplerin
‘hand’ el elin eller ellerin
‘Bits sak ulz. kwzumn kuwizlar kwzlarwn
‘stalk’ sap sapuin saplar saplaruin
H / manifestations
can you isolate in (9)? —
At a purely descriptive level, we can say that the Turkish genitive singular is
formed by the addition of either [in] or [um] to the nominative singular, the
nominative plural by the addition of either [ler] or [lar] to the nominative
singular, and the genitive plural by the addition of either [in] or [wm] to the
nominative plural. We could streamline these statements by construing the
nominative singular as the common base (or ROOT), to which the suffix
[er]/[lar] is added to signify plural, and the suffix [in]/[vum] added on to
signify genitive. A SUFFIX is, of course, a morpheme concatenated to the right
of some base; we assume that the concept of PLURAL needs no explanation;
GENITIVE refers to the form of the word, or CASE, used to indicate posses-
sion, a function expressed in English by the preposition of or the possessive
‘s, as in John’s.
The description we have just offered is quite satisfactory from the per-
spective of MORPHOLOGY, the branch of linguistics that studies the struc-
ture of words.
In particular, the Turkish words in (9) have a structure ROOT (+ PLURAL

SUFFIX) (+ GENITIVE SUFFIX), where the parentheses indicate optionality
(not all words are plural or genitive), and the order of concatenation is as
given. If we look at the data from the perspective of phonology, however, we
will inevitably notice the vowel alternations in the plural and the genitive
suffixes. In chapter 2 we set ourselves the task of reducing surface alternations
to a common lexical representation, accounting for the mismatch between
lexical and surface representations by means of phonological rules. We will
now attempt to do this for the Turkish data we are considering.
In order to elucidate what is actually going on in Turkish we first need

to decompose the alternating segments into their relevant distinctive features:
you are well aware by now that IPA phonetic symbols are only a shorthand
for the feature substance of the corresponding sounds. You can see in (9)
that the alternating vowels are [i] and [w] in the genitive suffix (cf. [in] vs.
[uin]) and [e] and [a] in the plural suffix (cf. [ler] vs. [lar]). In (10) we dis-
play the composition of these segments in terms of the features [thigh] and
[+back]:
(10) 1 ie
Pricey Oe sty eee
[back] — + —- +
The two features [thigh] and [tback] are sufficient to differentiate the four
Turkish vowels, and the remaining features are therefore redundant — the
system only has two degrees of height, for instance, and [+low] plays no
role.
If you examine the table in (10), you will notice that the difference
between the two vowels that alternate in each suffix concerns the feature
[+back]. The obvious question is why in these suffixes the feature [+back]
takes on the value “+” in some words, and the value “—” in others.
Pursuing our assumption that the reason for the alternation is phonological,
not morphological, we shall look at the feature composition of the vowel in
the root. Notice that the suffix consonants are identical in both alternants,
and therefore cannot be responsible for the alternation.
We could try to attribute responsibility for the alternation to some other

factor, for instance to the consonants of the root, but we will not consider
this obviously incorrect hypothesis, to save ourselves time and effort.
Instead, we will consider the relationship, if any, between the vowels
in the root and the vowels in the suffix. Sometimes this relationship is one
of absolute identity ([ipin], [saplar], etc.), but sometimes it is not ([elin],
[kuuzlar], etc.). Does this mean that the reason for the suffix alternation does
not lie in the root vowel after all? Not necessarily. Notice in particular that,
while the relationship between the two vowels in question is not always one
of total identity, it is systematically one of partial identity, as we illustrate
in (11):
(11) ipsam SG. pls quir eyrk tn ku, z2lyay se

[thigh] [+high] [-high] [-high] [-high] [+high] [+high] [-high]
[—back] [—back] [+back] [+back] [—back] [—back] [+back] [+back]
You will notice that the values of the feature [thigh] need not be uniform
across the root and the suffix. The values of the feature [tback], however,
always are: [tback] harmonizes throughout the Turkish word, and con-
[+back] harmon- sequently we say that Turkish has [tback] harmony for vowels. We should
izes throughout
of course check some more data before making such a general statement,
the Turkish word
but we will once more save ourselves the time and the energy.
Lexical Underspecification
Having discovered the essence of the Turkish alternation, we will now propose
a formalization for it. The vowels in [ipin] and [sapwm] will be represented
as in (12) — to facilitate identification, we provide full IPA transcriptions at
the top:
(12) eos nqto ley Sigh wony

[dorsal] [dorsal] [dorsal] [dorsal]
[thigh] / [+high] [-high] / [+high]
[—back] [—back] [+back] [+back]
You will notice that in the genitive the suffix vowel is consistently [thigh],
but its value for [tback] varies according to the value of this feature in the
root vowel: this is of course simply a restatement of the [+back] vowel har-
mony of Turkish. The most straightforward formalization of this situation
involves leaving the feature [+back] lexically unspecified in the genitive and
plural suffixes: the lexical representations of these suffixes will therefore be
left blank for [+back], but the genitive suffix will of course be lexically [+high],
and the plural suffix [—high].
The lexical representations of [ipin] and [sapwin] will consequently be as

follows:
(13) se i a all | So tee poe: |

[+high] [+high] [-high] [+high]
[—back] [+back]
Notice that, while the root vowels contain the appropriate specifications for
[+back] (compare *[urp], *[sep], respectively: by convention, incorrect forms
are starred), this feature is missing from the suffix vowels. We know, how-
ever, that in the phonetic representation of these words the suffix vowels are
[-back] and [+back], respectively, through agreement with the root vowel,
as represented in (12) above. How can this target be attained?
The answer is straightforward in the context of an autosegmental approach.

Surface representa- In particular, the most direct strategy involves borrowing the specification
tions borrow the
for [+back] from the neighbouring vowel, along the lines of the assimilation
specification for
processes we examined in chapter 4:
[+back] from the
neighbouring
vowel S a p uw n
(14) i P i n
[thigh] [+high] [-high] — [+high]
back] feck)
This analysis thus provides a principled explanation for the fact that the suf-
fix vowel always shows up with the same value for [tback] as the root vowel.
Clearly, the process also takes place in forms with a chain of suffixes, as we
now illustrate for the pair [iplerin], [kuizlarunn]:
(15) Le Dla ¢ CfLl a Ve ctl ke, 4, Zila Gi Cy pL
[thigh] [-high] _[+high] [thigh] . [-high] [+high]
[dorsal] [dorsal] [dorsal] [dorsal] [dorsal] [dorsal]
[-back] Lebeeel
The prediction of the approach is that the number of vowels following the
root will be irrelevant to the operation of the process of back harmony. This
prediction is obviously correct for the data examined. It is also correct for
comparable data not discussed, but that once more we could and ought to
test: you know from chapter 2 that the accounts phonologists come up with
only have the status of hypotheses.
[Vowel Disharmony
The analysis as it now stands predicts that vowels will always have the same
value for [+back] in the suffix(es) and in the stem. Consider, however,
forms like the following:
Phonological Processes Involving Vowel Features
(16) a. iki ‘two’ b. ikigen ‘two-dimensional’

altur ‘six’ altuugen ‘hexagonal’
yedi ‘seven’ yedigen ‘heptagonal’
sekiz ‘eight’ sekizgen ‘octagonal’
The forms in (16b) are obviously made up by suffixation of the stems in (16a)
with the suffix -gen. The form gen of this suffix is predicted when the stem
contains front vowels: quite simply, the [—back] specification of these vowels
will trigger the [—back] specification of the vowel [e] in gen. When the base
contains back vowels, as it does in altu, we would of course expect a back
vowel in the suffix also: *altugan. This is, however, not the case, since this
suffix invariably turns up with the front vowel [e], regardless of the specifica-
tion of the vowel(s) of the stem for [tback]: altwigen. The form altwegen there-
fore exhibits vowel disharmony, since the value of [tback] is not identical
in all its vowels:
(17) a@stelt Gyr ice pee oh
[dorsal] [dorsal] [dorsal]
[+back] [—back]
Even more striking instantiations of vowel disharmony are provided by such

bisyllabic suffixes as -vari ‘-like’, -lejin ‘during’, -istan ‘-land’. Consider the
following forms:
(18) a. ef, ‘partner’ b. efvari ‘partner-like’

balwk ‘fish’ balutkvari ‘fish-like’
fil ‘elephant’ filvari ‘elephant-like’
kedi ‘cat’ kedivari ‘cat-like’
sabah ‘morning’ sabahlejin ‘during the morning’

gedge = ‘night’ geelejin ‘during the night’
akfam ‘evening’ akfamlejin ‘during the evening’
ermeni ‘Armenian’ ermenistan ‘Armenia’

arap ‘Arab’ arabistan ‘Arabia’
kazak ‘Kazak’ kazakistan ‘Kazakistan’

hint ‘Indian’ hindistan ‘India’
afgan ‘Afghan’ afganistan ‘Afghanistan’
madgar ‘Hungarian’ madgaristan ‘Hungary’
In ermenistan, the first vowel of the suffix, [i], harmonizes with the vowels
of the stem: all the vowels are [—back]. However, in arabistan this same vowel
shows up disharmonic: [i], instead of the expected [wu]. Worse still, the second
suffix vowel, [a], clearly does not harmonize with its predecessor [i] in the
same morpheme. If it did harmonize, the morpheme would systematically
be -isten, but it obviously is not.
Vowel disharmony is also observable in stems. So far, we have provided
stems which are either monosyllabic, and therefore vacuous for harmony,
or polysyllabic with harmony. Indeed, polysyllabic harmonic stems are the
norm, as we further exemplify in (19):
(19) balwk ‘fish’ bilezik ‘bracelet’

jatak ‘bed’ dilek = ‘wish’
julan ‘snake’ bilet ‘ticket’
alturm ‘gold’ endife ‘worry, anxiety’
ifkembe ‘tribe’ salak ‘stupid’
salataluik ‘cucumber’ hugjar ‘cucumber’
You can easily verify that in the forms in (19) all the vowels have the same
specification for [+back]: they are all either [+back] or [-back]. However, dis-
harmonic roots also exist — indeed, their number is not inconsiderable. We
offer a small sample in (20):
(20) anne ‘mother’ fijat ‘price’

hamsi ‘anchovies’ adet ‘item’
mezat ‘auction’ takvim ‘calendar’
haber ‘news’ vazijet ‘position’
elma ‘apple’ battanije ‘blanket’
fakirlik ‘poverty’ jemekhane ‘dining hall’
basit ‘simple’ ikametgah ‘residence’
sefalet ‘poverty’ serinkanlut ‘cool-headed’
The stems in (20) exhibit a mixture of back and front vowels, and therefore
contradict vowel harmony.
We will account for disharmony through the idiosyncratic lexical specifica-

tion of [+back]. We exemplify with -istan, for suffixes, and with adet ‘item’,
for stems:
(21) i> st). ca. =H de <d e t

[—back] [+back] [+back] [—back]
The presence of the appropriate specification of [+back] in the lexicon blocks

any further association, since Turkish [tback] harmony, like other harmony
systems, only operates on underspecified representations.
The No-Crossing Constraint
Forms like altwgenler ‘hexagonals’, with the familiar plural suffix -ler added
to the base altugen ‘hexagonal’, raise an interesting question. In particular,
we know from our discussion in section 4 above that the Turkish plural suf-
fix has two alternants, -ler and -lar, as a result of vowel harmony. This being
so, we might expect vowel harmony to be triggered from the root, the strongest
morpheme both morphosyntactically and semantically. Therefore, we might
expect the form “altwgenlar, erroneously.
The fact that we get altwegenler instead shows that the harmony cannot have
been induced by the root. This situation is indeed systematic for all harmonic
suffixes:
(22) altuigenler ‘hexagonal’ (pl.) ermenistanlar ‘Armenia’ (pl.)

adetler ‘items’ elmalar ‘apples’
takvimler ‘calendars’ fijatlar ‘prices’
vazijetler ‘positions’ _ jemekhaneler ‘dining halls’
If you examine the forms in (22) carefully, you will notice that in all cases
each prespecified vowel (an OPAQUE VOWEL) starts off a new harmony
Each prespecified domain. To make your task easier, we make this explicit in (23):
vowel (an
OPAQUE
VOWEL) starts off
(23) altugenler ermenistanlar
a new harmony adetler elmalar
domain takvimler fijatlar
vazijetler jemexhaneler
In (23) the vowels emboldened are prespecified in the lexicon as [+back] or

[-back]. On the reasonable assumption that the unemboldened vowels are
unspecified for [tback], they manifestly take the value for this feature from
the immediately preceding vowel. The question we are asking is why they
do not take it from the root vowel(s) instead.
Let us investigate what such an outcome would formally entail. We shall
use the form altugenler as a test case. In (24) we provide the autosegmental
configuration of the alternative, but non-existent, *altugenlar, which would
indeed be derived if the harmony trigger were located in the root:
(24) der ituiursti.eie

ent ml alan
[-high] [+high] [-high] _ [high]
PPouiRsi gorIoas
In (24), the [+back] of the first vowel of the root has spread to the second
vowel, as predicted. Its spread to the third vowel has been blocked by the
prespecified value [—back] associated with this vowel in the lexicon, in line
with our previous discussion. The question now is why [+back] cannot in
fact associate to the last vowel, in -lar: we might expect that it could, given
the fact that this vowel does not carry any [+back] prespecification. If it did,
we would of course get the illegitimate form “altwgenlar.
The reason the structure in (24) is not possible is that it contains a crossing of
lines on the [tback] plane. In particular, the line linking [+back] to the suffix
vowel crosses the line linking this vowel to the prespecified value [—back].
This configuration infringes a general, inviolable principle of autosegmental
theory, the NO-CROSSING CONSTRAINT:
(25) No-crossing constraint:

Association lines may not cross Association lines
may not cross
(= “No-Crossing
The rationale for the No-Crossing Constraint is straightforward. Constraint”)
Consider (24) again. This representation in fact contains the implicit claim
that the specified feature [+back] both precedes and follows the specified
feature [—back]. The reason for this may not be obvious, since it would appear
that in this representation [+back] only precedes [—back]. In its line indeed
it does. However, once the multidimensional nature of autosegmental rep-
resentations is taken into account, the contradictory claim we referred to comes
through. In particular, [+back] is associated both with a vowel that precedes
the vowel associated with [—back] (precisely as it should) and with a vowel
that follows the vowel associated with [—back]. This contradiction accounts
for the illegitimacy of this type of representation, encapsulated in the No-
Crossing Constraint.
Because the No-Crossing Constraint rules out the representation in (24),
the correct derivation will be as in (26):
(26) @!. viltr (rev gid etivnhaneooet
[-high] [+high] [-high] [-high]
disaster
bledSeo CI
ea iidgyeall [dorsal] adage
ee
[+back]
ae
[-back]
In (26) there is no crossing of lines. Instead, the prespecified feature [-back]

spreads from -gen to the vowel of the plural suffix, correctly yielding the
allomorph -ler.
FE} German Umlaut
Turkish is likely to be rather an exotic language for most of the readers

of this book, who may be tempted to think of vowel harmony as some-
thing equally out of the ordinary. As it happens, however, languages closer
to home, such as German and English itself, exhibit essentially the same
phenomenon.
Consider the following German forms:
(27) Singular Plural

Bruder Brlulder ‘brother’ Brlylder Briider
Sohn Sfo]hn ‘son’ Slolhne Séhne
Tochter T[o]chter ‘daughter’ T[ce]chter Téchter
These singular—plural alternations illustrate the well-known German phe-

nomenon of vowel mutation, or UMLAUT, which involves the fronting of
the vowel in the plural (as you can see in (27), vowel fronting is indicated
by a dieresis in the German spelling):
(28) Plain vowel Unmlauted vowel

[u] ly]
[o] [o]
[5] [ce]
Not all German plurals are umlauted, however:
(29) Blume Bl[ulme ‘flower’ Bl[u]Jmen Blumen

Rose R[olse ‘rose’ R[o]sen Rosen
Rof R[olé ‘horse’ _R[o]sse Rosse
Indeed, plural umlaut appears as an idiosyncrasy in contemporary German.

Historically, German plural umlaut is reducible to an account similar to
the analysis of Turkish. In particular, the class of nouns that underwent umlaut
had a plural suffix -[i] in Old High German, some thousand or so years ago.
In (30) we trace the development of the umlauted plural of Sohn ‘son’, slightly
idealized for ease of exposition:
(30) [+back] [—-back]
[dorsal] [dorsal]
Se Oe ie at
Oo
You can see that all we have to assume is that the feature [—back] of the
plural suffix -[i] spread leftward, and dislodged the adjacent [+back] value
associated lexically to the root vowel. There are two obvious differences
between the German vowel umlaut and the vowel harmony of Turkish. First,
the direction of association in German umlaut is the reverse of the direction
of association in Turkish harmony: right-to-left in German vs. left-to-right in
Turkish. Second, the German umlaut involves the substitution of a lexical
feature: Turkish disharmony is of course incompatible with this procedure.
what wayTurki
on of a lexi
The German plural marker -[i] eventually evolved into schwa, completing
the change to the modern form [zona], where the motivation for the umlaut
has obviously been obscured.
A similar situation arose with such derivational suffixes as the adjective-
forming suffix -lich or the adverbial -ig:
(31) Tlold ‘death’ tlo]dtlich ‘deadly’

Br[uJder ‘brother’ brlyldertlich ‘brotherly’
v[olll ‘full’ vicelll+ig ‘fully’
Here the trigger of the umlaut survives into contemporary German: /i/. How-
ever, the umlaut process is no longer automatic, since other suffixes similar
on the surface fail to trigger it:
(32) Mlolde ‘fashion’ mlo]ld+isch ‘fashionable’

R[uJhe ‘silence’ r[uJhtig ‘quiet’
DIST nen tETC d[o)rttig ‘of that place’
Notice that German has two phonetically identical suffixes -ig, reminiscent of
English -ly: the umlaut-triggering adverbial in (31) (vollig), and the adjectival
formative in (32), which does not trigger umlaut (dortig). We can encode this
difference in the contemporary grammar by including an additional [—back]
feature in the lexical representation of umlauting suffixes, as follows:
(33) [-back] [-back]
[dorsal]
[—cons]
a? aech
You can see in (33) that this [-back] autosegment is not linked to any of the
In German sounds that make up the suffix -lich. Instead, it is an extra feature that FLOATS
umlaut, a [—back]
in the lexical representation of the suffix, to the left of the [-back] associ-
autosegment
FLOATS in ated to the vowel. Consider in this light the derivation of tédlich:
the lexical
representation (34) [+back] __Ebackl [-back]
Geir ee [dorsal]
[—cons] [—cons]
[labial]
[+round]
Explain how the introdi

The crucial aspect of the derivation in (34) is that the feature [—back] re-
sponsible for the umlaut in the root is unassociated in lexical representation.
Therefore, in modern German only lexically floating features undergo asso-
ciation in the course of the derivation, in contrast to Old High German, where
we saw in (30) that lexically associated features were liable to further associa-
tion derivationally. Suffixes that do not trigger umlaut despite containing
a high front vowel simply do not include such a floating feature in their
lexical representation:
(35) [+back] [-back]
oe ae
Se
[-cons] [-cons]
beat
[labial]
[+round]
in” ed i'Sch
In the purely umlauting plural forms in (27) above, we assume that the
floating autosegment [—back] constitutes a lexical entry by itself. This makes
perfect sense, in as much as the umlaut effected by this autosegment con-
stitutes the only mark of the plural in nouns like Bruder (pl. Briider) or Tochter
(pl. Téchter) (NB the e preceding the final r in the spelling corresponds to
phonetic schwa, which can be considered a phonetic artifact, and therefore
we omit it from the representation):
(36) [+back] __ Pack
Peal ae
[—cons]
[labial]
[+round]
Be ibe wdt
u
[J English Plurals
The regular rule of English plural formation is simple enough: essentially,

add -s, -[z], to the singular: see chapter 19 for specific discussion.
However, English has a number of idiosyncratic plurals, which have to be

memorized — unsurprisingly, children make mistakes on these for quite some
time.
One set of irregular English plurals involves a process of vowel fronting
One English similar to German umlaut, as we illustrate in (37):
irregular plural
involves a process
of vowel fronting
(37) goose [gus] geese [gis]
similar to German tooth [tu] teeth [tid]
umlaut
As in German, the process is entirely opaque in the contemporary language.

In Old English, however, there was a plural suffix -[i], again similar to German.
In addition, but irrelevantly for our present concerns, the root vowels in these
words were [o] in the singular and [e] in the plural, as still reflected in the
spelling (more on this in chapter 8):
(38) Singular Plural

gos gosi
to8 toi
What happened next is easy to guess. Specifically, the [-back] feature of the
suffix spread to the root vowel, thus fronting it:
(39) [+back] [-back]
fdoreall [dorsal]
[—cons] [—cons]
[labial]
[+round]
s Oo s i
2
Subsequent developments involved the unrounding of all front round vowels

and the loss of the inflectional suffix -[i]: [ges]. At a later stage, the mid vowels
le] and [o] underwent raising to [i] and [ul], respectively, to yield the contem-
porary [gis] and [gus]: we examine this raising process in detail in chapter 8.
In the modern language, it is possible to analyse the phenomenon along
the same lines as its German counterpart in (36) above, that is, by means of In the modern
a floating [-back] plural marker: language, the
phenomenon can
be analysed by
(40) [+back] [-back] means of a float-
ing [—back] plural
ee marker
[—-cons]
[labial]
[+round]
Zz u Ss
We must of course assume that the front round vowel of the output [gys]
is subject to an unrounding procedure, a synchronic reflex of the historical
event, to yield the surface [gis].
The processes of umlaut and vowel fronting in modern German and English,
respectively, are undoubtedly less well behaved than vowel harmony in
Turkish: their lexical incidence is basically unpredictable, and therefore the
synchronic analysis demands a considerable amount of abstractness. The three
phenomena are, however, essentially identical in kind, and thus reducible
to basically the same analysis.
features are sufficient to distin

each the primary and the second.
approximations of the entire cardir

clear that languages may choose
n
features, some of which will then
LS |
oN
Phonological Processes Involving Vowel Features
Key Questions
How do the phonetic parameters Explain the No-Crossing Constraint.

relevant to vowels convert into How does this constraint account for
phonological distinctive features? disharmony in harmonic systems?
Which of the distinctive features What is meant by “umlaut"?
defining vowels relate to the gen- How do processes operative in Ger-
eral place of articulation features and man and English resemble those of
how? Turkish?
What is meant by [+ATR]? How What is the historical explanation for
does this feature distinguish pairs of umlaut in German?
vowels? Explain how afloating autosegment
What is meant by “lexical under- can be called upon to explain umlaut
specification"? How does it help to in modern German and English.
explain harmony phenomena?
Further Practice
More Turkish
In section 4 we discussed backness harmony in Turkish vowels and displayed

a feature matrix for four vowels. Turkish, in fact, has eight vowels:
i wy oo
= tao. 8
(i) Which further feature(s) will allow for the additional four sounds?
(ai) Show the complete feature matrix of Turkish vowels.
Now consider the following forms:
Nom.sg. Gen.sg.
‘face’ jyz jyzyn
‘stamp’ pul pulun
‘village’ koj kojyn
‘end’ son sonun
(iii) List the genitive suffixes.

(iv) Following from what you learned in this chapter, can you postulate
one single underlying representation for the genitive suffix?
(v) Show how each surface form is obtained.
Now consider the data below:
Nom.pl. Gen.pl.
‘face’ jyzler _—jyzierin
‘stamp’ pullar pullarwn ©
‘village’ kojler keojlerin
‘end’ sonlar sonlarum
(vi) Explain these cases.
Finnish Vowel Harmony

The Finnish vowel system is as follows:
teaeY. u
e @ O
ze a
(i) List the distinctive features of the vowel sounds.
Many Finnish suffixes occur in two different forms according to features of

the stem vowels. Consider the examples listed below (notice that some of
the vowels are “transparent” to harmony):
talo-ssa ‘in the house’ kylee-ssee ‘in the village’

turu-ssa ‘in Turku’ keede-ssee ‘in the hand’
pori-ssa ‘in Pori’ venee-ssze ‘in the boat’
porvoo-ssa ‘in Porvoo’ helsinni-ssee ‘in Helsinki’
tuo-ko ‘that?’ taeme-koe ‘this?’
tuo-ssa-ko ‘in that?’ tae-ssae-ko ‘in this?’
naise-lta ‘from the woman’ _ tyte-ltee ‘from the girl’
sisare-lta ‘from the sister’ velje-ltee ‘from the brother’
(ii) What are the alternating forms of the suffixes meaning (a) ‘in’, (b)
interrogative and (c) ‘from’?
(iii) | Which feature undergoes harmony?
(iv) What causes the harmony?
(v) | Which vowels are transparent to the process?
(vi) The data can best be explained if we assume one of the values of the
harmonizing feature to be filled in by a default rule. Which is the
default value?
(vii) Write the default rule.
(viii) Write a rule to provide the non-default value.
(ix) In what order must these two rules apply?
CHA
In chapter 5 we provided the inventory of the cardinal vowels developed

by Daniel Jones as a set of reference points to aid with the description of
the vowels of the world’s languages: sixteen cardinal vowels in all, eight
primary and eight secondary. We saw that these vowels can be produced
mechanically by any healthy human by following a set of simple articu-
latory instructions, or by targeting certain perceptual points defined by ref-
erence to the articulatorily established vowels. We pointed out that the
idealized nature of the cardinal vowels makes absolute sameness with the
vowels of any natural language unlikely. Nevertheless, the vowels of nat-
ural languages are whenever possible identified with a neighbouring car-
dinal vowel, for obvious ease of reference, while of course making explicit
in the description what the differences are between the cardinal vowel and
the vowel in question. In the present chapter, we adopt this strategy to sur-
vey the complete inventory of English vowels.
Variation in English
The vowels of Eng-
lish are subject to
We have already cautioned in chapter 5 that the vowels of English are sub- remarkable vari-
ject to remarkable variation world-wide. This means that there is no single ation world-wide
English vowel system or inventory, but, rather, very many. This situation
170 The Vowels of English
obviously does not make easy the presentation of the English vowels in a
unified chapter.
The strategy we shall adopt is as follows. We will build our discussion
on the set of cardinal vowels we presented in chapter 5 — indeed, we shall
follow a similar order of presentation. For each IPA vowel relevant to
English we shall list a number of English specimens, ranked from closest to
most distant, each with the appropriate geographical or social identifica-
tion. Our goal is, of course, not to survey the wealth of pronunciations of
English vowels, but to enable readers to home in on those they are most
familiar with, in particular their own, and in this way identify the vowels
experientially — vowels, a bit like wines, can only be properly understood
when they are tasted.
Some accents of English are of course more “standard” than others, and
Some accents of accordingly they are better known. Two of these accents actually stand out
English are of as being widely recognizable: North American GA and British RP.
course more
“standard” than
others
we intro
These two accents will therefore play a central role in our discussion, but
we will make incursions into other varieties when useful. It goes without
saying that these other varieties are in all as worthy as their better-known
counterparts: in language, as in biology, what we each have is by definition
best, prestige and social recognition aside.
As a preliminary to our descriptions, we will give an overview of the main
accents of English across the world, and of the principal characteristics of
each. English, of course, originated in Britain, and consequently it is here
that we find the most fragmentation in accent and in dialect. The main accents
in the British Isles, with their respective chief distinctive traits at present,
are as follows:
¢ Scotland: rhoticity, no distinctive vowel length, tendency for pure

vowels, [j] in new, aspiration in when, only dark Is, no distinct vowel in
foot relative to food.
¢ Northern England: mostly non-rhoticity, occasional linking and intrus-
ive r, distinctive vowel length, tendency for pure vowels, [j] in new, drop-
ping of h, no aspiration in when, no distinct vowel in cup relative to put.
¢ South West England: rhoticity (even hyper-rhoticity), distinctive vowel
length, diphthongization, [j] in new, dropping of h, no aspiration in when.
¢ London Cockney: non-rhoticity, linking and intrusive r, distinctive vowel
length, dropping of h, glottal stop for t, other plosives, and even fricatives;
The Vowels of English 171
extensive vowel shifts, no aspiration in when, vocalization of final 1, vowel

nasalization before nasal consonants, dental fricatives replaced by labio-
dentals (three — free).
e RP: non-rhoticity, linking and intrusive r, distinctive vowel length,
diphthongization, [j] in new, no aspiration in when, [a] in bath set, pre-
glottalization of stops, simplifications of diphthongs in some environments.
e Estuary English: a variable accent intermediate between RP and Cock-
ney, widespread over south-eastern England and spreading to other parts
of the south.
¢ Southern Ireland: rhoticity, full range of vowels before r, no dental fricat-
ives [8, 6] (dental stops [t, d] instead), intervocalic t weakening (t > [t] >
[r]), aspiration in when, [a] as only reduced vowel, clear / in all positions,
“dark” r ([x]), [e] and [o] monophthongs.
¢ Northern Ireland: rhoticity, no distinctive vowel length, no [u], intervocalic
t tapping ([r]), tendency for pure vowels or centring diphthongs (see sec-
tion 8 below), clear / in all positions (except in Belfast), retroflex rhotic
r ([q]), aspiration in when.
RHOTICITY refers to the occurrence of the sound represented by r in all

positions, whereas in non-rhotic accents r only occurs before vowels (we talk The expression
“non-rhotic”
about HYPER-RHOTICITY when rs turn up where they shouldn’t, as in the
refers to the
pronunciation of china as china[1]). DISTINCTIVE VOWEL LENGTH means absence of [4] in
that at least some vowels are kept apart by their length, only or principally: all positions but
one member of the pair will be short, and the other long. DIPHTHONG- before a vowel
IZATION refers to the tendency of many vowels in many accents of English
not to be uniform from beginning to end, the way pure vowels are. LINK-
ING r and INTRUSIVE rrefer to pronouncing an r between a non-high vowel
and a following vowel, whether or not that r is pronounced when the word
is said in isolation, and whether or not it appears in the spelling: cf. hair and
nails (“linking”), law-r-and order (“intrusive”), and the like.
heck the features of your

Y
Of the accents just listed, the Scottish vowel system clearly stands apart:
in fact, it lies closest to the cardinal vowels. Consequently, we will make The Scottish vowel
system lies closest
special use of it in our description, even though numerically it is a minor-
to the cardinal
ity accent. The accent of northern England (itself varied, like most others, vowels
but we cannot go into many details here) is in some respects half-way between
Scottish and southern English, and we will also find it of use at times, with
special reference to the accent of (West) Yorkshire. The main accent of
Britain is, of course, RP, the prestige of which indeed spreads beyond the
white cliffs of Dover. RP is probably the most studied and best described
of all English accents, and therefore we will make ample reference to it
throughout the chapter. While related to RP, the London accent diverges
in a number of important traits, most particularly its traditional Cockney
variey. Finally, the English south-west converges with North American Eng-
lish in many of its features, not least in its rhoticity and its dislike of low
rounded vowels. These traits are shared with Irish English, itself divided
into the accents of Northern Ireland and the Irish Republic, the former
related to Scottish.
North American English of course has the largest number of speakers
of any variety of English. Its accent is far less diversified than that of the
English is far less old metropolis, for obvious reasons of more recent history. Indeed, it is only
diversified in
in the east, where the colonists first settled, that anything resembling the
North America
than in the
heterogeneity of England can be observed, and only to a limited extent at
British Isles that. While some regional and individual variation is inevitable in a coun-
try of the size and demographic weight of the USA, only four of its accents
are worth differentiating for our present purposes:
e Eastern New England: non-rhoticity, linking and intrusive r, centralization

of the vowel of father, rounded vowel in cot (and caught) distinct from
cart (and, of course, cat), occasionally [j] in new.
¢ New York City: non-rhoticity, linking and intrusive 1, “she back
vowel in father and often in nice, same vowel in cot as in cart, no aspira-
tion in when.
¢ The South: non-rhoticity, no linking or intrusive r, monophthongs
diphthongize and some diphthongs monophthongize, [j] in new, south-
ern drawl.
¢ General American: rhoticity, no [j] on new, cot (and possibly caught) with
the (unround) vowel of cart, flapping of intervocalic ts, aspiration in when,
dark | after vowels.
The r of the There are of course other accents in the area, but they are not very dis-
spelling corres- tinct, and we will ignore them to keep things simple. The most notable dif-
ponded to an ference between GA and its three counterparts concerns its rhoticity. This
r-sound in all
is a by-product of history. In particular, the r of the spelling corresponded
positions in all
varieties of English to an r-sound in all varieties of English until the eighteenth century, when
until the eigh- non-rhoticity began to hold sway in the London area, from where it gradu-
teenth century ally spread to other neighbouring zones (the process is still ongoing), and
across the Atlantic. There, it took hold on the eastern seaboard: New York
City, Boston and eastern New England in general, and also the southern states,
through such main centres of population as Richmond or Charleston. How-
ever, since the Second World War a process of uniformization in the direc-
tion of GA has been under way, and some of the old local accentual traits
are on the retreat, to different degrees in different areas and social classes,
although they are still by no means extinct. Further north in North America,
the accent of Canada can be safely subsumed under the label GA, at least
for our present purposes, with the notable exception of one specifically
Canadian phenomenon to which we will refer in due course.
Elsewhere in the world, English is spoken as a first language in three large
countries of the southern hemisphere, each of them endowed with its own
variety:
e Australian English: non-rhoticity, linking and intrusive r, raised front

lax vowels (bad, bed, bid), heavy diphthongization in see and Sue, [j] in
new.
e¢ New Zealand English: non-rhoticity, linking and intrusive r, heavy cen-
tralization of the vowel in bid, other front lax vowels raised (bad, bed),
heavy diphthongization in see and Sue, [j] in new.
¢ South African English: non-rhoticity, no linking and intrusive 1,
strengthening of r usually into a tap, contextual centralization of the vowel
in bid, other front lax vowels raised (bad, bed), far back vowel in father,
[j] in new.
Although the three countries are reasonably far apart geographically, their
accents share important traits, to the extent that it is often possible to refer
to them together under the label “southern hemisphere English”.
There are of course many other parts of the world where English is
spoken, many Caribbean islands among them. However, we obviously have
to draw the line somewhere, and we will ignore these further varieties in
order to keep the chapter within manageable limits.
The Four Corner Vowels
As we have already said, in our exploration of the vowels of English we

will adopt the same order of presentation as for the cardinal vowels in
chapter 5, for similar reasons. We shall therefore start with the four corner
vowels [i], [ze], [a] and [u]. Remember, crucially, that these vowels are
relatively easy to define articulatorily.
Let us remind ourselves of the position of these vowels in our familiar
chart (figure 7.1).
et u8
4e a5
Figure 7.1 The four corner vowels
We will now review each English vowel in detail, starting with the corres-
pondents of cardinal vowel no. 1, [i], as in heat, seed or key. The correspond-
ents of this vowel in English are as follows: ?
e [i] in Scottish English and the English of other Celtic areas

e [i] both in conservative or careful RP and in the GA short variant (heat)
e otherwise slightly diphthongized in RP and GA (seed, key) and in
accents of southern England (see section 6 below), particularly in the
longer variants
e clearly diphthongized in Australian English and in Cockney (see section
6 below).
Due
Figure 7.2 The vowel of heat

our own prot of this vowel. Do you reckon you

ronounce a
We have just made reference to short and long variants. Indeed, in GA,
in RP and in many other accents of English (but not in all), vowels auto- In many accents
matically vary in length according to the nature of the following consonant: of English, vowels
automatically vary
they are longest if there is no such consonant (key), and shortest if the fol-
in length accord-
lowing consonant is a voiceless obstruent (heat), with intermediate degrees ing to the nature
of length otherwise (seem, seed). This length variation takes place irrespect- of the following
ive of the identity of the vowel. The IPA transcription system includes a spe- consonant
cial length diacritic “:” for long vowels, hence k[i:]. In turn, half-long vowels
are assigned the diacritic “'”: s[i']m. To keep the representations simple,
however, we will only make use of these diacritics where length is directly
relevant to the discussion. The IPA transcrip-
Going now over to cardinal vowel no. 4, [ze], we already know that is low tion system in-
and front. The most common accents of English, among them traditional RP, cludes the special
length diacritic “:"
southern English and GA, do not have this precise vowel. In accents that
for phonologically
do, or almost, the spelling can be disconcerting: long vowels, and
a half-length dia-
e the vowel of hat in typical Yorkshire and Southern Irish accents: [het] critic for vowels of
e the vowel of heart in Yorkshire, Australia, New Zealand and the tradi- intermediate length
tional (non-rhotic) Boston accent: [heet]
e the vowel of hot in the North Central area of the US: [het].
&
Figure 7.3 The vowel of hat in Yorkshire and Southern Ireland
You can see that the spelling needs to be interpreted in the context of each
The spelling needs
particular accent. For instance, the vowel in both Yorkshire and Boston heart
to be interpreted
can sound like the vowel in Yorkshire hat, although the RP and GA pro- in the context of
nunciation of these two words would be quite distinct. The point of the each particular
present exercise is, of course, not so much to untangle the relationship be- accent
tween spelling and sound as to provide reasonable illustrations of the basic
sounds of English, and of their relationship with the cardinal vowels.
The more standard vowel sounds in heart and scotch will be presented below.
In some accents of English, among them traditional RP and varieties of GA,
the vowel in hat, while close to cardinal vowel 4, is a raised [ez], in effect a
half-way vowel between cardinals 4 and 3: [el].
The occurrence of this vowel in RP at the time of the formation of the IPA
alphabet led its promoters to assign the special “ash” symbol [z] to it, quite
exceptionally, since variants of cardinal vowels are usually indicated by
diacritics. As you know from chapter 5, though, we have decided to adopt
the ash [z] for cardinal vowel 4, and the plain “a” [a] for the central low
vowel, to avoid ambiguity. In addition to the arguments we offered in chap-
ter 5, we must point out that the intermediate pronunciation that originally
warranted [ze] is now probably a minority pronunciation throughout Eng-
lish. In particular, over the past few decades, the traditional RP [z] of hat
has been lowering back to [z], and in the pronunciation of many RP speakers
this vowel is now probably as good a representative of cardinal vowel no.
4 as is the Yorkshire [ze]. Moreover, the vowel of hat is raised to cardinal
vowel [e] in much of the southern hemisphere and, possibly diphthongized,
in Cockney. The raising can go further in North America, also usually
accompanied by diphthongization. Indeed, in New York City, for instance,
the vowel in bad can overlap with the vowel in beard, discussed in section
8 below. We now list the English correspondents of cardinal vowel 4:
¢ [e] or [e] in traditional RP and GA

e [z] more and more in RP
e [e] in the southern hemisphere and in Cockney (here possibly diph-
thongized)
e [el], [e] or even a lax [i] (See section 4 below), all with diphthongization,
in North America.
(See Figure 7.4 on the next page.)
your pronunciatior with those listed
Obviously, of the instantiations listed above only the second corresponds to

the cardinal vowel [e] per se, the others falling outside its range. Indeed,
Figure 7.4 The vowel of hat across accents
as we have already cautioned and will be seeing all along, lack of identity
between English vowels and cardinal vowels is the common situation.
Turning now to the back vowels, in chapter 5 we used the words father
[fada] and hoot [hut] to exemplify cardinal vowels nos. 5 and 8, respectively.
As usual, the English vowels tend not to coincide exactly with their car-
dinal counterparts, which have a somewhat more extreme articulation. In
particular, English [a] tends to be slightly advanced with respect to car-
dinal vowel no. 5, and English [u] slightly lowered relative to cardinal vowel
no. 8. Moreover, English [u] is advancing towards the centre in a number
of accents.
The range of correspondents of cardinal vowel no. 5 in English words like
father is as follows:
e a literally back articulation [a] in New York City and, in particular, in

South African English
e aslightly advanced variety [q] in RP and in GA
¢ a central variety [a] in Scotland and in some accents in the US
e front varieties [ze] in many parts of northern England (Yorkshire, for
instance), in Australia and New Zealand, and in eastern New England.
Figure 7.5 The vowel of father

In turn, the English correspondents of cardinal vowel no. 8 are as follows:
e back and slightly lowered [u] in conservative RP (possibly slightly cent-

ralized), in GA short variants (hoot), and in southern Irish English
¢ centralized towards [#] in Scotland, in the southern hemisphere, in the
US south, and in popular speech in England, advanced into the front area
in some accents
¢ also centralized towards [u] along the US Atlantic coast as far north as
New York City, and in much of the central midland
e diphthongized to some degree in long variants in GA (mood, whom) and
across the board in most other accents
¢ clearly diphthongized in Cockney and Australian English (see section 6
below).
61c
Figure 7.6 The vowel of hoot
This completes the presentation of the English correspondents of the four

corner cardinal vowels.
Intermediate Primary Vowels
We shall now turn our attention to the correspondents of the intermediate

primary cardinal vowels: [e] and [e] at the front, and [o] and [9] at the back.
Cardinal vowel no. 2, [e], is represented to various degrees of approx-
imation by the vowel in late, as follows:
e literally [e] in Scottish English: late [let], lay [le]

¢ perhaps a little lower in Yorkshire accents, and in GA in words like
vle]cation
¢ diphthongized otherwise in GA, and generally in RP (see section 6
below), where it is also a little lower ([e])
¢ lowered further (in addition to diphthongizing) in Cockney and in the
southern hemisphere.
Figure 7.7 The vowel of late in some accents
Cardinal vowel no. 3, [e], is fairly closely related to the vowel in let or red
in many accents, but not in all:
e usually [e] in Scottish English and Yorkshire English

e slightly higher in GA: [e]
e even higher in RP, but not as high as [el]: [e]
e diphthongized to various degrees in the American south.
10
FM
Figure 7.8 The vowel of let

For the back primary cardinal vowels, cardinal vowel no. 4, [o], partly paral-
lels its front counterpart [el]:
e literally [o] in Scottish English (coat [kot], low [o])

e slightly lowered and possibly advanced in Yorkshire accents
e tending to diphthongize in GA (see section 6 below)
¢ consistently diphthongized in RP (see section 6 below)
¢ the first part of the diphthong can be centralized ([e]) and unrounded
([3]) — we return to this matter in section 6 below.
Figure 7.9 The vowel of coat in some accents
The correspondences of cardinal vowel no. 6, [9], are as follows:
e [5] in words like caught in Scottish and Yorkshire English: [kot]

¢ a lower [9] or [p] in GA: [kot] /[kot]
e canbe lowered further to [p] in North America, from where it can unround
into [a]: [kot], [kat]
¢ in RP, raised and pretty heavily rounded: [k9t] (the underscripted dia-
critic “,” signals extra roundness)
e can be diphthongized towards schwa from [p]/[9] in New York City, in
the American south and (when not prevocalic) in Cockney.
Vv
FO
OO
40
Figure 7.10 The vowel of caught
It may be worth pointing out that the RP vowel of caught is similar in

quality to the Spanish vowel [o] in loco [loko] “mad” — it is half-way
between cardinal vowels nos. 6 and 7: Spanish [o] obviously does not cor-
respond to cardinal vowel no. 7 either. The RP vowel of caught is, however,
considerably longer and tenser than the Spanish vowel of loco, and has more
lip rounding.
We round off the section with a chart of the primary cardinal vowels, and
the charts of the related vowels of Scottish, GA and RP English alongside,
to facilitate cross-comparison (figure 7.11).
Primary cardinal vowels Related Scottish vowels

1 u 8 1 u
Ze 07 e O
3€ 26 € A)
4e a5
a
Related GA vowels Related RP vowels
i Urirad u
e O
oD
e € >
7 3
& a x
a a
a +
Figure 7.11 Four sets of basic vowels
£3 More Lax Vowels

We shall now examine four additional non-central lax vowels of English.
The first such vowel is the one in words like cot or rock. We have already
mentioned that Scottish English has a vowel equivalent to cardinal vowel
no. 6 here, similar to the French vowel of cotte ‘overalls’. In Yorkshire accents
this vowel is remarkably like secondary cardinal vowel no. 13, [p]. You will
recall that secondary cardinal vowels are produced by reversing the lip posi-
tion of their primary counterparts. This means that secondary cardinal
Secondary cardinal vowel no. 13 will be identical to primary cardinal vowel no. 5 in all but lip
vowels are pro-
rounding — secondary [bp] is rounded, whereas primary [a] is not. The set
duced by revers-
ing the lip position
of correspondents of this vowel across English accents is as follows:
of their primary
counterparts e literally [bp] in Yorkshire English: [kot]
e also usually [p] in GA before r in such words as horrid, orange, forest, and
after w (w[pv]ter, w[p]sp, w[v]tch): the latter also in midwestern and west-
ern areas
e [p] generally in eastern New England and parts of the coastal US south
e slightly raised in RP and similar accents: [p] to [9]
e unrounded to [a] in the English south-west and in southern Irish English
e also unrounded to [a], and then often advanced to [q] or even [a], in
most words in GA and in most of the US south
¢ further centralized to [a] in New York City (and possibly diphthongized)
and in the north central area of the USA.
2
i
Figure 7.12 The vowel of cot
_Avempt ie pronunciation of.‘ot wi

fal an, [9], to et a favour of th
The next two vowels we will consider are not cardinal vowels, primary
English has two or secondary. Instead, they can be construed as particularly lax pronuncia-
vowels that can
be construed as
tions of [i] and [ul], as we shall explain.
particularly lax The front lax vowel occurs in such words as hit or lick. If you compare
pronunciations of these words with their close phonetic correlates heat and leek, you will
[i] and [u] notice that the vowel in heat involves considerable tensing of the vocal appar-
atus, in a way that the vowel in hit or lick obviously does not. This tensing
is often thought to go hand in hand with an advancement of the root of

the tongue, captured by a positive specification of the distinctive feature
[ATR] introduced in the preceding chapter. According to this interpreta-
tion, the forward movement of the root of the tongue would be present
in the [i] of heat and leek, and absent from [t], its lax counterpart in hit or
lick. Laxing and lack of tongue root advancement result in a certain degree
of lowering relative to [i], and also in a certain amount of centralization
(figure 7.13).
bi
2e
Figure 7.13 The vowel of hit in GA and RP
On simple visual inspection of figure 7.13, we could be tempted to inter-

pret the new vowel [1] as a raised, retracted version of [e], namely [e]. This
interpretation would, however, be at odds with the disparity in tensing
between the two vowels, since in English [e] is tense, and [1] is lax. Note
also that tensing tends to push vowels towards the edges of the vowel chart,
and [1] is characteristically centralized. Moreover, as we show below, and
again in the next chapter in more detail, [1] pairs up phonologically with [i],
not with [e] or [e]. We conclude from all this that [1] is the lax counterpart
of [i]. Because [1] plays a distinctive role in the vowel system of English, it
is endowed with its own individual phonetic symbol, in fact two, since an
alternative to [1], also accepted by the IPA, is [1].
The correspondences of [1] across English accents are as follows:
e [1] in GA and RP
e raised in Australian English: [1]
¢ considerably lower and/or more retracted in popular Scottish accents
¢ centralized to [z] in New Zealand
e centralized to [¢] in some contexts in South Africa, and raised to almost

[i] in others
¢ diphthongized generally, and centralized to [], except before a velar, in
the US south.
Figure 7.14 The vowel of hit across accents
The back counterpart of [1] occurs in such words as hood, put or look. We
will analyse it as a lax, [-ATR] version of [u], matching the relationship
between [1] and [i]. The vowel in hood and its companions is usually tran-
scribed as [vu], although the symbol [@] is also countenanced by the IPA. As
in the contrast between [1] and [i], [u] is lower and more central than its tense
counterpart [ul].
u8
o7
Figure 7.15 The vowei of hood in GA and RP
The proximity of [u] to [o] in figure 7.15 is reminiscent of the proximity of

[1] to [e]. As in the case of the [1]-[e] pair, however, this spatial closeness
does not correspond to a phonetic or phonological relationship, for precisely
the same reasons.
The vowel [u] is missing from Scottish and Northern Irish English.
Scottish English substitutes [u]/[u]. In many other accents, [u] is under-
going centralization, and unrounding, a pronunciation [y] being far from
uncommon nowadays (the underscripted diacritic “” stands for loss of round-
c
ing). For completeness, we now list the correspondences of [u]:
¢ traditionally [uv] in RP, GA and most other accents (but not in Scottish
and Northern Irish English)
¢ progressive tendency to centralize and unround in many accents
* some tendency to diphthongize in North America.
Figure 7.16 The vowel of hood across accents
The last vowel we shall examine in this section occurs in words like hut,
up or cuff in GA and in Scottish English. It resembles [a], the secondary vowel
that corresponds to the primary vowel [5]: [a] is [5] without the rounding.
The English vowel is, however, slightly advanced vis-a-vis its cardinal coun-
terpart, hence [al]:
e advanced [a] in GA and Scottish English: [a].
Figure 7.17 The vowel of hut in GA and Scottish English

As we said in chapter 5, the same pronunciation ([a] or [4]) existed in RP

before the Second World War, but its position has now shifted. We exam-
ine the new RP vowel in the next section. In northern England this vowel
simply does not exist, words like hut, up or cuff having the same vowel as
foot: [vu]. }
Central Vowels
In present-day RP, and in RP-like accents, the vowel in hut, up or cuff cor-
responds to a raised central low vowel, for which we said in chapter 5
that the IPA official table provides the symbol [e] (= an inverted “a”).
Figure 7.18 The vowel of hut in RP
Confusingly from a phonetic perspective, this RP sound [e] is still usually

Confusingly from transcribed with the symbol [a], like its backer and higher GA and Scottish
a phonetic per-
counterpart.
spective, the RP
sound [e] is still
We now list this and other correspondents, to be added to the Scottish
usually transcribed and GA [a] or [4] we introduced at the end of the previous section:
with the symbol
[A], like its backer ¢ acentral raised low vowel [ev] in RP and RP-like accents
and higher GA
“
e raised to schwa [a] in Wales
and Scottish
counterpart e further raised to [9] in much of the American south.
Figure 7.19
ier
The vowel of hut across accents
] Which vowel do youproduce in I
Moving up within the central region, we enter the area of schwa, [a]. We
mentioned in chapter 5 that schwa is extremely frequent in English, repres- Schwa is
enting as it does the stressless, “reduced” pronunciation of most vowels, °*tremely frequent
although some stressless vowels reduce to [1] or [u], rather than to schwa. —" S"slish
We now offer a small sample of English words containing schwas, with the
letters that represent the phonetic schwa underlined:
ago collect telephony atom atomic column hippopotamus
Next, we incorporate schwa into our chart of English central vowels, delib-
erately keeping its spatial range somewhat fuzzy.
Figure 7.20 Schwa added
The realization of schwa is remarkably uniform across accents, perhaps

precisely because of its loose range. However, some Scottish accents sub-
stitute [a] for schwa. In rhotic accents, schwa can be “r-coloured”: [a]. The
expression R-COLOURED refers to the [1]-type quality superimposed on a
vowel, usually achieved by curling the tip of the tongue up, in a gesture of
retroflection.
The last central vowel we will present constitutes a strong version of schwa,
and occurs in words like bird or lurk. This is in fact the vowel that we referred
to in chapter 5 as likely to be used by English speakers for the rounded
vowels [9] and [ce] of other languages. This vowel is usually transcribed as
[3] when non-rhotic, and as [3] when rhotic. The symbol [3] used to share
the centre of the vowel space with [a] in the IPA chart. However, in the
latest version of the chart the symbol [3] is allocated a well-defined posi-
tion, as a mid-low unround central vowel, with the mid-high central vowel
space assigned to [9].
Figure 7.21 Central mid-vowels
Practise the sound

in e.g. about.
The implication that [3] is a mid-low vowel seems appropriate for RP, but
not necessarily for other accents. The range of correspondents of [3] is in
fact as follows:
e [3] in RP, most RP-like accents, and eastern New England

e r-coloured [3] in GA and other rhotic accents, Scottish excepted; also (curi-
ously) in some non-rhotic accents
e raised to [9] or even [i] in Birmingham and Liverpool (England) and much
of the southern hemisphere
¢ centralized and rounded in New Zealand, and possibly raised: [ce] to [6]
e a diphthongized [3] in traditional New York City and some US south-
ern states (see section 8 below).
Figure 7.22 The vowel of bird

Try out all these va

" possibiliti
It is worth noting that the lexical incidence of this vowel is severely restricted
in Scottish English, many varieties of which allow practically all lax vowels
to precede a final r, in striking contrast with most other accents of English.
We have now completed the inventory of the simple vowels of English.
We have seen that many of these vowels have a tendency to diphthongize
— we examine diphthongs in the next section. Despite this diphthongizing
tendency, it is useful to consider some of the vowels as primarily pure, and
others, still to be reviewed, as primarily diphthongal. Paradoxically, we will
see that some diphthongs actually tend to be pronounced monophthongally
in some accents, and some pure vowels tend to diphthongize.
At this point it will be helpful to chart all the basically pure vowels, of
which there are twelve in RP and eleven in GA (also twelve if we count in
the r-coloured schwa). We will display these systems in parallel with the
cardinal vowels to make cross-comparison easier (remember that RP [ev] is
usually (mis)transcribed as [a], the IPA symbol for the unround mid-low
back vowel).
11 y9 16W Us
i u
¢ 9
E ;
exe |,
a
+
Figure 7.23B RP pure vowels

41 1c
410
a
+2
Figure 7.23C GA pure vowels
Finally in the section, we tabulate all the RP and GA vowels we have exam-
ined, against the classificatory criteria we have used (table 7.1).
Table 7.1 Parametric classification of RP and GA pure vowels
Front Central Back
Round Unround
Tense Lax Tense Lax Tense Lax Tense Lax
High i I u U
Mid Ge ae a, & 22 A
Low ze e D qa
[4 Homogeneous Diphthongs
In our discussion of “pure” vowels above we have frequently commented

that some particular vowel has a tendency to diphthongize in some accent.
In the remainder of the chapter we turn our attention to this type of
phenomenon.
We said above that words like late and coat have the mid-high primary
cardinal vowels [e] and [o] in in Scottish and Yokshire English, and to a
limited extent in GA. Elsewhere, words like these have diphthongs.
Thus, in the majority accents the vowel in lay or raid is not a pure vowel,
with a constant quality throughout, but rather a diphthongal vowel, with
the sound quality changing half-way through. The phonetic tools we have
at our disposal allow us to transcribe this vowel as [e1]. This is a vowel of
non-steady realization: its first phase corresponds more or less to cardinal
vowel no. 2, [e] (usually a little more open: [e]), and its second phase to the
now familiar vowel [1]. The articulation of this complex vowel, or DIPH-
THONG, in fact glides from [e] to [1]. The standard transcription [e1] obvi-
ously suggests two independent vowels and, to this extent, it is misleading. The transcription
[el] obviously sug-
One way around this problem involves the addition of a tie bar, thus [er].
gests two vowels
However, we already know that phoneticians favour simple representations and, therefore, it is
and often ignore phonemically irrelevant phonetic detail. Indeed, in the specific slightly misleading
case of English diphthongs, the inclusion of the tie bar is the exception rather
than the rule in the literature.
One obvious characteristic of the vowel [et], besides the non-steady
nature of its sound quality, is its relative length, as you will find out if you
compare the pronunciation of the vowels in such pairs as late ~ let, raid ~
red, sale ~ sell, tames ~ Thames, etc.
You may be inclined to think that the difference in length between [er] and
[e] inevitably follows from their difference in complexity: according to this
construal, a composite vowel would inherently take longer to pronounce than
a simple one. As it happens, however, English has simple vowels that are
RP English has
as long as complex vowels like [et]. One such long vowel is the low back
simple vowels that
vowel [al] of father, which is pronounced long with a pure, steady sound in are as long as
most accents: f[la:]ther. complex vowels
~—
Paradoxically, though, [et] is never transcribed [em], presumably on the

grounds that the two symbols “e”, “1” already take up two spaces in the
line.
The range of correspondences of [er] is as follows:
e [er] in RP, in Southern England in general, and generally in GA

e [er] in Cockney and in the southern hemisphere
¢ monophthongal [e] in Scottish and northern English
¢ monophthongal [e] ([e] in a few words like great) in parts of the US south,
rediphthongized as a centring diphthong before a consonant in some areas
(see section 8 below).
Figure 7.24 The vowel of late
The diphthong corresponding to [e1] at the back is [ou]. This diphthong

turns up in many accents of English in such words as coat or foam ([kout],
[foum]), but GA is reported to exhibit a certain tendency to a monophthongal
realization, in particular when the vowel is short ([kot]). This monophthong,
identifiable with cardinal vowel no. 7, is general in Scottish English, and also
in much of northern England. When diphthongal, the first phase of the diph-
thong is actually central in many accents. The specific range of diphthongal
correspondents of this sound is as follows:
¢ [ou] or [0°] in GA, particularly in long variants: dome, low (shorter vari-
ants as in coat can be monophthongal)
¢ lowered to [pu] before a final / in London and the southern hemisphere:
dole [dput] (or even [dp:4] in South Africa)
¢ central start in the middle Atlantic and western Pennsylvania areas of
the US: [eu]
e¢ unround central start in RP: [su]
¢ further lowered start in London and the southern hemisphere, [eu],
undergoing fronting in younger speakers.
Figure 7.25 The vowel of coat
The high front vowel in heat, seem or see is also intrinsically long in most
accents — in Scottish English vowels do not carry any intrinsic length dif-
ferences, however.
We have already said that the quality of this vowel varies from pure [i:]
through [ti] to [31]. The geographical distribution of the diphthongal real-
izations is as follows:
* tendency to diphthongize to [ii], particularly in longer variants (seem, see)

in GA and in accents of southern England
e definitely diphthongized and lowered as far as [31] in Australian and
Cockney English.
Figure 7.26 The diphthongized vowel of heat

The long high back vowel [u:] undergoes diphthongization under the same
circumstances as its front counterpart [i:]: [vu], [ou] or [su], hence [fuud],
[foud] or [fsud] for food. The distribution of the diphthongal realizations of
this sound is as follows:
¢ tendency to diphthongize to [vu], particularly in longer variants (mood,

who), in GA and in accents of southern England
e centralized, diphthongized and lowered as far as [su] in Australian and
Cockney English.
Figure 7.27
Nie
The diphthongized vowel of food
The “breaking” of long vowels into diphthongs is not unusual across lan-
Diphthongization, guages, as if the articulators got tired or bored of maintaining the same sound
or “breaking”, of
throughout and opted for change. Indeed, this is the origin of [er] and other
long vowels is not
unusual across
English diphthongs — English [et] was a monophthong [e:] at some previ-
languages ous stage.
Of the diphthongs we have discussed in this section, we are considering
both [fi] and its variants, and [vu] and its variants, as realizations of the pure
vowels [i:], [uz]. By contrast, we are considering [et] and [ou] as primarily
diphthongal. The reason for this different treatment lies in the respective fate
of these vowels in the more prestigious accents, contemporarily and histor-
ically: we have in fact seen that the four vowels can be realized as diph-
thongs or monophthongs, depending on the accent.
~ The diphthongs [e1] and [ou] are homogeneous inas much asboth phases
ofthe diphthong are close in articulatory position and share the lip gesture. The diphthongs
In the next section we examine three other primary diphthongs of English, [ex] and [ou] are
homogeneous in
which are heterogeneous. Insection 8 we will examine a final set of English as much as both
diphthongs, which mainly arise as a result of contact with [i]. phases-of the
diphthong are
dose in articula-
Heterogeneous Diphthongs tory position and
share the lip
gesture
The set of heterogeneous English diphthongs has three members, instanti-
ated in the words buy, bough and boy.
A common realization of the diphthong in buy has a central low vowel
[a], in its first phase, and a high front lax vowel [1], in its closing phase, hence The set of hetero-
(ai). Forms with this diphthong include buy, eye, sigh and ice. The heterogeneous English
geneous nature of [a1] hinges on the fact that itcombines a low central vowel diphthongs has
articulation ({a]) with a high front vowel articulation ([i]). The obvious three members
question that arises is how such a divergent combination could ever have
arisen. The answer has already been hinted at: vowel sounds, including vowel
sounds that make up diphthongs, can and do move around the vowel space
as time goes by. Thus, unlikely as it may seem, the historical source of the Vowel sounds can
and do move
diphthong [ai] is a long vowel [i:] similar to the [i:] of heat in contemporary
around the vowel
English. This vowel is attested to have lowered to [es] (perhaps after a stage space as time
as [is]), eventually yielding [a1] after centring and further lowering: all this goes by
is discussed in more detail in chapter 8 below. To make these matters clear,
we now represent the evolution of [i:] into [ar] in the vowel chart (the num-
bers correspond to the stages).
In Canada, the vowel [a] of [at] undergoes raising before a voiceless

consonant: bite [bast]. This phenomenon isalso reported for Virginia and coastal
South Carolina. A similar realization [31] (or a lower [er]) in all positions is
typical of rural accents in southern England and in eastern New England.
Curiously, the diphthong [s1] (or something very much like it) substitutes
in traditional New York City speech for the more commonly pure vowel
[31] of bird before a consonant.
We now list the correspondents of [at] throughout the English-speaking world:
e [at] in RP and GA
e [at] in Cockney and much southern English urban speech, in the south-
ern hemisphere, and in New York City
e [a1], sometimes monophthongized to [z] (or even [¢]) in the north of
England
e [a] generally in the American south
e [er] before a voiceless consonant in coastal South Carolina
e [s1] in rural southern England and eastern New England, and before a
voiceless consonant in Canada and Virginia
e [st] before a voiceless consonant or a voiced stop in Scotland.
Figure 7.29 The vowel of bite
Is the sound you produce in bite a diphthong? Wh

diphthongs listed above doesyour diphthong most|
The second heterogeneous English diphthong is [au], as in how or gout,

with the same first phase as [a1], [a], but with a high back rounded lax vowel
[u] as a second phase. The disparity of lip gesture between the two phases
makes this diphthong more heterogeneous than [ar]. The source of [au] mir-
rors the source of [at], that is to say, a long high back vowel [u:] evolved
into [au] through the stages [uu], [ou] and [au] (we respect the common
transcription of the first phase of this diphthong as a schwa in the histor-
ical literature).
Figure 7.30 Historical evolution of [au]
The diphthong [au] exhibits variation reminiscent of that of [ar]:
e [au] in RP and GA
¢ [gu] in the US south and in popular speech in southern England, par-
ticularly in Cockney, where it can monophthongize to [zx]
¢ [au] in South African English
¢ [vu] in Scotland, in coastal South Carolina before a voiceless consonant,
and in rural speech in general
e [su] before a voiceless consonant in Canada and Virginia: doubt [dsut].
The last of the three heterogeneous diphthongs general in English is [ou],

represented in such words as boy or voice. The closing phase of this diph-
thong is identical to the closing phase of [ai]: [1]. The opening phase is
standardly represented as [3], hence [1]. Notice that the [5] in [o1] differs
considerably from the [3] of RP caught, which we saw is significantly tenser
and longer. Indeed, the [5] vowel in [51] approximates to cardinal vowel
no. 6, [9], to a degree substantially greater than the vowel in caught in many
accents. While unusually uniform throughout the English-speaking world,
the diphthong [91] still exhibits some variation:
e [or] in most accents, including GA and RP

e [v1] in parts of the south of England
e [eer] or [p1] in parts of Ireland and the US
¢ possibly [31] in New York City, in which case it merges with the tradi-
tional [31] of bird (bird also has [31] in the US deep south).
Figure 7.32 The vowel of voice
You can see that most variants of the diphthong [o1] straddle the vowel
space from back to front, uniquely so among the five standard English diph-
thongs — all the other diphthongs are localized exclusively in either the front
or the back area ([et] vs. [ou], respectively) or involve a rising move-
ment from the centre ([ai], [au], [su] /[eu]), at least in the standard accents.
Figure 7.33 Diphthongs from the centre

The diphthong [51] is also the only diphthong of contemporary English whose
historical origin is exclusively diphthongal: boy, for instance, was already
[bor] in Middle English, in contrast with the diphthongs in time, town, hate
and boat, which at some point were all simple vowels: [i], [ur], [ar] and [o:],
respectively.
El Centring Diphthongs
We shall now examine a final set of English diphthongs, in which the move-
ment is towards the centre of the vowel space.
Consider the following words:
pier poor
dare door
The final r in the spelling of these words is purely orthographic in non-rhotic

accents such as RP, which are defined by precisely this trait. The question
now is: how exactly are these words pronounced in these varieties, granted
that the final r is only orthographic?
The typical non-rhotic pronunciation of the forms in the first column above
is [pia] and [dea], respectively. Two aspects of this transcription are note-
worthy. First, the final sound is a schwa, [a]. Second, the preceding vowel
is lax: [1], [e], respectively. These two properties give a clue to the histor-
ical evolution of these sounds. In particular, before its disappearance in the
eighteenth century, r (a trill at the time) caused diphthongization of the pre- Before its dis-
ceding vowel, and then laxing, so that [pir] pier, for instance, came to be pro- ee
nounced [pier], and correspondingly for the other forms in question. Clearly, fo ied
after the r went, the schwa remained the sole trace of its former presence. diphthongization
The net effect of this process in RP and the other non-rhotic accents is an _ of the preceding
additional set of “centring” diphthongs, that is to say, diphthongs with an _ vowel, and then
initial phase somewhat towards the edges of the vowel space and with the /2xing
central vowel [a] (often lowered as far as [ev] in word-final position) as a clos-
ing phase.
In RP there is
a maximum of
four centring
We now exhibit the four centring diphthongs exemplified at the start of — diphthongs
the section.
Need
Figure 7.34 Centring diphthongs
We have already commented on the [1a] of pier. The back counterpart of [1a]
is [va], as in poor. In RP, forms like door also had a centring diphthong [5a]
until the post-Second World War period, but are now more commonly pro-
nounced with the long tense steady vowel of caught, hence [d9]. However,
in other non-rhotic accents (eastern New England, New York City, US
south) [da] persists. In RP, the vowel of poor, traditionally [va], as we just
said, is also merging with [9] in many speakers, hence [p9:]. The RP tend-
ency to monophthongize centring diphthongs extends to the front mid
vowel: dare is more and more pronounced monophthongally as [de:]. By
contrast, some other non-rhotic accents exhibit a greater number of cent-
ring diphthongs. For instance, in the traditional accent of New York City
the vowel in star is [aa], and in that of eastern New England the vowel
in square is [ga]. In US southern states, the front lax vowels [i], [e], [el]
break into [1a], [ea], [zea] in such entirely r-less words as bid, bed or bad,
with the result that bid and beard may become homophonous. On the other
hand, the RP tendency to monophthongize centring diphthongs is also
present to varying degrees in other accents: for example, in New York City
all centring diphthongs can become long monophthongs: [11], [e:], [az], [5:]
and [ur].
Breaking and laxing before [1] also took place after the set of non-centring
diphthongs [a1], [av], [er], [ov] and [51]. This resulted in the development of
the centring TRIPHTHONGS [ars], [aval], [eta], [ova] and [dra], as in fire, hour,
layer, lower and coyer, respectively. Of these, the first two ([ata] and [aua])
are considered triphthongal more often than the rest, perhaps more for
reasons of morphological makeup than of phonetic realization as such. This
said, it is not unusual for triphthongs to simplify in many accents, either
by breaking up into a diphthong followed by a simple vowel ({at-a], etc.)
or by smoothing up the initial diphthong into a monophthong, as happens

more and more in RP: fire [faa], tower [taa] or [taa], and so on, and similarly
in the American south. A further step in this development involves the
monophthongization of the resulting centring diphthong: [fa:], [ta:], etc.
of contrasts implied in the primary and se

simple vowels, perhaps only three can be cla
case is the English version so extere as

and (al. Of the others some are raised ¢
the phonological featre ATR,

el leads to a certain degree of
of [i], is i fact somewhere between
The present maximal list of een vowels ene sipady state <
diphthongs) is augmented by the so-called centring diphthongs, the
of which we explored. Breaking and laxing after non-centring diphthongs
gave rise to centring triphthongs.
Kie"y> -Q-uens Tons
1 List the basic differences between 7 In which positions in a word do we

RP, GA and standard Scottish find schwa in English? Is schwaa dis-
English. tinctive vowel in English?
2 What is rhoticity? 8 What is a diphthong? Describe the
3. What is meant by “distinctive vowel articulation of a diphthong.
length"? 9 How accurate is the IPA alphabet
4 What are “linking” and “intrusive” for the transcription of English
/r/? Why are there two terms? vowels?
5 How do most English vowels differ 10 What is the origin of the four cent-
from the cardinal vowels? Which ring diphthongs of English?
variety is the exception to this?
6 What is meant by “tense” and
“lax" vowels? What is their phono-
logical feature correlate?
Further Practice
Homophones
The following data contain a number of homophones. They also contain a

number of homographs (same spelling, different pronunciation). Some of the
words pair both ways.
(i) Identify and transcribe the homophones.

(ii) Give the alternative pronunciations of the homographs, also in phonetic
transcription.
pier dear read bead fair red reed pear peer bread breed
pair rite lead fare led choir right bough = quire rough row
bow dough grown pare sore doe saw sow so groan blew
aught blow blue court ought caught gone lone done loan dun
soar write ruff seize _ fleas fleece seas kernel dew berry: )f key
colonel quay bury due cow cough shower flower _ beer bear
Phonetic Transcription
Transcribe the following sentences into normal orthography:
pn 9 klia det ju kn si fa mailz

fast Imprefnz kaunt fai a lvt
je Oo last pssn ai ekspektid ta si hia tader
OI aldia Oat Or 39 Is flat Is toutli udikjalas

hi dagz evii moinin hwareva 0a weda
ail heef ta get samwan Im ta fiks da suf
Put the sentences below into IPA transcription:
(i) for your own accent

(ii) for a standard accent you are familiar with.
That man is far more important than the prime minister

How many professors does it take to change a light bulb?
All work and no play makes Jack a dull boy
They arranged all the Christmas presents around the tree
It is a truth universally acknowledged that a single man in possession of a
good fortune must be in want of a wife
Mr Salteena was an elderly man of 42 and was fond of asking people to stay
with him
Sound to Spelling
Below are a number of vowel sounds. How many spellings in English

words can you find to correspond to the sounds listed?
[at] [er] [a:] [e] [ix] [av] [ov] [jul [5:] [al
THE TIMING TIER AND
THE GREAT VOWEL SHIFT
In this chapter you will learn about:

m An extension of the autosegmental model.
m Alternations which appear to be inexplica
English.
A way of differentiating be
same features.
m The introduction of
m How processes r
In previous chapters we have introduced and substantiated the autosegmental

nature of phonology. In particular, we saw that the very common phenomenon
of partial assimilation between segments can readily be understood if we
assume that each distinctive feature is free to act independently of the other
features it may be associated with: this freedom of action allows each feature
to influence a neighbouring segment irrespective of whether other associated
features do so. In chapter 7 we saw that vowels (English vowels in particular)
can be long or short, and the question arises of how this contrast should
be formalized. In the present chapter we introduce an additional, and most
important, tier into our autosegmental model. This tier is made up of abstract
“timing units”, which associate with bundles of features to indicate their length.
In particular, a short segment will be associated with one timing unit, and a
long segment with two. Conversely, a single timing unit can be doubly asso-
ciated with two values of a feature, as is the case with affricates. The introduc-
tion of the timing tier makes possible a reasonably transparent analysis of the
English Vowel Shift, one of the most important processes affecting vowels
in English. At the end of the chapter we will suggest that the timing tier
The Timing Tier and the Great Vowel Shift 205
constitutes the baseline which supports the remainder of the autosegmental

structure.
A Puzzle with Affricates
In chapter 1 we introduced the affricate consonants [{] and [dg], instantiated

in the English words church and judge, respectively. We explained then that an
affricate is articulated as a stop with a partial release equivalent toa fricative.
Phonologically, the question arises of whether affricates (and other complex

sounds) constitute in fact one or two segments. Notice that the IPA symbols Phonologically,
[f] and [d3] could be interpreted either way: they correspond at the same the question arises
of whether
time to one segment ([t{] and [dg], respectively) and to two segments ([t] + affricates (and
[f] and [d] + [3], respectively). The phonological evidence is also equivocal: other complex
some facts suggest that affricates are monosegmental, whereas other facts sounds) constitute
suggest that they are bisegmental. in fact one or two
Let us examine the evidence for monosegmentality first. English words segments
may not start with two obstruents, setting aside the sequence s + obstru-
ent, which we discuss in chapters 10 and 16. Indeed, when native speakers
of English encounter such words as tsar, psychology or pterodactyl, with an
initial cluster in the spelling, they normally simplify the cluster: [s]ar or [z]ar,
[slychology, [t]erodactyl. The clusters [ts], [dz], [ps], [pt], etc., are, however,
perfectly acceptable in many other languages: in French, for instance, there
is [ps]ychologie ‘psychology’, [ps]eudonyme ‘pseudonym’, [ps]aume ‘psalm’,
[ptlérodactyle ‘pterodactyl’, [ptlolémaique ‘ptolemaic’, [dz]ar ‘tsar’, [dzligane
‘gypsy’, [ts]é-[ts]é ‘tse-tse’, etc.
The simplification of these clusters in English leads us to expect that the
pronunciation of word-initial [ff] and [dj] would also be problematic for
English speakers.
The fact that it is not problematic provides obvious support for a mono-
segmental analysis of these sounds. Indeed, in The Sound Pattern of English
Chomsky and Halle analysed affricates as [-continuant] (= stop) sounds with
the added feature [+delayed release]. The specification [+delayed release]
206 The Timing Tier and the Great Vowel Shift
defines the second, fricative, phase of [ff], [d3], thereby keeping these sounds
distinct from the corresponding plain stops [tl], [d].
"Would you say thatthe d

we discussed in chapte
a single segment?
There are other facts that suggest that [t{] and [ds] are bisegmental, how-
ever. We just mentioned that word-initial sC- clusters are allowed in Eng-
lish. However, s + affricate clusters are not: *[sif], *[sdg]. Similarly, English
words can begin with a single obstruent followed by the liquids [1] and [r],
with certain restrictions which we will discuss in chapter 10: think, for instance,
of pride, plate, trip, crate, clear, with stops, and fry, fly, thrive, slit, shrink, with
fricatives. Affricate + liquid clusters are, however, never found: *chroke, for
instance, is not possible. These two gaps in the distribution of segments would
obviously follow from an analysis of affricates as two segments.
Additional evidence for a bisegmental analysis of affricates comes from the

choice of plural and past tense allomorphs. An ALLOMORPH is a contextual
variant of a morpheme - the term is coined in the mould of “allophone”,
which, as we explained in chapter 2, designates the variants of a phoneme.
A noun like match forms its plural with the same allomorph as forms end-
ing in plain [f] (match[1z], bush{1z]). Similarly, verbs ending in [ff] and in [f]
select the same past tense allomorph (match[t], mash[t]), different from the
allomorph selected by forms ending in a non-continuant coronal like pat and
pad (patt{1d], padd|id]). Pat and pad also select the short allomorph in the
third person singular (pat[s], pad[z]). All this again falls into place under the
analysis of the affricates as [t] + [f] and [d] + [3], respectively.
In order to break the stalemate between the monosegmental and the

bisegmental interpretations of [t{] and [d3], we need to enrich our formal appar-
atus, in the manner that we shall explain in the next section.
The Timing Tier
The machinery in question involves the autosegmentalization of segment

length. We shall develop this formalism in connection with vowels first, before
returning to the issue of affricates.
We have been marking long vowels with the diacritic “1”, which suggests
a doubling of the length of the segment it modifies. In the pair [ze], [ze],
therefore, [a] stands for the shorter vowel and [ze] for the longer vowel.
As we explained in chapter 5, the addition of diacritics to phonetic sym-
bols denotes a difference in the sound represented equivalent to the use of
an independent symbol: [i], for instance, can in principle be construed as
being as different from [i] as [i] is from [1].
Following this logic, the members of the pair [z], [z:] would not stand in
any closer relationship than the members of pairs like [z] and [np], or [a]
and [y]. This conclusion is, however, patently wrong, because [z], [p] and
Ly] are vowel sounds of different quality, whereas the pair [ez], [ze:] contains
one and the same vowel sound, albeit with two different lengths.
In order to achieve the correct result, we shall enlist the help of the autoseg-
mental formalism. You will recall that this formalism allows us to separate
out (conceptually, functionally and graphically) the various elements that
make up a sound we perceive as a unit: our description of English nasal
assimilation in chapter 4, and of Turkish vowel harmony and English and
German vowel fronting in chapter 6, showed how individual features can
(and do) change association loyalties without affecting other features with
which they are also associated at some level. We shall now extend the auto-
segmental approach to the formal expression of segmental length.
; Se hae . : We create a spe-
Segmental length is a matter of timing: when we say that [ze] is twice a
as long as [ze], we are simply saying that [ae:] takes twice the time to say aS nit of inne
[ze]. Suppose then that we create a special autosegmental unit of timing, — which we write
which we shall write as “X”, in line with standard practice. The difference as "Xx"
between [ze:] and [z] will now be represented as in (1):
(ipiiras De ax bialX
ze
[eer] [ee]
The interpretation of these diagrams is straightforward. Formally, in (1a)

the vowel is associated with two timing units, whereas in (1b) precisely the
same segment is only associated with one timing unit. The suggestion is, of
course, that the delivery of the segment in (1a) should take twice as long
as the delivery of the segment in (1b). This statement is correct, provided
we replace “delivery” with “abstract” or “intended” timing, since in real life
the actual delivery of sounds can be (and often is) affected by non-linguistic
factors.
Can you think| yf

_ factors do not affe
This mismatch motivates the distinction forcefully drawn by Chomsky

Chomsky draws between “performance”, that is, language in the real world, and “compet-
a distinction
ence”, that is, the language system that underlies performance, and which
between “perform-
ance" and
must be assumed to be permanently present in the brain, in some form. In
“competence” line with this, the Xs that make up the “timing tier” need to be construed
as units of phonological (and thus abstract) timing, and are not intended to
correspond to actual phonetic measurements. The expression TIMING TIER
we have just used refers of course to the tier made up of such Xs.
The Xs in the Applying this formalism to some real data, consider the English contrasts
TIMING TIER are between long and short vowels in such pairs as beat ~ bit, or boot ~ put. These
units of phonolo-
contrasts can be formalized straightforwardly by means of the timing tier,
gical (and thus
abstract) timing
as follows:
QyeidetlXs Kan DicuXore Mex
i I u U
beat bit boot put
Notice that these configurations provide an unambiguous representation

of the two component parts of these or any other vowels: their quantity and
their quality. In particular, the first member of each pair is shown to be both
long, as expressed by the two Xs in the timing tier, and tense, as expressed
by the symbols “i”, “uw” in the melody tier - MELODY isa generic label
referring to quality, that is, phonetic substance, as against quantity, that is, The label
length. The second member of (2a) and (2b), on the other hand, is short (it MELODY refers
to quality, that
only carries one X) and lax (cf. “1”, “u’”). The independence of these two aspects
is, phonetic sub-
of a single sound is confirmed when we consider the pronunciation of pairs stance, as against
of words like cart and cat in Yorkshire accents where there is no [ze] ~ [a] quantity, that is,
contrast: length
(3), 2X7 X xX
ze ce
cart cat
We can now return to affricates. The introduction of the timing tier pre-
dicts the possibility of the mirror-image relationship between timing elements
and melody elements, as represented in (4):
(4) X
y
/\ Z
The schema in (4) implies a timing of a single segment, but an internal com-
position of two melodies — precisely the configuration we will give to the
affricates [if] and [ds]:
(on Xx X
fish d 3
PHONOTACTIC
The configurations in (5) capture the fact that affricates are simultaneously CONSTRAINTS
monosegmental, with a single X slot, and bisegmental, since they involve a are the restrictions
on the distribu-
dual sequential articulation. Each of the two melodies of this sequence will
tion of sound
of course obey its own “phonotactic constraints’: PHONOTACTIC CON- sequences in the
STRAINTS are the restrictions on the distribution of sound sequences in the words of any
words of any given language (we return to this matter in chapters 9 and 10 given language
below).
A Strange Set of Vowel Alternations in English
In chapter 6 we examined the fronting processes involved in some German

and English plurals (Br[u:]der ‘brother’ vs. Brly:lder ‘brothers’, and g[ur]se
vs. glitlse, respectively), and attributed them to the presence of a floating
autosegment [—back] in the lexical representation of the plural suffix. In the
same chapter, we showed that the vowels of each Turkish word normally
exhibit a uniform specification for the feature [tback].
These processes are very natural: they simply involve the spreading of a
certain feature value to neighbouring segments, as admirably captured by
the autosegmental formalism. Consider now the following pairs of English
words:
(6) divine divinity

serene serenity
sane sanity
hink carefully ab
s. Jot
Each of the pairs in (6) consists of an adjective and a noun derived from the
adjective by the addition of the suffix -ity. Both members of each pair have
stress on the same orthographic vowel, emboldened in (6) (stress will be dis-
cussed in chapters 11 to 13). Despite the constant spelling, these vowels altern-
ate phonetically according to the following patterns:
Pilate soz Lo]

ixtee~ [el]
ler = le)
Two things are noteworthy here. First, the length of the vowels in each
column is uniform: in the first column the vowels are long, and in the second
column short (English diphthongal vowels are, of course, intrinsically long;
the extra length of [i:] is represented by the appropriate diacritic). An alterna-
An alternation
tion based. purely on length is very natural:
based purely on
length is very
natural (8) Sibir 22]
[ex] ~ [el]
[ex] ~ [ex]
Clearly, though, the English alternations in (7) do not only involve length.
Another example of a natural melodic alternation is the above-mentioned
vowel fronting of German, repeated in (9), where the symbol [y] represents
the [-ATR] counterpart of [y]:
(9) [ul] r a
lu] ~ [I
lo] ~ [ol]
[Lela]
By contrast, the English alternations in (7) are bizarre in the extreme. This
bizarreness may lead us to expect such alternations to be relegated to a _Alternations
marginal and restricted set of words, along the lines of the goose ~ geese set. Bi as “
However, the pairs in question, further illustrated in (10), run well into their A
hundreds: extreme
(10) severe severity inspire inspiration

fertile fertility profane profanity
deprave depravity impede impediment
vain vanity incline inclination
wise wisdom brief brevity
invite _ invitation recite recitation
The situation we are describing obviously poses a considerable challenge to

the phonologist, whose working assumption of necessity is that the sound
patterns of natural languages are principled, rather than random. We will
meet this challenge in the remainder of the chapter, providing additional
evidence for the timing tier along the way.
Z¥ Short ~ Long Vowel Alternations
One puzzling aspect of the divine ~ divinity, serene ~ serenity and sane ~ sanity
pairs in (6) is the constant spelling of the phonetically alternating vowels:
i, e and a, respectively.
We know from previous chapters that the relationship between sound and
spelling in English is often fairly remote, for historical reasons. In particular,
English spelling was reasonably close to the pronunciation until the fifteenth
English spelling century, but many English sounds have undergone considerable evolution since.
was reasonably
However, the spelling has typically remained unchanged, especially after
close to the
pronunciation
the invention of printing towards the end of that century, for the obvious
until the fifteenth complementary reasons of printers’ inertia and readers’ habit. This means
century that some current spellings reflect more the pronunciation of English before
that time than in our time. We will now see that this historical dimension
sheds crucial light on why the phonetically alternating vowels in (6) have
the invariant spellings i, e and a.
At the time we are talking about, the alternations in question exclusively
In the fifteenth involved length, as we illustrate in (11). The sound originally correspond-
century the alter-
ing to the final e in the spelling had already been lost then, and so we paren-
nation between
divine and divin-
thesize it to avoid confusion:
ity exclusively
involved vowel (11) divli:]n(e) divlilnity
length serle:]n(e) ser[e]nity
s[ee:]n(e) — slae]nity
All the relevant vowels in (11) have the structural representation in (12), where
[-cons] abbreviates [—-consonantal], which of course formalizes vowelhood.
Notice, importantly, that the contrast between the two columns is carried
by the timing tier, since the melody is constant:
C2) fax CX X
[—cons] [—cons]
Now, the natural assumption is that divine and divinity share the morpheme
divin, serene and serenity share the morpheme seren, sane and sanity share the
morpheme san, and so on. Therefore, we must decide which of the two
configurations in (12) is included in the lexical entry of the respective forms:
the non-basic configuration will be derived from the basic one by rule. The
two alternatives we have available are as follows:
tothe ee Be Xe Xl ee eX
[—cons] [—cons] [-cons] [—cons]
In a., a short vowel becomes long, while in b. the opposite process takes place.
The scope of the rule in (13a) will need to be restricted to word-final posi-
tion: we want the rule to lengthen the /i/ in divin(e), but not in divinity, where
it is followed by the suffix -ity. This approach produces the desired results
in the alternating vowels:
(14) a. div/i/n(e) > divlix|n(e) by rule (13a)

b. div/i/nity = div[i]nity rule (13a) not applicable
Once this procedure is in place, however, the last vowel in forms like trim,
pin, acid (cf. acidity) and very many others will also lengthen, in defiance of
the facts:
(15) tr/i/m > *tr[i:]m

p/i/n > *pli:ln
ac/i/d = *acli:|d
We could of course mark these forms as exceptions to rule (13a): indeed,

phonological rules often do have exceptions, a fact of life we must simply
accept. Countenancing exceptions by the thousand, however, clearly borders
on the perverse. Countenancing
exceptions by the
thousand borders
on the perverse
This problematic result is avoided if we adopt (13b) as the rule respons-

ible for the alternations. On this analysis, the lexical form will contain the
long vowel, /i:/ in the current example. The short alternant [i] will now be
derived by (13b) in forms where /i:/ is followed by at least two other vowels,
as indeed is the case in (10) above (we return to the issue of vowel shorten-
ing in chapters 15 and 18):
(16) div/i:/nity > divli|nity by rule (13b)

div/ix/n(e) = divlit|In(e) rule (13b) not applicable
This new procedure works as well as the procedure in (13a) in cases where
there is an alternation. In addition, it circumvents the undesirable effects of
(13a) on trim, pin and similar non-alternating forms, which simply fail to meet
the environment of (13b), and therefore will remain unchanged. It is of course
possible in principle that this positive result may be offset by some negat-
ive development elsewhere, but this does not seem to be the case here.
The Great Vowel Shift
Our account of the alternation between long and short vowels affecting a
Our account of sizeable set of English forms crucially relies on the availability of the timing
the alternation
tier. We illustrate the alternation again in (17):
between long and
short vowels
affecting a size- Cee tN. |
able set of English
forms crucially
relies on the avail- [—cons] [—cons]
ability of the tim-
divine divinity
ing tier
serene serenity
sane sanity
The qualities of the related vowels in Modern English are of course sig-
nificantly different from their counterparts in Middle English, that is, in the
English that resulted from the mixture of the purely germanic Old English
with the Norman French of the conquerors, and which towards the end of
the fifteenth century became Modern English. Indeed, we deliberately chose
to go back to a time when the alternation only involved length in order to
shed some light on the apparently capricious state of affairs found in Modern
English, repeated in (18) as a reminder:
(18) Modern English vowel alternations:

diviai|ne div[t]nity
serlit]ne serle]nity
slei]ne —_slee]nity
The problem with the modern English situation is that the long vowels
[at], [ix], [er] do not correlate phonetically with their predecessors [ix], [ex],
[zx], respectively, in any obvious way — by contrast, the short vowels have
only undergone laxing, a minimal and highly natural change: [1], [e], [el,
respectively. The solution to the puzzle posed by the quality of the long
vowels can again be found in the history of the language. In particular, the
Middle English long vowels [i:], [ex], [ex] and [eer] (and correspondingly for
the back vowels, although the situation is more complex with the back If the change from
set) underwent a series of changes which ultimately led to their modern the Middle English
long vowels [iz],
incarnations [at], [i], [er]. If such processes are followed step by step, the
[ex], [ex] and [az]
motivation for each individual change becomes apparent. to their modern
The reconstruction that we are about to present is based on history, but incarnations [ai],
we have divided up historical stages where we deemed it necessary for main- [ix], [el] is exam-
taining the clarity of the exposition. We shall start the discussion with a dis- ined step by step,
the motivation for
play of the relevant set of the long vowels in Middle English:
each individual
change becomes
(19) [ir] divine apparent
[er] serene
[ex] meat
[z:] sane
The series of changes which eventually yielded [ai], [ix] and [e1] are known
collectively as the GREAT VOWEL SHIFT. The first stage of the Great Vowel
Shift involved the diphthongization of [ix] into [e1]. This process took place
in two steps. In the first step, the second timing slot of [i:] dissimilated in
tenseness from the the first element [i]:
(200) iz > i divii:|n(e) — divfit|n(e)
The autosegmental formalization in (21) brings out the exclusively segmental

nature of the process. Notice in particular that the number of timing slots
is not altered:
Ci tx. Xx xX xX
[radical] [radical] [radical]
[+ATR] [+ATR] [—ATR]
This formalization makes it obvious that the process involves melodic break-
ing, or diphthongization, a tendency which is still active in various contexts
in Modern English, as we saw in chapter 7.
The second step in the process of diphthongization of [i:] into [er] involved
the lowering by one degree of the melody associated with the first timing
slot of [it]:
OD) willine Seen divlitln(e) — divlet|n(e) divine
The result of this process in the set of English vowels at the time is as follows:
(23) [er] divine

[ex] serene
[ex] meat
[eer] sane
Notice that the system no longer has a high vowel [i:]. On the other hand,
there is considerable phonetic proximity between the relevant vowels of divine
and serene, [et] and [e:], respectively. The distance between these two vowels
was increased as [e:] raised to [i:]:
OA) iews—> oi: serle:|n(e) — ser|ix|n(e) serene
We now have the system in (25):
(25) [er] divine

[ir] serene
[ex] meat
[eer] sane
There is of course no [e:] in this system. This gap was filled as [e:] raised
to [ez]:
Q6) "e-> “ez mle:lt — mlex|t meat
The system is now as in (27):
(27) [et] divine

[iz] serene
[ex] meat
[eer] sane
In this system there is no [e:], a gap filled as [ze] raised:
(28) e > & sleex]n(e) — sfex]n(e) sane
The sum total of the changes we have considered led to the system in (29):
(29) [er] divine

[iz] serene
lex] meat
[ex] sane
We offer a summary of the various changes so far in (30):
(30) Input i divlit}n(e),

er €! x! serle:|n(e), mlex]t, sleex]n(e)
Output el div[ei]n(e),
iS eer serlit]n(e), mlex]t, slex]n(e)
Notice that the process involves a one-degree raising of all the vowels that
remained long. In turn, the first member of the diphthong [ii], not being able
to rise any higher, lowered instead, also by one degree. The chain of raisings
i < et < ex: was apparently triggered by the lowering of [i()] to [e(1)] - it
is as if the vacuum thus left in the [i] slot had had a suction effect, dragging
up all the remaining vowels by one degree. Appropriately, this type of chain
reaction is referred to as a “drag chain” (also “pull chain”).
The system in (29) is still not quite that of Modern English, and we must
now examine the remaining changes.
The diphthong [er] (remember, originally from Middle English [i:]) first
centralized to [ar] (thus transcribed with a schwa in the historical literature),
and eventually lowered to its Modern English position [az]:
(Sl) el “> al >, al diviet|n(e) — diviatjne — diviatr]n(e)
In turn, [e:] raised to [i], with the concomitant merging of the meat and serene
vowels:
G2) .Ch to E mie:lt — mli:t meat
The slot left vacant by [e:] was occupied by [e:] (from Middle English [z:])
through raising:
(33) eh->.e! sler]n(e) — slex]n(e) sane
Finally, this long [e:] diphthongized to [er]:
(34) es 7~ e! s[e:]n(e) — sler]n(e) sane
In (35) we display all the stages of the Great Vowel Shift, with the changes
at each stage emboldened and boxed in:
(35) divine serene meat sane

er i we Stage 0
er €t ei Stage la
Stage 1b
Stage 2
Stage 3
Stage 4
Stage 5
Stage 6
As would be expected from a historical sequence of events, not all these

changes have taken place in all English-speaking areas. This failure is of course
The final output an important source of difference between accents. We hope that our exposi-
effectively com-
tion has brought out the considerable degree of phonetic naturalness of each
presses the chain,
hence the puzzling
of the steps in the Great Vowel Shift. This naturalness is obviously obscured
nature of the in the final output, which effectively compresses the chain, hence the puz-
vowel alternations zling nature of the vowel alternations in Modern English. Importantly for
in question in us here, the processes we have examined provide strong confirmation of the
Modern English autonomy of the timing tier in relation to the melodies: the alternation in
length is quite independent of the changes in the melodies.
[4 The Synchronic Reflex of the GVS. Vowel Primes

and Vowel Processes
The slightly idealized historical account of the Great Vowel Shift we pres-
ented in the previous section attempts to make sense of the outcome of
the shift, which permeates the contemporary phonology of English: cf. the
samples in (6) and (10) above. We must now come up with a formalism
to capture these contemporary alternations. We will operate on the crucial
We operate on assumption that the lexical representations have not changed from Middle
the assumption to Modern English. This assumption is argued for in SPE and is accepted
that the lexical
by many.
representations
relevant to the The three rules in (36) account for the three changes in height that underlie
Great Vowel Shift the three alternations, provided that the rules apply simultaneously:
have not changed
from Middle to
Modern English
(Ge), 4. [-high} >“ tehigh!/ ei serene
—low
b. LHow] - [4low]/ ae sane
—high
c. [thigh] — [+low] ia divine
The problem with these rules is that they state the changes in a piecemeal
fashion, despite their obvious structural unity. An autosegmental version of
these rules will not improve matters.
thi assertion by writing
This failure could be interpreted as an indictment of the standard binary

approach to features. One proposal of some interest designed to overcome
problems of this kind formalizes the whole range of vowels as the product — One proposal of
of the three basic vowel sounds [al], [ul], [i], the most natural set of vowels: 50Mé interest for-
: k ? lizes the whol
la], [ul], [i] are the quantum vowels, as we explained in chapter 5. Oe eee A
: range of vowels
as the product of
the three basic
vowel sounds
[a], [ul], [i
The alternative approach in question replaces the features of the standard

theory with the three basic vowels as undecomposable phonological units.
Let us represent these three vowel primitives with the symbols <a>, and
 — we are maintaining the convention that symbols in square brackets
represent sound, and introducing the ad hoc, but crucially different, nota-
tion of angled brackets for the three vowel primitives. In isolation, these three
vowel primitives yield the vowels [a], [i] and [u] directly. So far, therefore,
nothing very spectacular has been achieved.
Let us now turn to the mid vowels, and to secondary vowels like the
rounded front vowels and the back unrounded vowels. Under the vowel-
primitive approach, these objects will have the representations in (37):
7). ioleh leh ofl 1; fo) jo) ly¥Jndu)., dela. [xhe),ete.

<a le PAU pele ele
The representations in (37) bring out the two essential components of the
approach: (i) all vowels are built out of the three quantum vowels, whether
by themselves or in combination with each other; (ii) such combinations
crucially involve a hierarchization of the primitives, such that [e] results

from the dominance of over <a>, [e] from the dominance of <a> over
, and so on — the notational expression of dominance is, obviously, ver-
tical precedence. The not infrequent interpretation of cardinal vowel [y] as
[iu] by untrained English speakers to which we referred in chapter 5 above
could of course be taken as evidence for the vowel-prime analysis we are
presenting.
Let us now analyse the vowel changes which make up the Great Vowel
Shift in this framework. The structure of the vowel melodies that are input
to the Great Vowel Shift will be as in (38):
(38) [a] [e] [i] Phonetic representation

<a> Phonological representation
<a>
After the shift, the respective composition of these vowels will be as in (39):
(39)sqxhe] li [a] Phonetic representation

 <a> Phonological representation
<a>
This alternative formalism does not in fact make the process any clearer.
Consider first the raising of the mid vowel [e] to [i]. The formalization of
this process as the loss of <a> from the representation of [e] is appealing, and
arguably does shed light on the event. It will therefore not be unreasonable
to expect an equivalent operation in the raising of [a] to [e]. Here, however,
the formalism forces us to insert the element into the bare input repres-
entation <a>, at the same time demoting this original <a> in the dominance
ranking to prevent the formation of the mid-low vowel [e]. Finally, the muta-
tion of [i] to [a] requires the brute force substitution of <a> for . In the
end, thus, no unified pattern emerges, and the approach is not particularly
well supported by these data. We will consider a related proposal for the
analysis of vowels in chapter 17.
The SPE Account

We now offer a summary of the ingenious pre-autosegmental formalization
of the Great Vowel Shift advanced by Chomsky and Halle in The Sound Pattern
of English. To keep the exposition simple, we shall adapt Chomsky and Halle’s
rule to the set of front vowels we have been discussing.
Chomsky and Halle split the process of vowel shift into two parts. In the
first part, the value of the feature [thigh] is inverted in non-low vowels: Chomsky and
Halle split the pro-
cess of vowel shift
(40) [ahigh] — [-ahigh]/ into two parts
Let us dissect rule (40) to ensure full understanding. The restriction of the
scope of the rule to non-low vowels is appropriately formalized by the inclu-
sion in the environment of the specifications [—-consonantal] (= vowels) and
[low] ([-consonantal, low] = non-low vowels, namely, [i] and [e]). The pro-
cess these vowels undergo involves the inversion of the value for [thigh]:
an input [ahigh] becomes [—ahigh] in the output. By convention, the value
of each GREEK LETTER VARIABLE (@, 8, y, etc.) is arbitrarily set as + or The value of each
—, independently for each variable. Accordingly, if we interpret the a in the Se
; cae cs VARIABLE (a, B,
input of rule (40) as + (thus defining the vowel [i] in our present context),
: y, etc.) is arbitrar-
we must also interpret 0 as + in the output. Consequently, the rule will express ily set as + or
the change from [+high] [i] to [-high] [e]: -o =—, ifa=+.Onthe other hand, — -, independently
if we interpret as —, rule (40) will change [-high] [e] to [+high] [i]: -a=+, for each variable
if « =—. All this is of course in accordance with the ordinary rules of algebra
and logic.
hese results accord with theprinciplesof
The overall effect of rule (40) will be an interchanging of /i/ and /e/, pre-
cisely as desired:
(41) /i/ 7> e div/i:/ne — _ divle:|ne

Je/ > i ser/e:/ne —_ serli:]ne
Work through theunifie le in(40) aain,interpreting O

_as + and then as ~, to make sure you follow theprocedur
The second part of Chomsky and Halle’s Vowel Shift rule inverts the value
for [tlow] in non-high vowels:
(42) [Blow] ~— [-Blow]/

—cons
—-high
Bas
),inter reti we did in rule
ng
Now work through ru
Dee | Bee
The mechanics of rule (42) parallel those of rule (40), with the appropriate
substitutions of the affected and contextual features. Notice the crucial use
of a different Greek letter variable, [ahigh] in rule (40) and [Blow] in rule
(42), to encode the independence of the respective values. The effect of rule
(42) will be the exchange of /e/ and /x/:
(43) /e/ > ez divie:|jne — diviz:|ne

33] Peer Oceate s/x:i/ne —- sfer|ne
At the moment, we have two rules formalizing the vowel shift process:
(40) and (42). However, these two rules can be reduced to a single rule by
factoring out [-consonantal] as the common input, as in (44):
(44) [-ohigh] / Branch a. = (40)

ohigh
—low
LeoDS|a teallsd kh Branch b. = (42)
Blow
—high
The BRACES enclosing the two branches signal participation in a single

BRACES signal rule. Crucially, the motivation for the unification goes beyond notational
participation in a
expediency: it relates to the fact that the two processes in question take
single rule
place in strict succession in the order given, with no other rule intervening
between them, as could happen, in principle, if each process were formal-
ized independently.
We wind up the section with a summary of the effects of the two

branches of rule (44), ordered in strict succession:
(45) i
| Branch a. (cf. rule (40) )
e
e
| Branch b. (cf. rule (42) )
ze
Notice that the relative success of Chomsky and Halle’s formalization

hinges on the breaking down of the lowering of [i] to [a] into two steps, cor-
responding to branches a. and b. of rule (44), respectively. The output of the
first branch feeds the second branch. In this way, each step corresponds to
a natural change in height: [i] — [el], and [e] — [ze], and conversely.
~ Check how th
fined in sectior
tion are The availability of
the timing tier
allows us to for-
The Vowel Shift rule in (44) is of course to be understood to apply only malize length as
to long vowels. The availability of the timing tier allows us to formalize such two timing slots,
without interfering
length as two timing slots, without interfering with the melody.
with the melody
EJ Further Repercussions of the Vowel Shift
The Great Vowel Shift has interesting repercussions in unexpected areas of

the English vocabulary, and we shall now mention some of these briefly. The Great Vowel
In chapter 6 we discussed the vowel fronting that occurs in a small hand- Shift has interest-
ing repercussions
ful of English plurals: goose — geese, tooth — teeth (also foot — feet, with addi- in unexpected
tional laxing of [u] to [u] in the singular). Now consider the following areas of the
additional plural alternations: English vocabulary
(46) mouse [maus] mice [mats]

louse [laus] lice [lats]
The alternations in (46) involve the weak element of the diphthong, and are
therefore strange from a phonetic perspective. When the historical evolu-
tion of the alternations is investigated, however, they are seen to be essen-
tially reducible to the same analysis as goose ~ geese, etc.
in the light of what

why these alternatio
as goose ~ geese.
The historical antecedent of [a1] was [i], and the historical antecedent of
[au], [us]. This means that in Old English the first stage of the plural process
in (46) must have involved the fronting of the singular vowels by the plural
suffix -i, the very process that we saw in chapter 6 affecting goose ~ geese, etc.:
(47) muisti > myisi
Rounded front vowels obviously underwent unrounding, the plural suffix

-i disappearing also at some point:
(48) my:si — mits
We now have the contrasting forms m[u:]s (singular) and mli:]s (plural). These
forms then underwent vowel shift, hence their respective modern counter-
parts m[au]s and m[ai]s, all in the expected manner.
A further set of alternations involving Vowel Shift concerns a set of so-called

STRONG VERBS, where the expression “strong” refers to a vowel alternation
in the stem which contrasts with the more common inflectional realization
of past tenses (cf. fine — fined, found — founded):
(49) find [farnd] found _ [faund]

bind [baind] bound [baund]
wind [watind] wound [waund]
grind [graind] ground [ground]
The interest of this group lies in the apparent reversal of the direction of the
backness shift: compare mouse > mice with find — found.
Historically, the alternation in (49) also arose from the spreading of vowel
features from the suffix to the stem. The past tense suffix was -on at the time.
The specified features [+back] and [+round] spread from the suffix -on to
the stem, via association of the [... Je nodes of the suffix vowel to the
[-consonantal] node of the stem vowel. The process is formalized in (50):
Re bn
(50) TL na
[-cons] [-cons]
The effect of this rule on findon is as in (51):
(51) [-back] [+back]
ae ak
[-cons] : [—cons]
Bo st [labial],
[+round]
f i nd oO n
u
The vowels [i] (in the present) and [u] (in the past) subsequently underwent
vowel shift. By then, the suffix -on had disappeared, and the backing of the
root vowel remained as the only mark of past tense.
Synchronically, we may consider relating the process to the familiar one
of plural formation: mouse > mice = found — find.
This solution obviously involves the derivation of the present from the
past. From Saussure’s perspective that the signifier is independent of the
signified, there is, of course, no reason to assume that the form of the pre-
sent is basic. However, work by another great linguist, Roman Jakobson, sug-
gests that morphological and semantic “markedness” usually go hand in hand
—~ MARKEDNESS is the nominalization of MARKED, a label which suggests
The label ‘less expected’, ‘more complex’, ‘less natural’, and the like (see chapters 17
MARKED suggests
and 19 for further discussion). If morphological and semantic markedness
‘less expected’,
‘more complex’, go hand in hand, we would expect the semantic unmarkedness of the pres-
‘less natural’, and ent, revealed in its more general scope, to correspond to a morphologically
the like underived form. This naturally leads to an analysis of the strong preterites
with a floating feature [+back] as a past marker.
Sp |Bune mal details of the
The formalization of Modern English umlaut by means of floating features

was introduced in chapter 6, and you can refer back to it if your memory
needs jogging.
E) Multidimensional Phonology: The Skeleton
In this chapter we have encapsulated the temporal dimension of phonological

structure in the timing tier. In particular, we have shown that melodies can
exhibit different lengths, irrespective of the identity of the melody: vowel
melodies, for instance, can be short or long, and consonant melodies can enjoy
full independence or be part of an affricate. In the current formal model,
Our present these situations are represented as mismatches between the timing tier and
model allows for
the melody tier: a melodic element can be associated with one or two timing
mismatches
between the tim- slots, and a timing slot can be associated with one or two melodic elements.
ing tier and the We represent these various configurations in (52):
melody tier
(52) a. One timing slot:
(i) One melody (ii) Two melodies
Xx Xx
a tTeef
b. Two timing slots:

(i) One melody (ii) Two melodies
xX xX Xe
V a
||
te
In both (52ai) and (52bii) there is a one-to-one correspondence between timing

slots and melodies. In (52aii) and (52bi), however, there is a mismatch between
the two tiers, as corresponds to affricates and long vowels, respectively.
We have seen the timing tier to be made up of a succession of Xs, the abstract
unit of timing. The timing tier and the melody tier make up a plane: geo-
metrically, two lines necessarily make up a plane. The association lines relat-
ing the melodies to the timing elements run on this plane. As we know from
this and previous chapters, however, melodies do not constitute indivisible
lumps, but are in fact constellations of distinctive features. In turn, distinct-
ive features enter certain universally determined dependency relations. This
means that the structures in (52) are only shorthand abbreviations for their
full versions in (53). Note in (S3aii) that the proper representation of affricates
contains not only one timing unit, but also one feature [+consonantal], from
which the two successive phases [+continuant] and [—continuant] branch off
in the [+continuant] tier:
Soa sew/ as a ii. tf

X X
| |
[—cons] [+cons]
[dorsal]
[-high] [-cont] [+cont]
[How] [coronal]
[+back] [-anterior]
bi? & [az] ii. [t] [f]

XX X
‘2 |
[—cons] [t+cons] [+cons]
[dorsal]
[-high] [coronal] [coronal]
[+low] x a [—cont] [+cont]
[+back] [+anterior] [-anterior]

You can see that phonological structures resemble the mobiles that hang down
Phonological struc- from the ceiling over the cots of babies, or, perhaps more accurately, the pull-
tures resemble the
outs that come out of some pages in some books. These figures are easier
mobiles that hang
down from the
to see and comprehend in the real three-dimensional space than on the two-
ceiling over dimensional drawings of the pages of a book, and we must ask you to make
babies’ cots a special effort with your imagination really to understand their structure.
Notice that the timing tier makes up the baseline of the structure. Con-
comitantly, the timing tier is often given the label SKELETON, suggestive
of its basic function in sound structure: the rest would simply be the flesh.
The timing tier, or Correspondingly, the timing units can be referred to as SKELETAL SLOTS.
SKELETON, consti-
tutes the baseline
of the edifice of
sound in language
the sound (the melody tiers). In he case

one correspondence betweena sl
shown to occupy only one

[+continuant]) in the melody.
about by conflicting evide
or bisegmental: the two-lev.
of data in the chapter concert
and melody tiers has been i

section we speleeout the
Reery. Cr ues
tet UOnr s
1 Rehearse the arguments for a mono- 7 Which two types of process are
segmental and a bisegmental ana- involved in the alternations of the
lysis of the affricate. Great Vowel Shift?
2 How can an autosegmental analysis 8 On which tier does vowel shorten-
solve the affricate problem? ing take place?
3 What are the “timing tier" and the 9 How are these alternations good
“melody”? evidence for the two-tier approach?
4 Define the term “phonotactic con- 10 What is the basis of the analysis of
straints”. vowels involving vowel primitives?
5 What is odd about the pervasive 11 What is the purpose of Greek letter
vowel alternations of English as ex- variables? What is the purpose of
emplified in such pairs as divine~ braces?
divinity, etc.? 12 What do you understand by the
6 How and why does English spell- term “marked"?
ing disguise the oddity of these
alternations?
Further Practice
Turkish
The following alternations are found in Turkish between formal speech and
casual speech:
Formal Casual
kahja kaxja ‘steward’
fihrist __fizrist ‘index’
tahsil tarsil ‘education’
kahve ka:ve ‘coffee’
mahsus matisus ‘special to’
(i) | What is the nature of the alternation?

(ii) Does the alternation affect the timing tier or a feature tier?
(iii) Give an autosegmental account of the process.
Vowel Shifts
Albanian
The diagrams below represent the shift of a number of Proto-Indo-European

(PIE) vowels to their Modern Albanian descendants. The overall effect is that
PIE /a/, /o/ and /u/ have become [9], [e] and [i] respectively in Albanian,
and the diphthong /au/ has become [al].
fi} ea /yf ee /a/

le] <— /o/ <€ /o/{al
fh
/a/
[la] < /au/
Give an account of the feature changes which occurred in the course of these
shifts.
New Zealand English
Compare the New Zealand English pronunciation of the vowels in the fol-
lowing words with those in standard northern hemisphere varieties.
bit [bit]
bet [bt]
bat [bet]
(i) | Which vowels are affected?

(ii) How does this vowel shift compare with the Great English Vowel Shift?
(iii) Provide a diagram to represent the shift.
London English
The relationship between various RP English diphthongs and their Estuary

(= popular London) English and Cockney equivalents is represented in the
diagrams below:
RP li el al or
\ \ \ \
Estuary li i OT he fF
\ \ \
Cockney al al DI Ol
Describe the relationship between the three accents as involving a vowel shift.
~
c_tegioeor
ras moat nn 7
ayCorviww is hemwanre of herentsd

‘ a ews he ahkornwidd Afiet’ rye visstah the
: i titer ache od PwriwAn ae

é Oe
Vox ee ee ee ee
Vowel Shite”
Hild lewey «6 gnividuet an eign sortt stip asewiod isan
aeoaailalt
The dlagretw |~ ‘ cont the «itr tf 4 cunnher ef Protoinge-Bagg

PLE) vonwets to thelr Movtern Albanian Jesoertdants Che overalleblet!
Malas. fos aw (os jvet bo cae hee
ony ih highs hs te i '< pocwrnie fa}
oq’ wy é- -
a i aa
(or an gooo ui 17 ro one ‘om wh iiecanad ines i

a?) **p
sw
tow Zeir~] Paget

a
Coupon: Gre New Lewt-al Magis pores quiet ai the v«

lwwine wens ive* (es 2 Mataere nla am eR
Sit. he
bet (ee;
hal [Bee
if) Wat wiihy ws Ale eto eal?
= ae
q id) rte yf De chin vay. a nee wiih '
ety Prue Gh) ® «Leayratiy 10 eeppegeeet thre ih

: yi
} London frggives ‘pein

ee
Tie relotloreli¢n leap own vartcupe RP Tall
Ss, « ay ae
gapsebay Turion) Peli aucCockney nwo
Piag reer Dhow
SUPRASEGMENTAL STRUCTURE
In part I we formalized speech as an array of timing slots, each supporting

a cluster of distinctive features. We saw that this mode of representation allows
a remarkably natural and constrained account of many of the phonological
processes present in languages, assimilation in particular. In part II we
show that this structure needs to be augmented in two ways. First, there is
a whole dimension to speech that involves a type of singing (“intonation”
and “tones”), and this requires the enrichment of the model with a tone plane
independent of the segmental planes, or cluster of planes. The tone plane
thus connects to the skeleton from a different angle to the segmental planes,
and contributes to the configuration of phonological structure in a manner
reminiscent of a book, with one spine and several pages fanning out from
it. The reason for the postulation of a separate tone plane is, of course, that
tones (by themselves or in intonational phrases, or “tunes”) engage in their
own processes, quite independently of the segments. In addition to the tonal
plane, we need to augment phonological structure in a second way. In par-
ticular, the slots linearized in the skeleton enter specific, family-like relations
with each other: more technically, they group themselves into constituents.
We know they do because otherwise a vast number of phenomena in
natural languages would be entirely arbitrary, indeed incomprehensible, but
they become quite simple and natural as soon as such suprasegmental, or
“prosodic”, structure is brought into the picture. We will come across two
levels of prosodic structure. The first level is made up of “syllables”, each
syllable with its own internal constituency. We shall see that the linear
distribution of segments is constrained in ways relatable to syllable con-
stituency, and that syllable constituency also plays a crucial part in delimit-
ing the scope of phonological processes. The second level of prosodic
structure gathers syllables into “metrical feet”, and hierarchizes the metri-
cal feet included in such grammatical constituents as the simple word, the
compound word or the phrase. The main, but by no means exclusive, man-
ifestation of metrical feet is “stress”, which materializes as prominence or,
perhaps more perspicuously, as the main anchor for intonational associa-
tion. While at first sight more abstract than segments, tones have played a
major role in the shaping of the autosegmental approach to phonology, and
of the principles that regulate such an approach, and therefore they well
deserve the attention of the phonologist.
iS eh £ oe rv > om> = 2 ev...
ae e* & - a
4 -
i: aaa a r 7 ;
ee
. - x
ae stamina ARALJe.
oe
griheorrrue dogs, 2lole grinit lo parte 26 Bs ice besilennoal ow 1

amoiis-rumensentioctu sboru esl! tell wee aW. egraiegt svitmiiteib to 29
Asse
Isoigoinnaric- ot to nein to tacos hurdsuends brig.Istutem vided
aw fl hag al miuncney of noialiaiags xonougas! at insesiq ¢ eR
2 gro featd e7ew oat m henemngus of of ebton oustourita aisld4 tt |
“noiinaani”)peiycue lo ogi « covleewtd teas dysoq2 Of NORNSAID Sis
sorely omits cities ishLunt 48 io ireommctohans orld geniupen eit ro
onslq sno! sft .2atekg io wehivin 2 pormlq isinsmgse ort to
2 rsh eee alt Gl uiayip: ie wear: «4 enon najelota ofltot2
qaqenm Ar LAH? aipatoualte i f roitmugivaas od ot Ra EN
ott ive BOLTKeT coye™ Mtovoe bes an Ono Atber dead 60%
ist eter 108
to ai onelg 40d steinyse 6 (17 pobishaieog of To mote
>
* ofl re Syrm ae (vai 36. ay ell
igsecsisariat mtto eovis ern:at
ener oti oF ebb! ni PAIGITEIIE|sds’a \iinebragqobst aliup.e
Of hesn 31
anu Al agen baense-s tt patente lowgelonarlg tmenrgss
apbbais asi Hort ais me? Vw s ossoie ovtant beniynanil abnoat a
eiraudietes atic axalgecaads quarg Psi Hisoisost wsorn srodtoAbe ast

na oN nati i oder teav 8 sabre ovethoe! ab. yond 4NUME
td oldie enai babel seriou vidi os

odbivow esyaugne
We Jeinetixseet ste 26 foge ze ieee bese, oleparrie stip TOs
Ger) eegte Go daw pai seni, ohi-ota Niggord 2 ountsuie bib
fone ,“zaldalhis” Wo qu sbers ob level tah of? oupoers oiboaeg10as
tea0i! odd tari) 24 noha i ana PROD iuvsini awe eit ditw sidalf
no aldaiires ui Sloe yew of bowievlenos et eimeagee to pid
fisatiel mi i169 inte 6
aa o fi Crnaitiane aidaltye teil bns \onsu
sibostay tc eyvot foose afl mee iTg . teatgolonoriq to oqae §
Sfonteaf) cosidsiateid Ore a lasixvera” cint esldallys mstliag ou
at dorow uae orf} as einspiifenes iaitammsty foveal tsbufortt 3sot Gi
-~skeg ovierloxs anner on vd tud mista “dT sentde, self to pow bruogtt
co orebaeare at easiieneten Hsulve “eaeata et 499) Isohient to
“CRS an ti!) sori nin ot en sglawouniqersg g1om eqn
5 tevelq svar aantol 2ineingueued) tomede otorn tigte test 46 sie
gpigecepclanr ay
bru .vankanordy of zpxomeg xs,Laterasntggaaale ohto.
Ibe Sailssuited bes shsungas 8 Abbe ateluger telt aolqioning
penn et
CHAPTER NINE
THE SYLLABLE
C h.a.p t.e. Obie.

In this chapter you will learn about: a
A curious bias in the phonolo xy of children's early utterance
= The need for grouping sounds together into units we call
m The terminology used in th i
m The constituents that make
= The interaction be
In previous chapters we have illustrated the distinction between lexical and

phonetic representations with processes like assimilation of voice and of place
of articulation, vowel harmony, or umlaut. In essence, all these processes
involve the action of one segment on another — or, more precisely and more
technically, of one feature on another.
i
atures in LL _
t
Some of these processes provide direct evidence for an autosegmental

representation made up of various interconnected levels, or “tiers”, sig-
nificantly different from traditional unilinear transcription, and from stand-
ard writing.
ae
236 The Syllable
In this and subsequent chapters we will see that many phenomena relevant
to the phonology of languages cannot be properly understood unless we enrich
the representation with certain structures which are abstract to the extent
that they are not, and by definition cannot be, pronounced directly in the
way that segments are. These structures express relations between (concrete)
elements. We assume that these structures are real because they can reason-
ably be thought to be behind many of the phonological phenomena observ-
able in languages.
The Shape of Children’s Early Utterances
At a certain point in a child’s early development we dignify its utterances

as “words”. In the very early stages of life an infant goes through a crying
and wheezing stage, to which “cooing” (contented vocalizations) is then added.
At around four or six months comes the period of “vocal play”, during which
the child experiments with a multitude of possibilities of what might be con-
strued as consonant articulations. Then, at around six or nine months of age,
the child starts “babbling”, that is, combining consonants and vowels into
recognizable words: [baba], [mama], [dada], [nana], and so on.
The following utterances by “Timmy” are representative of the early

speech of many children (the corresponding adult forms are given on the
right):
(1) [p*al ball [11 months]

[pe] book [11 months]
[gaz] duck [11 months]
[2a] hi [11 months]
[hak:'a] key [11 months]
[ka] kitty [11 months]
[beer] bird [15 months]

[ger]. cow [15 months]
[ka] cup [15 months]
[ger] dog [15 months]
[ka:] girl [15 months]
[2am:a] moon [15 months]
The Syllable 237
[pee] baby [16 months]

[pz] block [16 months]
[pee] boat —_[16 months]
[k‘ak"i] cookie [16 months]
[2adz] flower [16 months]
[namz] Simon [16 months]
These words obviously consist of a consonant followed by a vowel (for

instance, [pe], [ge:]), or of a string of such consonant + vowel pairs
([namee], [k‘ak*i]). Note that they could just as well have consisted of a string
of just consonants or just vowels (say, [bmk] or [ase], respectively), or of a
vowel + consonant sequence ([zp], [e:g], and so on). You may think that
the reason these possibilities do not materialize is that the adult forms the
child hears do not have this structure. However, if you pay close attention
to the data, you will notice that the adult forms are systematically more com-
plex than those of the child, who only retains the consonant + vowel pairs.
For instance, adult ball is reproduced as [p"a] by the child, not as [ap"], adult
duck is reproduced as [ga:], not as [aig], and so on. It is as if all the child is
bothered about is the sequence consonant + vowel.
At this point we could conclude that the speech of this child is simply
peculiar. Indeed, there are some children (not many) with speech problems
which require the services of the speech therapist. However, a child’s
speech is only deemed problematic when the “defect” persists beyond a
certain, reasonably mature, age — were this not to be the case, we would all
have needed the therapist in our early years: the distortion of adult speech
by children is common and typical. What is relevant for us here is that such Children's distor-
tion of adult
distortion obeys certain unwritten and unconscious laws, of which the sample
speech obeys
in (1) above isa fairly representative output. certain unwritten
and unconscious
- laws
OU onesomee real chil li to
ee the opportunity
speech with your ow!
Before we start investigating these laws, we shall look briefly at a

curious phenomenon involving the pronunciation of English words by
Japanese adults. Japanese has been importing and is still importing many
words from English. However, when the Japanese pronounce these words,
238 =‘The Syllable
they do not pronounce them in quite the way English speakers do, as we
illustrate in (2):
(2) English Japanese

Christmas kurisumasu
text tekisuto
club kurabu
dress doresu
glass (for drinking) gurasu
disc disuku
slum suramu
plus purasu
bolt boruto
grotesque gurotesuku
Essentially, Japanese adults seem to be doing what we saw English children

do: aiming at pronouncing consonant + vowel strings, rather than strings of
consonants or strings of vowels. However, the way they go about achieving
this result is different - while we saw that English children drop the conson-
ants they hear in English adult speech to make words more pronounceable,
we are now observing that Japanese adults prefer inserting vowels into the
English consonant clusters. It thus seems that Japanese adults are guided by
the same unwritten and unconscious laws as English children, but imple-
ment them differently.
The laws we have been alluding to govern the way segments are
abstractly connected in clusters known as SYLLABLES. Segments that are
Segments need
not consistent with these laws fall overboard, so to speak, and are not pro-
to be members
of a SYLLABLE nounced: segments need to be members of a syllable if they are to be pro-
if they are to nounced. It is the purpose of this and the next chapter to present and discuss
be pronounced the conditions for such membership, that is, to spell out the conditions under
which segments can be incorporated into a syllable.
Structure of the Core Syllable

We have seen that young children’s syllables tend to be made up of a con-
sonant and a vowel. This obviously means that the syllabic structure avail-
able to the child is rather rudimentary compared with the syllabic structure
that is available to adults, at least in English. We formalize the structure of
The Syllable 239
the child’s typical syllable in (3) with the help of a tree. A TREE is just a visual
representation of a network of hierarchical relations. The informal symbol A TREE is a visual
“C” stands for a consonant, that is, a segment defined as [+consonantal], and representation of
the symbol “V” for a vowel, that is, a [-consonantal] segment. You can see in a network of hier-
archical relations
(3) that such C and V are gathered under the umbrella of o, the Greek letter
“sigma” equivalent to the Roman s, here a shorthand for the word “syllable”:
Dovid?
/\
C V
The structure in (3) is simply telling us that a syllable is made up of a con-

sonant followed by a vowel. This structure is, of course, abstract, but none
the less real. For instance, the structure in (3) explains the rendering of adult
book as [pee] by the child in (1) above — quite simply, the syllable template
in (3) acts as a filter to the more complex input [buk]:
ayo
The Japanese adaptations of English words we illustrated in (2) above sug-

gest that something similar is happening in the Japanese adult language.
The component parts of the syllable are given special names, to make it Syllables contain a
easier to identify them and to refer to them. Let us assume at this point that vowel, the core or
NUCLEUS of the
syllables always contain a vowel, the centre or NUCLEUS of the syllable —
syllable, preceded
reality is a little more complex, but this assumption will do for the time being. by a consonant,
In the forms in (1) and (2) above, this vowel is preceded by a consonant, which acts as the
which acts as the ONSET of the syllable. The core syllable, conveniently ONSET
illustrated by the child’s utterances in (1) and the Japanese renderings of
240 The Syllable
English words in (2), is thus made up of a nucleus preceded by an onset.

We represent this more detailed structure in (5):
(5) Oo
O(nset) N(ucleus)
|
S
|
V
We now apply this schema to our database in (1) and (2) above:
PeeT
(6)
Ors Nogew 2ORLIN Op eN oe sO
el eee aioe
Pane k a os m «&
biti Ss Oo o Oo
sn
ON
ered sere
ar Or “N GN
se: Ce Ni
in ladop
i ee
cd a i dl Sip a ee
The label “O” on [p], [k], [n] and [m] in a. and on [t], [k], [s] and [t] in b.
simply indicates that these consonants fill the role of onset in their syllable.
In turn, the label “N” indicates that the vowels it dominates fill the role of
nucleus.
The CV syllable is legitimately considered the CORE SYLLABLE. Some
The CV syllable languages only have CV syllables. More commonly, languages allow for syl-
is the CORE
lables of greater complexity: this is obviously the case in adult English, and
SYLLABLE
Japanese also allows for other types of syllables besides the ones we have
mentioned, as illustrated by abusutorakuto ‘abstract’, sutoraiki ‘(labour)
strike’, aisukuriimu ‘ice cream’, burakku ‘black’ and doragon ‘dragon’.
The core syllable is, however, found in every language, whether or not the
language also allows more complex syllables.
The Syllable 241
The cross-linguistic omnipresence of the core syllable offers an interest-

ing parallel with its also invariable occurrence in early child language. The
correlation between the frequency of a sound in the sound systems of the
world’s languages and the order in which sounds and sound structures are
acquired by children was insightfully drawn over fifty years ago by Roman
Jakobson, one of the giants of modern phonology and linguistics in general,
to whom we have already referred in chapter 8.
Sonority and the Syllable
So far we have encountered the universal core syllable, made up of an onset,

usually a consonant, and a nucleus, usually a vowel. We must now motiv-
ate this particular configuration.
In chapter 3 we mentioned the fact that some sounds are more sonorous
than others. We explained at the time that being more sonorous means hav-
ing more sound (sonus means ‘sound’ in Latin). The obvious implication is Being more
that some sounds have more sound than others. There is no paradox here: sonorous means
having more
it is just that the word “sound” is being used in two different senses, to sig- sound
nify ‘segment’ and to signify ‘quantity of sound’. Another way of putting it
is to say that not all segments (bundles of features, as you know) carry the
same amount of sound.
It is not hard to see that vowels have more sonority than consonants. After
all, when we want to attract someone’s attention we typically shout “oy!”,
“eh!”, etc., with a fairly open vowel carrying the call. The point is that we
would not dream of shouting [p], or even the fricative [f], or the liquid [I].
Notice also that if we shout out for help it is the vowel we will lengthen for
emphasis: heeeelp, not hhhhelp, hellllp or helpppp. The reason for all this is that
vowel sounds carry at a greater distance than consonants, precisely because
vowels are more sonorous than consonants. Segments are
arranged within
We know by new that the segmental structure of the core syllable is
the syllable in
CV, rather than, say, CC, VV or VC. The arrangement of segments within such a way that
the syllable is therefore such that sonority goes up from the onset to the sonority goes first
nucleus, and then down to the next onset, like the ground in hilly terrain. up and then down
We represent this situation in (7):
242 The Syllable
(7) Sonority profile of speech:

N N N More sonorous
O O O Less sonorous
Given the greater intrinsic sonority of vowels, it is not surprising that vowels,
rather than consonants, should occupy the syllable nucleus, and consonants,
rather than vowels, the onset.
The sonority profile outlined in (7) replicates the reality of human speech,
Human speech
essentially a succession of vowels punctuated by consonants, or, from a more
consists in essence
of a succession of reductionist perspective, a long modulated vowel sound regularly interrupted
vowels punctuated by consonants, rather as the meat is regularly constricted by knots ina string
by consonants of sausages. The alternation of opening and closing that constitutes speech
thus underpins the core status of the ON (or CV) syllable.
E¥ The Coda
We will now see that syllable structure can be considerably richer than we
have been allowing.
Consider the following words of adult English:
(8) captain active septic

rustic kaftan rectum
It should be intuitively obvious that each of these words is made up of two

syllables (more on this in chapter 10). Clearly also, the segmental strings that
make up the words in (8) do not fit into the syllable template in (4) above,
as we now illustrate for captain:
(9) o Oo
ON Onan
Kee iP (2 ey)
You can see that [p] and [n] remain outside our current syllable structure.
We have already said that segments that are not affiliated to a syllable fail
The Syllable 243
to be pronounced. However, the prediction that [p] and [n] are not pronounced
in captain is obviously false for adult English (child English may well real-
ize captain as [katt], precisely because in such child language only the core
CV structure is available). In technical parlance, syllables act as “licensers”
of phonetic material. We can draw a useful analogy with the “licensing” of Syllables act as
humans by the issue of a birth certificate. Without a birth certificate, a per- “licensers” of
son does not officially exist, and in the case of the segments the unlicensed ees
segment will simply not be pronounced. The fact that captain is [keeptin] in
English therefore means that the /p/ and the /n/ are both licensed, and
consequently they must be part of a syllable.
The inclusion of the /p/ and /n/ of captain in a syllable implies that English
syllable structure admits more segmental material than the core syllable in
(4). In particular, English syllable structure allows for a consonant after the
nucleus. This new constituent is known as the CODA, a label we will abbre- 4 consonant after
viate to “Cd” (reserving “C” for “consonant”, as we have been doing so far): ~ ete oe
stitutes the syl-
lable CODA
(10) o
ON Cd
The addition of the coda to syllable structure allows for the representation
of captain as in (11):
(11) Oo rhe
O N Cd OniINviGd
Kae Pp feet
You can see that the orphan consonants of (9) are now affiliated to the syl-
lable coda. This correctly predicts that these consonants will have a phonetic
realization.
f the other words in (8).

244 The Syllable
The Rime
In (10) and (11) above we connected the subsyllabic constituents O, N and

Cd directly to the syllable node o. We shall now see that there are reasons
to make the structure of the syllable slightly more complex.
Consider the words in (12):
(12) pat bat cat fat

[pect] [beet] [keet] [feet]
Can youfind any
These words clearly rhyme with each other (we assume you are familiar with
the device of rhyming in English verse). Compare these forms with those
in (13):
(13) “a pat “pan pad #e pack

[pet] [peen] [peed] [peek]
by epat sie pit pot put
[peet] [prt] [ppt] [put]
The words in the two sets in (13) do not rhyme with each other, even though
they exhibit the same amount of phonetic overlap as those in (12): [at] in
(12), [pee] in (13a), and [p... t] in (13b). Clearly, thus, rhyming is not just a
matter of amount of overlap. Rather, the overlap must affect a certain posi-
tion within the syllable, namely, the sequence N + Cd: the onset simply does
not enter into the computation of rhyme.
The joint participation of the nucleus and the coda in poetic rhyming is
not coincidental: the same pairing plays a crucial role in many phonological
The nucleus and
processes, ‘as we shall see in due course. This leads to the replacement of
the coda group
together in a con-
the flat structure of the syllable in (10) with a structure where the nucleus
stituent called and the coda group together in a constituent RIME, with the spelling dif-
RIME ference conveniently highlighting the phonological relevance of the construct.
The enriched structure of the syllable is displayed in (14):
The Syllable 245
(14) o
O
/\ RGme)
NoCd
In (15) we match this new syllabic template to the segmental sequence [pzt]
of pat:
(15) o
O
/\ RGme)
Ls ] d
Pp ee +*~—O
The structure in (15) includes all the standard syllabic subconstituents: O,
R, N and Cd, organized as shown. As we know from our brief discussion
of child phonology, simpler syllables also exist. In the next section, we review
the four most basic structures of syllables found in the world’s languages.
[4 Basic Syllable Typology
We now have two types of syllable in our repertoire, namely, ON and ONCd,
which we are still assuming correspond to CV and CVC segmentally. The
nucleus is, of course, essential to the syllable, in the way that the heart or
the brain is essential to the human body: without a nucleus there is simply
no syllable. By contrast, syllables without an onset do exist, whether or not
with a coda. The four most basic types of syllable are, therefore, as follows:
(16) a. ON b. ONCd e. .NCd dN

The core syllable
trating thesestruct template can

undergo two mini-
mal changes: first,
it hed it
Of the structures in (16), we have seen (16a) to be the core syllable. The Ste et it
non-core syllables are in effect minimal deviations from this core syllable. can take ona
In particular, the core syllable template ON can undergo two minimal coda
changes. First, it can shed its onset:
246 The Syllable
(17) ON >.N (equivalently at the segmental level, CV > V)
Second, it can take on a coda:
(18) ON — ONCd (equivalently at the segmental level, CV — CVC)
The representation of these two alternatives in terms of Cs and Vs is as in

(19):
a Try out hisalternative formalism, forpractice.
(19). tas CS fae NV.

be O-> CPV
The effect of these two rules on the core syllable CV is shown in (20):
Q0)4CV => V by (19a)

CV = CVG by (9b)
CV =>VC _ by (19a) and (19b)
The degree of complexity of each of the four basic syllable types can be read
directly from this formalism. In particular, the core syllable CV is simplest,
The core syllable because it is not subject to any particular operation. The VC syllable is the
CV is simplest, the
most complex because it requires the application of both rules. Finally, the
VC syllable is the
most complex, the
two other syllables (V and CVC) have intermediate complexity, each being
two other syllables derived by means of one rule.
(V and CVC) have We now provide RP English examples of each of the four syllable types
intermediate we have described:
complexity
)srlesnced ual
(21) o
O N © N-Cd N
aa: ee ean Deel 3
law lawn awl awe

The Syllable 247
a (
of these syllable types?
& :
English syllable structure can of course accommodate more complexity

than is present in these four types. We start addressing this matter in sec-
tion 9 below, and take it further in chapter 10.
The differences in complexity between the four basic syllable types ren-
dered explicit by our formalization in (17) to (19) above are made explicit
in the implicational hierarchy in (22) (“>” = ‘implies’):
V
O27) Vea > CV
Ui
The hierarchy in (22) is backed up by fact over a wide spectrum of super-

ficially unrelated areas: children acquire the different syllable types in pre-
cisely the order predicted by the hierarchy; the presence of a more complex
type in any one language presupposes the presence of its simpler counter-
part(s); syllable-related historical change tends to go in the direction of greater
syllable simplicity; in languages with a rich range of syllable patterns, simpler
syllables are more frequent, both statically, in the inventory, and dynamic-
ally, in actual language use; and so on.
We complete this part of the discussion listing a few languages that

instantiate each of the four basic syllable types. Notice that the sample confirms
the implicational hierarchy in (22):
(23) CV only Senufo Hua

(W. Africa) (Papua New Guinea)
CV, :¥ Maori Cayuvava
(New Zealand) (Bolivia)
CV CVG Klamath Arabic
(N. America) (Middle East, N. Africa)
CV, V,CVC, VC French Finnish Spanish English
248 The Syllable
The Nature of the Syllable

In section 3 above we referred to the role of sonority in the syllable. In par-
ticular, we said that the essence of speech is a sequence of elements of high
sonority, so far embodied in vowels, interrupted by elements of lower son-
ority, embodied in consonants. Correspondingly, syllables are made up of a
segment of high sonority flanked by segments of lower sonority. A syllable
is therefore a cluster of sonority, defined by a sonority peak acting as a struc-
A syllable is a tural magnet to the surrounding lower sonority elements — the analogy of
cluster of sonor-
crystals in geology or of pearls in biology may be helpful in this connection.
ity defined by a
sonority peak
In turn, speech is a sequence of such sonority clusters, as we represent in (24):
(24) Oo Oo oO
Oqui
Joadspraye
R O
voivirdye oxxdof
R
BH O R
we ony /\
fallste abel een
(C) BV (G) (C) V © () V ©
The pivotal role of the sonority peak in the syllable becomes more obvious
if we formalize the whole syllable as a projection of its nucleus, as in (25a) /
otherwise equivalent to (25b):
(25) a. N” b. o
N’ R
N N
CN. ec CG M.S
5 the
The Syllable 249
If you are familiar with Chomskyan syntax, you will recognize in (25a) the
so-called “X-bar” template, adapted to syllables. If you are not, you simply
need to notice that in (25a) the syllable margins (that is, onsets and codas)
are in effect formal excrescences of the nucleus N. In particular, the “trunk”
of the tree in (25a) is made up of layered copies, or PROJECTIONS, of the
nucleus node N. The highest projection, N”, constitutes the syllable node,
which branches off to its left as the onset. The intermediate projection, N’,
constitutes the rime node, branching off to its right as the coda. As we said,
these higher nodes are supported by the nucleus node, N, which is in turn
projected from the vowel, the head of the syllable — the HEAD of a constituent
is the core element of the constituent, that is, the element without which there The HEAD of a
would simply be no constituent. conse Werte -Ue
For our immediate purposes, it will not really matter whether we formalize eee io
syllables with the schema in (25a) or with the perhaps more familiar one in
(25b), since we have introduced the “X-bar” formalism only to help you grasp
the nature of the syllable.
What does matter is that you should have reached a reasonable level of under-
standing of both the nature and the basic structure of the syllable, in pre-
paration for the further intricacies we will be examining in the remainder
of this chapter and in chapter 10.
E} Complex Nuclei
In chapter 8 we saw that vowel length is contrastive in English.
trastive"
We provide further examples of vowel length oppositions in (26):
(26) a. peak — [pitk] b. pick [pik]

pool [pu:l] pull [pul]
caught — [ko:t] cot [kot]
The vowels in the words in (26a) are long, and the vowels in the words in
(26b) short. The remainder of the sounds are identical in each pair, and there-
fore each contrast is minimal. Of course, the long vowels are also tense, and
250‘ The Syllable
the short vowels lax, but this additional difference is not contrastive by itself,
since contrasts like [pik] vs. [pr:k], etc., are not possible in English.
Long vowels are by definition associated to two timing slots. Strictly speak-
ing, syllable structure is built on the timing tier, not on the segments them-
selves as we have been doing so far, to keep the representations simple: as
we said at the end of the previous chapter, the skeleton constitutes the base-
line of the edifice of sound in language. Now, long vowels cause a mismatch
between our present syllable structure and the timing tier. In (27) we dis-
play two alternative interpretations of this mismatch:
(2a eawas Oo
In (27a), the timing slots associate to the syllable terminals one to one from
left to right, irrespective of the feature content of the segments. The con-
sequence is that the second part of the long vowel links to the coda, and no
syllable terminal remains for the slot of the final /t/ to link to. In (27b), the
feature content is taken account of, strictly keeping vowels for the nucleus.
The consequence is that only one of the two slots of /i:/ can be incorpor-
ated into the structure. Given the function of syllable structure as a segmental
|| The Syllable 251
licenser, (27a) predicts that the /t/ will not be pronounced, while accord-
ing to (27b) the vowel will be pronounced short.
Neither prediction is, of course, correct: if /t/ were not realized phonetic-
ally, beat would sound like bee, and if the vowel were short it would sound
like bit, after the concomitant laxing, or as impossible D[iJt, if there was no
laxing.
The obvious course of action to achieve an exact match with the timing
units involves augmenting the number of syllable terminals. The question
is whether such an increase ought to affect the coda or the nucleus. If we
assign one part of the /i/ to the coda (as we did in (27a)), we will conflict
_ with the typical pattern of association of [-consonantal] segments with the
nucleus. While we will see in the next chapter that this pattern does some-
times break down, it is all the same quite robust. Accordingly, we fit the
long vowel into the syllable by augmenting the nucleus to two units: We fit long vowels
into the syllable
(28) by augmenting
S the nucleus to
PK two timing units
_ The representation in (28) nicely captures the sameness of beat and bit with
_ regard to both syllable structure and melody identity (abstracting away vowel
tenseness, which in English we assume follows from vowel length, in line
with our findings in chapter 8): the two words only contrast with regard to
the number of skeletal slots in their respective nuclei. On the other hand,
bee and bit also differ with regard to syllable structure, since bit has a coda,
_ but bee does not.
252 =‘The Syllable
A nucleus with two slots also accommodates diphthongs, since we assume

that a diphthong is equivalent to a pair of vowels, both in the melody and
A diphthong is in the skeleton:
equivalent to a
pair of vowels,
both in the
melody and in the
skeleton
|
xX x
a —»
4 t
We hope that the need in English for nuclei with two timing slots, along-
side nuclei with only one slot, is by now quite clear.
EE} Complex Onsets
The syllable template we have available at the moment (maximally CVVC)

cannot account for the English forms in (30):
(30) plot blot flag

press brick frock
clock glad
crack grill
trap dress _ thrill
slum
shrill
The Syllable 253
In particular, the first two segments in the forms in (30) are consonants,
and our current syllable template only accommodates one consonant in the
onset.
iagram that this is
The obvious solution to the problem posed by the consonant clusters in

(30) involves allowing for onsets with two timing slots, in the way that we
have allowed nuclei with two timing slots. Augmentations of this kind man- WE allow for
ifestly involve more structural complexity, but the ultimate goal of phono- aie oe oe
logy, and of linguistics in general, obviously has to be the modelling of
reality, not the attainment of economy of representation at all costs.
We formalize the proposed augmented onset in (31), in each of the two altern-
ative modes of representation we have at our disposal:
631}e yea. o b. N”
R N’
N
| N
Cpe Cee
We have now enriched the structure of the syllable with two-member nuclei
and two-member onsets. In the next chapter we will encounter an apparent
need for two-member codas also.
The Sonority Hierarchy
The inventory of English branching onsets in (30) above contains a number

of gaps. In particular, only the consonant clusters in (32) actually occur:
(32) [pl] [bl [fl] [sl] [kl] [gl]

[pa] [ba] [fa] [62] [tx] [da] [fa] [ka] [gal
254 The Syllable
Why do consonant combinations other than those in (32) not turn up in the
onset, including, for instance, combinations which simply reverse the order
of the two actually co-occurring consonants, such as [Ip], [4p], etc.?
If you examine the clusters in (32), you will notice that the second timing
slot is always filled with a liquid, namely, [I] or [4]. In turn, the first slot
accommodates all the obstruents of English but the affricates ([tf], [d3]) and
the voiced fricatives ([v], [5], [z], [3]).
The pairing of obstruents and liquids in the onset is obviously subject

to some further restrictions (there is no [61] or [si], for instance), but we
will postpone discussion of these in order to maintain the continuity of the
exposition.
We could of course formalize the confinement of English onsets to
obstruent + liquid clusters simply by writing the appropriate requirements
directly into our syllable templates, as we do in (33), with “Ob” and “Lq”
as informal abbreviations for obstruent and liquid, respectively:
Gah a: 0 b. N”’
R N’
On
|
aN N
iS
Ob Lq Ob Lq
The problem with these syllable templates is that, while they do indeed do
their job, they only do it by brute force.
The Syllable 255
Thus, why should the two slots of complex onsets need to associate precisely
with an obstruent and a liquid, and in this order? The answer to this and
many other questions concerning the constraints on the distribution of seg-
ments within the syllable is found by reference to the SONORITY SCALE,
also known as the “Sonority Hierarchy”, a universal ranking of segment classes
on the basis of sonority. We will build up the sonority scale in stages, to
ensure that each step is motivated by explicit argument. At this point in the
exposition, we shall propose the embryonic scale in (34):
(34) Sonority scale (first version):

Most sonorous 4 Vowels
3 Liquids
2 Nasals
Least sonorous 1 Obstruents
The sonority scale is relevant to our present problem because, as you

already know, syllables constitute in effect sonority mountains: the sonor-
ity profile of a syllable rises, reaches a peak, and then falls. We illustrate
this profile in (35) for a syllable with a complex onset and a coda:
(35) a. 0 b. N’” Syllable structure
R N’
O
fo
N Cd N
Gin
cia
a SE
|
CCAir
aa ee Sonority profile
The sonority pro-
The sonority profile of the syllable is regulated by a universal principle file of the syllable
known as SONORITY SEQUENCING: must rise until it
peaks, and then
(36) SONORITY SEQUENCING: fall
The sonority profile of the syllable must rise until it peaks, and then fall.
256 The Syllable
As we show in (37), an onset such as lp, the converse of existing pl, would
violate sonority sequencing:
(37)—a: Go b. N” Syllable structure
N’
Ww
pe
ce
¢ 2,
aCc Z,
<=
—n
sat
<3—O ] —O
3
Sonority profile
The sonority trough in the onset in (37) obviously contravenes sonority

sequencing in (36). This is the reason for the absence of this sequence from
the inventory of onsets in English and other languages.
Sonority Distance
In (32) above you may notice that [ps] (stop + fricative), [pn] (stop + nasal)
and suchlike do not occur in English onsets, even though these sequences
are consistent with sonority sequencing.
Similar sequences do turn up in other languages, however. For instance,

in Greek they are part of words like [ps]lychologia ‘psychology’, [pnlefmonia
‘pneumonia’, [ks]enos ‘stranger’, sym[pnloia ‘accord’, kom|pslos ‘elegant’, edlksle
‘attraction’, etc. Is their absence from English therefore purely arbitrary?
If you work out the distance between [p] and [I] on the sonority scale
in (34) above, you will find out that it involves two points: liquids = 3,
The Syllable 257
obstruents = 1, therefore 3 — 1 = 2. On the other hand, [ps] scores a differ-

ence of 0 on this scale, and [pn] a difference of 1.
In many languages, the segments of complex subsyllabic constituents need

to satisfy a minimum of sonority distance. We shall formalize this require- We refer to the
ment through a MINIMAL SONORITY DISTANCE parameter. The setting requirement of a
certain minimum
of this parameter for English complex onsets will be as in (38), with refer-
of sonority differ-
ence to the sonority scale in (34) above: ence between the
segments of com-
(38) Minimal sonority distance in English onsets: plex subsyllabic
The minimal sonority distance between the two elements of an constituents
English complex onset is 2 as MINIMAL
SONORITY
DISTANCE
In the next chapter we will provide the reason for setting this sonority dis-
tance as minimal, rather than as a constant.
We will not go further into specifics here, for the issue is truly complex
across languages. However, we will emphasize that, while the sonority hier-
archy is essentially uniform across languages, the requirements of minimal
sonority distance between “sibling” elements in a given constituent need to
be specified language by language, and constituent by constituent (we are
using the term SIBLING to refer to nodes that are immediately dependent
on the same PARENT node; the terms “sister” and “mother” are also in use
in this connection). In our present context, it will suffice to know that the
architecture of the syllable in any given language is the joint product of invari-
able universal principles, such as sonority sequencing in (36), and parameters
with a fixed range of values, one selected by each language, as Minimal Sonor-
ity Distance in (38) has exemplified in our discussion.
258 The Syllable
foe are languages where syllables

where such syllables do not occur.
the consonant and vowel positions
“nucleus”, respectively. Obviously,
ate segment strings which consist o
syllable structure is richer than this
a further position after the nucle
the nucleus and the coda form
complexity can be ae on the basis of theputezee Ml

see of ies types being definabl in terms of oe
Key Questions
1 How does the syllable structure of 3 Which other constituent combines

children's early utterances and Eng- with the onset and the nucleus to
lish loanwords in Japanese differ make up a more complex syllable?
from that of adult English? What is the larger constituent
2 What is the structure of the core formed by this other constituent
syllable? and the nucleus?
The Syllable 259
4 Provide evidence for the rime as a 8 What is the “sonority scale"? List it.
constituent. 9 What is the role of sonority sequen-
5 What is sonority? What is its role in cing in the determination of syllable
syllable structure? structure?
6 How does the second timing slot of 10 How does minimal sonority dis-
a long vowel or diphthong fit into tance constrain onset formation?
syllable structure?
7 How does sonority determine the
structure of the complex onset?
Fodtstetee te Por ac th ce
Sonority
Arrange the following sounds according to their relative sonority:

fal Store i) fal.etfl le] Uz Im] [py] fi [d] [4]
Draw the sonority profiles of the following words. Identify the nuclei and
the syllable boundaries:
compass cocoa trigger asparagus acumen hippopotamus
Which of the following hypothetical words are syllabifiable in English? Explain

the reasons for the failures to syllabify.
[pnekil] [kepnil] [bizldin] [ldebsin] [nindip] [Idsiksel]
Spanish
In many varieties of Spanish [s] and [h] are in complementary distribution.

Consider the following examples:
ca[sla ‘house’ calsla[h] ‘houses’

calfh]co ‘helmet’ cen[sjo ‘census’
[sle[s]o ‘brain’ [s]e[s]o[h] ‘brains’
[sle[h]go ‘slant’ e[h]to[h] ‘these’
~ toh] ‘cough’ tols]e[h] ‘coughs’
Describe the distribution of the two segments taking into consideration all
the possible factors, including adjacent segments, word position and syl-
lable position.
260 ‘The Syllable
Yawelmani Vowels
Consider the following forms of Yawelmani verbs:
Future Passive Precative Dubitative

passive aorist gerundial
xilnit —_xilit xil?as xilal ‘to tangle’
maxnit maxit max?as maxal ‘to procure’
meknit merzkit mek?as metkal ‘to swallow’
sapnit sarpit sap?as satpal ‘to burn’
tannit ta:nit tan?as ta:nal ‘to go’
(i) Describe the alternations between the long and short vowels in the data.
(ii) Formalize the alternation on the basis of syllable structure.
SYLLABLE COMPLEXITY
ENGLISH PHONOTACTICS
Falling sonority in apé

m Further restrictions
onco
m The existence of non-vo
a Vowels in the onset.
m Refining the sono
m An explanationf
In the previous chapter we introduced the syllable, a structure with largely

universal characteristics that imposes a certain organization on the string of
segments. We saw that syllables are minimally made up of a nucleus, and
maximally of an onset, a nucleus and a coda, the latter two constituents being
conventionally gathered into a rime. We have been assuming throughout
that nuclei are filled with vowels, and margins with consonants. While this
correlation is essentially sound, it will have to be slightly relaxed in order
to take account of certain facts in many languages, English included. In this
chapter we also continue the survey of English phonotactics. English syl-
lables are useful because of their complexity, which will force us to enrich
our theoretical machinery: understanding the nooks and crannies of English
syllables will stand us in excellent stead to understand syllables universally.
262 Syllable Complexity
Complex Codas
In chapter 9 we motivated the existence of onsets and nuclei with two ele-
ments, respectively in sections 9 and 8. The data we will now present sug-
gest that we may need to extend this (minimal) constituent complexity to
codas.
Consider the following set of nonsense words in phonetic transcription:
(1) [kinp] [prmt] [pmt] [pmp] [kttp]

[kitk] [krpk] [prkn] [prpn] [prpll
We hope you will agree that none of these forms is possible in English. Clearly,
the reason is not simply that the forms are (deliberately) nonsense: another
nonsense word, blick, has repeatedly been proposed in the literature as a pos-
sible word of English — blick could, for instance, be the brand name of a newly
launched product, say, a new washing powder or a new soft drink. Indeed,
we must drawatripartite distinction in all languages between existing words,
non-existent but possible words, and non-existent and impossible words.
_ ible words?
Thus, the dimensions of possibility and existence need to be kept carefully

The dimensions of apart. Possibility has to do with phonological well-formedness, a matter
possibility and
directly related to the grammar of the language, while existence has to do
existence need to
be kept carefully with lexical inclusion, an essentially haphazard affair.
apart
You may think that a more plausible explanation for the impossibility of
the forms in (1) in English concerns the fact that they all have the structure
CVCC, since our present maximal syllable template only allows one C after
the nucleus, in the coda. Consider, however, the following set of also non-
sense forms:
Syllable Complexity 263
(2) [kimp] [pint] [stnd] [stld] [tmk]

[krpt] [keekt] [prlk] [pulp] [hilb]
[peeln] [wilm] [helf] [helv] [helf]
[helz] [hels] [benz] [bens] [hisp]
[hisk] [dest] [last] [kamm] [ho.6]
These forms can be pronounced by English speakers without any problem.

Therefore, we must conclude that, like blick, they qualify as possible words
of English. Crucially for our purposes here, they also have two consonants
after the nucleus. Indeed, parallels for the forms in (2) can easily be found
in the set of real words, as we show in (3):
(3) camp punt fond build pink

kept fact silk pulp bulb
kiln film shelf twelve Welsh
fells pulse lens fence clasp
risk west part farm forth
These facts are similar to those we came across in the previous chapter with
regard to the onset and the nucleus. Therefore, if we take these facts at face
value, we will need to accommodate a complex coda in our syllable tem-
We provisionally
plate, along the lines of (4):
accommodate a
complex coda
(4) 6 N” in our syllable
R
| |
N’
template
Pe(ed
N= N
/\
aC G.=€
| For the moment we will indeed adopt such complex codas, although at the
end of the chapter we shall see that there are reasons to favour a different
approach.
Incorporating complex codas into the English syllable still does not account
for the contrast in grammatical status between the nonsense sets in (1) and
(2) above, instantiating impossible and possible English forms, respectively.
Now, if you compare the two sets in question, the following facts about the
putative complex coda will come to your attention:
1 the liquid [1] only occurs as a left sibling, and similarly for [41] in rhotic
accents: [prlk], [helf], [karm];
2 when nasals occur as left siblings they assimilate in place of articulation
to their right sibling: [kimp], [pit], [tink];
3 when both siblings are non-sonorant stops, the right one is coronal: [kipt],
[keekt].
The restriction in 3 will be discussed in section 8 below. The restriction

in 2 follows from the rule of place assimilation in nasals we examined in
chapter 4. The restriction in 1 appears to mirror the restriction on onsets
we examined in the last section of chapter 9.
_ in question lat
The requirement of a sonority gap of at least two degrees between the two
elements of onsets is, obviously, relaxed in codas. In camp, pink, kiln or film,
for instance, the two coda consonants differ by only one degree in the sonor-
ity scale in (34) of chapter 9, which we now repeat as (5):
(5) Most sonorous 4 Vowels

3 Liquids
2 Nasals
In turn, in actually occurring kept or fact there is simply no sonority differ-

ence. This raises a question about the interpretation of sonority sequencing:
we would expect sonority sequencing to require sonority to fall in codas,
indeed as we stated in (36) in chapter 9. The fact that sonority does not fall
in kept or fact hints that either this requirement is too strong or that pt, kt,
etc., do not constitute complex codas after all. We will be addressing this
dilemma in the remainder of the chapter.
Non-Vocalic Nuclei
At this point you may feel tempted to take a leap and interpret the laxer
sonority distance requirement in codas as a hint that there are in fact no
sonority restrictions on English complex codas. Notice, however, that sonor-
ity does not rise in such putative complex codas in any of the forms we have
provided. Consider now the forms in (6), crucially bearing in mind that in
these words only the first vowel in the spelling has phonetic reality:
(6) prism
button
thicken
sickle
funnel
brother (in rhotic accents)
colour (in rhotic accents)
| Aword like prism does not strike the eye of the English speaker as odd in any
| way. Indeed, prism appears to comply with all the requirements built into our
syllable: the structure C(C)VCC of this and the other words in (6) is at first
sight syllabifiable as onset (simple or complex) + simple nucleus + complex
coda, thus apparently replicating the words in (3) above. However, closer
inspection reveals that, while the forms in (3) are monosyllabic, those in (6) are
bisyllabic. We make this difference explicit in (7) (the underscripted diacritic
“” indicates nucleus status, and “.” a syllable boundary, in IPA notation):
} (7) [pa.zm]
[ba.tn]
[61.kn]
| [st.K]]
[fa.nl]]
[b1a.03] (in rhotic accents)
[ka.l4] (in rhotic accents)
We do trust that your intuition agrees with ours on this matter.

Two questions arise at this point. The first question is why we are assign-
ing the final two consonants to the coda in the words in (3) but not in those
in (6) — compare silk with sickle, or think with thicken, for instance:
(8) a. Oo b.
R Rivetankls
N
eonsINE PCN
XX
| XX
i
Xe XR
aan
Se2 Ea) ei
lies
Soetae
GorSts eae kk Ose i.
ok a.
The second question is why (and how) we are assigning the word-final sonor-
ant consonant in (6) to the nucleus, in contradiction to the prerequisite we
have been operating with so far that nuclei are vowels.
The reason for the divergent syllabification of the forms in (3) and (6) is
that in the coda cluster [lk] of silk the first element, [I], is more sonorous
than the second element, [k] (3 vs. 1), whereas in the last two consonants of
sickle this relationship is reversed (1 vs. 3). This is relevant with respect to
the sonority profile of the syllable, which you know rises on its way to the
peak (= the most sonorous element in the nucleus), and then falls. We dis-
play this profile once more in (9):
(9) N More sonority

O Cd Less sonority
The principle responsible for this state of affairs is, of course, SONORITY
SEQUENCING, which we stated in (36) of chapter 9, and repeat here as (10):
(10) Sonority Sequencing:

The sonority profile of the syllable must rise until it peaks, and then
fall.
Clearly, a monosyllabic silk does comply with this principle:

(AE O-N-Eed
S
4 oO
n
3 oO
r
2 i
t
| | 1 y
Go TSat
A putative coda [kl] in the word sickle would, however, contravene Sonority
Sequencing, since the second element, [I], has more sonority than the first
one, [k]:
a2) -O; N -Cd

S
4 oO
n
3 O
fr
2 i
iE
| 1 va
Seas et
The sonority trough on [k] prevents [kl] from being a valid coda. Faced with
this situation, English (and other languages, but not all) relaxes the require-
ment that nuclei must be vowels, and allows sonorant consonants in the
nucleus under the circumstances, as already represented in (8b):
(13): eOpcJINeeOueN.
S
4 oO
n
3 oO
r
Zz i
t
| 1 yi
So, while it is true that English nuclei normally aim for a sonority minimum
English nuclei can of 4 and therefore normally contain vowels, they can lower this minimum
accept a sonorant
to 2 and accept a sonorant consonant under pressure from the sonority profile
consonant under
pressure from the
of the segmental sequence, precisely as happens in the cases we have just
sonority profile examined.
of the segmental
sequence
In section 4 we will present the opposite phenomenon, and argue that

English onsets can accommodate vowels.
Vowels in Disguise
Consider the syllabic structure of the words in (14):
(14) a. well wish wag

b. yes yob yum
You will most probably recognize a CVC syllable here, analysing the first
segment of these words (spelled w and y, respectively) as a consonant.
At this point we need to pause and ask ourselves what kind of consonant
these might be. Remember that consonants are divided into obstruents, with
a radical constriction to the airflow in the central passage in the mouth, and
sonorants, with the airflow also constricted in the central passage, but still
flowing out unimpeded through some other channel: the nose for nasals, or
the sides of the tongue for laterals, for instance.
Now, if the sounds spelled w and y in (14) were consonants, they would
most certainly not be obstruents.
What can they be, then? There is, of course, a further class of sonorants, rhotics,
but in these sounds the air is interrupted (albeit most fleetingly) at the front
of the mouth, and this is clearly not the case with w and y.
So, what kind of consonant can w and y be? If you turn to the literature, you
will find a variety of labellings: “glides”, “semivowels” and “semiconsonants”,
as well as “yod” for y in a diachronic context. Sometimes, w and y are also
included in the more general class of “approximants”, indeed a feature in
a system we will examine in chapter 17. Now, what exactly lies behind these
labels?
If you (artificially, and most unnaturally) lengthen the initial segment of
well, you will come up with the form [urel], and if you do likewise with yes
you will find yourself pronouncing [izes].
This simple experiment therefore reveals that w and y are really vowels, even
if particularly short ones — the diacritic “” is actually available to indicate Articulatorily, w
extra shortness, hence [ii], [i]. They are also pronounced witha higher tongue @"4¥ #"@ vowels
|° than in the words hoot and heat, in fact more or less literally as IPA [i] and
[u]. If w and y are phonetic vowels (despite the ambivalence of much of the
| literature on this matter), you may think that the forms in (14) above have,
| after all, no onset — only a complex nucleus made up of a high vowel ([ul,
| [i], regardless of the spelling) followed by a second vowel, as represented
| in (15):
(15) a. o baer’
| R
|
N’
BS N Cd N
fades |
Veo 2€ VV
/\ AG
Consider now the forms in (16):
(16) a. womb weep wade

b. yeast yawn you
words in (16) 1
In these forms the initial high vowel is followed by a long vowel, and there-
fore, following the logic of the argument, we would have to assume a three-
member nucleus:
(i/)tuai b. ING
Vet Nose Vatk) Ge
If the facts stopped here, we would have no alternative but to recognize

the existence of three-member nuclei, despite our overt disinclination to
increase the size of syllabic subconstituents.
As it happens, however, other facts take us away from the structures in (17),
and, concomitantly, from those in (15). These facts concern poetic rhyme and
the ban on [w] after labial consonants.
Z¥ Onset Vowels. The “OCP”
Let us compare the forms in (18) with their counterparts in (19), which replic-
ates (30) of chapter 9 minus slum and shrill, since these are best analysed
separately, as we will see in section 7 below:
(18) a. twin dwell queen Gwyn

thwart
swell
b. pew _ beauty cute argue
few view
mute
c. (in RP and most British-type accents)
tune dune
Thew
suit Zeus
news
(19) plot blot

press brick
flock
frock
trap dress
thrill
clock glad
crack _ grill
We know from the previous chapter that the forms in (19) have a complex
onset. Those in (18), by contrast, would have a complex nucleus according
to the analysis sketched in the previous section (cf. (17) above). If so, the
nucleus of forms like twine or queen would have three members.
This additional complexity in the nucleus cannot be welcome, but we will

have to accept it if life is indeed like that.
Life, however, is not quite like that. First, twin rhymes with tin, bin, sin,
etc., in the same way as play or pray rhymes with pay. The relevance of this
is that, as we know from the previous chapter, the criterion for rhyming in
English verse is that the segments in the syllable rime be identical. Therefore,
if the left-most high vowel in the words in (18) were part of the rime, twin
would simply not rhyme with tin, and so on. The fact that twin does rhyme
with tin therefore suggests that the high vowel represented byw in the spelling
is assigned to the onset.
The possibility of assigning high vowels to the onset receives specifically
High vowels can phonological support from the absence from the English native vocabulary
be included in the
of forms with the structure labial consonant + labial high vowel + another
onset
vowel. This sequence does occur in words of Spanish or French origin:
pueblo, bueno, poids, foie, etc., with a labial consonant followed by [we] or
[wa] respectively, as in the expressions “Pueblo Indians”, “Buenos Aires”,
“avoirdupois”, “foie gras”. In native English words, however, sequences
consonant + labial high vowel + another vowel are limited to the environ-
ments we listed in (18a). We transcribe these phonetically in (20), where we
are representing onset /u/ as [w] to highlight its non-nuclear status (a less
common alternative transcription is [uy], with the non-nucleus status signalled
by the underscripted diacritic “_”):
(20) a. [6w] [tw] [dw] [sw] b. [kw] [gw]
Notorious absentees from this list are, of course, the clusters [pw], [bw], [mw],
[fw] and [vw].
Can you a y
t be
— should no
In order to understand the bearing these gaps have on the role of [w] in
the syllable, we need to take note of the fact that [tl], [dl], [61] and [dl] are
also absent from English onsets. There are simply no words beginning with
such clusters in English ([0] and the other voiced fricatives do not occur
in branching onsets of any kind). Moreover, in a word like atlas, where one
of the clusters occurs word-medially, the [t] and the [l] do not join as a
complex onset. Instead, the [t] makes up the coda of the first syllable: at.las
(remember that “.” represents a syllable boundary). We return to this matter
in section 7 below, simply trusting for the moment that the syllabification
at.las does not conflict with your intuition.
The reason [tl], [dl], [61], [01] are banned from the onset, while [pl], [bl],
[fl], [kl] and [gl] are not, has to do with the fact that [1] shares the specifica-
tion [coronal] with [t], [d] and [6]: all these sounds are alveolar or dental,
that is, [coronal] in distinctive feature theory. Like many other languages,
English dislikes segments with an identical place of articulation in the same
subsyllabic constituent, a situation akin to phonological incest! The reason
for the non-occurrence of the onset clusters [pw], [bw], [mw], [fw] and [vw]
is of course similar, since both the sounds in these clusters have the feature
[labial].
The tendency of constituent siblings not to have similar places of The tendency for
articulation is stated formally in the picturesquely named OBLIGATORY constituent siblings
CONTOUR PRINCIPLE (OCP). The principle, and its label, originated in the not to have similar
context of tonal phonology (tone will be discussed in chapter 14). While places of articula-
tion is formalized
all-pervasive in various ways in all components of phonology, with the
as the OBLIGAT-
principled exception of the timing tier for obvious reasons, the OCP is sub- ORY CONTOUR
ject to a considerable number of exceptions, and its precise status therefore PRINCIPLE
remains uncertain.
The OCP is in fact more a tendency than a principle as such. We offer a pro-
visional and deliberately somewhat loose definition in (21):
(21) OCP (Obligatory Contour “principle”):

Similar melodies are disfavoured as constituent siblings.
If high vowels can, after all, occur in the English onset, we need to weaken
our current conception of the distribution of segment categories among the
Syllable margins
constituents of the syllable. In particular, we will now have to say that syl- are mainly (NB
lable margins are mainly (NB crucially, not only) occupied by consonants. In crucially, not only)
order to provide a formal basis for this situation, we need to introduce an occupied by
additional level in our sonority scale, as we do in (22). This version obviously consonants
supersedes its predecessor in (5) above:
(22) Sonority scale (second version):

Most sonorous 5 Non-high vowels
4 High vowels
3 Liquids
2 Nasals
We can now answer the question from the previous chapter as to why
we formulate the sonority distance restriction on English onset siblings as
a minimum of 2, rather than giving it a fixed value. The answer is that such
legitimate onsets as [tw] or [kw] have a sonority distance of 3 (NB not 2)
on the scale in (22): [t] and [k] are obstruents, and [w] (= [u]) a high vowel,
thus 4 — 1 = 3.
The conclusion that high vowels may be included in the English onset leads
to a new statement of English onset phonotactics, as in (23), where refer-
ence to the new sonority scale in (22) replaces the brute-force onset template
of (33) in chapter 9:
(23) a; Oo Db . IN
R N’
=
ao Goy
Sonority conditions: (i) y<4

(ii), -x S41
GH ie
Notice that (23) is more accurate than its predecessor, but this accuracy is
achieved at the cost of greater formal complexity, in the form of conditions.
Indeed, as is becoming increasingly evident and will be confirmed as we go

along, the formal description of English syllable phonotactics is a matter of
considerable intricacy.
Syllabification of [iu]
In the previous section we split the sequence [w] + vowel in words like twin
or quack between the onset and the nucleus. This analysis is compatible with
a formalization of the [w] and the following vowel as independent lexical
segments: in the same way that we assume that the [1] and the [ee] in crack
are lexically independent, the [w] and the [ze] in quack can be /u/ and /x/,
respectively, in the lexicon. At first blush, the situation with the sequence
[iu] in fume is analogous.
The sound [iu] occurs in the words in (18b, c) above, repeated now as (24)
(remember that in some accents, General American for instance, the forms
in (24b) do not have [iu:], but simply [u:]):
(24) a. pew beauty cute argue

few . view
mute
b. tune dune
Thew
suit Zeus
news
The vowel cluster [iu] is idiosyncratic in several respects. For instance,

articulating [iu] involves no raising of the tongue in moving from the first
to the second phase, contrary to what happens in the English primary diph-
thongs. Also, [iu] is the only instantiation of the sequence [i] + vowel after a
consonant: in a simple onset, [i] can be followed by pretty well any nucleus
vowel (see (14) and (16) above), but as part of a complex onset it only com-
bines with the nucieus [u:] (or its reduced versions [a] and [vu], where
appropriate), as the examples in (24) above illustrate.
Thus, English speakers avoid the sequence [i] + vowel (in one syllable)
unless the vowel is [u] and/or no tautosyllabic consonant precedes the [i]
(the prefix TAUTO-, which also occurs in “tautology”, signifies ‘same’).
One set of exceptions to this generalization, irrelevant for our immediate pur-
poses, will be accounted for in chapter 18.
The dislike of English for [i] in sequence with vowels other than [u] is
forcefully brought out by the English pronunciation of such Spanish words
as fiesta, siesta, (San) Diego, Santiago, etc. In Spanish, the vowel sequence is
diphthongal: [ie], [ia], etc. (we mentioned above that the underscripted dia-
critic “_” signals lack of syllabic independence). In English, however, such
vowel sequences are systematically pronounced with a hiatus ([i.e], [i.a], etc.),
except possibly in fast speech, where many of the restrictions we are dis-
cussing (although by no means all) appear to be flouted.
The apportioning of all post-consonantal [i]V sequences but [iu] to two dif-
ferent nuclei shows that English generally rejects the integration of the /i/
English generally of /iV/ into a complex onset. The question that obviously arises is why such
rejects the integra-
integration is not rejected in [iu]. One reasonable answer would be that [iu]
tion of the /i/ of
/iV/ into a com-
is a diphthong, since it seems sensible to assume that the very nature of diph-
plex onset. Such thongs prevents them from being split between two syllables: all uncontro-
integration is not versial diphthongs of English ([a1], [o1], [ou], etc.) are indeed tautosyllabic.
rejected in [iu] Now, if [iu] is a diphthong, we might expect it to be allotted whole to the
nucleus.
There is, however, specific evidence that the [i] of [fu] is assigned to the onset.
In the early seventeenth century, there was a diphthong [yl], with the syl-
lable peak’ on [1], and the second element [uy] as an offglide, in much the
same way as in the current primary diphthongs [au], [ou], etc.: indeed, there
seems to be a requirement for the first nucleus element to be the syllable
peak. By the end of the seventeenth century, however, the peak of [ru] had
been transferred to the second element, [u], and [i] raised to [i]: [fu]. All these
facts make it reasonable to construe present-day [{u] as a diphthong with a

lexicalized peak on [ul]. Given this analysis, the assignment of the [i] to the
onset follows automatically as a matter of formal necessity: from this per-
spective, [iu] will be [ju], where the symbol [j] indicates onset status, paral-
lelling the use of [w] for onset [u] ([y] is often used for [j] in North America,
rather confusingly from an IPA perspective).
This outcome is backed up by empirical evidence. Thus, English system-

atically lacks [ju] after onsets with two consonants, even in accents that admit
the sequence coronal + [ju] in the onset: compare bi[u] (NB not *bi[ju]) with
the possible I[ju]d (lewd), I[juJrid (lurid), and even I[julte (lute), I[juldicrous
(ludicrous), etc.
Assuming that onsets cannot have more than two elements, in English and
perhaps in other languages, an already binary onset such as [bl] will simply
leave no room for [j] — thus failing to be licensed, [j] will not surface. This
analysis crucially presupposes that the /i/ of /iu/ is assigned to the onset:
if it were assigned to the nucleus, it would have been licensed there, in some
way, and it would have been pronounced in all cases.
Further evidence for the assignment of [i] of [fu] to the onset comes from
the selection of the “weak” article allomorphs th[a] and a before [j] (a ewe,
th[a] ewe), as they always are before an onset (a/th[a] bow, a/th[a] blow), but
not otherwise (an/th[i] owl).
Finally, and most revealingly, a handful of receding accents do have [bliy],
and [ry] in general for [iu], with the first element of the sequence as the peak.
n be possible.
The behaviour of the [i] of /iu/ confirms that high vowels can partake in {igh vowels can
the English onset. In section 4 we showed that sonorant consonants can be partake in the
assigned to the nucleus. These two complementary facts obviously force the _English onset
relaxation of our early identification of ONCd with CVC.
[4 Onset Fulfilment
Our “parsing” of segments into syllables and into syllabic subconstituents

has been guided by the sonority profile of the sequence of segments — the
The word PARS- word PARSING denotes allotment of elements to constituents, and is well
ING denotes allot-
worth becoming familiar with. In a nutshell, English peaks normally require
ment of elements
to constituents a sonority level of at least 4 on the scale in (22) above, that is, they need to
be vowels. We illustrate in (25) with the words traffic, napkin and pumpkin:
(25) N } 1
East ce k n a p k I n DeAnm pk n
5 5 Qe——
7
BH it 2 5 Lea 4 2 dl ——Z
u> Rs — 2
ZPH
Segments to the left of the nucleus are parsed in the onset, to a maximum
of two, and segments to the right of the nucleus in the coda, at the moment
also to a maximum of two:
(26) a. Oo o Dees Oo ees o
PT
OPK eO ik OR OR OR OUR
sda
hs N Cd
eee di
NCd | NCd N Cd N Cd
Coa eee treks

Je. Wc
at poe
Ree et p AE pK lee
|
Lae eee Dene bed eee ee a2 1 eee
This parsing is, of course, conditional on compatibility with Sonority

Sequencing and the other general principles of syllabification. The lowering
of the sonority requirement on nuclei to level 2 on scale (22) that we dis-
cussed in section 2 is but a strategy to circumvent one specific problem for
Sonority Sequencing without paying the price of deleting segments, which
Languages show a would be the obvious alternative.
strong tendency
to assign an inter-
vocalic consonant
to the onset, as
encapsulated in
the principle of
MINIMAL ONSET
SATISFACTION An intervocalic consonant such as /f/ in traffic qualifies in principle as
an onset or as a coda. As it happens, languages show an overwhelming
tendency to assign an intervocalic consonant to the onset, as in (26a) above.

We express this situation in the principle of MINIMAL ONSET SATIS-
FACTION in (27):
(27) Minimal Onset Satisfaction:

Minimal satisfaction of onsets takes priority over satisfaction of codas
Following from Minimal Onset Satisfaction, a sequence... VCVCV... will

be syllapined ...... V.CV.CV ... >, and not... VC.VCV, ..
The syllabification ... V.CV.CV . . . of course tallies in with our finding in chap-
ter 9 that CV is the universal core syllable: a parsing ... V.CV.CV ... creates
two such core syllables, whereas a parsing... VC.VC.V ... creates none.
Specific evidence for the principle of Minimal Onset Satisfaction is also
available. Consider such words as carriage, car and cart. In non-rhotic accents,
there is no phonetic [4] in car and cart, since these accents by definition do not
admit [1] in codas. In words like these there is of course no other possible
parsing for the r. In carriage, however, the /1/ can also be parsed as an onset,
since it is followed by a vowel — the question is, will it indeed be parsed
as the onset of [1], or will it still be parsed as the coda of [a]? The fact that
non-rhotic speakers pronounce such an /1/ answers our question: the /1/
is parsed in the onset, for otherwise it would have no phonetic realization
in these accents. This result obviously supports the principle of Minimal Onset
Satisfaction in (27).
Onset Maximization. English Stop Allophony
Consider the forms in (28), with two or three intervocalic consonants:
(28) recline entreat comply surprise contrive

redress congress comfry pamlf]let conflate
recruit encroach actress undress poultry
include en[@]ral impress culprit portrait
The question that arises in connection with the intervocalic consonants in

(28) is whether they will all be parsed in the onset of the second syllable or
whether they will be split between the two syllables, and if so how.
Intuitively, the relevant syllable divisions in the words in (28) are as in (29):
(29) re.cline en.treat com.ply sur.prise con.trive

re.dress con.gress com.fry pam.[fJlet con.flate
re.cruit en.croach ac.tress un.dress poul.try
in.clude en.[6]ral im.press cul.prit por.trait
The formal procedures we have in place also allow for the parsings in (30),
still assuming complex codas:
(30) recline ent.reat comp.ly surp.rise cont.rive

red.ress cong.ress comf.ry pamlf].let conf.late
rec.ruit enc.roach actress und.ress poult.ry
include en[6].ral imp.ress culp.rit port.rait
Notice that the parsings in (30) do not incur any violation of Sonority
Sequencing, since the resulting syllables do not contain sonority troughs.
The parsings in (30) are obviously also compatible with the principle of
Minimal Onset Satisfaction in (27) above.
Intuitions are to be taken seriously in phonological (and other linguistic)

research, for they are often, if not invariably, the tip of the iceberg of phono-
logical (or linguistic) cognitive structure. The phonologist, however, wants
to find empirical confirmation of such intuitions, which may conceivably be
coloured, or even caused, by extralinguistic factors.
Empirical evidence for the parsings in (29) does indeed exist. This evidence
concerns some stop allophony, and it confirms the principle of Minimal Onset
Satisfaction (27).
In chapter 1 we illustrated the difference between English voiced and
voiceless stops with word-medial, rather than word-initial, instantiations.
The reason is that in word-initial position the contrast is not implemented
by means of simultaneous vocal fold vibration or its absence, respectively.
Rather, it is a function of the length of the time lag between the release of
the closure and the start of vocal fold vibration for the following vowel:
in “voiceless” stops the time lag is considerable, and in “voiced” stops very
short, or possibly null. Word-initially, therefore, English “voiced” stops are
actually voiceless! Word-initially,
English “voiced”
stops are actually
voiceless
Consider, for instance, the pairs of words in (31):
(31) “ar pie b. buy

tie die
cow guy
The period between the release of the closure and the start of vocal fold activ-
ity for the vowel is clearly longer in (31a), where it can be felt physically as
a puff of air known as ASPIRATION. Aspiration characterizes voiceless stops

Aspiration charac- in English, not only word-initially, but also word-internally in the onset of
terizes voiceless
a stressed syllable (we trust that you can identify the “stressed syllable” intu-
stops in English
itively; we examine stress in detail in the next three chapters): paper, repeat,
terminal, deter, cover, recall, etc.
~..
Now, the parsings in (29) above are confirmed by the aspiration test.
‘ . . Se
The aspiration test is only |

Which are these? y
Thus, in words like surprise or recline the obstruent stop is aspirated, to the
extent of rendering the following sonorant voiceless (recall that there are no
lexical voiceless sonorants in English and in most other languages).
The narrow phonetic transcription of the relevant sounds is therefore as in

(32), where the underscripted circle of course represents voicelessness:
(32) surlpilise re[kl]ine
It could be thought that devoicing is simply caused by the adjacency of

the two relevant segments, irrespective of their syllabic constituency: accord-
ing to this construal, syllable structure would be irrelevant to the spread of
voicelessness. Consider, however, forms like those in (33), already referred
to in section 4 above:
(33) atlas athlete
On the surface, the consonantal clusters [tl] and [61] are similar to their coun-
terparts in (32). However, if you observe your own pronunciation of [tl] and
[61] you will notice that [1] is now fully voiced, even though it also follows
a voiceless obstruent. This situation follows from the syllabifications in (34):
(34) atlas ath.lete
There is in fact a good reason for the contrast in syllabification between the
forms in (32) and the forms in (33): as we mentioned in section 4 above, the
OCP in (21) prevents the occurrence in the same constituent of clusters with
a similar place of articulation, such as [pw], [bw], [tl], [dl], [61], [Ol].
Restate tt
The facts of aspiration therefore show that the consonant clusters in (28) above
(recline, comply, etc.), in all respects similar to those in (33), are parsed as
complex onsets.
Likewise, most speakers may pronounce t glottally in atlas, but not in petrol,
for instance: the reason is, again, that in petrol /t/ would be parsed, not in the
coda, but in the onset.
The conclusion to draw is that intervocalic consonant clusters are allotted
to the onset unless prevented by the familiar conditions on syllabifica- In English, two
intervocalic con-
tion. We formulate this bias towards the onset as the principle of ONSET
sonants are parsed
MAXIMIZATION: in the onset unless
prevented by the
(35) Onset Maximization: syllabification
Maximal formation of onsets takes priority over formation of codas conditions
It would seem at this point that Onset Maximization subsumes our previ-
ous Minimal Onset Satisfaction principle in (27) above. This is, however, not
so, since the two principles can show a different strength of application. For
instance, in French the voiceless alveolar stop f is usually not pronounced
at the end of words (pet[i] ‘small’), although it can be pronounced when the
_ following word beings with a vowel (pet[it] enfant ‘small child’). One simple
way of formalizing this contrast involves a prohibition against parsing /t/
in the coda, similar to the prohibition against parsing /1/ in the coda in English
non-rhotic accents. Now, the fact that the final /t/ of petit is pronounced in
the phrase petit enfant indicates that Minimal Onset Satisfaction has impelled
the parsing of this /t/ as the onset of the next vowel: pe.ti.ten.fant. What is
interesting, and directly relevant to our present concerns, is that such parsing
does not take place in a phrase like petit roi ‘little king’, which is invariably
petlixloi (not *petlits]oi), even though the sequence [ts] is otherwise quite
legitimate in French onsets: [ts]ois ‘three’ (cf. pet[il [tJois ‘little three’!). The
reason for the opposing behaviour of the word-final /t/ in petit enfant and
petit roi is that, in French, Minimal Onset Satisfaction is operative across
words, hence pe.ti.ten.fant, but Onset Maximization is not, hence pe.tit.rot
(= .pe.tit. + roi.): Minimal Onset Satisfaction is therefore stronger than Onset
Maximization.
It may be helpful to construe Onset Maximization and Minimal Onset

Satisfaction as the maximal and minimal implementation, respectively, of a
more general ONSET FIRST PRINCIPLE favouring onsets over codas. This
A general ONSET result may be derivable from astill more fundamental principle favouring
FIRST PRINCIPLE
a maximal jump in sonority at the start of the syllable, and a minimal drop
favours onsets
in sonority at the end.
over codas
Explain
_ under: iple favouring a maxi
.
EE] No Complex Codas in English
The structure of the syllable as we have it at the moment is fairly symmet-

rical, and thus quite pleasing. Summing up, syllables are made up of two
constituents, the onset (optional) and the rime (obligatory), the latter made
up of the nucleus (obligatory) and the coda (optional). The melodic contents
of each subconstituent are restricted on the basis of sonority: nuclei prefer
vowels, although they accept sonorant consonants under duress, and margins
prefer consonants, although the onset can also accommodate high vowels.
Each subconstituent is maximally binary, at least in English. Further prin-
ciples governing syllabic parsing are Sonority Sequencing, Minimal Sonority
Distance (both related to sonority), Minimal Onset Satisfaction, Onset Maxim-
ization, and the OCP.
Some word-final consonant clusters are problematic for our present
approach. Consider first the forms in (36):
(36) clamp tent link

tend
help bolt milk

bulb _ gild
film kiln
These words seemingly contain a complex coda made up of a sonorant fol-

lowed by another consonant. The sequences in question clearly obey Sonority
Sequencing, since the first sonorant invariably has more sonority — if it had
less, an additional syllable would be formed, as we explained in section 2
(cf. sickle vs. silk). In fact, the clusters in (36) almost mirror their counterparts
in onsets, as we mentioned in section 1: complex onsets allow for a cluster
C + liquid, and the complex codas in (36) for a sequence sonorant + C.
The difference between the two cases therefore hinges on minimal sonority
distance: 2 in onsets (liquids = 3, obstruents = 1), but apparently nil in codas,
given forms like kept, fact, etc. This poses an obvious problem for Sonority
Sequencing.
A further difference between complex onsets and the putative complex
codas is that complex onsets occur freely word-medially, as well as word-
initially: indeed, the formation of word-medial complex onsets is favoured
by Onset Maximization (35).
By contrast, the distribution of complex codas is heavily skewed towards

word-final position, at least in English. This statement may appear to be wildly —_English complex
inaccurate, in the face of forms like parenthood, thankless, boldness and a great Cae
many others. However, we will argue in chapter 16 below that the clusters ig goa postion
in question are effectively also word-final in these cases: parent, thank(s),
bold, etc., are words in their own right. Another group of apparent counter-
_ examples has a stop between a homorganic nasal and an obstruent ~ the
stop can be analysed as a purely phonetic transition, that is, as non-lexical:
pumpkin and one or two other forms ending in kin (pumpkin is actually pulm)]kin
for many speakers), resumptive, redemption, assumption, and so on. Among
the few truly genuine examples of word-medial complex codas is arctic in
rhotic accents, but a pronunciation a[at]ic or [a:t]ic, with no [k] is far more
common.
The effective restriction of complex codas to the right edge of the word
is obviously puzzling. One way out of the conundrum involves abandoning
the complex coda analysis for the consonant clusters in question. This move
receives support from forms like those in (37):
(37) claim leak

hail _ like
bowl
file
The forms in (37) end in a consonant preceded by a long vowel or a diph-

thong. Again, this configuration is not found word-internally — the only genu-
ine counterexamples are the rather technical deictic, deixis and seismic: in words
like chamber, dainty, council, ancient, angel, poultry, shoulder and a handful
of others the “coda” sonorant shares its place of articulation with the onset,
which can therefore be argued to license it, in a way we will explain in
chapter 16.
The common denominator of (36) and (37) is, of course, the presence of
an extra consonant after an ordinary rime word-finally: this is the analysis
we will adopt. In particular, we will assume that the word-final consonant
We will assume of English words affiliates directly to the o node, not indirectly via the R
that the word-
node, as we would expect:
final consonant of
English words
affiliates directly (38) ">a. Oo b.
to the o node
This analysis has the advantage of allowing us to preserve the statement that
English codas are simple throughout the word: the coda is the consonant
licensed by R, and so the word-final consonant in (38) does not partake in the
coda. At the same time, such a final consonant remains inside the syllable,
and therefore the compliance of such clusters with Sonority Sequencing is
correctly predicted (cf. (8a) vs. (8b) above): the exception inevitably involves
final coronals (act, apt, and so on), and will be seen to directly (remember
that we said in section 1 that when both siblings in an apparent coda are
non-sonorant stops, the right one needs to be coronal).
As we have just hinted, coronal obstruents can follow any type of segment
word-finally, irrespective of sonority, to complicate the English syllable still
further. We show this in (39):
(39): a. ~ left raft lift

b. act apt concept erect
c. buzzed
d. lapse silks]
e. bread[6] wid[6@]
Coronals appear
Indeed, coronal obstruents appear to occur unrestrictedly on the right edge
to occur unre-
of the English word. A rather spectacular instantiation of extra word-final strictedly on the
coronal obstruents is offered by the forms sixths and contexts, with no fewer right edge of the
than three coronal obstruents after /k/, at least in careful diction: [stks@s], English word
[konteksts].
Situations of this kind pose a serious challenge to the constrained frame-

work we have been endeavouring to develop for the syllable.
We return to these issues and advance a solution in chapter 16.

[The Antics of /s/
The limitation of the English coda to one consonant that we have just pro-
posed appears to be contradicted by /s/, which can follow any coda con-
sonant word-internally: institute, transfer, abstain, instruct, conscript, obstacle,
etc.
Faced with these facts, we could think of accepting complex codas with /s/
as their second element. However, this solution would be problematic for
Sonority Sequencing as it stands at the moment, as we illustrate in (40) —
remember that stops and fricatives are level in sonority in the scale in (22)
above:
(40) clasp crisp tusk risk
You can see that, as a possible first element of the new complex coda, /s/
patterns with sonorants, even though /s/ is level with stops in the sonor-
ity scale in (5). One answer would be to split fricatives and plosives in the
sonority ranking, in a more precise sonority scale — objectively, fricatives
are indeed more sonorous than stops:
(41) Potential sonority scale:

Most sonorous 6 Non-high vowels
5 High vowels
4 Liquids
3 Nasals
2 Fricatives
Least sonorous 1 Stops
The /s/ + stop coda now obviously complies with Sonority Sequencing.
However, /s/ is the only fricative licensed in the English coda before non-
coronals, as attested by the impossibility of forms like *lefp, *lefk, etc. (in con-
trast to left, which ends in a coronal). Therefore, the scale in (41) still falls
short of solving the problem, and therefore we shall not adopt it.
The idiosyncrasies of /s/ extend to the onset. Consider the words in (42):
(42) a. ‘slum
b. small snow
€. spy sty sky
sphere
These forms apparently have a complex onset. However, the parsing of the
initial sC cluster in the onset would put some of our current generalizations If the initial sC
in jeopardy. cluster were
analysed as an
onset, some of
the standing gen-
eralizations would
be put in jeopardy
First, an onset sC would be incompatible with the setting 2 we have given

to the Minimal Sonority Distance parameter: in the forms in (42b) the dis-
tance between s andC is only 1 (obstruents = 1; nasals = 2), while in (42c)
the score is even (1, since voiceless stops and /s/ are both obstruents), all
according to the hierarchy in (5) above. Next, an onset sl, as in slum, would
violate the OCP, which we know bans *[tl], *[dl], *[61] from the onset. Like-
wise for an onset shr [fx], as in shrill, best analysed as derived from /s1/ by
assimilation, since [si] never occurs word-initially in English, even when
prompted by the spelling: Sri Lanka.
The problems for an onset parsing of sC clusters get compounded in forms

like those in (43):
(43) spray stray scrap

splay sclerosis
spew stew skew
squash
These consonant clusters can only be parsed in the onset if we admit three-
member onsets:
(44) [spi] [sta] [ska]

[spl] [skl]
[spj] [stj] [skj]
[skw]
ing |else speci
You are well aware that an increase in the size of syllabic subconstituents
is undesirable, on grounds of grammatical stringency. Moreover, the fact that
the first segment in such putative three-member onsets is always [s] would
be a coincidence.
The changes in onset sonority distance that took place in the historical
development of English also militate against the inclusion of /s/ in the onset.
In Old and Middle English, forms like knave and gnat were pronounced as
spelled, on a par with the contemporary German knabe [knaba] ‘boy’ or
gnadig [gnedic] ‘gracious’. In Early Modern English, the initial stop was lost
(although it survived in the spelling). By contrast, initial [s] was retained,
not only prenasally but, more generally, preconsonantally.
The behaviour of
inclusion in the onset.
/s/ is idiosyncratic
across the board,
in English and
in many other All the facts we have considered provide strong motivation for granting word-
languages initial [s] special status, in English and in other languages. We formulate a
specific proposal along these lines in chapter 16.
Chapter
In this chapter we have investigated fu

fication, concentrating specifically on
lish branching onsets and on codas. Branching onsetstypically consist of an
obstruent followed by a sonorant consonant. We came to the conclusion
lar hing codas, although an extra
Key Que tions
What are the basic constraints on Which two principles can be com-
apparent complex codas in English? bined to make up the Onset First
Under what conditions may con- Principle? What is the difference
sonants form nuclei in English? in the provisions of these two
Under what conditions may vowels principles?
occur in an onset? Which vowels? How does the possible word-
What evidence is there for high internal coda differ from its appar-
vowels being analysed as part of an ent word-final counterpart?
onset, when followed by another Some words end in clusters of
vowel, rather than forming part of three or four consonants. What is
a complex nucleus? special about these consonants?
What is the Obligatory Contour Can the extra consonants be con-
Principle (OCP)? What influence sidered part of a complex coda?
does it have on onset formation in 10 List the problems inherent in a
English? claim that sC clusters may form
What is unusual about the complex onsets.
sequence /iu/ in English? How is it
syllabified?
FU rete heir 2aPs fean Celt peace

French
French high vowels can be syllabified in two different ways ([y] is the sym-
bol used for [y] in non-nuclear position, in parallel with [w] for [u] and [j]
for [i]):
a. il joue [ilzu] ‘he plays’

jouer [Zwe] ‘to play’
jouable [3wabl] ‘playable’
il tue [ilty] ‘he kills’

tuer [tye] ‘to kill’
contribution [k3Stribysj5] ‘taxJ
contribuable [k5tribyabl] ‘taxpayer’
il lie [ili] ‘he ties’

lier [lje] ‘to tie’
colonie [koloni] ‘colony’
colonial [kolonjal] ‘colonial’
(i) Can you see any reason for the different syllabifications?
Compare the data in a. with those in b.:
b. plier [plie] *[plje] ‘to fold’

prier [prie] *[prie] ‘to pray’
clouer [klue] *[klwe] ‘to nail’
trouer [true] *[trwe] ‘to make a hole in’
cruauté [kryote] *[kryote] ‘cruelty’
influence [€flyds] *[€flyds] ‘influence’
(ii) What do these new facts suggest about the structure of syllabic sub-
constituents in French?
Southern Welsh Rime Phonotactics
Monophthongal Southern Welsh vowels, except for schwa, fall into groups,
long and short. The list is as follows:
{/WTOVSAaVa
Oro UL Oo,
In some circumstances the long and short vowels can be shown to be con-
trastive but in other situations only one or other group may occur. Consider
the following representative examples of monosyllabic words:
a. /di:n/ 4
man ,
/gwin/ ‘white’
/hemn/ ‘old’ /pren/ ree
/ta:n/ ‘fire’ /man/ ‘place’
/somn/ ‘speak’ /bron/ ‘breast’
/suin/ ‘noise’ /grun/ ‘ridge of ploughland’
/bitd/ ‘world’ /jet/ ‘gate’
/te:d/ ‘width’ “to!
/at/
/bo:d/ ‘to be’ /tok/ ‘sheepfold’
/ta:d/ ‘father’ /krut/ ‘boy’
/knu:d/ ‘crop’
/pri:d/ ‘earth’ /ritx/ ‘furrow’
/be:0/ ‘grave’ /hwe:x/ “S1x’
/ta:0/ ‘to kill’ ‘pigs’
/morx/
‘way’
/mo:0/ /hurx/ ‘sow’
/ki:t/ ‘hazel’ /gweit/ ‘better’
/tart/ ‘other’ /ho:t/ ‘hole’
‘dog’
/ki:/ /te:/ ‘place’
/da:/ ‘good’ /to:/ ‘roof’
/tu:/ ‘oath’
/pint/ ‘pound’ /gwer8/ ‘value’
/plant/ ‘children’ /golx/ ‘children’
/kusk/ ‘sleep’
(i) Is vowel length contrastive in all environments?

(ii) What conditions are there on the distribution of long and short
vowels?
(iii) Make a suggestion about the shape of the rime in southern Welsh.
(iv) Compare the possible rime in southern Welsh with that of English out-
lined in section 8.
THE PHENOMENON OF STRESS
RHYTHM
Chapt
sequences.
m Syllable prominence interpret
The difference in stress loc
Uy
In the previous two chapters, we argued for the enrichment of phonolo-

gical representation with abstract PROSODIC structure, over and above
the structure that corresponds to the linear arrangement of segments and to
the relations between the features inside the segments — the word prosody
ultimately derives from the Greek pros ‘to’ + didé ‘song’, with the mean-
ing ‘song sung to music’, or, more to the point here, ‘tone of a syllable’. So
far we have identified such prosodic structure with the syllable, and we
have gone to some lengths to present and justify the internal architecture of
this construct. In this and the next two chapters we will examine a level of
prosodic organization larger than the syllable. The principal manifestation
of this higher prosodic level is what is commonly known as “stress”, and
in this chapter we provide ample illustration of the phenomenon, conveniently
tapping the intuitions of the English speaker. The specific prosodic nature of
stress will become clear as the exposition proceeds, in this and subsequent
chapters.
The Phenomenon of Stress 295
Syllable Prominence
In line with the hands-on methodology we are adopting, let us try to gain
first-hand experience of the phenomenon of stress by pronouncing the
words in (1):
(1) increase imprint relay torment upset

contest contract escort decrease protest
contrast implant survey convict digest
Pronounce the
The task of pronouncing the words in (1) ought to be child’s play for any
fluent speaker of English, but it turns out to be tricky, for the simple rea-
son that each of the words in question can be pronounced in one of two
ways, which we represent impressionistically in (2):
(2) INcrease inCREASe

CONtest conTEST
CONtrast conTRAST
A eee eee ew eer eenensen ee eneneeeeseseseeee
Each of these pairs of words is basically made up of the same string of seg-
ments — indeed, the spelling is constant. The words in the first column of
(2) have, however, more prominence on the first syllable, while in those in
the second column the second syllable is more prominent. We hope that this
contrast will be obvious to any fluent English speaker, but we contextualize
it in (3) to remove any possible doubts:
(3) a. another tax increase b. taxes continue to increase

when is the beauty contest? when will the beauties contest?
an unexpected contrast they unexpectedly contrast
The words we are examining are of course being used as nouns in col-
umn a. and as verbs in column b. We are now seeing that this grammatical
296 The Phenomenon of Stress
difference goes hand in hand with a difference in the location of the word’s
highest prominence: the first syllable in (3a) and the last syllable in (3b). The
prominence in question is what we call STRESS. The data we have considered
thus show that one of the syllables of each English word is singled out as
Each English word the word’s stress carrier. .
has one of its
syllables singled
out as the word's
stress carrier Word Prominence
Let us take our awareness of stress one step further by considering the two-
word string in (4):
(4) time flies
We just saw that in each word one syllable has more prominence, that is to
say, carries more stress. The question now is whether one of the two words
in the sequence in (4) is also more prominent than the other, and, if so, which.
In a sequence
of words one of
the words carries
more stress than
the others
The answer will again come readily to any fluent speaker of English: the
second word, flies, is more prominent:
(5) time FLIES
We are, of course, assuming a “neutral” context, that is, a context that does
not involve a contrast with another word, as would be the case, for instance,
in space doesn't fly: TIME flies, with the emphasis on time for contrast.
The string in (6) looks very much like its predecessor in (4):
(6) time-flies
The meaning of time-flies is intended to parallel the meaning of the forms

in (7):
(7) mayflies
horseflies
greenflies
white flies
black flies
Time-flies is, therefore, the plural of time-fly, the potential name of a fly species.
Of course, no such object exists, but it could have existed and may well exist
in the future. This again shows that a language is much more than the words
in the dictionary, since time-fly, albeit a well-formed word in English, is not
likely to be found in any English dictionary.
The relevant question here is whether an utterance of time-flies will be

confused with an utterance of time flies. The two strings are in fact indis-
tinguishable with regard to both word identity and word order. They do
not, of course, look the same when we write them down, since in time-flies
we are separating the two words with a hyphen. However, it ought to be
obvious from our repeated comments that spelling conventions are simply
that, and cannot automatically be assumed to have a phonetic correlate. The
broad transcription in (8) confirms that the segments of the strings in (4)
and (6) are indeed identical:
(8) [tarm flaiz] “time flies” or “time-flies”
Puzzlingly, although the segments of the two phrases are identical, there
will be no confusion between time-flies and time flies when we say them.
At first sight, the statement that time-flies and time flies are pronounced dif-
ferently conflicts with their identical phonetic representation in (8). The conflict,
however, is resolved when we look for a difference beyond the makeup of
the segments.
While, as you now know explicitly, time flies has the main prominence on
flies (time FLIES), in time-flies time is more salient, assuming again a neutral
context:
(9) TIME-flies
As we have just hinted, flies in time-flies will be contrastively emphasized if

we are comparing such a fly species with another time-connected species:
I'm not talking about time-ants; it’s time-FLIES I’m interested in. Contrastive stress,
however, falls outside our present remit and, therefore, we will ignore it in
the rest of the discussion: essentially, any English word or syllable can be
stressed for emphasis.
Emphasis or no emphasis, the data in (4) and (6) reveal that in a sequence
of words one of the words has greater prominence than the others. We already
know that in each word one of the syllables is more prominent than the rest.
We shall give a common formal expression to these two facts by super-
imposing a specific hierarchical prosodic structure on syllables and on words.
We superimpose a This prosodic structure is functionally analogous to the prosodic structure
hierarchical struc-
that gathers segments into syllables and that we discussed in chapters 9 and
ture on syllables
and on words,
10, in as much as it also gathers elements into prosodic constituents.
functionally ana-
logous to the pro-
sodic structure
that gathers
segments into
syllables
The configurations of the two structures are, however, significantly differ-
ent, as we shall see. We present the prosodic structure associated with stress
in the next section.
Metrical Grids
We will now introduce a formal notation for the representation of the

prosodic structure we are proposing, and consequently for stress itself.
Consider once more the now familiar contrasts between (to) implant and (an)
implant, or time flies and time-flies. A straightforward representation of these
contrasts is given in (10) and (11), respectively:
(10) a. - b. - Stress line

es Foe ak Baseline
(to) implant (an) implant
(Li) a: - b. - Stress line

. i . . Baseline
time flies time-flies
These graphics can be interpreted at a glance: stressed elements are more

heavily starred than their stressless counterparts, a bit like higher-ranking
army officers being more heavily starred than officers of lower rank. Thus,
consider (10). In the baseline we mark with an asterisk all and only the seg-
ments that qualify to bear stress, normally the syllable heads — we said in
chapter 9 that the head of a constituent is the element defining the constituent:
the head of a syllable is, normally, a vowel. The reason that only syllable
heads can be stress bearers should be obvious in the context of the theory
of the syllable we presented in the previous two chapters.
ited to syllable heads.
Quite simply, we know that the syllable head constitutes the true core of
the syllable: it is its sonority peak. As a consequence, only syllable heads
qualify to bear stress, and it is this potential for stress that the baseline is
meant to formalize. In contrast to the stress potential formalized by the base-
line, the stress line signals the actual presence of stress on the element it Stress can be con-
singles out: a in (10a) and i in (10b), for instance. From this perspective, we silica as Wie Pro-
ae : : jection of a certain
can construe stress as the projection of certain syllable heads onto a higher aylablesheadonts
structural level (we came across the notion of “projection” in the context of a higher structural
syllable nuclei in chapter 9): formally, the baseline asterisk dominating the _ level
privileged syllable head is projected onto the stress line.
Diagrams (10) and (11) are made out of a series of rows intersecting with
a series of columns, in the style of a grid (more obviously so when the
number of rows and columns of asterisks is increased), hence the official
label METRICAL GRID. The exact import of the key word “metrical” will
be clarified as we go along. From now on, we will use metrical grids to for-
We will use malize the stress patterns of words and word collocations.
metrical grids
to formalize the
stress patterns of
words and word 3 Motivating Stress Constrasts
collocations
Having availed ourselves of a reasonable formal notation for the repres-

entation of stress, we will now try to find the factor behind the differences
between the two stress patterns in (2) (INcrease vs. inCREASe) and between
the two stress patterns in (4) and (6) (time FLIES vs. TIME-flies).
We have already pointed out that the different location of stress in the
words in the two columns in (2) can be attributed to the categorial opposi-
tion noun vs. verb: INcrease is a noun and inCREASE a verb. You may be
inclined to think that the contrast between time FLIES and TIME-flies hinges
on meaning: the stress shifts implementing contrastive emphasis in TIME
flies and time-FLIES are indeed related to meaning.
Note, however, that time flies has at least two different readings. The meaning
most likely to spring to mind is that time moves at great speed. The colloca-
tion has, however, a second possible meaning, just as real and legitimate, if
perhaps alittle less obvious.
This alternative meaning involves a command by the speaker to the hearer

to engage in the timing of flies (A: “What shall I do for fun?”; B: “Time flies!’”).
In this sense, time flies contrasts with time fleas or time ants, or, more real-
istically, with time horses or greyhounds. Of course, it also contrasts with kill
flies, collect flies, admire flies or even race flies.
What is directly relevant to the discussion is that this additional meaning

of time flies is not marked prosodically. The hearer must, therefore, interpret
time flies simply on the basis of the background and contextual information
available. This shows that there is no necessary connection between stress
and meaning. There is no neces-
sary connection
between stress
and meaning
A more productive line of inquiry into the stress difference between time
flies and time-flies involves grammatical structure. In particular, you will notice __ The stress differ-
that time flies (in either of its senses) is a sentence, with a subject, time,and ‘°" between
a predicate, flies, or alternatively a predicate, time, and a direct object, flies, ee che
depending on the sense (do not worry excessively if you are not fully con- grammatical
versant with this syntactic terminology). By contrast, time-flies is a (compound) structure
noun — we could, for instance, order someone to time time-flies. We provide
further illustration of the stress contrast between phrasal and compound con-
structions in (12):
(12)— green HOUSE GREENhouse

(the house is painted green) (where tomatoes grow)
black BIRD BLACKbird
(could be a rook) (the female of the species is brown)
playing CARDS PLAYing cards
(to pass the time) (in a pack)
Compounds, of course, abound in English, as we now (modestly) illustrate:
(13) child minder phone book dolls house

oak apple clothes horse horse box
house plant _ tie rack trouser press
alarm clock baby alarm coffee pot
In fact, English compound formation resembles sentence formation in having

no apparent limit. As Chomsky observed in his book Syntactic Structures,
which set the programme of generative grammar in motion in the late 1950s,
the number of sentences in English, or any natural language, is by its very
nature infinite. For a similar reason, most compounds we use in real life are
not in the dictionary.
Summarizing the discussion so far, we have seen that in words with more
than one syllable one of the syllables has more stress than the others, and that
in phrases or compounds one of the words also exhibits more prominence.
Moreover, we have seen that, in the data we have considered, the location
of prominence is not dependent on meaning, but, rather, on the category of
the word (noun vs. verb) or the word collocation (phrase vs. compound).
The Distribution of Stress in Personal Names
In this section we will examine a class of apparent counterexamples to the

stress patterns we have been observing in phrases, and we will suggest a
reason for the situation. Conveniently, this will involve the manipulation of
additional stress data.
Consider the English double names in (14), either attested or sufficiently
realistic:
(14) Sue Ann

Sarah Jane
Donna Jo
Peggy Sue
Maggie May
Mary Lou
Billy Jean
Peter John
Christopher Robin
In all the collocations in (14) the second name is more prominent (Sue ANN,
etc.), indicating that these constructions are stressed like phrases, rather than
like compounds. It is indeed a fact of English that some compounds receive
stress in the manner of phrases (and it is actually not clear how complex
names should be analysed syntactically).
First names can, of course, be followed by surnames:
(15) Sue Ann Cook

Sarah Jane Brown
Mary Lou Jones
rries the main
In these constructions, the main stress still falls on the right-most item, that
is, on the surname (Sue Ann COOK, etc.). Two further, related facts are directly
of interest here. First, the remaining two words (making up the double first
name) are not pronounced with equal prominence but, instead, one of them
bears more prominence than the other. We would expect this greater pro-
minence to be carried by the second word (Ann, Jane, etc.), on the basis of
the data in (14) above (Sue ANN, etc.). As it happens, however, it is the first
word (Sue, etc.) that bears this subsidiary prominence. This second fact is
intriguing in the light of the present discussion.
ee et
In order to solve the puzzle, we need to examine the grid representation

of the stress pattern in question:
(16) . Stress line 2

a e Stress line 1
cs = Baseline
Sue Ann Cook
Notice that we have now increased the number of stress lines in the grid,
to take account of the intermediate degree of stress carried by Sue. Indeed,
we will see as we proceed that the number of lines in the grids of word col- The number of
locations is in principle open-ended —it is simply a function of the number Cee
of words in the construction: the more words, the more grid lines. finckeelomie
Formally, the problem is why in the grid in (16) the line 1 stress should pumber of words
be located on Sue, since this stress is located on Ann when the double name __ in the construction
Sue Ann is said in isolation:
(17) a Stress line 1

ps Baseline
Sue Ann
We offer a solution in the next section.
[4 Stress Retraction under Clash
You will now see that our grid formalism provides an elegant resolution to
our paradox. Consider what the structure of the grid corresponding to the
string Sue Ann Cook would look like if Ann, rather than Sue, had a line 1
mark:
(18) m Stress line 2

Stress line 1
+ *+ +
Baseline
Sue Ann Cook
The grid in (18) contains a “stress clash”. There is a STRESS CLASH

There is a stress between two asterisks in any grid line when they are adjacent and there is
clash between two
no asterisk in between the corresponding pair of asterisks in the line imme-
asterisks adjacent
in any grid line
diately below. In (19) we show with the aid of a box that this is precisely
when there is the case in the grid in (18):
no asterisk in
between the cor- (19) = Stress line 2
responding aster- : - Stress line 1
isks in the line
rls Z Baseline
immediately below
Sue Ann Cook
The structure boxed in includes two adjacent asterisks in line 1, without any
asterisk separating the corresponding asterisks in the line immediately
below, the baseline.
o back to thediagram in (19)and pointatthe cla

English and lan-
guages with
similar metrical
English and similar languages strongly resist stress clash. The stress clash
systems strongly in (18) has of course resulted from the concatenation of the double first name
resist stress clash Sue Ann, which we saw in (17) has the main stress on Ann, and the surname
Cook, which bears the main stress of the entire construction Sue Ann Cook.
The clash between Ann and Cook is resolved by movement of the sub-
ordinate stress from Ann to Sue. From now on, we informally signal the
original location of the moved asterisk with an arrow. We also embolden
clashing asterisks, to make them more salient to the eye:
(20) $ * Stress line 2

= x * e = Stress line 1
wa . oo Baseline
Sue Ann Cook —~ Sue Ann Cook
Stress movement of this kind is very frequent in English, and is generally

seen as the result of the RHYTHM RULE in (21) - the motivation for the
label “Rhythm Rule” will be given in section 9: SiiSes moveRent
is very frequent
(21) Rhythm Rule: in English, and is
* * * * generally analysed
Ft Mee ee pt as the result of
the Rhythm Rule
You will notice that the left-most of the two clashing asterisks moves back
by one position: as a result of this movement, the output is free of clash.
The Rhythm Rule gives English much of its characteristic metrical fla-
vour, as even a modest survey of the phenomenon reveals.
Thus, the situation we just investigated with personal names is general in

the language. Compare, for example, the strings in the two columns of (22):
(22) a. South American b. South American music

Sydney Harbour Sydney Harbour Bridge
apple pie apple pie bed
In the shorter construction South American, the main stress falls on the sec-
ond word, American. As before, we would have expected this pattern to sur-
vive when the word music is added to form the longer phrase. Specifically,
the metrical structure of South American music ought to be a composite of
the metrical structure of South American and music. As we show in (23), how-
ever, this is not the case:
(23) * Stress line 3

< ‘ * Stress line 2
* * ; *) 4 Stress line 1
+ + * + + + a Baseline
* + * +
a. South American ‘’b. South American music
In South American music the second highest stress column rests, not on
American, but on South. This structure parallels that of Sue Ann Cook in (15).
Indeed, the reasons are the same, namely, avoidance of stress clash, in com-
pliance with the Rhythm Rule in (21) above:
(24) : Stress line 3

= 3 e Stress line 2
* 5 ri a 5 Stress line 1
* * + + * + + + + + + Baseline
a. South American b. South American music
By contrast, in South American modern music, American does retain its sub-
sidiary stress, simply because in this collocation American does not incur a
stress clash:
(25) = Stress line 3

: % Stress line 2
Ke - = $ Stress line 1
* + + + + + *+ + +
Baseline
In the construction in (25), made up of the shorter phrases South American

and modern music, the line 2 asterisks on American and music are separated
by a line 1 asterisk on modern. Consequently, no clash arises, and stress does
not have to move.
In this section, we have seen that stress clashes in English word colloca-
tions are resolved through movement of the first clashing asterisk to a word
further to the left. In the next section we show that leftward stress move-
ment can also happen inside words.
Word-Internal Stress Retraction
In all the cases we have examined, stress has ended up in a word to the left
of the word that originally bore it. Now consider the words in (26a) and
the collocations in (26b), made up of each of the words in (26a) followed by
another word:
(26) a. Suzanne b. Suzanne Morris

Michelle Michelle Pfeiffer
Heathrow Heathrow Airport
Berlin Berlin Wall
Dundee Dundee marmalade
Cornell Cornell hockey
Aberdeen Aberdeen Angus
All the simple words in (26a) have main stress on the final syllable (some
of the words admit idiolectal variation, but the patterns we are citing are
widespread):
(27) = Stress line 1

ee Baseline
Suzanne
When the words in (26a) appear in the larger collocations in (26b), however,
the subsidiary stress of the phrase invariably falls on their initial syllable:
(28) - Stress line 2

= is Stress line 1
eeis * * Baseline
Suzanne Morris
The high frequency of this stress retraction in real life can lead to perman-
ent stress shift. For instance, the pronunciation HEATHrow, with unetymo- The high ;fre-
logical stress on the first syllable, may nowadays be heard even when the "ry © eee
Rees ; : retraction in real
word is said in isolation or in non-clash contexts. fe Cxaiicad bs
Clash-triggered retraction is, of course, not restricted to proper nouns. For _ permanent stress
instance, the words in (29a) have a different stress contour in isolation and shift
in the collocations in (29b):
(29) a. antique b. antique chair

fifteen fifteen children
bamboo bamboo table
Here also, stress retraction in the collocation is impelled by stress clash:
(30) i = Stress line 2

= < fe Foe 2 Stress line 1
1 Ree - “aia e Baseline
antique antique chair — antique chair
Formally, the process falls within the scope of the Rhythm Rule in (21), clearly
The Rhythm Rule a fundamental principle for the organization of English stress.
i Meaeaa tes Interestingly, stress movement can cause the merger of the verb and
a pivotal principle : ; ;
P P : P noun patterns we mentioned in section 1:
for the organiza-
tion of stress in
English
Remind yo
patterns are.
(31) a. INcrease TAxes b. TAX INcrease

IMplant SYlicone SYlicone IMplant
DISpute BOUNdaries - BOUNdary DISpute
The collocations in a. are phrasal (Chancellor of the Exchequer: “What can

I do to avoid bankruptcy?”; Advisor to the Treasury: “Increase taxes!”), while
those in b. are compounds (TV newscaster: “The government is planning a
tax increase”). We know from section 1 that the verbs in (31a) normally exhibit
final stress:
(32) inCREASe taXAtion

imPLANT susPIcions
disPUTe inTENtions
In the phrases in (31a), however, the verb pattern merges with the noun
pattern in avoidance of stress clash, as we display in (33):
(33) y + Stress line 2

* 5 See Stress line 1
+ * + + * * + + Baseline
a. increase taxes — increase taxes
The stress clash in tax increase in (31b) obviously cannot be corrected by

leftward movement. It could be corrected by rightward movement of the
asterisk on in, but the English Rhythm Rule in (21) above does not provide
for rightward movement.
We return to this matter in the next section.

In all the data we have considered, the retracted stress has ended up in
the word’s first syllable, and therefore it would be possible for this to be
the decisive criterion in the choice of landing site for the moving asterisk.
However, the data in (34) reveal that the landing site is actually chosen on
grounds of asterisk adjacency in the line immediately below the clash: The criterion
responsible for the
choice of landing
(34) a: MonongaHEla b. MoNONgahela River pie iciactent
tyrannoSAUrus tyRAnnosaurus REX adjacency in the
line below the
clash
You can see that stress does not retract onto the word-initial syllable here.
The reason becomes transparent when we examine the grid:

" i < * Stress line 2
* a = ce Stress line 1
* + * + + *+ + *+ + + + + Baseline
Monongahela Monongahela River
The line 2 clash created by the concatenation of MonongaHEla and River in

(35) is again resolved by asterisk movement. The retracted asterisk lands on
the closest syllable with a line 1 asterisk. As it happens, this syllable is no
longer the word’s first syllable, but the non-initial syllable non.
FE] Retraction Failures: The Continuous Column Constraint
Phrases like antique chair, with stress retraction, strikingly contrast with com-
pounds like antique dealer, with no retraction in spite of the clash.
Consider the metrical structure of antique dealer:

. bY Stress line 1
ee * * Baseline
antique dealer
This structure contains a stress clash. However, the grid of antique dealer dif-
fers crucially from the grid of antique chair in (30) above with regard to the
position of the main stress.
Notice now what would happen if antique underwent stress retraction in

antique dealer:
(37) + z Stress line 2

‘ e ee “s Stress line 1
Pa + te Pere * * Baseline
antique dealer — antique dealer
The asterisk column on ti has now been broken. This situation is formally
unacceptable, since columns obviously need to be continuous if they are to
Metrical grid
have any real substance. We formulate this common-sense requirement in
columns must be
continuous, that
the universal CONTINUOUS COLUMN CONSTRAINT of (38):
is, they cannot
have gaps (38) Continuous Column Constraint:
Metrical grid columns must be continuous, without skipping lines
If you re-examine the antique chair grid in (30) above you will notice that
the metrical configuration of this construction after retraction does comply
with the constraint in (38), in contrast to the antique dealer grid in (37). Con-
sequently, antique chair will indeed undergo stress movement, but antique
dealer will not.
Stress clash also persists in constructions like sports contest, cash register or
house hunting:
(39) : Stress line 2

* ss Stress line 1
= = 2 Baseline
sports contest
Stress movement could be carried out in contest without contravening the

Continuous Column Constraint of (38) above:
(40) = ; Stress line 2

% —>* Stress line 1
be ae Baseline
sports contest
The output of this operation is, however, illegitimate in English: SPORTS

conTEST would imply that sports, not something else, are the contestants!
In German, by contrast, an analogous outcome is obligatory:
(At) meas b. a - Stress line 2

- * * - =5 ~*» Stress line 1
* * + + + + * * Baseline
Marschall Feld Marschall — Feld Marschall

‘marshal’ ‘field marshal’
The reason for the difference between the two languages lies in their respect-
ive formulations of the Rhythm Rule. In English, the Rhythm Rule only imple-
In English the ments leftward movement, as we made explicit in (21) above. In German,
Rhythm Rule only however, no such limitation exists, as we illustrate further in (42) with word-
implements left-
internal contrasts: ’
ward movement
(42) a. SICHTbar ‘visible’ b. UNsichtBAR ‘invisible’

ANziehen ‘to put on’ den ROCK anZIEhen ‘to put the
skirt on’
The fact that the English Rhythm Rule only sanctions retraction explains
the asymmetry of our familiar noun—-verb pairs with regard to stress merger
— the shift to word-initial position caused by clash in verbs is not matched
by a mirror-image shift in nouns:
(43) TAX INcrease
The clash in (43) could only be resolved by moving stress to the right: TAX
*inCREASe. However, the English Rhythm Rule does not provide for this
result, which we have just seen would be fine in German.
So far in the chapter we have been examining the mechanics of stress move-
ment. In the next section we offer a motivation for the phenomenon.
EJ] Rhythm
The label “Rhythm Rule” we have given to the rule responsible for stress
movement in (21) appropriately suggests a connection with rhythm.
Music, of course, has rhythm, and so usually does poetry. Consider the
following children’s poem by Spike Milligan, where we have marked the
strong beats with capitals:
(44) MAry PUGH was NEARly TWO

when SHE went OUT of DOORS.
She WENT out STANding UP, she DID,
and CAME back ON all FOURS.
A reading of this poem reveals an alternation of stressed and stressless syl-

lables, precisely the favourite rhythm of English. The favourite
rhythm of Eng-
lish involves
an alternation
between stressed
and stressless
syllables
The strong liking English has for binary rhythm has interesting (and
perhaps unexpected) consequences for more mundane activities than the
writing or reciting of poems. For instance, English speakers usually count
objects emphasizing the odd numbers and de-emphasizing the even num-
bers, as follows:
(45) ONE two THREE four FIVE six SEVEN eight NINE ten
You may think that this is the only possible way of counting (or at least the
only natural way), but a comparison with French reveals that this is not the
case:
(46) UN DEUX TROIS QUATRE CINQ SIX SEPT HUIT NEUF DIX
You will notice that French speakers emphasize all the numbers. The dif-
ference between the two patterns is nicely captured by our metrical grid
formalism:
(47) awk * * % *
+ * + + * + + + * +
ONE two THREE four FIVE six SEVEN eight NINE ten
b * *+ + + * + + + + +
UN DEUX TROIS QUATRE CINQ SIX SEPT HUIT NEUF DIXx
You can see that each numeral carries a stress of its own in both languages.
Moreover, a higher metrical layer is erected in English to provide the first
word of each pair with greater emphasis. This additional metrical layer is
missing from French altogether.
The respective behaviour is deeply engrained in the speakers of the two

languages. Given a string of nonsense syllables Ja, English speakers will also
assign alternating stresses, whereas French speakers will pronounce it flat:
(48) a. English: LA la LA la LA laLA la

b. French: LA LALA LA LALA
This difference can again be accounted for on the assumption that English
speakers build an additional level of metrical structure.
Segmental Evidence for Stress: Vowel Reduction
So far, we have been trying to gain familiarity with the reality of stress by
accessing what are sometimes subtle intuitions about the different degrees
of prominence exhibited by different syllables in words, or by different words
in word constructions. In this section, we complement this evidence with
some more easily accessible segmental data.
(49). a. Pat b. Patricia

Sam Samantha
prop propeller
prep preparatory
The forms in column a. are monosyllabic truncations of the words in col-

umn b., and therefore we would expect their segmental makeup also to be
a section of the fuller string — their spellings certainly are. However, the pro-
nunciation of the vowel is clearly different in the two contexts.
In particular, the identity of the vowel in the monosyllables in (49a) is unpre-

dictable ([z], [pv], [e], . . . ), but its correspondent in the polysyllables in (49b)
is invariably the reduced vowel schwa ([a]):
(50) a. Plelt b. Plaltricia

S[ze]m S[a]lmantha
prip]p prie]peller
prlelp prla]paratory
Crucially for our purposes here, the two sets of forms also differ in their
stress patterns. In particular, the lone syllable of the forms in (49a) is
stressed. This is to be expected, since under normal conditions fully mean-
ingful words include a stressed syllable, in English and in most other lan-
guages. This particular syllable is, however, stressless in the polysyllabic forms
in (49b), which are stressed elsewhere (PaTRIcia, etc.).
The reduction of stressless vowels to schwa (or to [1] or [vu] in specific
contexts) is one of the most characteristic traits of English. The reduction of
stressless vowels is
one of the most
characteristic traits
of English
In chapter 7 we saw that the articulation of schwa is less precise than the
articulation of other vowels. Schwa is also intrinsically weak, at least in Eng-
lish, in as much as it is less salient than other vowels and can even delete
under the appropriate circumstances: plaJ]tato, for instance, can become
p'tato, and so on. The use of schwa in stressless positions is therefore not
surprising, given the intrinsic association of stress with prominence, and
conversely.
Stop Allophony
The position of stress can also influence the pronunciation of consonants.

For instance, the weakening of /t/ (to a glottal stop in some British dialects The position
of stress can
or to a flap in American English and a number of other accents world-wide)
influence the
is directly contingent on stress. We saw in chapter 3 that the articulation of pronunciation
the flap involves the active articulator lightly hitting the passive articulator of consonants
as a consequence of a rapid ballistic movement. We also know from chap-
ter 1 that the glottal stop is a stop made with the vocal folds.
Word-initial /t/s obviously never weaken:
(51) ten
taramasalata
tabasco
etc.
Word-internal /t/s do weaken, but not indiscriminately, as illustrated in (52):
(52) a. cutting b. mastery c. attain

waiting after retort
The forms in a. can exhibit glottalling (cu[?]ing) or flapping (culrling), but

not so those in b. or c. Comparison of (52a) with (51) suggests that inter-
vocalic position is a precondition for /t/ weakening. This conclusion is
confirmed by the absence of weakening in the forms in (52b), where the /t/
is preceded bya fricative. The forms in (52c), however, do not undergo
weakening either, even though their /t/ is intervocalic. The difference
between (52a) and (52c) becomes clear when stress is brought into the pic-
ture: a further condition on weakening is that the vowel following the /t/
must be stressless. The full formalization of English /t/ weakening is thus
as in (53), where we have ignored some additional complexities to keep
matters simple:
(53) English /t/ weakening:

t > c OR 2 /[-consonantal] [—consonantal]
Condition: the second [-consonantal] does not support a grid column
The central role of stress in the weakening process is expressed by the con-
dition on rule (53). As a consequence, the presence of a flap ora glottal stop
provides indirect evidence about the distribution of stress, and a fortiori about
its existence.
y and ho\ ex |
els were all stres
We find more evidence for stress in the aspiration of voiceless stops. In

chapter 10 we made brief reference to the fact that English voiceless stops
are kept apart phonetically from their voiced counterparts by the puff of
air, or “aspiration”, that follows the release of the closure. As it happens,
the degree of aspiration of voiceless stops is not uniform across contexts:
(54) a. pend b. append c. happened d. spend

tale entail retail stale
It should not be difficult to verify that aspiration is weaker in (54c), and alto-
gether missing in (54d), after /s/. Saying that aspiration only occurs word-
initially will obviously not do, because (54b) has as much aspiration as (54a).
The relevant criterion is again stress: there is strong aspiration when the vowel
after the stop is stressed (and the preceding segment is not /s/). Once more,
therefore, segmental allophony provides a strong clue about the position of
stress, and consequently about its existence.
The overall message of this chapter has been that there is a hierarchy of
prominence among syllables in words, and among words in compounds and
phrases. This prominence hierarchy is grounded in rhythm, and we can appro-
priately represent it in a metrical grid. We have seen that the drive to pre-
serve rhythm can induce stress movement. We have accounted for the
differences in the location of stress between nouns and verbs, or between
compounds and phrases, by appeal to differences in grammatical category.
In the next chapter we shall investigate the actual mechanics of stress assign-
ment in simple words, first in English, and then in other languages.
showed that within words and in

ent than its neighbours. This
and a word(through its promin
metrical.grid
rmal way. An asterisk is placed
nuclei te being recorded
which bans gaps from the column of a'

which allows only leftward move
clashing asterisks can also mo’
vowels to sch Ww does not precedea stressed

vowel, and the ution oF rati Ops.
Key Questions
1 What is a metrical grid? What is its 8 Why does the Continuous Column
function? Constraint prevent some cases of
How is the stress pattern instru- retraction even though an apparent
mental in differentiating between clash occurs? State the terms of the
nouns and verbs, and between com- Continuous Column Constraint.
pounds and phrases? 9 What is the directional setting for
Is there any connection between the Rhythm Rule for English? How
word stress and meaning? does the Rhythm Rule operate in
Does the stress pattern of double German?
proper names comply with that of 10 What is the basic rhythm of Eng-
compounds or with that of phrases? lish? How does it differ from that of
What is the cause of stress French?
retraction? 11 What segmental evidence can be
What is a stress clash? adduced for stress in English?
7 Show how the Rhythm Rule works.
Further ‘Pravert Pee
Catalan
The seven vowels in a. occur in stressed syllables in Central Catalan, but in

unstressed position only the three in b. are found:
ae 4 u Dever u
e oO 3
€ 3
a
With this information in mind, work out where the stress falls in the fol-
lowing words (orthographic accents have been suppressed):
ull [uA] ‘eye’ ulleres [uAeras] ‘eye glasses’

camio [kamjo] ‘truck’ camionet [kamjunet] ‘truck’ (dim.)
cosa [koza] ‘thing’ coseta [kuzeta] ‘thing’ (dim.)
roda [roda] ‘wheel’ algu [alyu] ‘someone’
menja [menzea] ‘he eats’ menjar [monza] ‘to eat’
libre [AiBra] ‘book’ llibreta [AiBreta] ‘notebook’
eldia [aldia] ‘the day’ engany __ [angan] ‘deception’
pasta [paste] ‘dough’ pasteta [pasteta] ‘dough’ (dim.)
pruna [prune] ‘plum’ pruneta [pruneta] ‘plum’ (dim.)
On the basis of these data, can you suggest which reduced form corresponds
to each of the full vowels?
English Rhythm
Compare the words in column a. and the phrases in column b. below. Draw
the relevant grids for each of the words and phrases showing where retrac-
tion has occurred and why:
a. Piccadilly b. Piccadilly Circus

piccalilli piccalilli chutney
Mississippi Mississippi Delta
Hallowe’en Hallowe’en party
Mediterranean Mediterranean Sea
Poetic Rhythm
Consider the following snatches of English poetry. Work out their pat-
terns and say how they relate to the rhythms of speech we have been
demonstrating.
a. The curfew tolls the knell of parting day,

The lowing herd wind slowly o’er the lea,
The ploughman homeward plods his weary way,
And leaves the world to darkness and to me. (Thomas Gray)
b. Old Meg she was a Gipsy,

And liv’d upon the Moors:
Her bed it was the brown heath turf,
And her house was out of doors. (John Keats)
Cc. Eye of newt, and toe of frog,

Wool of bat, and tongue of dog,
Adder’s fork, and blind-worm’s sting,

Lizard’s leg, and howlet’s wing. (William Shakespeare)
d. Tomorrow, and tomorrow, and tomorrow,

Creeps in this petty pace from day to day,
To the last syllable of recorded time;
And all our yesterdays have lighted fools
The way to dusty death. (William Shakespeare)
e. ‘The time has come,’ the Walrus said,

‘To talk of many things:
Of shoes — and ships — and sealing wax —
Of cabbages — and kings —
And why the sea is boiling hot —
And whether pigs have wings.’ (Lewis Carroll)
CHAPTER T\
METRICAL PRINCIPLES AND PARAMETERS
m The general principles of stres

= How one single paramet
compounds.
Disregarding peripheral (
The role of the “metri
How foot structure re
In the previous chapter we surveyeda variety of stress patterns in English,

with the explicit aim of gaining familiarity with the phenomenon. We paid
special attention to the procedure by which these patterns are modified to
avoid clash. In the present chapter we turn to the procedures by which the
original stress patterns are arrived at. In the first part of the chapter we exam-
ine the stress patterns of English phrases and compounds, and then extend
the inquiry to simple words. In the second part of the chapter we show how
the varied stress patterns of the languages of the world are accounted for
simply by introducing slight variations in the procedure we originally set
up for English.
English Phrasal and Compound Stress

In chapter 11 we saw that in English phrasal collocations main stress falls
on the right, whereas in (binary) compounds the greatest prominence tends
to be on the left (remember that personal names pattern like phrases with
regard to stress).
322 Metrical Principles and Parameters
ist a few tokens in

1pounds, perso
Stress on the edge of some domain is a very frequent occurrence across lan-
Stress on the edge guages. This situation is straightforwardly formalized in (1):
of some domain is
a very frequent (1) END STRESS:
occurrence across Project the right-most/left-most asterisk
languages
It should be quite obvious by now that the expression “project the right-
The expression most/left-most asterisk” simply means that the last/first asterisk in a met-
“project the right-
rical line is copied onto the line immediately above, created for the purpose
most/left-most
asterisk" simply
if necessary.
means that the Consider, for instance, the sentence time flies from the previous chapter.
last/first asterisk in To keep the presentation simple, let us assume that there is only one base-
a metrical line is line asterisk for each component word — this simplification is of course made
copied onto the more plausible because the words involved are both monosyllabic. Phrasal
line immediately
stress can now be assigned to the collocation simply by applying End Stress
above
(1) in its “right-most” setting:
(2) : Stress line 1

+ + + +
Baseline
time flies -— time flies
End Stress [Right]
You can see that the last baseline asterisk, directly above flies, has projected
onto the newly created line 1. The resulting grid appropriately expresses the
fact that flies has greater prominence than time.
Were we to apply the same procedure to the compound time-flies, we
would of course obtain an identical output. This result would be incorrect,
since we know that in this and many other compounds the first word is more
prominent:
(8) Stress line 1

- ui Baseline
time-flies
=xplain what
Metrical Principles and Parameters 323
We obviously need to introduce some change in the procedure to bring about

this result.
Extrametricality
One possible way to get the correct stress in compounds would simply be
to switch the setting of End Stress in (1), that is, to select “left” for com-
pounds, rather than “right”:
(4) * Stress line 1

. : ; f Baseline
time-flies -— time-flies
End stress [left]
In (4), the left-most asterisk in the baseline (directly above time) is projected
onto line 1, which the projection procedure in fact creates.
This procedure can, however, be objected to on the grounds that it
requires multiple settings for the same parameter in a single language.
language b niet onable? -
An alternative procedure, consistent with the “right-most” setting of End

Stress for compounds too, involves concealing the right-most asterisk in the
input of (4). You will probably find this strategy rather baffling at this point,
but if you bear with us you will see that it receives considerable backing.
In real life, we could simply cover up the asterisk we want to render invis-
ible with a finger, thus ensuring that we ignore it in the ensuing computations.
Cover up the offending asterisk in the input . 2 and (4) above

r yourself what the result igs .
In the more stylized medium of scientific writing we need some symbol to

achieve the same end, and a pair of angled brackets (“< >”) is commonly The device known
as EXTRAMETRI-
used for the purpose. The device is given the name EXTRAMETRICALITY,
CALITY involves
because it excludes the element it acts upon from the computations in the excluding the
metrical grid. We provide a general statement of extrametricality in (5): element it acts
upon from the
(5) Extrametricality: computations in
Make the right-most/left-most asterisk extrametrical the metrical grid
(notationally:* — <*>)
Of course, extrametricality needs to be assigned in a specific context in each

particular case, since we clearly do not want the right-most/left-most aster-
isk of all lines in all grids to be extrametrical.
In the case we are currently discussing, extrametricality is only relevant to

compound structures, since we know that phrases and sentences do not require
it (indeed, do not allow it), at least in English.
The restriction that only elements that are peripheral in the appropriate
Only elements domain are allowed to be extrametrical has solid empirical grounding:
that are peripheral
extrametrical behaviour has been found to be associated with peripheral
in the appropriate
domain are elements in language after language. The restriction is also common sense,
allowed to be because, were unbridled extrametricality to be allowed, we could end up
extrametrical with any number of extrametrical elements anywhere in the string, in a quite
ad hoc manner.
The restriction in question is made explicit in (6), under the label PERI-
PHERALITY CONDITION:
(6) Peripherality Condition:

Only peripheral elements can be extrametrical
Let us now observe the effect of (right-most) extrametricality on the stress

of English compounds:
Cy = = <*> Baseline
time-flies — time-flies
Extrametricality
Extrametricality makes the asterisk on flies invisible, and consequently the

output of (7) now includes only one baseline asterisk. The action of End
Stress [Right] (1) on this structure places the main stress of the collocation
in the correct position:
(8) > Stress line 1

NIMs mrt 2) Baseline
time-flies — time-flies
End Stress [Right]
lyhow we.
The procedure we are adopting obviously yields the right results. Import-
antly, it also allows us to retain the setting “right” for End Stress through-
out the language in domains above the simple word. In fact, we will see
below that this setting also encompasses the domain of the simple word.
This is clearly an advantageous outcome, both formally and from the point
of view of learnability, as we will explain in section 9 below.
The Elsewhere Condition

The last baseline asterisk in compounds is a potential target for both End
Stress [Right] and Extrametricality [Right], but we just want Extrametrical-
ity to apply. Therefore, the interaction between the two procedures needs
to be regulated.
Suppose we were to allow End Stress [Right] to apply before Extra-
metricality:
(9) z Stress line 1

- = i . * <*> Baseline
time-flies — time-flies — time-flies
End Stress [Right] |Extrametricality
© you see anythin

YM
The result is very dubious from a formal point of view. First, can a domin-
ated asterisk be the target of extrametricality? Second, doesn’t the resulting
grid violate the Continuous Column Constraint in (38) of chapter 11? Third,
how is the grid in (9) to be interpreted anyway? In particular, which of the
two words is predicted to bear greater prominence, and why? All these diffi-
culties are avoided if Extrametricality applies before End Stress, precisely
as it does in (7) and (8) above. We could, of course, simply stipulate this
EXTRINSIC ordering. However, EXTRINSIC ORDERING of this kind is widely dis-
ORDERING is
favoured, and only accepted as a last resort when there is no viable alter-
widely disfav-
oured, and only
native (we deal with ordering in detail in chapter 18). We will now see that
accepted as alast in the case we are discussing a viable alternative does indeed exist.
resort The answer is contained in the formal relationship between End Stress
[Right] in (1) and Extrametricality in (5) in English compounds. We spell
out the two procedures in (10), to facilitate comparison:
(10) a. End Stress [Right]:

Project
the right-most asterisk
b. Compound Extrametricality [Right]:
Make extrametrical
the right-most asterisk
in compounds
Two rules can only conflict when they share the input, as (10a) and (10b)
do: the right-most asterisk. The two aspects relevant to the interaction
between these or any competing rules are:
1 the effect of the rule — here: “project” for end stress in (10a), and “make
extrametrical” for the Compound Rule in (10b);
2 the environment or context in which the rule applies — here: no context
for End Stress in (10a), and “in compounds” for the Compound Rule
in (10b).
Now, the effects of the two rules in (10) are obviously incompatible, since
they involve, respectively, the projection of the right-most baseline asterisk
and the assignment of extrametricality to this very same asterisk.
As for their environments, the environment of Compound Extrametricality

in (10b) is more restricted than the environment of End Stress in (10a): Com-
pound Extrametricality only applies in compounds, whereas End Stress has
no built-in limitations.
Over two thousand years ago, the Indian linguist Panini noticed a very
special relation between two rules when they produce incompatible results When two rules
produce incom-
and the environment of one rule is included in the environment of the other
patible results and
rule: in such cases, the more specific rule (with the more qualifications, the environment
and thus with the more detailed environment) applies first. In its modern of one rule is
incarnation, Panini’s principle is commonly referred to as the ELSEWHERE included in the
CONDITION, because the more general rule (with the fewer qualifications, environment of
and thus with the less detailed environment) applies in all the contexts where the other rule, the
more specific rule
the more restricted rule cannot, that is, “elsewhere”.
applies first
The Elsewhere Condition is active throughout phonology, and indeed other

areas of language. We formulate it explicitly in (11):
(11) Elsewhere Condition:

Given two rules such that:
(i) their inputs are identical
(ii) their outputs are incompatible
(iii) the environment of one rule is the same as the environment of
the other rule plus something extra
then, the rule with the richer environment is ordered first, and, if it
does apply, the other rule is skipped.
You can now see that our rule of Compound Extrametricality in (10b) will
Compound
apply before End Stress in (10a) simply as a consequence of the Elsewhere
Extrametricality
Condition — there is no need to stipulate this order. applies before End
Stress as a con-
sequence of the
ELSEWHERE
CONDITION
£9 Stress Assignment in Words
The stress pattern of French words is remarkably simple and provides a use-
ful entry into our discussion of word stress. The procedure responsible for
328 Maetrical Principles and Parameters
the stress patterns of English compounds that we discussed in section 2 also

applies to French individual words, with minor adaptations. We illustrate
the stress pattern of French words in (12). You should bear in mind that French
orthographic accents bear no relation to stress, and that the final e in words
like extreme (bracketed in (12) for ease of identification) usually has no phon-
etic correlate, although exceptionally it can be sounded as a “schwa” (in effect,
the round front vowel [ce] in French):
You can see that stress invariably falls on the right-most syllable, unless this
syllable contains a “schwa”, which is ignored for stress purposes. French
word stress can therefore be formalized as in (13):
(13) French word stress:

a. extrametricality: “schwa” on the right edge (if there is one)
b. asterisk placement: End Stress [Right]
All French words obey these simple principles, and therefore no more
needs saying about the matter in the present context.
English word stress is considerably more complex. You can geta taste of
this complexity from the small sample in (14), which replicates the French
one in (12) above:
(14) exTREme alloCAtion EXtra CANdidate oPInion
Can yousee a pattern to English stres:
In extreme stress falls on the final syllable, or, equivalently, on the second
syllable; in allocation it falls on the third syllable, which is also penultimate;
in extra and candidate it falls on the initial syllable, but if we count from the
end, stress will be penultimate in extra and antepenultimate in candidate; finally,
in opinion stress is located on the second or the antepenultimate syllable -
you can take your pick!
Faced with this situation, it is small wonder that until fairly recently each
English word was thought to have its own idiosyncratic stress. For instance, Until fairly
Daniel Jones, the one-time influential English phonetician to whom we recently, each
English word was
referred in chapter 5 in connection with the cardinal vowels, wrote that “gen-
thought to have
erally speaking there are no rules determining which syllable or syllables its own idiosyn-
of polysyllabic English words bear the main stress” (An Outline of English cratic stress
Phonetics, 1967: 248). In the context of this belief, and of the type of data
that supported it (cf. (14) above), the position of Chomsky and Halle that
“both the placement of main stress and the stress contours within the word
and the phrase are largely predictable from the syntactic and the non-
prosodic phonological structure of an utterance” (Ihe Sound Pattern of
English, 1968: 59-60) was obviously daring and provocative.
Indeed, Chomsky and Halle initiated the trend for in-depth research into
the regular stress patterns of languages. Such research has gone a long way Chomsky and
Halle initiated the
since then, and in what follows, as in the preceding chapter, we will base
trend for in-depth
our exposition on the metrical theory that developed subsequently. We are research into the
seeing that in metrical theory stress is conceived of as a network of promin- regular stress pat-
ence relations, formally represented as a metrical grid. terns of languages
Crucially, metrical relations are assumed not to be present in lexical repres-

entation, but to be filled in by the action of the metrical procedures.
All this will of course become progressively clearer as we go along.
Basic Stress Pattern of English Nouns
We shall start our empirical investigation of English word stress with the
set of nouns in (15). From now on, we shall mark stress by means of an acute
accent on the vowel that carries it:
(15) cinema platypus ténement impetus

algebra animal vénison filament
élephant chdcolate accolade customer
All the words in (15) have three syllables, the first of which is stressed, as
signalled by the mark on the corresponding vowel. This mark is, of course,
not included in ordinary English spelling, but is provided in standard dic-
tionaries, usually in the form of a preceding apostrophe, as part of the
phonetic information of each lexical item. Must we conclude from (15) that
English nouns are simply assigned stress word-initially? The data in (14) above
already gave a hint that this cannot be the case. Consider further the forms
in (16):
(16) asparagus
aluminium (British)
aluminum (American)
hypochondriac
metamorphosis
hippopdétamus
pantéchnicon
parallélogram
gloxinia
You can see that stress is not word-initial in these (longer) words.
Comparison of the sets in (15) and (16) reveals that in both cases stress
English noun is in fact antepenultimate: it is located on the third syllable from the right
stress is ante- edge of the word. Can we achieve this result from the metrical procedures
penultimate, that
we have .available at present? The answer is a clear “no”. In particular,
is, it is located on
the third syllable
the only metrical devices currently at our disposal are Extrametricality,
from the right which in English effectively nullifies the right-most grid element, and End
edge of the word Stress, which, also in English, enhances the right-most element in the met-
EE rical grid.
Application of Extrametricality and End Stress yields the wrong result for
the data we are considering. Let us first try the familiar “right-most” set-
ting for both rules:
(17) oa Stress line 1

* + + + *+ + + + Peale * e+ + + <*> Baseline
hippopotamus — hippopotamus —> hippopotamus

Extrametricality [R] End Stress [R]
The output, *hippopotdmus, does not match the correct pattern hippopétamus.
Let us experiment with a setting “left-most” also for both rules:
(18) ss Stress line 1

+ SE a ee cae + eek nos * Baseline
hippopotamus — hippopotamus — hippopotamus

Extrametricality [L] End Stress [L]
Again, the result is incorrect (*hippépotamus). Assigning opposite settings

to end stress and Extrametricality clearly does not help (*hippopotamus, *hip-
popotamus), and therefore we need to introduce some significant modifica-
tion in the procedure.
[4 The Metrical Foot
In the previous chapter we referred to the fact that English favours stress
in alternate syllables, as revealed in the typical pronunciation of a sequence
of the nonsense syllable Ja by an English speaker:
(19) lalalalalala...
Notice, however, that there is another way of achieving alternating rhythm

in the same sequence:
(20) lalalalalala...
The alternation of stressed and stressless syllables (equivalently, of S[trong]

and Wl[eak] metrical elements) is preserved in (20), but the pattern is
reversed: S—W in (19) and W-S in (20). As we will see, languages with altern-
ating stress employ one or other of these two opposite metrical patterns.
We shall now introduce some formalism and terminology. Let us call a
We call a pairing pairing of S-W or W-S syllables a FOOT. The word “foot” is also used in
of S-W or W-S constructs, albeit related, must be kept apart
versification, but the two
syllables a FOOT
firmly: we certainly do not wish to imply that ordinary language is verse
(unfortunately, there is no handy spelling contrast here as there was with
“thyme” vs. “rime”).
The first property of metrical feet, binarity, is implicit in the configurations

S-W, W-S, each made up of precisely two syllables:
(21) (SW) or (WS)
We are keeping to common practice and enclosing metrical feet in paren-

theses. The strong element of a foot is the HEAD of the foot: feet, like syl-
lables and other constituents, have a core, or “head”. Accordingly, we shall
refer to the two feet in (21) as LEFT-HEADED and RIGHT-HEADED,
Feet can be LEFT- respectively. Alternative terms, taken from classical metrics, are “trochee”
HEADED or
and “iamb”, respectively, but we shall avoid using these for the time being
RIGHT-HEADED
in order to maximize the transparency of the terminology. Our metrical
machinery has, therefore, now been augmented with the two constructs in
(22):
(22) left-headed foot =(S W)

right-headed foot = (W S)
We next need to give a grid interpretation to the abbreviations S and W,

since we know that the metrical grid is the formal device for representing
metrical structures:
(23) var * Stress line 1

left-headed foot =(**) Baseline
b. * — Stress line 1
right-headed foot = (* *) Baseline
You can see that the extension of the foot is indicated by means of ordinary
parentheses in the baseline, and the location of the foot head by means of The extension of
an asterisk in line 1. a foot is indicated
by means of ordin-
ary parentheses in
the grid’s baseline,
and the location
of the foot head
by means of an
asterisk in line 1
In the next section we will make use of binary feet to account for the basic
stress pattern of English words.
Main Word Stress in English
The procedure in (24) assigns the correct main stress to the English nouns
listed in (15) and (16) above. We will be refining this procedure as we go
along, and therefore we number each successive version to allow you to keep
track; the word ALGORITHM is in common use to designate a self-contained
procedure:
(24) English noun stress algorithm (no. 1):

1. Make the last element extrametrical
2. Build a left-headed foot at the right edge
In (25) we illustrate this procedure with the noun asparagus:

+ + + + * + tats * Ee 2) <*> Baseline
asparagus — asparagus — aspara gus

(24.1) (24.2)
In the first step in (25), the final element of the baseline becomes extramet-
rical (cf. the angled brackets), paralleling hippopotamus in (17) above. In the
second step, a left-headed foot is constructed at the right edge, with the extra-
metrical element naturally outside the computation altogether. As a result
of these two steps, main stress is correctly assigned to the syllable pa. A sim-
ilar outcome obtains for the remainder of the nouns we are investigating.
The algorithm in (24) does not include all the steps you already know are
necessary to assign primary stress to English nouns. We list the complete
procedure in (26), again with the form asparagus as an illustration:
(26) English noun stress algorithm (no. 2):

1. Input a lexical form, with no metrical structure:
asparagus
2. Construct the grid baseline by projecting syllable heads:
See aeDaSelne
asparagus
3. Make the last element in the baseline extrametrical:
<7 cS) Baseline
asparagus
4. Build a left-headed foot at the right edge:

* Stress line I
* 1(* 7+) <*5 = Baseline
aspara gus
The algorithm in (26) works well for the bulk of both nouns (médicine,
magnanimity, etc.) and suffixed adjectives (medicin-al, magndnim-ous, etc.). It
does not, however, yield the correct stress pattern for verbs and unsuffixed
adjectives. We present a sample of these in (27):
(27) implicit insipid imagine _—_impéril

devélop endéavour admonish endémic
astonish deliver detérmine pellucid
Let us try out the procedure in (26) on implicit:
(28) é Stress line 1

ferred ® At ( *)<*> Baseline
implicit — implicit -—> impli cit
Extrametricality Footing
The output is incorrect: *#mplicit. This problematic result should not be sur-
prising, given that in the previous chapter we learnt that there is a stress
contrast between nouns and verbs (and, as we are now seeing, also between
suffixed and unsuffixed adjectives).
Fortunately, the correct result is not hard to get - quite simply, we sup-
press the extrametricality clause in verbs and unsuffixed adjectives: We suppress the
extrametricality
5 d clause in English
(29) his ppb ave ti Stress line 1 verbs and un-
(* *) Baseline suffixed adjectives
implicit — implicit
Footing
In (29) you can also see a left-headed foot at the right edge. In contrast to
what happens with nouns, however, the last syllable is not excluded from
the computation. The difference between verb and noun stress in English,
therefore, is the product of the restriction of extrametricality to nouns.
EE] Multiple Stress
Our present procedure yields one (left-headed) foot at the right edge of the
word. Now, in words like aluminium or metamorphosis, the vowels in the first
syllables are unreduced: [ze] and [e], respectively. We know from the pre-
vious chapter that stressless short vowels in English undergo reduction. The
fact that /e/ and /e/ surface unreduced in aluminium and metamorphosis
suggests, therefore, that these vowels are stressed. Yet, the most prominent
syllables in these two words are mi and mor, respectively.
The situation is further illustrated in (30), where we have highlighted un-
reduced vowels by underlining. Some of the words are admittedly, but irrel-
evantly, rather uncommon.
(30) hamamelidanthemum
mesembryanthemum
pelargonium
sanatorium
pimpinellifolia
serendipity
hippopotamus
The algorithm in (26) assigns stress to the antepenultimate vowel, and

therefore the full quality of this vowel is expected. What is still unexplained
is the presence of other unreduced vowels in each of the words.
You may have noticed that the unreduced vowels are located at equal inter-
vals to the left of the vowel stressed by the algorithm. This pattern provides
a clue as to what’s going on. You will recall that the basic rhythm of English
The basic rhythm involves an alternation of strong and weak elements throughout the word,
of English involves
not just at the right edge. If this is so, foot construction must sweep across
an alternation of
strong and weak the whole word, contrary to what we have been doing so far:
elements through-
out the word: Gil) 3 Stress line 1
* * +e * <*> *+ * * + © *"Na*> Baseline
English foot
construction hamamelidanthemum — hamamelidanthemum —->
thus takes place Footing
iteratively
a a * saaiels Stress line 1
+ + @ ) (je 2) <*> € ) @ )) @ A) <*> Baseline
— hamameli danthe mum — hama meli danthe mum

Footing Footing
You can see that in (31) foot construction takes place iteratively from right
to left, where ITERATIVE means ‘repeated’ or ‘recurring’.
As a result of the right-to-left iteration of foot construction, the syllables

ha, me and dan of hamamelidanthemum end up supporting foot heads, and,
therefore, their vowels do not reduce. These additional peaks of prominence
are referred to as SECONDARY STRESSES. Additional peaks
The existence of secondary stresses leads us to modify clause 4 of the stress of prominence are
referred to as
procedure in (26) above, along the lines of (32):
SECONDARY
STRESSES
(32) 4’. Build left-headed feet iteratively from right to left
Note that the direction of foot construction needs to be stipulated language

by language. In particular, while we are constructing English feet from right The direction of
foot construction
to left, in other languages (some of them to be considered below) feet are needs to be stipu-
constructed from left to right. In English, left-to-right iterative foot construction lated language by
would yield a wrong contour in words with an even number of syllables: language
*rhinocéros.
At this point we have three (left-headed) feet on hamamelidanthemum, with

their heads on ha, me and dan (cf. (31) above). We know intuitively that the
stress on dan is stronger, but in (31) there is no structural difference between
this syllable and the two other foot heads, ha and me. This must mean that
the grid in (31) is incomplete, since grids need to provide a complete rep-
resentation of metrical structure. To make up for this shortfall, we submit
the structure in (31) to End Stress [Right]:
(33) + Stress line 2

- _ <s nae Stress line 1
(x =) EG ) (* <3) <*> & =) & 2) C =) <*> Baseline
hama meli danthe mum —> hama meli danthe mum

End Stress [Right]
We shall have more to say about End Stress further on, but for now we can
be satisfied with the result we have obtained: a grid for hamamelidanthemum
that signals greatest prominence on dan, and subsidiary prominence on ha
and me.
ther words
in
EK} Stress Typology: Metrical Parameters
We have now accounted for the difference between the antepenultimate

stress of English nouns (and suffixed adjectives) and the penultimate stress
of English verbs (and unsuffixed adjectives) through the simple strategy of
including an extrametricality clause in the metrical algorithm of nouns, but
not in the metrical algorithm of verbs. This reveals that significant differ-
Significant differences in the output can be due to a small difference in the grammar, open-
ences in the out-
ing the way for a parametric account of the metrical structure of languages
put can be due to
a small difference
in general. We spell out the details of such an account in this and the fol-
in the grammar lowing section.
In (34) we list the metrical possibilities we have available at present:
(34) Foot head location: Left/right

Construction direction: Right-to-left/left-to-right
Extrametricality: Yes(right/left) /no
The emboldenings in (35) specify the English settings for these parameters:
(35) Foot head location: Left/right

Construction direction: Right-to-left/left-to-right
Extrametricality: Yes(right/left)/no (nouns)
Yes(right/left)/no (verbs)
The settings required for English nouns yield the regular stress pattern of
all types of words in (literary) Macedonian, and the settings for English verbs
the regular pattern of all types of words in Polish (Macedonian is spoken
in the Former Yugoslav Republic of Macedonia, and Polish, of course, in
Poland). In (36) we provide a sample of words for each of these two lan-
guages (NB Polish y = [i]):
(36) a. Macedonian b. Polish

(= English nouns) (= English verbs)
‘miller’ ‘marmalade’
vodénigéar Sg. marmolad Gen. pl.
vodeniéari ‘Pi. marmolada Nom. sg.
vodenicérite PI. def. marmoladéwy Adj.
We illustrate the workings of the respective metrical algorithms in (37). Note

that we are assuming that all syllables need to belong to a foot, even if some
feet finish up monosyllabic as a result, as is the case with (vo) and (mar) in
(37) — we return to this matter below:
(37) a. Macedonian Polish
(= English noun parameters) (= English verb parameters)

vodeniéar Input marmoladowy
| Syllable nucleus
t+ SF F projection * Se Oe
ai marmoladowy
* * *<*> Extrametricality
vodeni éar
+ + * + +
et Footing tee Ae 25)

vodeni éar marmola dowy
* *
OC 2) > End Stress CM oe

vodeni Car marmola dowy
You can see that the present procedures assign main stress in the correct
place in both languages.
We can obtain other stress patterns by varying the settings of the para-
meters. Suppose first that we give the head location parameter the opposite We can obtain
setting from English, to produce right-headed feet: different stress
patterns by vary-
ing the settings of
(38) Location of foot head: Left/right the parameters
If we keep the settings of all other parameters as in (35) above, with extra-
metricality as for English nouns, we will once more derive the pattern of
Polish: word-penultimate main stress.
However, if we do not select extrametricality (as we don’t for English

verbs), main stress will be assigned to the final syllable, and secondary stress
to every other syllable to the left of the main stress. This turns out to be the
pattern of Aklan, a language of the Philippines, which we illustrate in (39).
Note that from now on some of the segmental transcriptions, taken directly
340 Maeetrical Principles and Parameters
from the literature, may not conform exactly to standard IPA practice,
although, fortunately, this has no effect on the points we make:
(39) Aklan (right-headed, right-to-left, no extrametricality):

bisa ‘kiss’
bisahi ‘kiss’ (Ref. imp.)
suyugu?un ‘servant’
matinamartn ‘being lazy’
Let us next reset the direction of foot construction to “left-to-right”:
(40) Direction of foot construction: Right-to-left/left-to-right
We shall illustrate the consequences of this resetting with data from two
Australian aboriginal languages, Maranungku, with left-headed feet, and
Yidin’, with right-headed feet (n” represents the palatal nasal [pn]; d’ =
palatalized /d/). Notice that neither language has extrametricality, and that
Yidin’ does not allow monosyllabic feet, perhaps in avoidance of clash — the
fate of monosyllabic feet is therefore best approached as parametric:
(41) Maranungku (left-headed, left-to-right, no extrametricality):

pan ‘friend’
tiralk ‘saliva’
mérepét ‘beard’
yangarmata ‘the Pleiades’
wongowutanawan’ ‘thunderhead’
(42) Yidin* (right-headed, left-to-right, no extrametricality):

galbi ‘catfish’ (Abs.)
gudaga ‘dog’ (Abs.)
wawad’int ‘see’ (Antipass. past)
mad*imdanad’in ‘walk up’ (Trans. antipass. pers.)
Extrametricality on the left edge is rather uncommon. It is, nevertheless,

attested in the native American language Winnebago, which has a basic
stress pattern otherwise analogous to Yidin’ (an alternative analysis of

Winnebago stress without extrametricality is also available in the literature).
We offer examples of Winnebago basic stress in (43) — notice that all vowels
have been projected onto the baseline, and that there are also no clashing
monosyllabic feet:
(43) Winnebago (= Yidin’ with left extrametricality):

waghighi ‘ball’
naana?a ‘your weight’
hochichinik ‘boy’
haakitujik ‘I pull it taut’ (Plain)
hakirujikshana ‘he pulls taut’
haakitujikshana ‘I pull it taut’ (Declined)
The data we have been reviewing back up our claim that small changes
in the setting of the given parameters can have a considerable effect on the
output. To the extent that this formal result matches the stress patterns of
the world’s languages, the parametric model of stress we are proposing
receives empirical confirmation. The model is also superior from the point
of view of learnability. In particular, it provides a very reasonable answer
to the thorny question of how humans (specifically, children) manage to work
out and learn the stress patterns of natural languages (some of them pretty
diabolical on the surface!) simply from exposure to data, and to do so ina
reasonably short space of time.
The approach
Thus, the approach based on Universal Grammar and parameter setting car- based on Uni-
ries the implication that the child learner intuitively knows what to look for: versal Grammar
left vs. right foot-headedness, and so on. In this way, the child can home in and parameter
setting carries the
on the data to achieve these modest goals without expending too much time
implication that
and energy, and the learning task becomes manageable. The Universal the child learner
Grammar cum parameter-setting model is of course also applicable to other intuitively knows
branches of linguistics, syntax in particular, although, naturally, the iden- what to look for
tity of the parameters varies from branch to branch.
We end the section with a tabulation of the various parameter settings we

have discussed and the names of the languages that instantiate each pattern:
(44) Language Foot head Construction Extrametricality

location direction
I r l-to-r/r-to-l yes no
l r
English nouns v J v
Macedonian v J v
English verbs v v J
Polish v v v
= Polish J J J
Aklan J v J
None? v J v
Maranungku J Jf S
Winnebago v v v
Yidin” v v v
Word-Level Stress: Line Conflation
In section 8 we saw that the selection of the right-most foot as the carrier of
The selection of the main word stress in English is achieved through asetting “right” for
the right-most
End Stress: hamamelidanthemum, not hamamelidanthemum. An analogous situ-
foot as the carrier
of the main word
ation obtains in Macedonian, Polish, Aklan and Yidin’, among the languages
stress in English is we have mentioned. We illustrate with Polish in (45):
achieved through
a setting “right” (45) _ Stress line 2
for End Stress iamelapstbes ee ad Stress line 1
CRG Cae) CC A) Baseline
marmola dowy — marmola dowy
End Stress [Right]
In languages like Maranungku and Winnebago, by contrast, the initial foot

is enhanced. We show this in (46) with the Maranungku word wéngowii-
tanawan* ‘thunderhead’ (grave accents indicate secondary stresses, and the
acute accent the primary stress):

3 Se, > Oy Stress line 1
(Sitstie(p2 ois S) (bu): Git) 2 ceBaseline
wongo wuta nawan” — wongo wuta nawan’
End Stress [Left]
Maranungku and many other languages are described in the literature

as exhibiting the predicted pattern of secondary stress in the surface. How-
ever, in languages like Macedonian the secondary stresses are not realized
phonetically. In other languages, like Polish or English, secondary stresses
do appear in the surface, but not necessarily where we might expect from
the iterative application of the primary stress metrical algorithm. In many
cases, therefore, our metrical procedure is generating more structure than
appears necessary.
nerates more struc-
Two strategies are available to resolve this problem. One strategy involves
enriching the model with an ITERATIVENESS parameter to control the
repetition of footing: the setting for this parameter would be positive for
Marunungku, so that feet be constructed over the whole domain, and neg-
ative for Macedonian, so that only one foot be constructed: (ydngar)(mdata)
and vodeni(Ccari)<te>, respectively. The alternative strategy, LINE CON-
FLATION, allows iteration across the board in the first instance, and then
deletes line 1 from the grid to dispose of all the feet but the one bearing the
main stress. We illustrate this procedure in (47) with Macedonian:
(47) i Stress line 2

come Es e Stress line 1
Neae trie a> ep ee (* Oye?Se Baseline
vo deni Gari te — vodeni éari te
Conflation
You can see that conflation does away with the line 1 of the input, and that
line 2 of the input becomes line 1 in the output as a result. Notice also that, Conflation does
away with line 1
as feet are beheaded, the parentheses enclosing the relevant asterisks in the of the input
baseline are deleted: decapitated constituents automatically cease to exist. The
reason for this lies in the indissoluble connection between a foot and its head:
a constituent is nothing but the domain of a head, and therefore it cannot
exist without its head. Conversely, of course, a head defines a constituent.
This mutual implication between heads and constituents is formulated in
the FAITHFULNESS CONDITION on grid structure in (48) — the word “faith-
fulness” naturally refers to the inseparability of the head and its domain:
(48) Faithfulness Condition:

Each grid con- Each grid constituent has a head (plotted in the line immediately above),
stituent must have
and each head has a domain (delimited in the line immediately below)
a head, and each
head must have a
domain When Line Conflation deletes line 1 in (47), the feet on vo and deni are deprived
of their heads, and therefore they automatically cease to exist.
The Line Conflation analysis is reminiscent of the Duke of York’s antics
in the nursery rhyme, marching his men up to the top of the hill only to
march them down again: we first create feet across the whole domain, only
to end up destroying all but one of them. The Iterativeness parameter may
therefore look like an obviously superior alternative: if nothing else, it is sim-
pler, and it will be favoured by Occam’s razor.
However, in the next chapter we will see that matters are a bit more com-
plex than they may have appeared so far, and that Line Conflation is in fact
supported by specific evidence.
At this point the question arises of whether the effect of End Stress is the
simple provision of a line-2 asterisk, as we have been assuming, or whether
it results in the creation of a full line-1 constituent.
ain what the orm differe

uld be.
The construction of a line-1 constituent for the Maranungku word in (46)

above would result in the grid in (49), with the line-1 asterisks enclosed by
a pair of parentheses:

¢ 5 *) Stress line 1
G. <isG@2) Ca baeBaseline
wongo wuta nawan’
The empirical evidence for this higher constituent is not overwhelming.

None the less, its existence does follow from the logic of the Faithfulness
Condition, and therefore we will include it in our representations from now

on. Some related discussion appears in chapter 16.
We have now completed our presentation of the prosodic structure per-
tinent to stress. In (49) above you can see that this structure consists of a
foot layer and a word layer. The foot layer is formalized by means of aster-
isks marking foot heads in line 1, and of pairs of parentheses delimiting the
foot’s extension in the baseline. All the feet considered up to this point
are maximally binary: they encompass a maximum of two syllables. The
line-2 asterisk corresponds to the head of the metrical word, and the pair of
parentheses enclosing the line-1 asterisks formalizes the inclusion of all the
feet in the metrical word domain. The head of the metrical word shows up
as the word’s main stress, and the heads of the remaining feet as secondary The head of the
metrical word
stresses, unless such feet are suppressed by Line Conflation, as we have
shows up as the
explained. word's main
stress, and the
heads of the
remaining feet as
Chapte secondary stresses
In this chapter we have ee
Stress (“right” for English: End

line 1 after extrametricality
a
Key Questions
What is “End Stress"? 6 How can the direction of foot con-

What is “extrametricality”? Why is struction be determined by the sites
the Peripherality Condition import- of the stresses in words?
ant in constraining extrametricality? How can changes to the settings of
State the terms of the Elsewhere the metrical parameters yield a wide
Condition. variation of stress patterns in lan-
How do the standard stress patterns guages? List the parameters discussed.
of nouns differ from those of verbs? What are the settings for English?
How does the stress algorithm deal What is the result of Line Confla-
with the discrepancies? tion? Why is it necessary?
What is the structure of the metrical State the terms of the Faithfulness
foot in English? What evidence do we Condition. How does it relate to
have for it? Line Conflation?
Further Practice
Araucanian
The stress pattern of the Penutian language Araucanian, spoken in Chile and
Argentina, is illustrated below (assume orthographic w and y to represent
non-nucleic [u] and [i], respectively): .
wulé ‘tomorrow’
tipante ‘year’
kimubaluwulay ‘he pretended not to know us’
elumuytt ‘give us’
eluaénew ‘he will give us’
(i) What are the parameter settings required to account for this pattern?
(ii) Show the processes through which stress is assigned, in the form of a
grid.
Warao
Compare the stress pattern of Araucanian with that of Warao, spoken in

Venezuela:
yiwaranae ‘he finished it’

nahorodahakutai ‘the one who ate’
yapurukitanehase ‘verily to climb’

enahoroahakutaéi ‘the one who caused him to eat’
(i) What are the parameter settings for Warao?

(ii) Show the processes in the form of a grid.
Weri
As in the cases of Araucanian and Warao, account for the stress pattern of
the New Guinean language Weri:
nintip ‘bee’ kulipu ‘hair or arm’

uluamit ‘mist’ akunétepdl ‘times’
Hungarian
Now show the settings for Hungarian, in the same way as you have done
for the three languages above:
bdldog ‘happy’
bdldogsa:g ‘happiness’
bdldogtalan ‘unhappy’
bdldogtalansa:g ‘unhappiness’
légeslegmegengesztelhetétlenébbeknék ‘to the very most irreconcilable
ones’
Western Aranda
In the Australian language Western Aranda, primary stress falls on the first
or second syllable in trisyllabic or longer words and on the first syllable
if the word has fewer than three syllables. Examples of the patterns are shown
below:
a. Consonant-initial words of three or more syllables

tukura ‘ulcer’
kutuntla ‘ceremonial assistant’
woratara place name
b. Vowel-initial words of three or more syllables

erguma ‘to seize’
artjanama ‘to run’
utnddawara place name
c. Bisyllabic words
kama ‘to cut’
ilba ‘ear’
wuma ‘to hear’
Allowing for an extrametricality clause in the algorithm, work out the set-
tings for Western Aranda.
SYLLABLE WEIGHT
FURTHER METRICAL MACHINERY
In this chapter you will

= The definition of “heavy
m Word stress falling closer to
syllable.
m Marking heavy syllable
regular footin
The account of English stress in the previous chapter only works for a
subset of nouns/suffixed adjectives and verbs/unsuffixed adjectives. This
shortfall should not come as a surprise, in view of our earlier warning on
the complexity of the English stress system. In this chapter we continue to
investigate stress, adding to the theory as we need to along the way. The
specific focus will be on the effect of syllable structure on stress assignment,
and its implications for syllable theory. In the final sections we extend the
metrical model to cases that fall outside the scope of the machinery as it stands
at present.
An Additional Pattern of Stress in English
Consider the two samples in (1), nouns in a. and verbs and unsuffixed adject-
ives in b.:
350 Syllable Weight
(1) a. agénda b. (to) incréase

amalgam (to) reply
asbéstos (to) recomménd
incisor oblique
meménto (to) despise
debénture inténse
incéntive (to) invént
Octdber (to) enthtise
amanuénsis (to) withdraw
In both sets stress falls one syllable further to the right than is predicted
by the procedures we gave in chapter 12, now summarized in (2) as a
reminder:
(2) English stress algorithm (no. 3):

a. Project syllable heads onto the baseline
b. Make the right-most baseline asterisk extrametrical (in nouns and
suffixed adjectives only)
c. Construct line 1 by building left-headed feet iteratively from right
to left
d. Construct line 2 by applying End Stress [Right] on line 1
As we explained at the time, the steps in (2) follow from the settings of the
relevant parameters, which we repeat in (3). Remember that we are crucially
assuming that lexical forms are bare of metrical structure:
Lexical forms are
bare of metrical (3) English stress parameter settings:

t
Me Extrametricality: Yes [right] | Nouns
No Verbs
Location of foot head: Left
Direction of foot construction: Right-to-left
End stress: Right Onto line 1
Syllable Weight 351
In (4) we illustrate these procedures at work in the forms agenda and

recommend, from (la) and (1b), respectively. The derivations incorporate
line conflation, anticipating the evidence we present in chapter 15 to the
effect that line conflation is also operative in English, contrary to our
practice so far, which we based on the secondary stresses of forms like Line conflation is
hamameélidénthemum: also operative in
English
(4) agenda Input recommend

5 wie a eke Baseline
agenda (2a) recommend
at s lildate Baseline
agenda (2b) recommend
: “i Stress line 1
CG st> (*)( *) Baseline
agenda (2c) re commend
. Stress line 2
‘ ss Stress line 1
Condy Sari (*)(* *) Baseline
agenda (2d) re commend
a i Stress line 1
(LAs Pit Rates ih) Baseline
agenda (Conflation) recommend
Obviously, the outputs *dgenda and *recémmend do not match up to the real
data agénda and recommend.
There are several things we can do next.
We could, for instance, treat the noun agenda as an exception to the extra-
metricality clause (2b). This move would not be too outrageous in prin-
ciple, since phonological rules can be subject to idiosyncratic exceptions, as
we will see in some detail in the last section of the chapter. However, the
exception-based solution cannot solve the problem posed by the wrong con-
tour *recémmend.
352 = Syllable Weight
The reason recommend remains problematic is that verbs are not marked for
extrametricality in the first place, and therefore making them exceptions to
extrametricality will by definition be inconsequential.
An alternative explanation for the stress patterns of the forms in (1) could
be that these forms are assigned right-headed feet, instead of the normal left-
headed feet:
(5) - : Stress line 1

Ci 2) Baseline
agenda recommend
This approach, however, goes against the reasonable assumption that

metrical structure is uniform across the board in each language, so that, for
Metrical structure example, in any given language all the feet are left-headed (or right-
is uniform across
headed), End Stress invariably affects the left (or the right) edge, and so
the board in each
language
on. Under this alternative proposal, English feet would sometimes be left-
headed (for instance, in aspdragus or implicit) and sometimes right-headed
(in agénda and recomménd). Remember that the uniformity assumption has
very positive implications for learnability, which would be lost, or at least
considerably weakened, if we were to set the parameters word by word.
Before we go on to offer a solution, it is worth pointing out that the words

in (1) are as regular as their predecessors in chapter 12 — forms in the mould
of (1) are legion, and do not feel odd or exceptional in any way: on the con-
trary, it is *dgenda and *recommend that feel un-English. The procedure by
which stress is assigned to the words in (1) must therefore be as general as
the stress procedure responsible for their predecessors.
Syllable Weight and Metrical Accent
Up until now, syllable structure has played no part in the construction of

the metrical grid, beyond syllable peaks providing baseline elements. This
Syllable Weight 353
omission needs to be corrected. The syllable structure of the forms in (1) above
is as follows:
(6) a.gén.da in.cr[iz]se

a.mal.gam re.pl[ar]
as.bés.tos re.co.mménd
in.c[a1].sor o.blLiz]que
me.mé€n.to des.p[at]se
de.bén.ture in.ténse
in.cén.tive in.vént
Oc.t[6u].ber en.th[jur]se
a.man.u.én.sis with.dr[5:]
Notice that in these words the stressed syllables are all HEAVY, in that their
rimes contain either a long nucleus ([i:] in increase, [at] in reply, etc.) or a
coda ([n] in agenda, [n] in recommend, etc.).
By contrast, in all the words we considered in the previous chapter, the stressed
syllable (antepenultimate for nouns and penultimate for verbs) was followed
by a LIGHT syllable, with neither a long nucleus nor a coda:
(7) cime.ma as.pa.ra.gus

algebra a.lu.mi.ni.um
élephant a.li.mi.num
pla.ty.pus hy.po.chén.dri.ac
la.by.rinth me.ta.mor.pho.sis
as.te.risk hi.ppo.pd.ta.mus
For ease of reference, we now tabulate the correspondences between stress

and rime structure we are focusing on:
354 Syllable Weight
(8) Rime structure and stress in English:

Stress
Antepenultimate Penultimate Final
Nouns:
Penultimate light J
Penultimate heavy J
Verbs:
Rime
Structure
Final light v
Final heavy v
The connection between the location of stress and the structure of the rime
The connection obviously needs to be encoded in the grid. In order to achieve this, we pro-
between the
ject the baseline asterisks of heavy rimes onto line 1. The word ACCENT is
location of stress
and the structure commonly used to refer to this projection, perhaps not very helpfully, given
of the rime is the other functions of the word, among them the designation of the graphic
encoded in the mark signalling stress.
grid through the
device ACCENT
We illustrate the accenting procedure in (9) with agenda:
(9) z Stress line 1

+. ee * * <*> Baseline
agenda — agenda — a.gen.da
Baseline and Accent
Extrametricality
In (9) all the syllable heads (including the /e/ of gen) supply an asterisk to
the baseline, as usual. In addition, the baseline asterisk dominating the /e/
of gen, a heavy syllable, projects onto line 1 before footing (2c) is activated.
This line-1 asterisk on gen defines the head of a foot. Now, we know from
the previous chapter that heads imply constituents, and constituents heads,
Syllable Weight B55
as encapsulated in the Faithfulness Condition in (48) of chapter 12, which

we now repeat in (10):
(10) Faithfulness Condition:

Each grid constituent has a head (plotted in the line immediately above),
and each head has a domain (delimited in the line immediately below)
It follows from the Faithfulness Condition that the output grid in (9) effect-
ively contains one foot, with its head already explicit on gen:
(11) * *
+ *+ <a + (2) <a
agenda — a.gen.da
Faithfulness
Notice that, instead of making accent make the head explicit in line 1, we
could make accent supply the corresponding parenthesis in the baseline,
the line-1 asterisk then being automatically provided by the Faithfulness
Condition in (10):

or ee ae * (*) <*> Baseline
agen.da — agenda — a.gen.da
Accent Faithfulness
You will observe that the ultimate effect of the two procedures is identical.
Consequently, we will have nothing further to say on the matter here.
In order to remain notationally neutral, however, we shall enter in our

grids both the asterisk and the parenthesis whenever we activate the accent
procedure.
If footing were to disregard the pre-existing line-1 asterisk and its line-0
parenthesis, the effects of accenting would obviously be nullified, and the
accent procedure would be futile:

yaltaac tt. (* *) <*> Baseline
agen.da — a.gen.da
Footing
356 Syllable Weight
In order to avoid this result, we enrich the theory with a FREE ELEMENT
A FREE ELEMENT CONDITION limiting the action of metrical algorithms to the building of
CONDITION limits
structure:
the action of met-
rical algorithms to
the building of (14) Free Element Condition: |
structure Only metrically free elements may undergo metrical construction
If the pre-existing head mark on gen is to be respected by the footing pro-

cedure, in keeping with the Free Element Condition in (14), and if English
feet are necessarily left-headed, then the pre-existing head mark supplied
through accenting needs to be aligned with the left edge of the foot:
Explain exactly why the accentual he

aligned. _ 2
(15) 3 Stress line 1

* (*) <*> Baseline
a.gen.da
Notice that the ensuing foot is DEGENERATE (it has its growth stunted!),
with only one syllable. This result shows that falling short of foot binarity
is preferable to contravening the Free Element Condition or the English set-
ting of the headedness parameter.
We have now successfully accounted for the penultimate stress of agenda

and similar nouns in (1a) through resort to SYLLABLE WEIGHT, that is, by
computing syllables with a complex nucleus or rime as heavy, and assign-
ing them accent. The analysis also accounts for the final stress of the verbs
and unsuffixed adjectives in (1b), allowing for the general exclusion of these
categories from extrametricality marking:
(16) + A * Stress line 1

aM 5 to Rte. ante Ga) tade) Baseline
re.co.mmend — re.co.mmend ~ re.co.mmend —>
Accent Footing
Syllable Weight 357
e Stress line 2
> * * Stress line 1
Srp woe *+n*isaed(®) 1ieBaseline
— re.co.mmend — re.co.mmend
End Stress Conflation
Stress systems that require accenting of heavy syllables, as English does, are
said to be QUANTITY-SENSITIVE, and stress systems that do not require __ Stress systems
accenting of heavy syllables are said to be QUANTITY-INSENSITIVE. oe oe ey
syllables are
QUANTITY-
SENSITIVE, and
stress systems that
do not accent
heavy syllables
are QUANTITY-
We wind up the section with an update of the English stress algorithm, INSENSITIVE
incorporating both the appropriate accent clause and conflation:

b. Make the right-most baseline asterisk extrametrical (in nouns and
suffixed adjectives only)
c.. Accent all (metrical) heavy syllables
d. Construct line 1 by building left-headed feet iteratively from right
to left
. Construct line 2 by applying End Stress [Right] on line 1
f. Delete line 1
The Word-Final Consonant
The algorithm in (17) appears to run into difficulties when the data set of
verbs and unsuffixed adjectives is extended.
Consider the forms in (27) of chapter 12, which we now repeat in (18) syl-
labified:
(18) im.plicit in.si-pid imagine im.pé.ril

de.vé.lop en.déa.vour ad.mé.nish en.dé.mic
as.t6.nish de.li.ver de.tér.mine pe.llu.cid
358 Syllable Weight
The words in (18) end in a consonant ([nJ, [f], etc.), and therefore their
final syllable is heavy according to the definition of heavy syllable we
gave in the previous section. Therefore, following our present reasoning,
such a syllable ought to receive accent and carry stress, but it manifestly
does not.
This situation is obviously related to the observation we made in chapter

10 that the distribution of English consonants word-finally differs from
their distribution word-medially.
In particular, we saw then that the regular coda consonant can be followed
by one extra consonant word-finally, in a manner only restricted by sonor-
ity sequencing (leaving aside some cases with two obstruents):
(19) camp punt fond pink fence

build silk pulp bulb belt
shelf twelve pulse kiln film
farm fort ford corn harsh (in rhotic accents)
fork harp serve herb surf (in rhotic accents)
You will remember that we proposed affiliating this additional consonant

directly to the syllable node.
Suppose now that the analysis is extended to all word-final consonants,

whether or not they are preceded by another consonant. If this step is taken,
We shall assume
the last consonant of English words will systematically lie outside rime struc-
that the last con-
sonant of English
ture, and the last syllables in the words in (18) will be light. We illustrate
words lies outside this situation with imagine and recommend in (20) and (21), respectively, with
the rime the non-rimal final consonant separated from the rime by a dash (we will
propose an alternative in the next section):
Syllable Weight 359

rad ne tuts Baseline
ima.gi-ne — i.ma.gi-ne
Footing, etc.
(21) 5 = Stress line 1

* * * * * * + (*) Baseline
re.co.mmen-d — re.co.mmen-d —~ _ re.co.mmen-d

Accent Footing, etc.
You can see in (21) that the final syllable of recommend remains heavy even
after the exclusion of the word-final consonant from the rime. In imagine in
(20), on the other hand, this exclusion makes the final syllable light: gi. As
a consequence, this syllable is parsed as the weak element of a left-headed
foot, just as we showed in (20).
E¥ Long Vowels in the Last Syllable

So far we have been assigning systematic right-most extrametricality to nouns
and suffixed adjectives. The examples in (22) appear to confirm that this extra-
metricality is assigned irrespective of syllable weight, indeed of the num-
ber of word-final consonants:
(22) a. instrument b. significant

labyrinth revérberant
détriment éxcellent
interdict ignorant
asterisk irréverent
catapult déminant
Unexpectedly from this perspective, the nouns in (23) have final stress:
(23) chimpanzée seventéen

enginéer macaréon
brigadé6on margarine
brigadier magazine
smitheréens referée
millionaire questionnaire
360 Syllable Weight
Crucially, all the forms in (23) have a long vowel in their final syllable: English
English extramet- extrametricality is therefore blocked by a long vowel, although obviously
ricality is blocked
not by a coda consonant.
by a long vowel
An update of the English stress algorithm is desirable at this point:

b. Innouns and suffixed adjectives, make extrametrical the right-most
baseline asterisk if it dominates a syllable with a simple nucleus
Qa Accent all (metrical) heavy syllables
d. Construct line 1 by building left-headed feet iteratively from right
to left
e. Construct line 2 by applying End Stress [Right] on line 1
f.. Delete line 1
Puzzlingly, there are nouns with a long open final syllable which do not
There are nouns exhibit the expected final stress:
which end in a
long open syllable
and do not exhibit
the expected final
stress
(25) buffal[ou] Malib[u:]

calic[ou] caérib[u:]
mosquit[ou] Méxic[ou]
Kikty[u:] jujits[u:]
albin[ou] command|[ou]
tomat[ou] potat[ou]
wind[ou] méad[ou]
A way of accounting for this situation involves treating these word-final

vowels as short underlyingly (buffal/o/, etc.), to allow them to induce
extrametricality. After stress has been assigned, these vowels will undergo
lengthening:
Syllable Weight 361
(26) buffal/o/
<o> Extrametricality
rar) Stress (NB left-headed footing)

o: Word-Final Lengthening
ov Diphthongization
In (27) we formulate the rule of Word-Final Vowel Lengthening, somewhat

informally. Notice that only high and mid vowels lengthen — the final low
/z/ of algebra, for instance, reduces to [a] (compare algebraic, where that /z/
is word-internal and becomes [et], by Vowel Shift and Diphthongization):
(27) Word-Final Vowel Lengthening:

Vos Vr / #
—low
The existence of rule (27) obviously makes the surface length of word-final
vowels predictable, and therefore their underlying representation as short
becomes plausible, indeed desirable on our familiar economy tenets.
Our repeated warning that English stress assignment is by no means a

simple matter is certainly receiving ample confirmation.
Moras
Implicit in the discussion throughout has been the irrelevance of onsets to
syllable weight. Consider in this connection the contrast between the two
sets of words in (28):
(28) a. aggregate b. agénda

réprimand incisor
instrument vidla
In (28a) the penultimate syllables have a complex onset, but a light rime. In
(28b), they have a heavy rime, but the onset is simple, or even null, as in viola.
362 Syllable Weight
Show how this

. :of the words in
Now, while all the penultimate syllables in (28a) involve some form of internal
complexity, only the forms in (28b), with a complex rime, bear penultimate
stress. This confirms that it is only the rime that is relevant to the determina-
Only the rime is tion of syllable weight.
relevant to syllable
weight
“How doe
weight follow?
The conclusion that only rimes determine syllable weight does not follow
from our present formalism. Compare, for instance, the syllable structures
of gre in aggregate and gen in agenda:
(29) o o
R R
N
| N
NAN
| | x Xe
TI
aie’ —WaAballet
giob we ea
There are as many skeletal slots in gre as in gen (three), and yet only gen
behaves as a heavy syllable.
Suppose then that we suppress the skeletal slots that correspond to the onset:
(30) a. b.
=i
gq 4 9
w—
Z—
——xX—
o
Syllable Weight 363
We can now compute syllable weight simply as a function of the number

of skeletal slots in the rime: a syllable will be heavy if and only if its rime
contains more than one skeletal slot. We spell this out in (31):
(31) 0 0
R
| R
|
N
| N
| 1X = light o
| 2X = heavyo
e r
|
¢ g
||
eon
The syllable gre now contains only one skeletal slot, and therefore it is
reckoned as light. By contrast, gen includes one additional slot, and there-
fore it is reckoned as heavy.
In the approach we are now developing, skeletal slots are referred to

Two moras define
as MORAS. Traditionally, this term designates a basic unit of classic a heavy syllable,
versification, and is also used in the phonological analysis of Japanese. Moras and one mora a
are conventionally represented by the Greek letter “uw” (“mu” [mju] in light syllable
English):
(S20 ae 0 Oo
R
| R
N
| N
|
u ul
gS Yr
|
e S e n
In the mora-based trees in (32) we have left the onset consonant

unaffiliated, to make the presentation simpler. However, as we pointed out
364 Syllable Weight
re in
in chapters 9 and 10, all segments need to be part of prosodic structu
order to be licensed.
We can incorporate the onset consonant or consonants into syllabic struc-

ture in one of two ways — we can link them either to the first mora or to the
syllable node directly, without any mora intervening:
(33) rat o b. oO
N N
|bow |Ul
J GP
a
Pease ce Tl
Sae
Either way, of course, the onset consonant will have no mora of its own,
and therefore it will not contribute to syllable weight.
The replacement of skeletal slots with moras is generally assumed to make
the intermediate R and N nodes superfluous:
(34) a. Oo ib: Oo
IN Wow
i Wow
ge
oamec n
||
cane n
The structure in (34a) keeps moras (attached to o) formally distinct from seg-
ments (all attached to w). In the alternative structure in (34b), by contrast,
onset segments are attached directly to 6, with no p intervening.
One final question that needs addressing concerns the source of the
moras themselves. One reasonable answer involves lexical projection from
vowels, and structural projection from consonants that are preceded by a
vowel and followed by a consonant, as is /n/ in agenda — this procedure is
often referred to as WEIGHT BY POSITION:
Syllable Weight 365
65) Peat PT OESFi
a re na
mice a gen?
ane otf
da = —>
WBP
a gen da
The final consonant in words like imagine will of course not project a mora,
since it is not followed by another consonant:
(36) “GE SRS ap
i ma gi n(e)
This structure obviously makes its predecessor in (20) redundant, since

the final —n does not contribute weight.
[4 Foot Structure and Universal Rhythm

The metrical parameters we have been using define the following symmetric
inventory of feet:
(37) Symmetric foot inventory (6 = syllable; L = light o):
left-headed Right-headed
Quantity-sensitive ( L) (Lo)
(3) (2) Regular
Quantity-insensitive (50) (oo) Degenerate
(s) (s)
366 Syllable Weight
Notice that in quantity-sensitive systems the head’s sibling, if there is one,

must be light.
Given the inventory in (37), we would expect each of the patterns it con-
tains to have an even chance of being realized across languages. However,
recent research has shown that reality does not live up to this expectation.
In particular, only the patterns in (38) have been found strongly represented
in the world’s languages:
(38) Asymmetric foot inventory (H = heavy 0):

Left-headed Right-headed
Quantity-sensitive CL) (L o)
(H) (H)
L is of course equivalent to , and H to pu. Therefore, from a mora per-

spective, the quantity-sensitive feet in (38) are constrained by the set of require-
ments in (39):
(39) Constraints on quantity-sensitive feet:

a. Left-headed feet have precisely two moras
b. Right-headed feet have at least two moras
c. The head’s sibling in a right-headed foot has precisely one mora
The point of these constraints is that syllable sequences that do not comply
with them will fail to be footed. There is, however, some evidence that the
degenerate (L) foot may be allowed parametrically: we saw in chapter 12
that such feet seemingly exist in Aklan and Maranungku, for instance (cf.
(ma)(tind)(marun) ‘being lazy’ and (mére)(pét) ‘beard’, respectively).
In quantity-insensitive systems, the moraic makeup of the syllables plays
no role in foot structure — indeed, it is reasonable to assume that in these
systems all syllables are monomoraic by definition.
Syllable Weight 367
You can see in (38) above that, in the model we are discussing, foot type is
severely restricted in quantity-insensitive systems. In particular, only left-
headed binary feet are allowed for in these systems.
The lopsided foot system in (38) is grounded in limitations of rhythm per-
ception by humans. Specifically, when humans heara string of elements of The lopsided foot
system in (38) Is
even duration but uneven intensity, they perceive a succession of units made
grounded in limi-
up of a strong element followed by a weak element, that is, a succession of tations of rhythm
left-headed feet. By contrast, when they hear a string of elements of even perception by
intensity but uneven duration, they perceive a succession of units made up humans
of a short element followed by a long element, that is, a succession of right-
headed feet.
The formalization of this contrast in terms of moras is as in (40):
(40) a. Left-headed foot: (o oO)
Mow
b. Right-headed foot: ] A
otientt
These structures correspond to the maximal feet in the inventory in+ (38) above.
The justification for their non-degenerate minimal versions ((6) in both
+ Mow
types of feet, but (66) in right-headed feet only) is less clear.
me
368 Syllable Weight
We formalize the perceptual foundation of metrical rhythm in (41) as the

We formalize IAMBIC-TROCHAIC LAW - remember that “iamb” and “trochee” are the
the perceptual
names given in classical versification to right-headed and left-headed feet,
foundation of
metrical rhythm
respectively:
in the I|AMBIC-
TROCHAIC LAW
(41) Iambic—Trochaic Law:
a. Elements of uneven intensity (and even duration) pair up as left-
headed feet (= trochees)
b. Elements of uneven duration (and even intensity) pair up as
right-headed feet (= iambs)
The noted differences in frequency of foot types in the stress systems of the
world obviously support the asymmetric foot inventory in (38) above. Note,
however, that this is not the only possible interpretation for the rarity of some
foot types. In particular, this rarity could also be attributed to performance
factors related to the realities of human perception encapsulated in the
The rarity of some Iambic-Trochaic Law. From this perspective, the Iambic—Trochaic Law
foot types could
would be irrelevant to the competence grammar that we must assume
also be attributed
to performance
underpins metrical systems, and the parametric system in (37) above could
factors be maintained as a model of metrical competence level in spite of the typo-
logical facts reported.
Non-Rhythmic Stress
Binary feet are, of course, the formal incarnation of alternating rhythm, as

we now remind ourselves of with the help of hamamelidanthemum:
(42) * * %
Stress line 1
Baseline
( ey) CG 2) é =) <*s
hama meli danthe mum
Without left-headed binary feet, dan or me could not be stressed, since End
Stress can only stress one syllable on the word edge, as you know.
Syllable Weight 369
Some languages do have asingle surface stress located on the edge, with
or without extrametricality. One such language is Bengali, spoken in Bengal
(Bangladesh and Indian West Bengal):
(43) Bengali stress:

apon ‘personal’
dhdépa ‘Washerman’
apnar ‘your own’ (honorific)
batforik ‘annual’
Onufpron ‘pursuit’
paromanobik ‘atomic, molecular’
a“
The pattern in (43) can obviously be derived by End Stress, without

recourse to footing.
In yet other languages, the derivation of the single surface stress is less
straightforward. Consider, for instance, the following data from (Khalkha)
Mongolian, the official language of the Mongolian Republic:
(44) Mongolian stress:

bosgu:l ‘fugitive’
baria:d ‘after holding’
xoyerduga:ir ‘second’
garaisa: ‘from one’s own hand’
ali ‘which’
x6tabara ‘leadership’
Your first reaction to these data will probably be utter bewilderment.
Closer inspection reveals a rather simple pattern, however: Mongolian

stress is carried by the first long vowel of the word or, in the absence of
long vowels, by the first syllable.
370 = Syllable Weight
This type of pattern poses an obvious challenge to our metrical model, and
we now attend to it.
The distribution of stress in (44) suggests that in languages like
In languages like Mongolian stress is not a direct consequence of rhythm. Therefore, it can-
Mongolian stress
not be formalized through binary footing. Consider, for instance, xétabara,
is not a direct
consequence of
where there are no fewer than three stressless syllables after the stressed
rhythm. Therefore, syllable: rhythmic stress would of course favour *xétabara. Similarly, xoyar-
it cannot be for- dugd:r ought to have been *xoydrduga:r.
malized through
binary footing
o anyofthe otherwords in (4
Ms
A quite convincing analysis of Mongolian stress runs as follows. We first

subject all long vowels to accenting:
(45) Rac * “* — Stress line.J

* * * * (* (# * (*) (*) Baseline
garaisai — 8a fai sar —+ $a rai Sai

Accenting Faithfulness
We then apply End Stress [Left] to enhance the first of these accents:
(46) Stress line 2

me Ser mOtLeSSmunjen|
ete) CS), * (*)(*) Baseline
ga raisar => ga TaiSat
End Stress [Left]
Finally, we apply line conflation to dispose of all secondary stresses:
(47) F Stress line 2

oe * Stress line 1
SO) *(*) * Baseline
gayaisat — garaisa:
Conflation
The result is gard:saz, exactly as required.

Syllable Weight 371
The procedure is also successful in forms with no long vowels. In such

forms the accenting clause will be inoperative. As a consequence, there will
be no accent-induced marks on stress line 1 — indeed, no stress line 1 at all
at this stage.
In the final stage, End Stress [Left] will simply enhance the word’s initial
syllable, as we show in (48):
(48) + * Stress line 1

et F SF FE + F + Cc + + =) Baseline
xOtebsara — xOtabsara — xOtabara

End Stress [Left] Faithfulness
Ei] Unbounded Feet
The development we have just introduced also accounts for languages

where stress falls on the first or the last heavy syllable (again independently
of rhythm), but where the polar opposite syllable is stressed in the absence
of heavy syllables.
This apparently more complex pattern can be illustrated with the Uralic lan-
guage Selkup:
(49) Selkup stress:

kd ‘winter’ sori ‘white’
Amirna ‘eats’ qorkitil ‘deaf’
qéltsimpati ‘found’ pynakisd: ‘giant!’
kananmi: ‘our dog’ karman ‘pocket’
qumé:qi ‘two human beings’ qumit ‘human beings’
turtsigqo ‘to work’ uttsikkak = ‘I am working’
urtsS:mit ‘we work’ uttsikké:qi ‘they two are
working’
372 Syllable Weight
The Selkup pattern is in one respect the mirror image of the Mongolian pat-
tern: Selkup stress falls on the last (rather than the first) of a sequence of
long vowels. However, in words with no long vowels, stress falls on the ini-
tial syllable, just as in Mongolian.
The Selkup pattern can be derived straighforwardly by constructing
unbounded left-headed feet — notice that, crucially, we did not construct feet
in Mongolian: more on this below. As is implied by the name, the number
The number of baseline elements in an UNBOUNDED FOOT is undefined:
of baseline
elements in an (GO) rae w lon ee Stress line 1
UNBOUNDED Baseline
Ce) ai) (* * *)
FOOT is
urtsormit amirna
undefined
The left-headed unbounded feet in (50) are anchored on the heads supplied
by the accenting procedure, or, in the absence of such heads, on the base-
line. At word level, Selkup End Stress has the opposite setting to Mongolian
— “right” in Selkup and “left” in Mongolian:
(51)..¥a: : Stress line 2

i at D 4 Stress line 1
Cm aay) + (* =)" Baseline
urtsormit — ulrtsormit
Conflation
Dee wee Stress line 2

(*) : Stress line 1
ects) (* *. *) (Baseline
amima — amirna
Conflation
Once more, we have achieved the desired result through the familiar
parameters, simply enriched with unbounded feet.
We saw above that in Mongolian there is no foot construction.
In particular, Mongolian feet originate in the inevitable relationship

between foot heads and foot boundaries encapsulated in the Faithfulness
Condition. In (52) we offer the whole derivation of gararsaz:
Syllable Weight 373
(52) ¥ Stress line 2

eS Sag hin) P Stress line 1
Ste ree) Pree) * (*) * Baseline
garaisa: — garaisat — garatsa! — ga raisa!
Faithfulness End Stress [L] Conflation
Crucially, the first syllable remains unfooted.
EE] Idiosyncratic Accent
We have remarked in several places that exceptions to phonological rules

are far from uncommon. Importantly, we shall now show that even excep- Even exceptional
forms are subject
tional forms are subject to the overall system of rules and principles.
to the overall
It would be most surprising to find an unsuffixed noun in English with system of rules
the primary stress outside the customary THREE-SYLLABLE WINDOW and principles
on the right edge, allowances being made for some additional factors that
we will examine in chapter 15. The three-syllable window is conveniently
instantiated by the chain derivation médicine, medicinal, medicindlity, where
stress can be seen to move rightwards to keep within the window (we
observed a similar situation in the Macedonian data we presented in chap-
ter 12). In the possible médicinelessness, the violation of the three-syllable win-
dow is only apparent, since suffixes like -less and -ness are systematically
excluded from the domain of stress, as we will see in chapter 16. A poten-
tially irreducible case of three-syllable window violation is provided by a
word like Kalevala, the name of the Finnish national epic. The question is,
how will English speakers (radio and TV presenters, for instance) stress
this word? The answer is that they can assign it penultimate stress, Kalevdla,
by lengthening the penultimate vowel to [a:], or antepenultimate stress,
Kalévala, by keeping the penultimate vowel short and ultimately reducing
it to schwa. What would have been unexpected, indeed quite astonishing, The device of
would have been for the word to have received initial stress (Kdlevala), in lexical accent
defiance of the three-syllable window. Crucially, though, this pattern is in allows a natural
interpretation of
itself neither absurd nor impossible: it is indeed the pattern the word has
the ultimately
in Finnish, a language with systematic word-initial stress, like Bengali. lawful behaviour
We will now show that the device of lexical accent allows a natural inter- of exceptional
pretation of the ultimately lawful behaviour of exceptional stress. Consider stress
the real English patterns in (53):
374 Syllable Weight
(53) a. Kentucky b. Berlin

Mississippi violin
The forms in (53a) ought to have had antepenultimate stress (Kéntucky,

Mississippi), because their penultimate syllable is light (the ck and pp clus-
ters are purely orthographic), and their last syllable extrametrical, as they
are nouns.
The short vowel in the last syllable of the forms in (53b) ought also to have
been extrametrical, and therefore stressless.
The facts just mentioned reveal that the patterns in (53) contravene the rules
of English stress. The question is whether this contravention is wild (as
*Kdlevala would have been) or still constrained by the general principles.
Irregular stress can in fact be reconciled with the regular stress procedures,
which remain in force in all cases. In particular, we shall assume that the
vowels stressed in the forms in (53) idiosyncratically carry an accent in lex-
ical representation:
(54) ¢ (*
Kentucky Berlin
Vowels with a lexical accent are effective foot heads before the derivation
Vowels with a lex- even starts. We illustrate in (55) for Kentucky. Notice that when nuclei are
ical accent are
projected onto the baseline, the preassigned asterisk is raised to the next line
effective foot heads
before the deriva-
up, to preserve differentials:
tion even starts
(55) 3 ; i tart Stress line 1

¢ Tans 5 Ae) *(*)igt>, teaG@a@) om Baseline
Kentucky — Kentucky — Kentucky — Kentucky > Kentucky >
Baseline Faithfulness Extrametricality Accent
Syllable Weight 375
y Stress line 2
mois CG 2*) * Stress line 1
CE - * (*)<*> Baseline
— Kentucky — Vacuous + Kentucky > Kentucky
Faithfulness Footing End Stress [R] Conflation
You can see that the desired result follows automatically. Berlin and violin
in (53b) above obviously suggest that extrametricality is blocked by lexical
accent.
Given this approach, a form like *Kdlevala is underivable. Let us see what
happens if we hypothetically provide its first vowel with a lexical accent: “form like
*Kdlevala is
underivable in
(56) 4 e English
Kalev/ze/la or Kalev/a:/la
The respective derivations will be as follows:
(57). a. - os r Stress line 1

ee ite ete (i+ .* <*> Baseline
Kalev/z/la — Kalev/z/la > Kalev/z/la >
Baseline Extrametricality
* Stress line 2
Sask oo * Stress line 1
EYEE =. Set cee ae eae, ee *(* *) <*> Baseline
— Ka lev/x/ la > Ka lev/x/ la > Kalev/x/ la
Footing End Stress Conflation
b. % = = Stress line 1
( (eRe ere (Parr? Fs” paseline
Kalev/a:/la > Kalev/a:/la > Kalev/a:/la >
Baseline Extrametricality
3 = 2 = Stress line 1
Gt Bera (*\o@Qi<ts # 1%1 *). <*> Baseline
— Ka lev/a:/la > Kalev/a:/la — Kalev/a:/la
Accent Footing End Stress, etc.
376 Syllable Weight
You can see that the lexical accent on Ka has no effect on the final output.
If so, forms like *Kdlevala simply cannot exist in English. This outcome confirms
the suitability of lexical accent for the formalization of stress exceptions. It
also vindicates line conflation, which is obviously needed to dispose of stresses
Line conflation is that are unwanted in the surface: Kalevala, for instance, can surface as Kalévala,
needed to dispose
but not as *Kalévala, with the lexical accent surfacing as a secondary stress.
of stresses that are
unwanted in the
surface
The hypothetical English situation we have just discussed actually exists

in Macedonian. In the previous chapter we saw that regular Macedonian
stress is antepenultimate. There are, however, a few cases of penultimate or
final stress in modern loanwords, exemplified in (58):
(58) a. literatura ‘literature’ b. citat ‘quotation’

romantik ‘romantic’ autobus ‘bus’
konsumator ‘consumer’ . restoran ‘restaurant’
Upon suffixation, stress moves rightwards whenever it needs to in order to

remain inside the three-syllable window, as is the case in the forms under-
lined in (59):
(59) literatura romantik konsumator

literaturata romantikot konsumatorot Sg. def.
literaturi romantici konsumAatori PL
literaturite romanticite konsumatorite PI. def.
This situation parallels the Kalevala case in (57) above, and can be
approached in a similar manner.
The final case we shall discuss concerns Vedic Sanskrit, the language of the
Veda, the ancient sacred texts of the Hindus. This case is related to the English
and Macedonian cases we just described, but has an interest of its own.
Syllable Weight 377
The data in (60) illustrate the stress system of Vedic Sanskrit (hyphens
signal morphological divisions, and § represents a voiceless alveopalatal
fricative):
(60) Sanskrit stress:

a. aSv-a-nam ‘horses’ b. aSv-a-vat-i-nam ‘having horses’
dev-a-nam_ ‘gods’ pad-vat-i-nam ‘having feet’
You may think that these patterns are truly beyond the powers of our familiar
procedures. However, we have come up against similar situations before (the
stress patterns of English are not famous for their simplicity), and have con-
sistently been able to bring them to heel after a bit of hard thinking.
Let us suppose that some Sanskrit morphemes idiosyncratically carry an

accented vowel lexically, as we indicate in (61):
(61) + + + + + * * + +
re. a hoe © Cae S

aSv-a-nam_ aSv-a-vat-i-nam dev-a-nam_pad-vat-i-nam
Of the morphemes in (60), the root asv ‘horse’ and the suffixes -a and -nam
are therefore accented, and the rest accentless. Once we take this simple step,
we can account for the Sanskrit system by means of procedures we are already
familiar with.
In particular, the Sanskrit system obeys the parameter settings we proposed

for Mongolian earlier on: End Stress [Left], with no foot construction. We
illustrate in (62) and (63) for dSvanam and devdnam, respectively:
(62) F Stress line 2

+ + + + + * © + a) * Stress line 1
iis COO CC) a" sag eabaseline

asvanam — asvanam — asvanam — asvanam
378 Syllable Weight

*+ *+ (* a + Line ]
+ +
+ (2) (3) + (*) (*) + ( * Baseline

* (G es
devanam — devanam — devanam — devanam

We do not think it will be unreasonable to take this rather spectacular result

as evidence that the metrical model we have been proposing is indeed on
the right track.
lables have branching rimes.S

whose rime does not branch ar
pair with), since by the

constituent, and conversely
final branching rime
account for this fac
Syllable Weight 379
ety Que
6S.Wl O.n Ss
What is the explanation for the 8 Under what conditions is noun
regular penultimate stress in some extrametricality suppressed?
nouns and final stress in some 9 Define the “mora". Which syllable
verbs? constituents are moraic?
2 What is meant by “syllable weight”? 10 What is “weight by position”?
Define a “heavy syllable”. 11 What statement about stress sys-
3 What is a degenerate foot? tems is encapsulated by the lambic—
4 What do we mean by “accenting” Trochaic Law?
in the context of stress assignment? 12 How do unbounded feet differ
5 How does the Free Element Con- from bounded ones?
dition prevent normal foot con- 13. What is the “three-syllable win-
struction overriding the effects of dow"?
accenting? 14 How can some apparently intract-
6 What are quantity-sensitive and able stress assignments be explained
quantity-insensitive languages? by lexical accent?
7 Why do final syllables with simple
codas not count as heavy?
Further Practice
Latin Stress
Classical Latin word stress is either antepenultimate or penultimate. Consider

the following data and say what parameter settings of the stress algorithm
will produce the correct stress patterns (hint: conflation is operative in Latin):
amicus ‘friend’ refércit ‘set over’ (3sg.perf)

agricolairum ‘farmer’ (gen.pl.) agricola ‘farmer’ (nom.sg.)
volucres ‘winged’ (pl.) pepérci ‘spare’ (1sg.perf)
simula: ‘snub nosed’ (fem.) magister ‘master’
Demonstrate the application of the stress algorithm to the data shown.

380 Syllable Weight
Cairene Arabic Stress
The dialect of Arabic spoken in Cairo has the following syllable types: light
CV, heavy CVV and CVC, and superheavy CVCC and CVVC. Stress may
fall on the final (as in a. below), penultimate (as in b.) or antepenultimate
(as in c.) syllables (conflation is also operative here):
a. katabt ‘T wrote’
jadgdgaat ‘pilgrimages’
sakakiin ‘knives’
b. Samialti ‘you (fem.sg.) did’
hadaani ‘these (fem.du.)’
katabta ‘you (masc.sg.) wrote’
mudarris ‘teacher’
martaba ‘mattress’
katabitu ‘they wrote’
fJadgaratun ‘tree’
?adwijatuhu ‘his drugs’
c. buxala ‘misers’
kataba ‘he wrote’
fadgaratuhu ‘his tree’
?adwijatihumaa ‘their drugs’
Making use of the machinery you have available, show the relevant para-
meter settings for the assignment of stress in Cairene. (Hint: all rime ele-
ments are stress bearers.)
Guajiro Stress
Consider the following sets of data from Guajiro, a language spoken in

Venezuela and Colombia (adjacent vowels are tautosyllabic; all segmental
symbols are IPA):
a. ipa ‘stone’ b. tatfé?e ‘my ear’

e?rahda ‘to watch’ atyhaa ‘to know’
mi?ira ‘party’ a?jataasy ‘she works’
ifi ‘a well’ o?jot6waa ‘to cut’
a?lanaa ‘to fell’ irama ‘deer’
iffi ‘be sour’ o?unysy ‘she goes’
so?u ‘her eye’ a?wanaahaa ‘to change’
sa?anyin ‘she weaves’ aréepa ‘corn bread’
Syllable Weight 381
c. sapdaatapy ‘shoelace’ d. dasajawaa ‘to speak’

akamaha ‘to smoke’ éemerawaa ‘to rest’
Ousahaa ‘to kiss’ fipynaahee ‘from above’
atpanaa ‘rabbit’ éirakawaa ‘to turn one’s eyes
to something’
Ottahaa ‘to distribute’ fiinalu?u ‘at the bottom’
ha?jumtlery a fly’ Jéemeraain ‘she rests’
(i) Is there an extrametricality clause in the stress algorithm of Guajiro?

(ii) If yes, what are the criteria which determine extrametrical elements?
(iii) | What is the foot structure of Guajiro?
(iv) Is Guajiro quantity-sensitive?
(v) What is the Guajiro setting for End Stress?
(vi) Does conflation apply?
(vii) Show how the settings of the Guajiro stress algorithm derive the stress
of following words:
fe?uhdasy ‘it is needed’ a?jalahaa ‘to cry’

tousahyin ‘J kiss’ ko?6i ‘beehive’
In the last three chapters we have reviewed stress, but still have not said
specifically what stress is, having instead relied on general expressions like
“prominence” or “salience” to identify the phenomenon. In chapter 11, we
saw that vowels which do not support metrical structure in English tend to
be pulled towards the centre of the vowel space, often becoming schwa:
Canada vs. Canddian. We also mentioned several phenomena affecting con-
sonants equally related to stress, like the aspiration of voiceless stops and
/t/ weakening. All these segmental facts provide evidence about the loca-
tion of stress, and, by implication, about its existence. However, this evidence
is indirect by its very nature. More to the point, it is only partial, since the cor-
relations are by no means perfect: full vowels can be stressless (cf. raccdon
in many accents), stressed vowels can have the quality of vowels plausibly
analysed as reduced in other contexts (cf. [1] in distinguish), aspirated voice-
less stops occur in some stressless contexts (cf. tendcious), and so on. A more
direct, and particularly salient, diagnostic for English stress is intonation,
as we shall now explain. First, however, we must prepare the ground
by describing succintly what intonation is. An understanding of the
Tonal Phonology 383
mechanics of intonation allows us to connect with other tonal phenomena

that are absent from English, but are important in a world-wide context, and
we will also discuss them in the chapter.
The Phenomenon of Intonation
Any English utterance, irrespective of length, can be delivered in a variety

of tunes. Consider such simple monosyllables as yes or no. Suppose some-
body says to us “Are you ready?”. We answer “yes” or “no” to express agree-
ment or disagreement, respectively, with the proposition “you are ready”
contained in the question (the term PROPOSITION is used to refer to the
logical content of utterances).
The way we will utter “yes”, “no” or any other words will be very differ-
ent from the flat, monotonous delivery typical of Daleks, robots or computers.
Human delivery of language will almost certainly be “intonated”, with the

pitch rising and falling as we say the words, in a kind of simplified singing. Human delivery
of language will
In (1) we give three particularly common manifestations of such intonated
almost certainly be
delivery in English. We represent the modulation of the voice impres- “intonated", with
sionistically, by means of falling or rising lines reminiscent of the lines an the pitch rising
orchestra conductor draws in the air: and falling as we
say the words
(1) a. yes Bir yes c.

\
yes
\
ho ho
\
ho
384 Tonal Phonology
In (1a), the falling line represents the falling pitch of the tune characteristic
of statements, a mode of delivery expressing agreement with the proposi-
tion (“yes, I’m coming”) or disagreement (“no, I’m not coming”).
In (1b), by contrast, the line (and thus the tune it represents) rises, to signal
incompleteness — (1b) can therefore be paraphrased as “yes/no, but I’m sur-
prised you asked”, or “yes/no, and so?”. Finally, in (1c) the tune first falls
and then rises, approximately conveying the meaning “yes/no, but...”.
The simple set of data we have just examined appropriately demonstrates

English intonation.
Our data also provide us with a first approximation to the physical substance
of intonation, namely, singing-like movements with our voice which man-
ifest as TUNES: melodies made up of a certain pitch pattern. Finally, we have
TUNES are seen that each of the tunes conveys a certain meaning, or meanings, which
melodies made up
are often difficult to express in ordinary words: this difficulty provides an
of a certain pitch
pattern. Each tune
obvious functional motivation for intonation.
conveys a certain
meaning, or
meanings, which
are often difficult
to express in ordi-
nary words
The Mechanics of Intonation

In order to understand how intonation works, we first need to under-
stand its physical mechanics. Fortunately, this is not a difficult task. We have
already hinted at the involvement of voice in the process. You will recall
Tonal Phonology 385
from chapter 1 that “voice” is the technical term used in phonetics to refer
to the vibration of the vocal folds. We said at the time that vocal fold vibra-
tion is induced indirectly, by the action of the air coming out of the lungs,
much as a flag is caused to flap by the wind. Remember, in particular, that
we cannot make the vocal folds vibrate directly: vibration happens by itself
as a result of the physical dynamics caused by the air flowing through.
In order for such aerodynamics to come about, we need to position the

vocal folds in a particular way: not too far apart, not too close together, not
too tense, not too lax, and we do have control over these gestures. We also
have the capacity to regulate the thickness of the vocal folds, and through
this thickness regulate the rising and falling of the intonational tunes: a thick
set of vocal folds will produce a lower pitch than their thin equivalent, in
the way that a guitar’s thick strings give a lower note than its thin strings.
We have already said that intonation is nothing but a hummed melody,

albeit a radically simplified one. In a nutshell, then, INTONATION is modu-
lated pitch. Different patterns of modulation correspond to different intona- INTONATION is
modulated pitch
tional contours, or “tunes”, and we gave three of these in (1) above. We must
warn at this point that, like stress, English intonation is a very complex area,
and therefore our discussion in the following sections will have to be
confined to the essentials. Different patterns
of modulation cor-
respond to differ-
ent intonational
The Primitives of Intonation
contours, or
TUNES
In (1) above we represented the variations in pitch that make up intonational
contours by means of falling or rising lines, which we compared with the
lines drawn in the air by the conductor’s baton. These lines are useful in
that they can be interpreted immediately, and accordingly they have been
made frequent use of in the literature on intonation, in a manner reminis-
cent of phonetic symbols for segments.
386 Tonal Phonology
You already know that phonetic symbols are merely a convenient shorthand,
and that the real substance of segments is their distinctive features. In the
case of intonation, we will assume that the pitch movements impressionis-
tically evoked by the lines do not constitute its real phonological substance
either. Instead, we will propose that the underlying components of the intona-
tional melodies, equivalent to the distinctive features in segments, are pitch
The underlying levels, essentially H and L, representing “high” and “low” pitch, respectively.
components of
This is a bit like a simplified musical scale: instead of the seven musical notes,
the intonational
melodies are pitch from C to B, intonation would only have two, H and L. Also, the phonetic
levels, essentially substance of H and Lis less precisely defined than the substance of their
H(igh) and L(ow) musical correlates: the only general requirement is that H be realized at a
higher pitch than L.
There is of course much more than this to the way we use our pitch when
we talk. For one thing, different languages (different varieties of languages
even), as well as different individuals, have their own characteristic pitch
ranges. In addition, pitch has an important “paralinguistic’” function -
PARALINGUISTIC means interacting with language without being part of
it. For instance, when we are excited we expand our pitch range (and speed
up our speech rate, etc.), whereas when we are bored or depressed we reduce
the phonetic difference between our Hs and Ls; indeed, we bring down the
pitch of the Hs noticeably. All this is undoubtedly important for human expres-
sion and interaction, but it falls outside language as such, and therefore it
is beyond the scope of intonation: it is paralinguistic, rather than linguistic.
For strictly linguistic purposes, we will operate for the moment with the
assumption that the component elements of intonation are H and L. Not sur-
prisingly, matters are reallya little more complex, but we will abstract away
these complexities for the time being in order to keep the exposition clear.
Tonal Phonology 387
E¥ Autosegmental Intonation
If the basic components of intonational melodies are not the slant lines, but
the primitive elements H and L, then our original figures in (1) must be re-
written as in (2), where the lines are meant as association lines, rather than
as impressionistic representations of pitch movement:
(2) Varo Hsb BaskLecH HSL ee
yes yes yes
Th Leh bie? ert
ho no ho
The configurations in (2) closely resemble the autosegmental structures that

we introduced and discussed in chapters 4 and 6 in connection with such
phenomena as the assimilation of nasals in English or the vowel harmony
of Turkish, and therefore they must be given a similar interpretation.
First, you will notice that the primitives H and L responsible for intona-
tion are assigned to a specific autosegmental tier, distinct from the tier where
we represent the segments (more rigorously, of course, a cluster of tiers, one
for each feature).
We will refer to the tier housing H and L as the TONE TIER, and to the ele-
ments H and L as TONES - the word “tone” is of course included in the word _ We refer to the
“intonation”: in-ton-at-ion. You can see in (2) that the tone tier contains lin- "*" owe Hand
ear sequences of the tones H and L, in a manner that parallels the lineariza- ree Pee fa
tion of the segmental distinctive features in their corresponding tiers: [+back] _glements H and L
[—back], [+round] [—round], etc., for features; HL in (2a), LH in (2b), and HLH as TONES
in (2c), for tones. Also as is the case with features, H and L integrate into
the structure by means of association lines. In particular, H and L associate
to segments specifically licensed to carry this association, usually syllable
388 Tonal Phonology
head segments. Strictly speaking, tones of course associate to the skeletal

slot associated to the nuclear vowel. Thus, remember that in chapter 8 we
identified the timing tier as the baseline of the whole autosegmental struc-
ture — the skeleton acts as a central distributor for all the autosegmental tiers
(naturally, excluding those corresponding to dependent features), as the spine
of a book acts as a central distributor for all the pages of that book:
©) gd Dees lake CTL Lab lonaliner
XXX XXX XXX Skeleton
yoers y es ya e's Superordinate segmental tier
To keep the representation simple, however, we will ignore this formal detail
and continue using the abbreviated graphics of (2).
Weare now ready to interpret the structures in (2) (or (3)). (2a), for instance,
is telling us that the vowel [e] of yes is pronounced with a high pitch fol-
lowed by a low pitch, and similarly for the vowel [ou] of no. Naturally, the
time interval between the start and the end of these or any other vowels is
very short. Consequently, it would be impractical to raise the pitch for H,
sustain it there for a while, stop vocal fold vibration, and then go through
the same routine for L.
The structure in (2a) is instead implemented by a simple pitch movement,

modulated from a high start to a low finish as we articulate [e], precisely
as the impressionistic lines in (1a) indicate. What we are now arguing is that
Intonational con-
tours are a phon-
such contours are a phonetic by-product of the simpler tonal primitives, which
etic by-product are the real targets for pitch.
of the simpler
tonal primitives,
which are the real
targets for pitch
Tonal Phonology 389
Stress and Intonation
In this section, we shall show that the autosegmental behaviour of intona-

tion provides specific evidence for stress.
So far we have been analysing English intonation with the help of mono-
syllabic words. Obviously, though, English also has a great many polysyl-
labic words. In our monosyllabic words yes, no, all the tones of the tune were
associated to the lone nucleus, and the question arises as to how such tones
will associate to the several syllables of polysyllables. Will all the tones con-
verge on the first syllable of the word? On the last syllable? Will they asso-
ciate syllable by syllable in an orderly fashion? If so, will the association
proceed from left to right, or from right to left? In any event, after each tone
has been assigned to one syllable, will any excess syllables receive a tone
and, if so, which?
The configurations in (4) offer a first approximation to phonetic reality
and provide us with the key evidence we need to begin to answer these
questions:
(4) a H L H | ie
for.tu.nate.ly un.doub.ted.ly kan.ga.roo
b. -L H L H \/
for.tu.nate.ly un.doub.ted.ly kan.ga.roo
The configurations in (4a) and (4b) (corresponding, respectively, to the state-

ment and plain question tunes) puzzlingly reveal a diversity of patterns of
tone association: in the third column the two tones converge on the final
syllable; in the first column, they are as far apart as they can be, landing on
the opposite edges; finally, the pattern in the middle column is similar, but
with the first syllable skipped. At first sight, it doesn’t seem possible to come
up with a unified formal account of these patterns.
sais We —e saan
390 Tonal Phonology
If you scrutinize the patterns more closely, you will indeed find a com-
mon thread: in all cases, the second tone associates to the word’s last syl-
lable, while the first tone links to the syllable that carries the main stress.
The privileged status of the main stressed syllable with regard to tone asso-
ciation is confirmed by the data in (5):
HL Ha bale Hig Hath H L Hk

(S)nyac
congratulations
ng ewhere inate ste aaa
derisory juvenile lampoon presidency residential
b. LA iglag le ea Deere H [en
congratulations
sa 0¢freon lle alah
derisory juvenile lampoon presidency residential
yo ene on
ti
shown, and suggest
The systematic choice of the stressed syllable as the anchor for the associ-
The systematic ation of the tonal melody, irrespective of that syllable’s linear position, pro-
choice of the
vides particularly strong evidence for the reality of stress: if nothing else,
stressed syllable as
the anchor for the
pitch is measurable, and therefore we can now test syllables for stress
association of the instrumentally.
tonal melody pro-
vides particularly
strong evidence [4 Non-Lexical Tones
for the reality of
stress One obvious question outstanding concerns the syllables we have been rep-
resenting as toneless to the left of the stressed syllable.
- Why should this be q
The question arises because intonation is all-pervasive in the utterance: as

The incorporation
of M into the set we said, humans do not speak like robots or computers, and, therefore, all syl-
of tones all but lables will come out intonated. This means that syllables that do not receive a
completes the tone by the procedure as it stands need to receive a tone in some other way.
inventory: in a We shall assume at this point that initial syllables unmarked for tone are
minority of lan-
pronounced at a pitch intermediate between H and L, which we will tran-
guages, the mid
tone M is split
scribe as M, for “mid”. The incorporation of M into the set of tones is widely
into a high mid supported cross-linguistically, and all but completes the tonal inventory: in
and a mid proper a handful of languages, the mid tone M is split into a high mid and a mid
SSS proper.
Tonal Phonology 391
Some of the structures in (4) and (5) above have toneless syllables sand-
wiched between two toned syllables, and we must now describe how these
syllables are intonated. One logically possible answer would be that one of
the available tones spreads to such syllables — we have already observed left-
to-right spreading in the vowel harmony of Turkish, and right-to-left spread-
ing in the plural umlaut of German and English. However, the mechanics
of intonation are not quite like the mechanics of vowel assimilation. In par-
ticular, intonational tones do not appear to spread or, if they do, they do so
to a very limited extent: we have just seen that initial toneless syllables have
their tone assigned by default, rather than by spreading. In the case of medial
syllables sandwiched between two toned syllables, no default tone is pro-
vided either. Instead, the pitch of the left tone changes gradually into the
pitch of the right tone over the tonally empty space, a phenomenon tech-
nically known as INTERPOLATION, which we now illustrate in (6) (notice The gradual
the affinity with the lines of our initial impressionistic notation): change of pitch
over a tonally
empty space
(6) between two
i
Pitch contour toned syllables
sient? | is known as
for.tu.nate.ly for.tu.nate.ly INTERPOLATION
| | Phonological representation
H L L H
Three Types of Intonational Tones
So far we have slightly simplified the tonal melody, for presentational con-
venience. As it happens, however, there are three types of tone included in
each tune. We will refer to these three types of tone by the reasonably trans-
parent labels “word tone”, “phrase tone” and “boundary tone”, corres- Tones come Li
ponding to the “pitch accent”, “phrase accent” and “boundary tone” of much eee
of the literature. We illustrate the distribution of these three tone types in iphiace Coa
(7), again with the statement tune, previously simplified to H*L, but now “boundary tone”
given its full representation H*L'L”:
Cy. @: re |i
un.doub.ted.ly
b. ie shale
congratula
/ tions
392 Tonal Phonology
The richer, and more rigorous, representation in (7) includes the three tone
types we are now introducing: the word tone (H*), the phrase tone (L), and
the boundary tone (L”). The phrase tone bears the graphic mark “” and the
boundary tone “””, while the asterisk singles out the word tone, or, more
rigorously, the dominant component of the word tone — English word tones
can in fact be made up of two tones, graphically linked with a plus sign: all
six of H*, L*, H*+L, H+L*, L*+H and L+H can therefore be word tones in
English, in different tunes, of course. In each complex tone one of the tones
is the dominant tone (graphically, it bears the asterisk), and as such it will
associate to the stressed syllable (dominant tones are commonly referred to
as “accented” tones in the literature).
The transparent terminology we are using provides a key to the function

of the three tone types, and therefore to their linear pattern of distribution.
The boundary tone associates to segmental material on the edge of the domain
where intonational association takes place. The word tone associates to the
word’s stressed syllable. Finally, the phrase tone associates to material posi-
tioned shortly after the word tone, or, if no such material is available, to the
same material as the word tone. The functions of the word tone and the bound-
ary tone are obvious, as they signal the presence of words and boundaries,
respectively. In turn, the postulation of the phrase tone is justified by the
richness in pitch that gravitates to the last word tone in every utterance, as
we will explain in the next section.
EE Sentence Intonation
Up to this point we have illustrated intonation with isolated words. Clearly,

though, most English utterances consist of more than just one word. So, sup-
pose that, instead of uttering just “yes”, “no” or “undoubtedly”, we say “the
dark clouds in the sky threaten imminent rain’. The point about this or any
long utterance is that it can be intonated in all the ways we have illustrated
for single words (and, naturally, like the single words, also in many we have
not illustrated).
How are tunes to be associated to the segmental material of longer utter-
ances? The full intonational representation of the utterance we have just given
is as in (8):
Tonal Phonology 393
(8) a P4135 Ft HS be-k*
ees aa
the dark clouds in the sky threaten imminent rain
VY
Let us carefully dissect this structure. First, you will recognize our familiar
statement tune H* L'L” on the right edge: its three tones are associated to
the vowel of the right-most word rain, a monosyllable. If the right-most posi-
tion were occupied by a polysyllable, the three tones would of course dis-
tribute themselves over the whole word:
(9) Hen He i,Hts od? A ee”
the dark clouds in the sky threaten rain imminence
Notice that the word tones associate to the main-stressed syllable of each
word (the small stressless function words are known as CLITICS), and the
boundary tone associates to the material on the edge of the domain. In turn,
the phrase tone associates to material lying between the last word tone and
the right boundary tone, here the syllable mi in imminence. The sequence last
word tone + phrase tone + boundary tone obviously carries the main load The privileged
status of the tone
of the tune, and is often given the label NUCLEAR TONE.
complex at the
In (9) we have H*s as prenuclear word tones, but we could of course sub-
end of the utter-
stitute another tone, to get a slightly different tune, naturally with a slightly ance warrants
different meaning. We exemplify in (10) with L*: the special label
NUCLEAR TONE
often given to it
(10) Eee a Le poole
Li oe Vibe
the dark clouds in the sky threaten imminent rain
Other variations are possible in the prenuclear tone sequence, to express vari-
ous nuances of meaning. However, combinations of prenuclear tones are
limited in comparison with the variety possible in the nuclear tone itself, or
in the combinations of prenuclear and nuclear tones. Stress corresponds
to rhythmic prom-
inence, formally
E} Tone Languages represented as
greater column
height in the
We have now reached a point at which we can fully understand the pre- grid, and also
cise nature and mechanics of stress, the question with which we opened the manifested in a
chapter. Quite simply, stress corresponds to rhythmic prominence, formally number of areas
represented as greater column height in the grid. This metrical prominence
394 Tonal Phonology
has manifestations in a number of areas, ranging from the distribution of

intonational tones over the segmental string to the specific realization of
some segments, for instance, vowels (cf. vowel reduction), ts (cf. t aspiration
and weakening), etc. The close connection between intonation and stress is
expressed by the association of word tones to stressed syllables.
There are other functions of tone, besides intonation, which are extremely
common in the world’s languages, even though they do not occur in
English or most European languages. In particular, in “tone languages” tones
serve to differentiate lexical items or to express morphological functions.
In turn, in “pitch accent languages” tones mark each word with a fixed
tune. We examine these other functions of tone in the remainder of the
chapter.
Consider the words in (11), taken from a dictionary of Mandarin Chinese,
the main Chinese language, nowadays referred to as Putonghua in the
People’s Republic. Notice that we temporarily revert to the impressionistic
method of representing tones, in an effort to keep the presentation clear
(“~” indicates absence of pitch change, a situation referred to as “level tone”):
(11) shu ‘write’ shu ‘sorghum’ shu ‘technique’ shi ‘category’

e / \
moc/feeln| of mo ah’ plary, mo ‘end’ mo ‘smear’
x1—- 4
sunset’ ,
x1/ /
exercise
s
let\ 4
play y X1yA 4
wash ,
a / \
bao ‘wrap’ bao ‘hail’ bao ‘report’ bao ‘treasure’
ae fale , / 1 , N ie y v ‘
rang ‘shout rang ‘avert rang ‘give way’ rang ‘earth ,
What is striking to the Western eye here is that the words in each line are
kept apart not by their segments, which they all share, but by their tonal
structure, which in turn is uniform in each column. It is rather as if the
English yes or no (or shoe, which is segmentally reminiscent of the Chinese
words in the first line of (11)) were different words when we pronounce
them with the different tunes, not just different ways of pronouncing the
same word.
In Mandarin
Chinese and other
TONE LAN-
GUAGES tones
In Mandarin Chinese and other TONE LANGUAGES, tones differentiate
encode lexical lexical entries. By contrast, in STRESS-AND-INTONATION LANGUAGES
distinctions like English, tones provide the functional or emotional meaning of the
utterance (NB not even of each individual word in the utterance).
Tonal Phonology 395
English inyou
In most African languages and in a number of native American languages

tone also has a morphological function, similar to the function of suffixes in In most African
languages and
European languages. For instance, in English we form the past tense (essen-
in a number of
tially) by adding the suffix -(e)d to the present, so that from refuse we get native American
refused, and so on. In the Nigerian language Tiv, on the other hand, the dif- languages tone
ference between the general past and the recent past of vende ‘refuse’, for also has a
instance, is implemented through a change in the tonal makeup: vénde vs. morphological
function
vendé, respectively. Notice that tone is conventionally marked on the vowel
that bears it with diacritics resembling the marks commonly used for ortho-
graphic “accent”: an acute accent “” indicates a high tone, and a grave accent
“”" a low tone.
Do a count ° F the
in the text.
Pitch Accent Languages
We will now see that Japanese is neither a “stress-and-intonation language”,

like English, nor a “tone language”, like Mandarin Chinese, but, rather, a
“pitch accent language”. The slightly awkward expression PITCH ACCENT
The expression
LANGUAGE refers to a situation where a single fixed tonal melody is asso-
PITCH ACCENT
ciated to each word. In the Tokyo variety of Japanese, the melody in ques- LANGUAGE refers
tion surfaces as (L) H (L) in nouns, with the material enclosed by the to a situation
parentheses only present when the circumstances are favourable, as we will where each word
see. You can get a first flavour of the system from the minimally contrastive has a single fixed
pairs in (12): tonal melody
associated to it
(12) (Tokyo) Japanese surface pitch accent contrasts:

HE LH
hashi ‘chopsticks’ hashi_ ‘bridge’
mochi_ ‘birdlime’ mochi ‘durability’
ima ‘now’ ima ‘living room’
396 Tonal Phonology
mushi ‘disregard’ mushi ‘insect’

shiro ‘white’ shiro ‘castle’
tsuyu ‘dew’ tsuyu ‘rainy season’
tabi ‘socks’ tabi ‘trip’
ame = ‘rain’ ‘ame = ‘candy’
The point about these words is that they always turn up with the same
tonal melodies. This makes the Japanese system distinct from a stress-and-
intonation system like that of English: in a stress-and-intonation system hdshi
would simply be the statement realization of both ‘chopsticks’ and ‘bridge’,
while hashi would be its question counterpart.
Explainwhy Japanese
s — _
The three sets of words in (13) provide further evidence on the system
(-ga is an enclitic particle marking the nominative: an ENCLITIC isa clitic
that follows its “host”, that is, the word that supports it):
(13) a HLL
hashi-ga_ ‘chopsticks’
mochi-ga_ ‘birdlime’
b.ctitHik
hashi-ga_ ‘bridge’
mochi-ga ‘durability’
erat EL
hashi-ga ‘edge’
mochi-ga_ ‘rice cake’
There are three surface melodies here: HLL, LHL, LHH; HLH does not occur,
and cannot in fact occur. In order to account for these facts, we shall make
the following assumptions about the Japanese pitch accent system:
(14) Assumptions about the Tokyo Japanese pitch accent system:

(i) The lexical tone melody is L”H*, where, as expected, H* is the
_ tone associating to the word’s accented syllable
(ii) If the word has no accented syllable, H* associates to the word’s
last vowel
(iii) H* spreads leftwards
(iv) Tone bearers only support one tone
Tonal Phonology 397
(v) Syllables still toneless at this stage are assigned L by “default”

(we expand on this notion in the next two sections)
(vi) An unassociated L” deletes at the end of the derivation
Following from the assumptions in (14), the derivations of the forms in (13)
will be as in (15). Notice that the form for ‘rice cake’ has no accent. This is
another key difference between pitch-accent languages, like Japanese, and
stress-and-intonation languages, like English, where all non-clitic words need
to carry stress:
(15) *birdlime’ ‘durability’ ‘rice cake’
* *
Lexical accent mochi-ga mochi-ga mochi-ga
* *
Tone association mochi-ga mochi-ga mochi-ga
L” H* Sows a
+ *
|
Tone spread mochi-ga mochi-ga mochi-ga
L” iz L” H*
eeert
+ +
ae
L default mochi-ga mochi-ga * mochi-ga
L*H* LL L* H*L L” H*
* *
Stray L” deletion mochi-ga mochi-ga mochi-ga
on ere L” Ht
mdchiga mochiga mochiga

398 Tonal Phonology
You can now see that the Japanese system cannot be analysed as a tonal
system along the lines of Chinese. In particular, given the two tones H and
L and four syllables, a Chinese-type system allows for sixteen tonal con-
tours: HHHH, HHHL, HHLL, HLLL, LLLL, LLLH, LLHH, LHHH, HHLH,
HLHH, HLLH, HLHL, LLHL, LHLH, LHLL, LHHL.
Of the sixteen combinations just listed, only HLLL, LHLL, LHHL and
LHHH show up in Japanese: indti-ga ‘life’, kokérd-ga ‘heart’, atamd-ga ‘head’
and néziimi-gd ‘mouse’. This limitation would be entirely arbitrary in a tone
system, but is principled in the pitch accent system we are describing: a word
made up of four syllables can have the accent on the first syllable (inoti-ga),
on the second syllable (kékéré-ga) or on the third syllable (atdmd-ga), or be
accentless (néztimi-ga).
A handful of European languages commonly classified as “pitch accent”

(Swedish/Norwegian, Lithuanian and Serbo-Croatian) are in fact quite dif-
ferent from Japanese. Among other things, these languages have a fully fledged
intonational system along the lines of English. By contrast, intonation is con-
siderably simplified in Japanese and in tone languages, for obvious reasons.
In particular, because both tone and intonation are implemented through

pitch, a full intonational melody would inevitably distort the tonal melody.
Consequently, in tone and pitch accent languages intonation tends to be
The intonation confined to relatively simple devices, for instance a contrast in the right bound-
system of tone
ary tone or an alteration of the pitch range, with a wider pitch range sig-
languages and of
pitch accent lan-
nalling a question.
guages is relatively One of the persistent themes of this book has been and is the auto-
simple segmental nature of phonology. As it happens, autosegmental theory was
developed in the context of the tonal phonology of African languages. In
the next two sections we review the mechanics of tone association in these
languages, both for its own intrinsic interest and as a useful way of over-
viewing the general principles of autosegmental phonology.
Tonal Phonology 399
Principles of Autosegmental Association
The repertoire of tone melodies of Mende, a language spoken in Sierra Leone,

includes H, L, HL, LH and LHL. Revealingly, these melodies turn up The tone melodies
irrespective of the number of syllables in the word, as the sample in (16) of Mende turn up
illustrates: irrespective of the
number of sylla-
bles in the word
(16) H b
kS ‘war’ kpa ‘debt’
pélé ‘house’ bélé ‘trousers’
hawama_ ‘waistline’ kpakali ‘tripod chair’
Ht tH
mbt ‘owl’ mba ariCe
ngila ‘dog’ fandé ‘cotton’
félama ‘junction’ ndavula ‘sling’
LHL
mba ‘companion’
nyaha ‘woman’
nikili ‘groundnut’
‘associatio
Data such as those in (16) provide strong support for the autosegmental
approach to tone. Notice, in particular, that all the tones of each tune are
realized regardless of the number of syllables in the word: if syllables out-
number tones, then the last tone spreads onto the as yet tonally vacant syl-
lables, and if tones outnumber syllables, then the excess tones are dumped
onto the final syllable.
The tonal and

segmental tiers
This mode of behaviour demonstrates the autonomy of the tonal and seg- are mutually
mental tiers, and hints at the principles that govern their interaction. These autonomous
principles were originally formulated as in (17):
400 Tonal Phonology
(17) Original principles of autosegmental association:

1 Association Convention: associate tones and syllables one to one
from left to right
2 Well-Formedness Condition: at each stage in the derivation, all syl-
lables are associated with at least one tone, and all tones are asso-
ciated with at least one syllable
3 No-Crossing of Lines: association lines may not cross
Not all these principles have endured subsequent testing. Principles 1 and
3 have proved to be quite resilient, apart from a few cases of right-to-left
tone association. Principle 2, however, has subsequently been abandoned:
the total association effects illustrated by data like those in (16) above are
now interpreted as the result of language-specific rules, rather than of a uni-
versal Well-Formedness Condition inducing automatic spreading of tones
The Well- onto vacant syllables (cf. Mende hawdmd ‘waistline’) and automatic dump-
Formedness
ing of excess tones on already toned syllables (cf. Mende mba ‘companionr’).
Condition induces
automatic spread- Shona, a language of Zimbabwe, also shows the three association prin-
ing of tones onto ciples in action. In Shona, the suffixes -es, -er and -a show up either with a
vacant syllables high tone, as in (18a), or with a low tone, as in (18b) (the dots on the left-
and automatic hand side of the forms indicate that the stems concerned must be preceded
dumping of excess
by a prefix, a detail irrelevant to the present discussion):
tones on already
toned syllables
(18) a. ...téng-és-4 ‘sell’
. téng-és-ér-4 ‘sell to’
b. ...é@réng-és-A = ‘make read’
.eréng-er-a ‘read to’
We could set up one lexical entry for each tonal variant of each suffix, but
this duplication would obviously be uneconomical. Worse still, it would
obscure the fact that the suffixes have no fixed tone.
A more satisfactory analysis draws on the parallel case of Turkish vowel

harmony we discussed in chapter 6.
All we need to do in Shona is omit the tonal information from the single
lexical entry of the suffixes:
Tonal Phonology 401
a9, TH iE
XXXX XX XX X
HE TE TI
ten. es a
Notice that, while the stems -teng- and -ereng- carry their own tone (H and
L, respectively), the suffixes are lexically toneless. Upon concatenation, the
Well-Formedness Condition (17.2) will cause their association to the adja-
cent tone, here the only one available:
(20) H L
XXXX XX XXX XXXXX XXX
tengtest+terta ereng+test+a
This outcome ostensibly confirms the appropriateness of the association con-

ventions in (17) above.
Mysteriously, after the future prefix -ngd ‘could’ only low tones show up,
even in stems with a lexical H, such as téng- (cf. (19) above):
(21) Fk.
oiiga
AN
lenge +s +a
The solution to this puzzle lies in the Obligatory Contour Principle, formu-
lated in (21) of chapter 10 in connection with syllabification as “Similar
melodies are disfavoured as constituent siblings.” As we hinted then, the
impact of the OCP extends throughout the phonology. We offer a suitably
general reformulation of the OCP in (22):
(22) Obligatory Contour Principle:

Similar adjacent elements are disfavoured
We said at the time that the OCP is not a principle as such, but rather a "he OCPisa _
latent force motivating some of the rules and principles of languages, some- nee oie
what as the shifting of the earth’s inner matter motivates volcanoes to jijiee and
erupt: the shifting is not the eruption. In the Shona case we are discussing, principles of
the OCP motivates “Meeussen’s law” to dissimilate adjacent H tones, as in _ languages
many other African languages.
402 Tonal Phonology
ow the OCP will account for the
The autosegmental formalization of Meeussen’s law is as in (23):
(23) Meeussen’s law:

H-~ 0 /H__
Quite simply, Meeussen’s law resolves the OCP violation through the dele-
tion of the second H. The effect on the string ngdténgésd is as follows:
(24) HH = |
nga-tenge sa nga-tengesa
You can see that the deletion of just one H in the autosegmental tonal tier
deprives no fewer than three syllables of tone. These syllables do not of course
reassociate to the remaining H — if they did, Meeussen’s law would be vacu-
ous. Rather, these syllables are eventually supplied with L, the default tone
in bitonal languages. The notion of default has already cropped up in pre-
vious chapters: it is an important concept, and we will return to it in earnest
in chapter 17. Notice, importantly, that the observed effect of Meeussen’s
law on no fewer than three syllables would be very awkward to account for
in a non-autosegmental model of phonology.
Try to produce one such a non-a tose
Floating Tones
We mentioned above that some lexemes are purely tonal. We will now intro-
Some lexemes are duce two such tonal lexemes and show that their behaviour can contravene
purely tonal the Well-Formedness Condition in (17.2) above. It follows from this that this
condition cannot be upheld as a universal principle of tonal association.
Consider first the following forms from the recent past tense of Tiv, a lan-
guage of Nigeria, as we have already indicated:
Tonal Phonology 403
(25) a. 6ngd ‘heard’ b. vendé ‘refused’

yévésée ‘fled’ ngohdérd ‘accepted’
If you look closely at these forms, you will notice that the first tone differs
across the two sets: it is H in a., but L in b. By contrast, the second tone is
H throughout. Finally, if there is a third syllable, it invariably has L.
Check that ou seethese patterns before you go on,
It will be helpful to compare the forms in (25) with their counterparts in

the general past tense, which we display in (26) (the segmental allomorphy
in the verb meaning ‘heard’ is irrelevant to the discussion):
(26) a. !ungwa ‘heard’ b. vende ‘refused’

lyévésé ‘fled’ ngohord ‘accepted’
The tone in the second syllable is now L throughout, while the other tones
are the same as in the recent past. In addition, the forms in (30a) exhibit an
initial “!”, which we will disregard for the time being.
_ Again, ens
ure at you seethese p tte
oe
The tonal contrast between the two tenses in the second syllable suggests
that H is the (purely tonal) marker of the recent past, as follows:
(27) Tiv recent past:

L H LH
vende —+ vende
If there is a third syllable, it obviously does not receive its tone through spread-
ing of this right-most H, as the Well-Formedness Condition in (17.2) above
would predict, but rather by default assignment of L, similarly to what hap-
pens with the output of Meeussen’s law in Shona (cf. (24) above):
(23) oH HH HHL
yevese — yevese -» yevese

404 Tonal Phonology
If the Well-Formedness Condition in (17.2) were indeed a universal prin-

ciple governing the association of tones and segments, the output would have
been *yévésé, with spreading of the recent past H onto the final syllable.
The ungrammaticality of *yévésé thus provides evidence against the Well-

Formedness Condition in (17.2), and shows the need for its replacement by
language-specific rules in those cases where it was thought to be active.
At this point, we shall return to the general past forms in (26). The third
tone in /yévésé obviously also corresponds to the default L. The second tone,
invariably L, could also be interpreted as default, or, alternatively, as the mor-
pheme of the general past: after all, in (25) we identified second-position
H with the recent past morpheme. However, the “!” at the beginning of
the forms in (26a) prompts a different solution. The symbol “!” stands for
a DOWNSTEP, a term that refers to the realization of a following H or
The term DOWN- sequence of Hs at a lower pitch than the previous Hs — this lowering is not
STEP refers to a
so extreme as to turn Hs into Ls, however: tone differentials are maintained.
lowering in pitch
of the following Downstepping is commonly thought to be triggered by a preceding
H tone(s). FLOATING L, that is, an L which is not associated with any melody.
Downstepping is Unlinked as it is, for reasons to do with the specific configuration and con-
commonly thought straints of the language, this L still influences the tonal realization by down-
to be triggered by
stepping the following H — the symbol “!” is just a diacritic commonly used
a FLOATING L,
that is, by an L
to represent this situation (the IPA symbol is 1).
which is not asso- The presence of downstep in (26a) provides the key to the analysis of the
ciated to any general past. All we need to do is to assume that the morpheme of the gen-
melody eral past is a floating L prefix — not a suffix, which we saw the evidence in
(26) above looks equally consistent with:
(29) boo Disk
yevese vende
This floating L prefix downsteps the lexical H of yevese, the Ls in the addi-
tional syllables being eventually supplied by default. In vende there can be
Tonal Phonology 405
no downstep, since the stem tone that follows the prefix L is also L. The L
on the final syllable of the surface form vénde is provided by default, in line
with the general procedure.
The obvious question to ask at this point is why the initial L does not asso-
ciate to the first syllable ye, in line with the association convention in (17.1)
above, to create an LH contour. This association would of course pre-empt
the downstep, since downstep is only induced byafloating L.
In order to answer this question we need to acquaint ourselves with a device

known as “the cycle”, which constitutes the subject matter of the next chap-
ter. In the meantime, you should notice that this failure of the floating L to
associate provides further evidence against the Well-Formedness Condition
in (17.2) above.
vocal pitch is Heed not to aiaMe orgie

to express the speaker's intention in —
tion, to express doubt or Supe 5
1e-bearing unit). These

for stress, since the lec
406 Tonal Phonology
wards, except for a first syllab

Syllables that are still toneless r
Key Questions
1 What is meant by “intonation”? 7 What are the four uses of tone in

How is it manifested? language? How does intonation
2 Whatis a “tune” in speech patterns? differ from the other three uses?
3 What are the three basic compon- 8 What is “pitch accent"?
ents of the intonational melody? 9 List the three original principles of
4 In what way is intonation evidence autosegmental association. Which
for stress? of these has been found not to be
5 What is “interpolation”? well supported?
6 What are the three tone categories 10 What is Meeussen's law and how
making up intonational patterns? is it connected with the OCP?
What is a “nuclear tone”? 11 Describe downstep. What triggers it?
Further P T-ac¢e
-€ Beve
Margi Tones
Consider the following forms from the Chadic language Margi, spoken in
Nigeria:
sal 4
man te +ari (def.) 3 salari
kim ‘meat’ +ari — kutmaéri
Tonal Phonology 407
?imi ‘water’ +ari > ?imjari

ku ‘goat’ +ari — kwari
ti ‘mourning’ +dari =>: ijari
hu = ‘grave’ +ari — hwari
uri . ‘fire’ +ari —> wu?wari
(i) Explain how the tonal melody of the suffixed forms is obtained from
the simple ones.
(ii) Formalize the process using autosegmental machinery.
Ci-Ruri
In some languages morphemes may be lexically marked as bearing a tone.

In the Bantu language Ci-Ruri, verb stems are underlyingly either un-
marked or marked on the first vowel of the stem. In the data presented below,
the surface tone is located according to the following principles:
1. On the penultimate vowel if that syllable is lexically marked

2. Elsewhere on the syllable following the lexically marked syllable
Unmarked syllables receive default tone.

The following examples show the result of these principles:
a. Unmarked
inf. stem final vowel
oku gura [okt gur a] ‘to buy’
oku sakir a [Okt sakir a] ‘to help’
oku sorotor a [Okt sorotor a] ‘to pull out’
Marked
H
oku tem a [okt tém a] ‘to cut’

H
oku sumik a [okt sumik a] ‘to tie’

H
oku surumur a [okt strumur a] ‘to untie’

408 Tonal Phonology
(i) Show how tone is acquired by the forms in a.
Object prefixes are all marked:
b. H
oku gu gura [okt gu gur a] ‘to buy it’

HH
oku gu tema [okt gu tém a] ‘to cut it’

EIeEt
oku gu surumumurir a [okt gu surumurir a] ‘to untie it’
(ii) Bearing in mind the provisions of Meeussen’s law in (23), show how
the patterns in b. emerge.
ADVANCED THEORY
In the previous two parts we have laid the foundations of the edifice of phono-
logy, essentially providing the autosegmental organization (together with the
phonetic backup) in part I and the prosodic structure (stress and syllables)
in part II, where we completed the autosegmental picture with the tonal tier.
Now, in part III, we shall elaborate the model further, addressing three issues
in particular. The first issue concerns the effect of grammatical structure on
phonology. The specific question is whether grammatical structure (mor-
phological structure in particular) provides domains for the application of
phonological rules. One answer is that it does for some such rules (the “cyclic”
rules), while other rules simply apply in the maximal domain (the word).
It has also been argued in the literature that word construction can be sen-
sitive to the previous action of phonology, and therefore that phonological
rules need to be interleaved with word-formation rules: the model of “lex-
ical phonology” provides a theoretical framework for the expression of this
and other related aspects of word building. The model works less well in
phrasal domains, where phonological domains were soon found to be
at variance with syntactic structure. As a consequence, a theory of specific
phonological domains, related, but not identical, to syntactic domains, was
developed: it is commonly known as “prosodic phonology”, somewhat
confusingly. It is a moot point whether this theory must be extended to phono-
logical domains below the word to resolve some mismatches with morpho-
logical structure, the so-called “bracketing paradoxes”. A second issue we
deal with in part III concerns the structure of the lexical forms, in particu-
lar their composition in terms of distinctive features. The combination of
features is limited in two ways: intrinsically and extrinsically. Intrinsically,
the features themselves are organized into a “feature geometry”, by which
they make up a web of dependencies, expressive of their autosegmental
behaviour. Extrinsically, some combinations of features are disfavoured by
the “markedness” conventions, designed to capture the degree of natural-
ness of phonological systems. While the markedness conventions (like the
feature geometry) are universal, they interact with the systems of specific
languages through the device of “underspecification”, by which predictable
feature values are left out of lexical representations. Unspecified features are
of course predicted to be transparent to autosegmental spreading, and we shall
410 Advanced Theory
see that in some cases such transparency is only achievable if we assume that
certain (terminal) features are monovalent. The third issue we will address in
this part is whether the best way of formalizing phonological relationships
between forms is by means of rules and derivations (as has been typical of
generative phonology) or by means of surface constraints, as proposed in
Optimality Theory. Rules are crucially ordered, and OT constraints are like-
wise ranked. The innovation brought in by OT is the claim that constraints
are universal (the rankings, not the constraints themselves, are responsible
for the language-specific grammars) and that they are violable, satisfaction
of higher-ranked constraints taking precedence. Cyclic application and level
ordering of rules is replicated in OT through “correspondence” constraints
that relate surface forms to one another.
MODES OF APPLICATION
THE CYCLE
In this chapter you will learn a

m Some problems with the resol
solution. es
™ Two modes of phonologi
We ended the previous chapter with a question about the lack of association
of the first L to the first syllable in the Tiv general past form /yévésé. This
form is derived by prefixing the general past tonal morpheme L to the stem
yevese, lexically marked for H (the Ls in the last two syllables are both default).
The question arises because of the general autosegmental principle that tones
associate to vowels one-to-one from left to right, as formulated in the asso-
ciation convention in (17.1) of chapter 14. This convention leads us to expect
that the prefixal L- tone will associate to the initial syllable of the stem, ye,
and the lexical tone of the stem, H, with the second syllable, ve. However,
the output *yévésé is incorrect. Instead, the prefixal L- remains unassoci-
ated (= floating) and downsteps the H associated to the initial syllable ye:
!yévésé. The question is how this result is achieved in the face of the said
universal convention.
412 Modes of Application
We said at the time that the answer involves the cycle, or, more precisely,
the cyclic mode of application of (some) phonological rules. In this chapter,
we motivate this formal device and spell out its mechanics, with the appro-
priate exemplification. Following our usual strategy of building on the
familiar and intuitively obvious, we shall first motivate the cycle with a
set of the English data involving stress retraction which we discussed in
chapter 11. After providing the analysis of Tiv downstep, we will show that
the rules of English word stress also apply cyclically, as does a rule that
accounts for the shortening of English vowels in certain contexts. Some loose
ends of English stress will be tied up as we go along.
Staged Grid Construction
In chapter 11 we examined phrases like Stizanne Morris or Sarah Jane Bréwn,

and pointed out that the left-most of the two stresses in these constructions
originates in a syllable further to the right: Suzdnne (not *Stizanne), Sarah Jane
(not *Sdrah Jane), and so on.
7 Whai s the rea
You will remember that the reason for the mismatch is that in the longer
phrases the stress on the left undergoes leftward movement in order to avoid
clash — Suzdnne Morris — Stizanne Morris, etc.:
(1) e a Line 3 phrase head

+ - co 3 Line 2. word heads
te + * * + €
oT ine'l “foot heads
pea ie eae * * Line 0 baseline
Suzanne Morris — Suzanne Morris
Now, in the case of Sarah Jane Brown, we did not concatenate the three
We did not con- words that make up the construction simultaneously. Indeed, had we done
catenate the three
so, the stress clash would have been unresolvable:
words that make
up Sarah Jane
Brown simultane- (2) 2 Line 3 phrase head
ously . - . Line 2. word heads
‘g * = Line 1 foot heads
er na 3 Line 0 baseline
Sarah Jane Brown
Modes of Application 413
ayexactly why the clash in (2) isu resol able
Notice that the line-2 clashing asterisk on Jane in (2) has no landing site in
line 1 to go onto, and therefore it cannot move. However, if we first form the
metrical grid of the double first name Sarah Jane, and then add the grid of
the surname Brown to this grid, the desired result becomes attainable. This
is precisely what we did in chapter 11, where we simplified grid structure
slightly so as to not to complicate the presentation unnecessarily. We replic-
ate the procedure in (3), where we are assuming a line-3 asterisk in Brown
at the stage it is added to Sarah Jane, on the reasonable assumption that only
phrases can concatenate to phrases:
Pas me Line 3 phrase head

* *
Line 2 word heads
* *
Line 1 foot heads
+ + +
Line 0 baseline
Sarah Jane
b. Line 4 long phrase
head (SJB)
*
Line 3 short phrase
heads (SJ; B)
* * * *
Line 2 word heads
* cy * *
Line 1 foot heads
* + * + + + * +
Line 0 baseline
Sarah Jane Brown -—> Sarah Jane Brown
End Rule [R]
The clash between Jane and Brown in (3b) can now be resolved by leftward
movement, in the familiar way:
(4) Line 4 long phrase head (SJB)

+ fi x Line 3 short phrase heads (SJ; B)
* * * Line 2 word heads
* *
Line 1 foot heads
+ + +
Line 0 baseline
Sarah Jane Brown
As we explained at the time, the weaker of the two clashing asterisks moves
leftwards onto the nearest asterisk in the line immediately below, here the
line-2 asterisk in the column on Sa. Resolution of the remaining clashes is
blocked by the Continuous Column Constraint in (38) of chapter 11, which
we now repeat in (5):
(5) Continuous Column Constraint (= (38) in chapter 11):

Metrical columns must be continuous, without skipping lines
The relevance of the stress movement in Sarah Jane Brown to the subject
We build the of this chapter is that we have built the metrical grid in stages, first on the
metrical grid in shorter phrase Sarah Jane (naturally with prior grid construction on the indi-
stages, as if the
vidual words Sarah and Jane), and then on the longer construction Sarah Jane
word sequence
were organized in
Brown. It is as if the word sequence were organized in layers of Chinese boxes
layers of Chinese (equivalent to Russian dolls, an object perhaps more familiar to many), with
boxes the grid being constructed box by box outwards from the core, rather than
all at once across the board.
The structure in question is represented in (6):
(6) [[[Sarah] [Jane]] [Brown]]
The square brackets in (6) are the Chinese boxes of the analogy, and reflect
the morphosyntactic makeup of the construction. In particular, the bracket-
ing in (6) captures the following facts: (i) the basic elements in the longer
phrase are Sarah, Jane and Brown; (ii) Sarah and Jane make up a unit, or CON-
STITUENT;; (iii) this constituent makes up a larger constituent together with
Brown, namely, the longer phrase Sarah Jane Brown. Notice that the structure
we are proposing appropriately matches our intuition that the surname Brown
combines with the whole of the double first name Sarah Jane, rather than
just with Jane.
Once the multilayered structure in (6) is in place, the ground is clear for the
application of phonological rules (in the case just discussed, grid construc-
tion) in the manner we described. This mode of rule application is referred
to as CYCLIC, and the procedure that drives it as the CYCLE — words like This mode of rule
“cyclic” and “cycle” appropriately carry an implication of repetition. English SEpesHOes
phrasal
!
grid construction, therefore, 2 takes place cyclically, t first in the smallest Bi pee
CYCLIC, and the
inner constituents, and then in successive layers of constituent structure, until procedure that
the outer, largest constituent is reached: the thought of a Lilliputian Samson drives it as the
trapped in the innermost box and making his way out by pushing down the CYCLE
walls of the embedded boxes one by one after the appropriate phonological
processes have taken place in that box may be helpful in this connection.
Cyclic Tone Association
We shall now show that our Tiv !yévésé problem has a ready solution in the
context of rule cyclicity. You will recall that the problem is that, were tones
to associate to vowels one-to-one from left to right, as the Association
Convention (17.1) of chapter 14 says they must, the prefixal L- tone would
finish up associated to the initial syllable ye, and the tone H of the root would
finish up associated to the middle syllable ve:
gybiive A LH
> : : etc.
yevese yevese
nplete hederivation in(7).
You know, however, that this is not what happens. Instead, the L- remains
unassociated (= floating) and downsteps the H associated to the initial syl-
lable: !yévésé (you must recall that downstep involves relative pitch lower-
ing of high tones). The correct configuration is therefore that in (8):
(8) LH
etc.
yevese
In chapter 14 we attributed downstep to a floating L immediately before the

downstepped H. Therefore, it is essential that the L which is left-most in (8)
remain unassociated. The question is how this result can be achieved in the
face of the said universal convention that free tones associate to free tone-
bearing units one-to-one from left to right. A few languages associate tones
from right to left instead, but clearly this is not the case here either.
As we anticipated at the time, the problem can be resolved by carrying

out tone association cyclically. Let us first consider the makeup of !yévésé
prior to tone association:
(Oye |ThEF
yevese
In (9) there are two layers of structure, namely, the inner layer of the stem
and the outer layer of the tonal prefix: we are therefore defining word-
internal cyclic domains on a morphological basis, just as in the preceding
section we defined word-external cyclic domains on a syntactic basis. We
will introduce some important refinements on this in chapter 16, but for the
time being this approach will do.
Let us next examine the mechanics of tone association in the two domains
of /yévésé, on the assumption that Tiv tone association is cyclic. Tone asso-
Tiv tone associa- ciation will first take place in the stem yevese, the form’s innermost constituent
tion is cyclic ~ we are emboldening the brackets enclosing the currently active domain to
make the procedure clearer:
(10) |L] H
yevese
In (10) you can see H associating to ye, the left-most syllable of the stem,
in compliance with the Association Convention in (17.1) of chapter 14. This
completes the first cycle: you will recall that Tiv does not allow multiple
association of tones, and therefore the remaining stem syllables, ve and se,
must remain toneless for the time being. The second cycle is defined by the
outer pair of brackets in (10) (the outer box, in our Chinese box analogy),
housing the prefix L-. The structure input to this second cycle is as in (11):
ir Toy et
yevese
The free tone L looks around for a vowel to associate with. However, this turns
out to be a futile endeavour. First, L cannot associate to the first syllable ye,
because this syllable is already associated to a tone, and we know that Tiv
does not allow multiple tone association. Second, L cannot associate to the
syllables ve or se either, because any such association would infringe the
prohibition against the crossing of lines which governs the association of
autosegmental elements universally:
(12) No-Crossing of Lines ((17.3) in chapter 14):

Association lines may not cross
While floating
status obviously
prevents a direct
If L- cannot associate to any syllable in (11), it has to remain floating. While Phonetic realiza-
floating status obviously prevents a direct phonetic realization, it is still ihe
compatible with an indirect manifestation as downstep, and this 1sprecisely ire ER gram
what happens in /yévésé. What is relevant for us here is that this output is _festation as
achieved through the cyclic association of the tone, providing further sup- _downstep
port for the cycle in phonology.
Non-Cyclic Refooting
In section 1 above we saw that the construction of metrical grids in English

phrases takes place cyclically. In section 8 below we will see that some of the
stress rules of English word stress also apply cyclically. However, it is not the
case that all rules (whether for stress or otherwise) apply cyclically: some rules
apply non-cyclically, that is, only once, in the largest domain. It is also pos-
sible for one and the same rule to apply both cyclically and non-cyclically.
land ee,
The cyclic or non-cyclic status of any one rule is therefore idiosyncratic, and
The cyclic or needs to be specified in the rule itself.
non-cyclic status
In this and the following sections we lay the foundations for the analysis
of any one rule
is idiosyncratic of some of the rules of English word stress as cyclic. The first step involves
the analysis of the patterns of secondary stress in English words. Consider
the stress pattern of the forms in (13), which include some US place names:
(13) Winnipesdaukee
abracadabra
Kalamazéo
hullabaléo
Tatamagéuchi
The forms in (13) are, crucially, monomorphemic, and therefore they cannot
involve any word-internal cycle. The issue that interests us at present con-
cerns the presence of two stressless syllables between the initial secondary
stress and the primary stress. In particular, our present footing procedure
predicts *Winnipesdukee, etc., with two secondary stresses (we assume that the
final vowel ee is short underlyingly, and thus subject to extrametricality):
(14) a. x Line 1
* + + * * Hie ee Ne <*> + + + (Gs <*> Line 0
Winnipesaukee — Winnipesaukee — Winnipesaukee +

Extrametricality Accent Footing
b. ‘ Line 2
* * * + * @ * ) Line ‘ls
Git 7) Cres Cre DO) <*> t Line 0

Winnipe saukee — Winnipe saukee — Winnipe saukee
Footing End Stress [Right]
The procedure in (14) matches the procedure we adopted for hamamelidan-

themum in chapter 12, with footing sweeping leftwards across the word. The
crucial difference between hamamelidanthemum and Winnipesaukee concerns
the number of metrical syllables to the left of the main stress, even in the
former (ha.ma.me.li) but odd in the latter (Wi.nni.pe). Because the number of
such syllables in Winnipesaukee is odd, the standing procedure creates a degen-
erate foot in the left-most syllable, as we have shown in (14). This outcome
does not match the evidence, since the second syllable of Winnipesdukee is
in fact stressless. In chapter 12 we saw that some languages do not admit
monosyllabic feet, but this constraint will not help here, since it will incor-
rectly predict *Winnipesdukee.
We did anticipate in chapter 12 that, although English words have second-

ary stresses in the surface, the location of these stresses does not necessarily
match the output of right-to-left footing, and accordingly we introduced a line
conflation clause in the English stress algorithm, to get rid of all secondary
stresses. The obvious paradox between this suppression and the presence of
such stresses in the surface (albeit not necessarily in the same place) needs
to be resolved now.
The solution we will provide involves an additional non-cyclic application

of footing, that is, an additional once-only application in the domain of the The derivation
of English surface
whole word. Anticipating the discussion in section 8, we will assume that
secondary stresses
the procedure responsible for primary stress in English is cyclic, that is, applies requires a non-
constituent by constituent. cyclic application
We resume the cyclic derivation of Winnipesaukee in (15), which starts with of footing
the last step of (14) above. To simplify matters graphically, we are adopt-
ing the informal practice of maintaining the word-level asterisk in line 2 after
conflation — the graphic advantages of this informal practice will become
apparent shortly:
(15) MF Line 2
(eae =) Line 1
(cS) G =)(>) a * * * (e) <*> Line 0
Winnipesaukee — Winnipesaukee
Conflation
The application of conflation in (15) completes the cyclic procedures of Eng-

lish word stress, which we now list in (16) in the appropriate order, as a
reminder:
(16) English cyclic stress procedures:

Extrametricality
Accenting
Foot construction
End stress
Line conflation
At this point we are in the outer box of our Chinese box analogy, in which
all non-cyclic rules apply. The first non-cyclic rule we will propose is foot
construction: remember that it is possible for one and the same rule to apply
cyclically and non-cyclically. A literal non-cyclic application of footing to
Winnipesaukee would obviously reverse the effects of conflation:
(17) ; Line 2
Tie Brig? Line 1
Tee Bel ant Min@ACisstey bhined
Winnipesaukee — Winnipesaukee
Footing
As before, we have generated one stress too many on the second syllable.
Of course, this stress clashes with the stress on the first syllable, and there-
fore it will not be unreasonable for one of these two stresses to go. The
obvious candidate for deletion is, however, the degenerate foot on the first
syllable, inconveniently so.
We shall propose a solution in two steps, both of them to be justified

independently as we go along. First, we shall postulate that the direction
of non-cyclic footing (henceforth dubbed “refooting”) is left to right. This The direction
change in directionality will naturally produce a different structure, and thus of non-cyclic foot-
ing is left to right
provide a concrete motive for the terminological split:
(18) = * Line 2
* aa Line 1
+ + *+ (*) <*> ce =) (4) (®) <a Line 0
Winni pe saukee -— Winni pe saukee

Refooting
In (18) you can see that the degenerate foot is now on the third syllable,
rather than on the first. You will also notice that refooting has not destroyed
the word foot previously constructed on sau. This is in line with the Free
Element Condition in (14) of chapter 13, which we now repeat in (19):
(19) Free Element Condition:

Only metrically free elements may undergo metrical construction
The second step in the derivation of Winnipesdukee involves the introduc-

tion of a non-cyclic rule of destressing, which we formulate in (20):
(20) Destressing:
*»QO/ * Linel
eG, warie0
ul
Destressing deletes
a monomoraic
Examine this rule carefully and state its precise effects. degenerate foot
when this foot
immediately pre-
Destressing deletes a monomoraic degenerate foot, that is, a foot with only cedes the head of
a light syllable, when this foot immediately precedes the head of another another foot
foot. The effect of Destressing in Winnepesaukee is as in (21):
(21) . ; Line 2
G pon) (* 2) Line 1
ie) Oi > (8°34) * Ora eine
Winni pe saukee — Winnipesaukee
Destressing
The conjunction of left-to-right refooting and Destressing provides a simple

account for many of the patterns of secondary stress in English:
(22) a. América c. abracadabra

Dakota Kalamaz6o
b. Alabama d. Apalachicéla
Califérnia hamamélidanthemum
Work through some of these words with the procedure

._ _
€ proposing.
In the forms in b. and d., the binary feet created by left-to-right refooting
exhaust the domain. In the forms in a. and c., by contrast, a degenerate foot
is created next to the primary stress, and then disposed of by Destressing
(20).
ZY Final Stress Retraction
The procedure we have just adopted helps us to account for the surface
violation of the three-syllable window in forms like péregrinate, with the main
stress four syllables from the end of the word. Like many other English
Many English words, verbs ending in the suffix -ate reject main stress in the final syllable
words reject and undergo asterisk retraction by a mechanism akin to our familiar Rhythm
main stress in the
Rule.
final syllable, and
undergo asterisk
retraction
Think of a few -ate-final verbs and see i they
from the final syllable. Find a few other word
a
yy
This retraction obviously cannot be triggered by asterisk clash in non-clash

contexts, most particularly when the word is said in isolation. The pattern
with retracted stress is, however, clearly more in sympathy with the over-
all rhythmic structure of English. Heathrow, for instance, is currently under-
going stress shift: Héathrow Airport is motivated by clash, but I’m leaving from
Héathrow is not. Maureen has consolidated the shift in many accents: in Eng-
land at least, most people don’t say Mauréen, although quite a few still say
Heathrow.
Whatever the reason for the noted dislike of word-final stress, the line-2
asterisk also undergoes leftward movement in words like peregrinate. To keep
the presentation simple, at this point we shall derive the retracted pattern by
means of the Rhythm Rule, which we formalized in (21) in chapter 11. The
triggering environment of the rule must of course be extended accordingly,
but we postpone the introduction of this technical detail until the next
chapter:
(23) cs ‘5 ohine 2
. *) ‘¢ See Lane
re ae Je). bine
peregrinate — peregrinate
Rhythm Rule
This word-bound application of the Rhythm Rule obviously needs to follow

both refooting and Destressing, and consequently it must also be non-cyclic:
if the Rhythm Rule did not follow refooting and Destressing the landing site
would not be on pe at all.
Explain why the Rhyth

-
conflation.
The pattern of stress retraction in péregrinate is actually not the most

common one in -ate verbs, as we illustrate with the forms in (24), apparently
more numerous than their initially stressed counterparts. Note that we have
deliberately excluded obviously prefixed forms from the list, to take account
of the possibility of such prefixes being systematically excluded from the domain
of secondary stress, for reasons we will discuss in chapter 16:
(24) capitulate, hydrégenate, pontificate, pacificate, matriculate, ejaculate,

hyp6thecate, anticipate, negotiate, alléviate, manipulate, retaliate, . . .
In these forms retraction stops on the second syllable to the left of the final
syllable. Our present procedure is unable to yield this outcome:
(25) e s ° . Line 2
+ * + + Line 1
+ + >
( 4) * Ce) ¢ rs) * (*) Line 0

he eke ks (*) Cc =)(EN(C2)
capitulate > capitulate > capitulate —> capitulate

Refooting Destressing Rhythm Rule
The output *cdpitulate is obviously incorrect. In order to obtain the correct

output, capitulate, we shall declare the first grid element extrametrical in this
and similar forms:
(26) ? 3! a Line 2
a lle we Line 1
<*> * *(*) <*> (* *) (*) <5 (* *) (*) Line 0
ca pitulate — capitulate — ca pitu late

Refooting Rhythm Rule
Countenancing lexical extrametricality of this kind is obviously not ideal,

Countenancing but it is imposed on us by the facts: we have already warned that the English
left-edge extra- stress system is oozing with complexity.
metricality in some
forms is imposed
on us by the facts
ET
It may be possible to avoid total lexical idiosyncrasy in the assignment of

such left-edge extrametricality, though: monomorphemic forms tend to
eschew it, while polymorphemic forms appear to favour it.
Vowel Shortening
The deletion of monomoraic feet by Destressing as in (20) can be related

to the universal predilection for binarity in feet. This predilection allows an
elegant analysis of vowel shortening in the divine ~ divinity pairs we dis-
cussed in chapter 8 above. At the time, we examined the formal mechanics
of these vowel alternations in some detail, both in Modern and in Middle
English, but we did not give any reason for the shortening of the vowel in
the longer form.
We further illustrate the surface alternations in (27):
(27) a. divine b. divinity [ar] ~ [1]

derive derivative
vile vilify
extreme extremity [ir] ~ [e]
brief brevity
compete competitive
profane profanity lex] ~ [ze]
chaste chastity
inflame inflammatory
Note that in all cases the long vowel in the shorter form corresponds to a short
vowel in the longer form. Suppose now that we add a moraic dimension to
foot binarity, such that each foot is not allowed to contain more than two We add a moraic
dimension to foot
moras. The feet in both (alge)bra and di(v/it/ni)ty obviously violate this con-
binarity, such that
straint on foot structure, since they each contain three moras, which we have
each foot is not
emboldened to make their recognition easier. The violation in di(v/i:/ni)ty allowed to contain
is repaired by the rule of Vowel Shortening, to which we give a moraic more than two
formulation in (28) (2 = foot): moras
(28) Vowel Shortening:

[-cons] [-cons]
he St epee eee
2
The effect of the rule in (28) is the shortening of the vowel /i:/ in the stressed
syllable vi, which subsequently laxes to yield di(v[1]ni)ty, with a bimoraic
foot. In words like algebra, by contrast, the violation of foot bimoraicity is
maintained, as it must be if the input melodic material is to be preserved.
Indeed, preservation of melodic material is apparently a powerful force in
Preservation of
English phonology, as we will have the occasion to see again below.
melodic material
is a powerful
fes the moraic binari ty

offeet. force in English
phonology
The framework we are proposing provides an explanation for the puzzling

vocalic contrast in (29):
(29) a. tlou]nal b. t[p]nic
Whatisthecontrast betwee n tona

pune?
Morphologically, both these forms derive from tone. The root vowel shortens
in tonic, however, although not in tonal. This contrast falls in with certain
independently motivated facts of extrametricality. In particular, in chapter 12
we saw that suffixes trigger extrametricality in English adjectives. However,
-ic is clearly an exception to this generalization:
(30) a. pérsonal b. masonic

medicinal atomic
maximal alcohdlic
significant titanic
vigilant psychopathic
sonorous Homeé€ric
magnanimous telepathic
rigorous moronic
tress pattern
n
The forms in (30a) are all stressed in the same way as nouns, with the final
syllable extrametrical, but those of (30b), with -ic, are stressed in the same
way as verbs, with no extrametricality.
The extrametricality difference between tonic and tonal and the bimoraic-
ity of feet conspire to explain the contrast in the root vowel of the two forms:
zs | V |
to nic (wor? Tay ig wl
aaa) (3) <*>
es *
* *
In both words the foot only contains two moras. If tonic kept the long vowel
of tone, however, its foot would be trimoraic.
On the other hand, there is no need to shorten /2:/ in tonal, where it instead
undergoes vowel shift and diphthongization into [ou], exactly as in the base
tone. This rather spectacular result provides obvious support for our approach.
fa Strict Cyclicity
There are well-known exceptions to Vowel Shortening:
(32) a. obesity, hibernate, isolate, probity, scenic, basic, phobic, anaemic,

deictic, Vedic, rhotic, hyphenate, dangerous, vaporous, cyclic (in
some accents)
b. rotary, notary, irony, decency, primary, primacy, papacy, vagary,
vacancy, regency, potency, piracy, secrecy
c. nightingale, Abraham, stevedore, dynamo, protocol, dinosaur,
boomerang, apricot, ivory, rosary, Averell, protege, Omaha,
Clytemnestra, Polaroid, Oberon, Aubusson, bijouterie, dinothere,
boutonniere, coterie, Obadiah, abelmosk, Novocain, rhododendron,
Locofoco, ovolo, overture, rotifer, souvenir, troubador, edelweiss
(in General American, also: wisenheimer, trilobite, dynasty,
Bolshevik, bechamel, vitamin, ocarina, ocotillo, conquistador,
nomenclature)
The forms in sets a. and b. are derived, and those in c. underived.
Some of the forms in (32b) also violate accenting: compare, for instance, vdcancy
with the expected *vacdncy.
The forms in (32b) are systematically spelled with a final -y, apparently the
The suffix /-i/ orthographic encoding of a suffix /-i/ which, most idiosyncratically, does
spelled -y does
not project a stress bearer. If so, the whole string cancy in vacancy will cor-
not project a
stress bearer
respond to one extrametrical line-0 element, and the va foot will only be
bimoraic, with no need for shortening.
nde tand afull:
Having thus regularized set b., only sets a. and c. remain as genuine excep-
tions to vowel shortening. Set c. is considerably more sizeable than set a. The
property common to the forms in c. is that they are all underived. The fact
that the majority of the exceptions to vowel shortening involve forms with
no internal morphological structure clearly cannot be coincidental.
Indeed, it has been observed in case after case in language after language
that some rules systematically apply in newly created environments, while
failing to apply in pre-existing ones. A case in point is the rule of vowel short-
ening (28), which we are seeing applies in divinity, derived from divine, but
not in nightingale, which is underived. An important hypothesis standardly
made in this connection is that the rules that behave in this way are pre-
cisely the rules that change structure and apply in a cyclic mode.
Structure-changing cyclic rules, therefore, only apply in environments derived

Structure-changing in that cycle — “derived” in the sense of having been created by a morpho-
cyclic rules can logical operation or, in some cases, by a structure-changing phonological
only apply in process. This generalization is encapsulated in the so-called PRINCIPLE OF
environments STRICT CYCLICITY, as follows:
derived in that
cycle, as encap-
sulated in the (33) Principle of Strict Cyclicity:
PRINCIPLE OF Structure-changing cyclic rules only apply in environments derived in
STRICT CYCLICITY that cycle, where “derived” = resulting from a morphological process
or, in some cases, from a phonological change
The Principle of Strict Cyclicity is an important pillar in the edifice of phono-

logical theory. It will directly account for the failure of the forms in (32c) to
undergo shortening if the long vowel of these forms is already present in
the lexicon.
On the other hand, the deviant behaviour of the forms in (32a) needs to be
formalized by brute-force individual exception marking: we have already
pointed out more than once that the existence of idiosyncratic exceptions is
a fact of life in phonology.
Non-Cyclic Accenting
In section 3 above we used left-to-right refooting to account for the initial

secondary stress in words like Winnipesdukee, abracaddbra or Kalamazoo, with
two stressless syllables before the main stress.
Consider now the forms in (34):
(34) Monongahéla
Valénciénnes
Atascadéro
Manantenina
In these forms the secondary stress falls not on the initial syllable, but on
the second syllable, the initial syllable remaining stressless.
The obvious difference between these forms and those in (13) above concerns
the weight of the second syllable, which is heavy in (34) (Monongahela) but
light in (13) (Winnipesaukee). It will not be unreasonable to attribute the differ-

ence in secondary stress placement between the two sets to this difference
in syllable weight.
The heavy syllables in second position in (34) will indeed receive stress
as a consequence of the accent clause of the English stress procedure. How-
ever, this accent clause is at the moment cyclic, and therefore it will be
followed by line conflation, which will dispose of the corresponding foot.
by cyclic
The obvious way forward involves adopting the same strategy as for
foot construction, and allow accenting to apply in a non-cyclic mode also.
Following from this, the non-cyclic derivation of Monongahela will be as
in (35):
(35) “4 ‘ Line 2
* * (* * *) Line 1
*+ (c + (is (*) ( =) Gye Line 0

+ + *+ (3) *
Monongahela -» Monongahela — Monongahela —

Accenting Refooting
- Line 2
G 4) Line 1
BC aCe Line 0
— Mononga hela
Destressing
Crucially, Destressing now affects the degenerate foot in the initial syllable.
In Winnepesaukee, by contrast, the first syllable is the non-clashing head of
a binary foot, as we saw.
At this point you may feel a little suspicious that our decision to reapply
There is independ-
ent evidence for
accenting non-cyclically, while convenient for the set we are examining, may
non-cyclic be ad hoc and have adverse consequences elsewhere. There is, however, inde-
accenting pendent evidence for non-cyclic accenting. Consider in particular the also
a monomorphemic forms in (36):
(36) Halicarndssus
osténtation
incarnation
incantation
These words also have a secondary stress, and thus no vowel reduction, on
vowels that need an input accent if they are to end up as foot heads.
Halicarnassus will indeed receive an accent on car in the cyclic phase.

However, as usual, this accent will be disposed of by conflation at the end
of this phase:
(37) Line 2
* + * + * + Line 1
Pag aa [Fa CC) Cre Line 0

Halicarnassus — Halicarnassus — Halicarnassus —>
Accent Footing
x * Line 2
en tase) Line 1
2) Cy (sy er> ae (an Line 0
— Halicarnassus — Halicarnassus
End Stress [R] Conflation
As a consequence of line conflation, the syllable car will come out of the deriva-
tion with no prominence at all unless we reapply accenting in the non-cyclic
phase:
(38) * # * Line 2
ane (Pag Caan) Line 1
ake Foplte (G).<ts (* *)(*) (*)<*> Line 0
Halicarnassus — Halicarnassus — Hali carnassus
Accenting Refooting
The bimoraicity of car prevents the operation of Destressing, and the vowel
will surface with a secondary stress.
The data we have considered provide reasonable evidence for the applica-
tion of accenting in both a cyclic and a non-cyclic mode. Cyclically, accent-
Accenting applies ing applies across the board without failure.
in both a cyclic
and a non-cyclic
mode in English
Non-cyclically, however, accenting is subject to a not inconsiderable num-

ber of exceptions: sérendipity, épistémic, Conestoga, Transylvania, Pennsylvania,
dnecdote, and so on. By now you are well aware that many phonological rules
do have exceptions.
We finish the section with an update of the English word stress procedures:
(39) English word stress algorithm (update):

Cyclic:
Extrametricality
Accenting
Foot construction
End Stress
Line conflation
Non-cyclic:
Accenting
Refooting
Destressing
Rhythm Rule
EA Word-Internal Stress Cycle

The rules of
English word pri-
mary stress apply We now have all the ingredients we need in order to show that the rules of
in a cyclic mode English word primary stress apply in a cyclic mode.
(40) a. péregrination b. originality

classification compatibility
oxygenation precipitation
tergiversation equivocation
mélioration capitulation
vitriolation syllabication
The stress patterns of the two columns of (40) parallel those in (13) and (34)
above, respectively: (40a) and (13) have initial secondary stress, while in (40b)
and (34) secondary stress falls in the second syllable. We attributed the con-
trast between (13) (Winnepesaukee) and (34) (Monongahela) to the presence
of a heavy syllable in second position in (34), but not in (13). In (40), how-
ever, the second syllable is light in both columns.
The contrast in secondary stress between the two columns in (40) could of
course have been accidental.
Close observation reveals that the bases of the forms in (40b) have their
primary stress on the second syllable: original, and so on. This is not the case
for the forms in (40a). Therefore, it will be reasonable to attribute the dif- The formal
ference between the two columns to the word-internal stress cycle. mechanics of the
The formal mechanics of the word-internal stress cycle differs in certain word-internal
stress cycle differs
respects from the more general mechanics of the cycle, which we presented
in certain respects
above in connection. with phrasal grid construction and tone association. In from the more
particular, no word-internal stress metrical structure is carried over to sub- general mechanics
sequent cycles, as we demonstrate in (41) with original, obviously derived of the cycle
from origin (we simplify the derivations innocuously):
(41) z
* (*) *
+ tty ¢ *) (é; <> (& <*>
origin — ori gin — ori gin — ori gin

Footing EndStress [R] Conflation
* + + * + +
(Ga 5B) * cht ( *) Qs (* 2) @=as + AO) es
original — origin al — origin al — origin al

Footing End Stress [R] Conflation
The preservation of the foot on (ori) in the second cycle incorrectly leads to
a degenerate foot on (gi): the resulting stress contour will be “original, as in
(41), or, even worse, *6riginal if footing were to maintain stress differentials.
In order to obtain the correct primary stress of original, we obviously need

to start the second cycle anew:
(42) 2
* (3) *
Ot Oty (* *)<*> (* *)<*> (24s.
origin — ori gin — ori gin — ori gin

* *
+ er oes
% Ok cts, (*) (# +) ats, (*) (* *) <*> Pi *) <*>
original — o riginal —> origin al — origin al

Non-cyclic binary refooting will next create a degenerate foot on the o

of original, irrelevantly so because it will be disposed of by Destressing. In
originality, however, left-to-right refooting would derive *origindlity on the
output of cyclic conflation:
(43) . ‘ Line 2
* eb * * Line 1
+ wate =) <> Cc =) (*) (e =) <*> (e Bye ( 2) <*> Line 0
originali ty — originali ty — originali ty

Refooting Destressing
The correct result originality could of course be generated by adopting left

extrametricality, along the lines of capitulate above. However, all forms in the
mould of (40b) would require this setting, and these are precisely the forms
where main stress falls in the same position in the source words: this cor-
respondence would go uncaptured if we simply stipulated left extrametricality.
Obviously, we need to go half-way between total preservation of prior

metrical structure and total loss. In particular, we need to keep a trace of
the primary stress of previous cycles. We will achieve this aim through a
twofold strategy. First, we shall construct each cycle on a different plane:
In the case of
(44) * Line 2 word-internal
Tene lol stress we con-
As Ea struct each cycle
ns Line 0 on a different
peer Oia x plane and inter-
origi nali ty pret all the sylla-
ee bles that support
HOH (FH CHS Line 0 line-2 aati a
Line 1 originality plane oe ood
* Line 2
Second, we will interpret all the syllables that support line-2 asterisks on any
such planes as heavy, and will therefore subject them to non-cyclic accenting:
(45) : Line 2
Line 1 original plane (only cyclic)
cai Galea Line 0
——_———*
origi nali ty
UE EEE
* (* * (* Hots Line 0
originality plane (cyclic
* * Line 1 and, here, non-cyclic)
* Line 2
The remainder of the derivation on the non-cyclic plane of (45) is as follows:
(46) Refooting Destressing

originali ty > originali ty — originali ty
* G * G i) <*> () & ay & 2) <*> + ¢ #) (* :) <*> Line O
* * + * *+ * Line il:
y z s Line 2
You can see that we have now attained the same stress pattern in originality
as in Monongahela, even though the relevant syllable of originality is not heavy.
In originality the cyclic application of the stress rules is of course a neces-
sary prerequisite.
EE] The Structure of the Word-Final Syllable
In section 5 above we accounted for the length alternations in pairs like divine
~ divinity by means of the rule of Vowel Shortening in (28), which we repeat
here as (47):
(47) Vowel Shortening

[—cons] [—-cons]
Ie Wire ee ee yeaaa
(* ey
We will now show that this rule also accounts for the vowel length contrast
in (48), where both alternants have the same number of syllables.
(48) a. deep b. depth

five fifth
dream dreamt
leave left
feel felt
sleep slept
heal health
Syllable structure is responsible for the alternation in (48). In chapter 10

we suggested that English rimes include a maximum of two timing units,
now interpreted as moras. However, in word-final position we observed the
possibility of an extra consonant, as in d[i:p], d[ep6], and the remainder of
the forms in (48).
Also in chapter 10 we saw that word-final biconsonantal clusters need
to comply with Sonority Sequencing (final coronal obstruents excepted, a
matter we discuss in chapter 16). Given that the domain of Sonority Sequenc-
ing is obviously the syllable, we decided to affiliate the extra consonant dir-
ectly to the syllable node. The moraic formalization of this configuration is
as follows:
(49) 9
d
Vi P
The incorporation of the structure in (49) into the syllabic repertoire of English
raises the question of why it is not permitted word-internally.
lynot permitted wordl-internally? Aren't

The extra conso-
nant affiliated to o
is only licensed at
One crucial observation we made at the time is that the extra consonant affili- the right edge of
ated to o is only licensed at the right edge of the word. In particular, two _ the word
or more such consonants cannot be licensed:
(50) *O
The licensing of exactly one extra element in the periphery is reminiscent

of extrametricality, a device we made ample use of in our discussion of stress:
by the Peripherality Condition (6) of chapter 12, “only peripheral elements
can be extrametrical”. Were the one extra consonant at the end of words
to be analysed as extrametrical, its peripheral nature would follow auto-
matically, and indeed an analysis of this kind can be found in the literature.
However, the analysis has clear disadvantages. At the conceptual level, it
is not self-evident why extrametricality, a device responsible for skipping
metrical elements in the metrical counting, should also be applicable to syl-
labic parsing, which is not metrical. Moreover, in the empirical arena, extra-
metricality of the extra consonant ought to make it impervious to Sonority
Sequencing, but we know that this is not the case in English.
What actually seems to be happening is that the edges of words have
special licensing powers, which materialize as metrical skipping © “extra-
metricality”) in the area of stress, as underparsing in the area of syllabifica-
tion (= “extrasyllabicity”), and so on. By way of compromise, we shall refer
to the special properties associated with word edges by the generic label
We shall refer “EXTRAPROSODICITY”, and mark the elements subject to such special licens-
to the special ing with angled brackets, as we did earlier with extrametricality:
licensing powers
of word edges
by the generic (51) Oo
label “EXTRA-
PROSODICITY”
Ww
d
Vi 
We wish to make it quite clear, however, that we are not proposing a

uniform implementation of extraprosodicity across the various structural
domains. In particular, we are not suggesting that the extraprosodic con-
sonant should be ignored by Sonority Sequencing, at least in English.
Once this simple framework is in place, the account of the vowel length
alternation in (48) becomes straightforward. In particular, the forms in (48b)
have a (suffixal) extra obstruent. If the vowel did not shorten (dli:p]th, etc.),
these forms would, illegitimately, contain either three-mora rimes or a non-
peripheral extraprosodic consonant, as we illustrate in (52):
(62) ai nO b. iG @ 26
In a., the rime contains three moras. In b., the extraprosodic element is not
peripheral. In c., extraprosodicity extends over two segments, the left-most
of which is not peripheral. In the face of these licensing failures, Vowel
Shortening (47) is called upon as a repair strategy. Shortening the vowel allows
us to have our cake and eat it — all the segmental material of the input can Shortening the
now be parsed, at the small cost of dropping one of the moras originally — V°Wé! allows all
: : the segmental
associated with the vowel: cnt ape
input to be
(53) o parsed, at the
small cost of
dropping one
a of the moras
d Sii-P <0>
In the structure in (53) the p and its mora are licensed, despite the presence
of the suffix -th in depth. Crucially, the addition of this suffix creates the envir-
onment for the application of Vowel Shortening (47), thus making the appli-
cation of the rule compatible with the Principle of Strict Cyclicity.
The structure in (53) presents us with a new challenge.
In particular, the sequence [p86] does not comply with Sonority Sequencing.
We pointed out in chapter 9 that the sonority profile of the syllable must first
rise, and then fall. In the cluster [p0], however, there is obviously no fall
in sonority — indeed, there will be a rise if fricatives are assigned a higher
sonority ranking than stops, as we briefly explored in (41) in chapter 10
above. As a consequence of this infringement of Sonority Sequencing, the
incorporation of [6] (NB not [p]) into prosodic structure in depth becomes
problematic.
Importantly, /6/ is a coronal obstruent, and we have already said that

coronal obstruents are allowed on the right edge of English words, with no
obvious restrictions. We shall give a formal analysis of this fact in the next
chapter, in the context of which the problem of the licensing of /@/ in depth
will be resolved.
In this chapter we looked agai

cussed in chapter 11. We show
resolved by simple retraction, the s
bringing together the individual we
unworkable grid. This problem is
first in smaller constituents, and ther
procedure affords a solution to the

for Fit int
fying thefloating low tonecee
vowels ins :pai

erivatio
Key Questions
1. What is meant by a “cyclic” mode 6 How can Strict Cyclicity be called upon
of rule application? to show that Vowel Shortening is a
2 How does cyclic rule application cyclic rule?
explain stress retraction in English 7 List the cyclic and non-cyclic clauses
phrases and tone association in Tiv? of the English word stress algorithm.
3 Can the cyclic or non-cyclic status 8 How does word-internal stress assign-
of rules be predicted from the way ment work?
the rules are formulated? Name a 9 How does Extraprosodicity, governed
non-cyclic rule. by the Peripherality Condition, prevent
4 How do we know that the word- the application of Vowel Shortening
bound application of the Rhythm in words such as deep (cf. depth)?
Rule is non-cyclic?
5 What is the Principle of Strict
Cyclicity?
Further Pr arerri ce
Finnish
Finnish has a rule of “assibilation” which converts [t] into [s] before a suf-
fixal [i], as we show in a.:
a. tilat-a ‘to order’ tilas-i ‘ordered’

halut-a ‘to want’ halus-i ‘wanted’
compared with:
tila ‘room’ eiti ‘mother’
(i) | What is the status of the assibilation rule with regard to cyclicity?
The language also has arule raising [e] to [i] in word-final position:
b. joki ‘river’ joke-na ‘river’ (essive sg.)

ziti ‘mother’ eiti-nze ‘mother’ (essive sg.)
kuusi ‘fir’ kuuse-na ‘fir’ (essive sg.)
koti ‘home’ koti-na ‘home’ (essive sg.)
Now consider the following forms:
c. vesi ‘water’ vete-nze ‘water’ (essive sg.)

keesi ‘hand’ kete-nze ‘hand’ (essive sg.)
(ii) |Why does assibilation not apply in b.?

(iii) Assuming both assibilation and e-raising to be cyclic, trace the deriva-
tion of the forms in a. and in c.
Slovak
Slovak contrasts long and short vowels at the lexical and phonetic levels,
but in addition the language has morphologically conditioned vowel
lengthening:
a. Nom.sg. Gen.pl.
blatt = bla:rt = ‘mud’
pivto piv ‘beer’
putto put ‘chain’
b. tfelto tfiel ‘forehead’
kol+9 kuol ‘wheel’
mest) mies ‘meat’
The evidence from b. suggests that there is a diphthongization rule in

Slovak affecting the long versions of the vowels /e/, /3/ and /z/:
The process of diphthongization does not apply to certain etymological loan-

words which otherwise behave exactly as native words:
c. legernda ‘legend’
afe:ra ‘affair’
meto:ida ‘method’
betorn ‘concrete’
This is not a matter of a native/loanword distinction, since other loanwords

do undergo diphthongization.
Can the assumption of cyclic status for diphthongization explain its failure
to apply to the examples in c.?
DOMAINS OF APPLICATION
LEXICAL AND PROSODIC PHONOLOGY
In this chapter you will lea

m Suffixes ignored by the a
Two classes of suffix, cre
In the previous chapter we showed that some phonological rules apply cyclic-
ally, that is, in every domain defined by morphological structure, and other
rules apply non-cyclically, that is, just once in the domain of the whole word.
We pointed out that the selection of mode of application (namely, cyclic vs.
non-cyclic) needs to be stipulated individually for each rule. For instance,
of the rules assigning stress to English words, Destressing and the Rhythm
Rule are specified as non-cyclic, and the remainder as cyclic, with the accent-
ing and footing subprocedures as non-cyclic also. In this chapter we present
and discuss some cases where morphological domains are ignored by cyclic
rules, and develop the model further to meet this challenge.
Three-Mora Feet?
In the previous chapter we aimed at limiting the English foot, and a fortiori
the rime, to two moras. As we showed then, this limitation accounts for the
shortening. of long nuclei, with the attendant vowel shift-related effects, in
pairs like dli:]p ~ dle]pth, div[ai]ne ~ div[i]nity, and so on.
Domains of Application 445
The data in (1) and (2) plainly contradict the state of affairs we have been
describing:
(1) deepness divineness _ lightly

fivehood chasteness _ slothful
dreamful vileness vagueness
leaveless extremeness_ delightful
creedless briefness painting
sleepless profaneness attainment
flightless soundness _ divinely
(2) breezily wearisome __leaderless

beautiful likelihood __ tidiness
flavoursome craziness warily
(1) deliberately includes some of the base words we used in chapter 15

to illustrate vowel shortening before a consonant cluster. In contrast to
what we saw then, the relevant vowels remain long in (1), notwithstand-
ing the apparent violation of the two-mora restriction on rimes they incur: There are many
words where
dli:p].ness, div[ain].ness, etc. Similarly, in the forms in (2) the relevant
vowels remain
vowels do not undergo shortening, even though the foot ostensibly contains long notwith-
more than the canonical two moras: (br[i:]z[i])ly, (I[at]kel[i])hood, and so on. standing the
apparent violation
DER
of the two-mora
restriction on
) checking rimes they incur
One immediate reaction may be to attribute the anomalous behaviour

of the forms in (1) to the word-internal position of the offending sequence,
in contrast to its word-final position in depth and similar forms from the
previous chapter. However, the licensing of the longer sequence word-
internally, but not word-finally, would run against the observation we
made in chapters 10 and 15 that it is precisely in word-final position (not
word-internally) that extra elements turn up, in English and in many other
languages. Indeed, nucleus shortening does take place word-internally in other
forms under the expected conditions:
(3) a. convene b. convention [ix] vs. [e]
contravene contravention
intervene intervention
conceive conception
perceive perception
receive reception
redeem redemption
describe description [ai] vs. [1]

inscribe inscription
subscribe subscription
transcribe transcription
detain detention [er] vs. [e] (NB no vowel shift)

retain retention
The forms in (3a) have long vowels or diphthongs before the word-final con-
sonant. In the forms in b., however, before a consonant-initial suffix, the
corresponding vowels are short. Why, then, do long vowels persist under
similar circumstances in the forms in (1) and (2)?
Violations of the Three-Syllable Window

Before we address the problem posed by the long vowels in the forms in
(1) and (2), we shall examine another puzzling, and, as it will turn out, related
phenomenon. Observe the distribution of stress in the forms in (4):
(4) a. impétuous b. impetudsity c. impétuousness

anonymous anonymity anonymousness
finicky finickiness
pernickety pernicketiness
uppity uppitiness
In the forms in (4a), which are embedded in those in (4b) and (4c), primary
stress has been assigned by means of last-syllable extrametricality and right-
to-left left-headed binary footing, the usual English procedure:
(5) . stg Line 1

* Hoy (* * omy (*) (* *No*> Line 0
impetu ous — impetuous — impetu ous —>

Accent Footing
. Line 2
Guay) a Line 1
(ees) e> SHES) Spine O
— impetu ous — impetu ous
You will notice that stress ends up three syllables (NB not, say, four) from
the right edge of the word, as a fallout of the basic English stress algorithm
as it stands at the moment. The restriction of the stress locus to one of the
three peripheral syllables is in fact quite common cross-linguistically, and
we have been referring to this state of affairs as “the three-syllable window”.
In chapter 15 we appealed to asterisk retraction to account for some surface
violations of the three-syllable window in English, as in péregrinate, for instance,
where metrical structure does comply with the window prior to retraction.
The suffixation of -ity in the forms in (4b) gives rise to an additional cycle
(on a new plane, as we know from chapter 15):
(6) Second cycle

7 PLAS ar eee Line 1
+ + + * tty (ce + +% tats (4) CG = NaS Line 0
impetuosi ty — impetuosity -— impetu osi ty

Accent Footing
a Line 2
Rey oe Oe . bine?
a) <*> ‘ene Hck palainen))
— impetu osi ty — impetu osi ty
The main stress lies further to the right in impetudsity than in impétuous, and
thus complies with the three-syllable window — *impétuosity would not
comply with it.
“Check that you see how the
ferently in the two version
<<
ARDLS
The forms in (4c) are also derived from their counterparts in (4a) by
suffixation, and we would expect the stress contours in (7):
(7) a. *finickiness b. *impetudusness

*pernickétiness *anonymousness
*uppitiness
In *impetudusness and *anonyméusness, in b., stress is attracted to the heavy

penult, while in *finickiness, *pernickétiness and *uppitiness, in a., the three-
syllable window is utilised to its full extent. None of these forms is, how-
ever, correct. Moreover, their correct counterparts in (4c) (impétuousness,
anénymousness, etc.) are all incompatible with the three-syllable window in
a manner which does not appear to be reducible to a retraction analysis, in
the way that péregrinate is.
Explain why retra

.
the stress pattern of
If you look carefully at the data, you will see that impétuousness and the
other forms in (4c), with stress outside the three-syllable window, retain
primary stress on the same syllable as their bases in (4a): impétuous, etc. What
seems to be happening, then, is that the basic English stress algorithm
The basic English simply does not reapply after the suffixation of -ness. This failure of the
stress algorithm
algorithm to reapply accounts both for the stresslessness of the heavy
does not reapply
after the suffixa- penults that precede -ness and for the violations of the three-syllable
tion of -ness, window.
and some other
suffixes
neck that this is the case in
The obvious question now is: why does the basic English stress algorithm
not reapply after the suffixation of -ness?
Cyclic and Non-Cyclic Affixes
Up until now, we have seen all the cyclic rules applying morphological layer
by morphological layer, from smallest to largest. By contrast, the non-cyclic
rules take no account of internal morphological structure and apply only
once, to the fully formed word right at the end of the procedure. What
we are now seeing is that this model oversimplifies reality, since cyclic
rules (for instance, the English primary stress algorithm) can systematically Cyclic rules sys-
tematically fail to
fail to apply in some affixal domains (for instance, in the domain defined
apply in domains
by -ness). This suggests that compatibility or non-compatibility between specified as
rules and domains with regard to cyclicity works in both directions: rules non-cyclic
have to be specified as to whether they apply cyclically or non-cyclically,
and domains (as defined by affixation) have to be specified as to whether
or not they trigger cyclic rules.
All rules and
all domains (as
defined by affixes)
are marked as to
whether they are
by me andsimilarforms.in
n (e) cyclic or non-cyclic
The list of suffixes incompatible with cyclic rules in our present data includes
-ness, -ly, -ful, -some, -hood, -less, -ment. We shall refer to these and similar
affixes as NON-CYCLIC AFFIXES. You must of course realize that in the
context of affixes the expression “non-cyclic’” means ‘only compatible with
non-cyclic rules’— rules can “cycle” (= apply in cycles), but affixes obviously
cannot: indeed, affixes cannot “apply” at all.
We shall represent the cyclic or non-cyclic status of any given rule or

any given affix by means of the subscripts “c” (for “cyclic’) and “nc” (for
“non-cyclic”), as we illustrate in (8) for the cases related to stress we have
been examining. You should bear in mind that cyclic status is, of course,
confined to the domain defined by the cyclic affix — all forms eventually
undergo the non-cyclic rules at word level:
(8) a. Rules: Extrametricality,

Accent, nc
Footing, nc
450 Domains of Application
End stress,
Conflation,
Destressing,,
Rhythm rule,. —
b. Domains: Stem, (by general convention)
-ity,, -al,, -ation,, -ify,, etc.
“NESS ne, LYnc, ~fUlnc, -SOME,,, ~hood,,, -less,,, -ment,,, etc.
With this background, let us look at the derivation of the forms impetus
(cyclic, as a base), impetuous (with cyclic -(u)ous), impetuosity (with cyclic
-ity) and impetuousness (with non-cyclic -ness):
(9) /impetus/ /impetus/
|TEP oe ee ea Primary Stress. tr

ee ye |
es impetus
Sas Sane 02 isan er SUE XATONY >< Pee

essteko >
| -U-0US, |
a hen impet-u-ous,
|qeceRee Cg ae eT Primoty stress Wa gee 3
impétuous impétuous
|Seti A Mae ah Suffixation= of) je.cass yx |

| “ity, -nessi |
impetuos-ity, impétuous-ness,,.
|!
Son tg ware ae ae Primary stress c
impetudsity impétuousness
The crucial difference in the derivational history of the forms impetudsity and
impétuousness is that only the former undergoes the primary stress proced-
ure in the domain defined by its outermost suffix. The contrast follows from
the fact that the rules of the English primary stress procedure are cyclic, and
the domain defined by -ity is also cyclic, while the domain defined by -ness
is non-cyclic: the mismatch between the cyclicity of the English primary stress
procedure and the non-cyclicity of the domain defined by -ness prevents the
application of this procedure in this domain.
The Interaction between Morphology and Phonology
We have just seen that our previous practice of strictly identifying cyclic
domains with morphological layers was an oversimplification, since some
morphological layers are non-cyclic, that is, do not define a domain of applica-
tion for cyclic rules.
In (9) we presented the derivation of impetuosity and impetuousness as
cumulative, in that we first entered impetus, then formed impetuous, and then
impetuosity and impetuousness (in the first and the second column, respect-
ively). The obvious advantage of this style of presentation is that the reader
can follow step by step how the word is being built. In itself, however, the
procedure is also compatible with a preformed word, inside which the cyclic
rules simply work their way through the appropriate cyclic domains. Indeed,
this is the practice we adopted for the cyclic derivation of !yévésé when we
presented the cycle in the previous chapter. There is an obvious, and crucial,
difference between the two approaches. If we build the word in stages, the
morphology and the phonology can apply in tandem, and each step in the
construction of the word be immediately followed by the application of
the relevant phonological rules.
By contrast, if the word is fully formed first, the phonological rules cannot
interact with word formation.
How can we possibly adjudicate between the two alternatives just

sketched? Consider the data in (10) and (11):
(10) a. arrive appraise b. arrival appraisal

construe rehéarse constrial rehéarsal
revive procure revival proctral
approve rent approval __ réntal
commit try committal trial
betray § withdraw betrayal withdrawal
(11) a. deliver b. *deliveral cf. delivery

abandon *abandonal abandonment
édit *édital edition
endéavour *endéavoural endéavour
devélop *devélopal devélopment
consider *consideral consideration
deposit *deposital deposit
promise *promisal promise
In (10b) we have listed a number of nouns formed from the verbs in (10a)
by the addition of the suffix -al. Such a derivational relationship is imposs-
ible for the forms in (11).
The contrast between (10) and (11) with regard to the acceptance of the suffix
-al can be attributed to one simple fact: all the base forms in (10) have final
stress, but none of those in (11) does. If final stress is a precondition on
The morpho-
-al suffixation (necessary but not sufficient, though: depart > *departal, etc.),
logy and the
phonology are
then arrive needs to have stress assigned before -al suffixation takes place.
interleaved, not This means that the morphology and the phonology are interleaved, not
segregated segregated. In particular, if they were segregated, with all the derivational
aa processes preceding the phonology, phonological information would not
be available to affixation. As a result, the derivations in both (10) and (11)

would be blocked if -al suffixation were indeed stress-sensitive, since under
the circumstances -al would simply not find the contextual stress it needs.
Alternatively, if -al suffixation were not stress-sensitive, both groups of
words would undergo -al suffixation, producing arrival, etc., and *deliveral,
etc., respectively. One or other outcome is of course incorrect for one of
the sets.
The derivation in (12) illustrates the interleaving of the phonology and

the morphology we are talking about:
(12) /arrive/ /deliver/
or eae nee ey PRAT SeteeS. ote ier oe ae >
arrive
|
deliver
1 Ts sanhi Pllon ae
| -al,
arrival -
Sea ae ce Primary stress,
|
arrival
Derivational affixes of the -al type are of course heavily lexicalized, in that
they cannot be added freely to stems to form new words.
Instead, affixes of this type belong in constructions that are, so to speak, fos-
silized, and thus most likely learnt as a block: notice that no English speaker
will ever dream of saying *departal or *arrivure, even though the exchange of
the two suffixes does not seem to be ruled out by any principle. This state of
affairs is to be expected if words like arrival and departure are indeed learnt
as one unit, rather than being composed by actively combining arrive and -al,
and depart and -ure, respectively. This being so, the argument from -al we
have just advanced for interleaving the phonology with the morphology could
be objected to on the grounds that the distribution of -al has been fixed his-
torically, and therefore falls outside the remit of the synchronic grammar.
Similar cases of interleaving involving clearly productive affixes, how-

ever, do exist in many languages. One such case in English involves the
colloquial (and unrefined) “infix” -bloody-, as in the expression I have to go to
uniBLOODYversity! — an INFIX is an affix that neither precedes nor follows
the stem, in the way that prefixes or suffixes, respectively, do: rather, it is inserted
in the middle of the stem. The point about this extremely productive infix
is that it cannot be inserted just anywhere, as the set in (13) clearly attests:
(13) *uBLOODYniversity
uniBLOODYversity
*univerBLOODYsity
*universiBLOODYty
You will notice that there are no fewer than four possible insertion sites, but
only the second one listed (between uni and versity) is viable. At first sight,
this restriction may appear totally mystifying.
Our knowledge of metrical structure allows us to find the answer: the

legitimate insertion site of -bloody- lies precisely between the two metrical
feet of the word: (uni)l(versi)<ty>. Indeed, it immediately precedes the
strongest foot (versi). This is in fact the preferred location, as we illustrate
in (14) with the longer form Apalachicéla, a place name in the USA:
(14) *ABLOODYpalachicola
?ApaBLOODYilachicola
*ApalaBLOODYchicola
ApalachiBLOODYcola
*ApalachicoBLOODYla
Apalachicola contains three feet: (Apa)(lachi)(co)<la>. What the set in (14) shows
is that the best position for the infix -bloody- is immediately before the strongest The best position
foot, the rightmost one: (co). for the infix
-bloody- is imme-
diately before the
strongest foot
The relevance of all this to our present discussion is that -bloody- infixation,
a morphological process, requires information about metrical structure, a
phonological process. This means that metrical structure needs to be assigned
to the base (university and Apalachicola in our examples) before infixation takes
place. Consequently, a phonological operation (namely, footing) needs to
precede a morphological operation (namely, -bloody- infixation), exactly as
predicted by the interleaving model of morphology—phonology interaction
we are presenting.
The Scope of Peripherality
We have come some way in our understanding of the relationship between

morphology and. phonology, but we still have not accounted for such con-
trasts as depth vs. deepness, with which we opened the chapter. We shall now
show that our present procedure cannot account for this type of data, and
therefore it needs refining.
The reason our present procedure falls short of explaining these data is
quite simple. You will recall that in the previous chapter we analysed the p
of deep as extraprosodic ([di:]), and construed this extraprosodicity as
the source of its licensing.
When th is added to form depth, p ceases to be peripheral, and therefore

it can no longer be extraprosodic. If the p of depth is not licensed by extra-
prosodicity, it will have to be licensed in the rime, but because of rime
bimoraicity this can only happen if one of the two vowel moras is lost, hence
the shortening of the vowel.
The problem now is that, if we extend the procedure to deepness, we will

be predicting *dle]pness. In fact, the correct outcome dli:|pness suggests that
p remains licensed by extraprosodicity in spite of its loss of peripherality.
How can this be so?
One logically possible answer would be that the Peripherality Condition on

extraprosodicity is simply wrong. However, the evidence for this condition
is so strong overall that it would be foolish to give it up.
Indeed, if we did give up the Peripherality Condition, how would we

explain that there is dle]pth rather than “d[i:]pth?
Following on from the discussion in the previous sections, you may next
think of relating the vowel contrast in depth and deepness to the by now fam-
iliar cyclic vs. non-cyclic dichotomy. Indeed, we have accounted for the
stress contrast between -ity and -ness formations by declaring -ity cyclic and
-ness non-cyclic. Could we use the same strategy now, assuming that -th is
also cyclic (we have of course already established that -ness is non-cyclic)?
Obviously, the implementation of this solution cannot be direct, since, as

a condition, the Peripherality Condition is not subject to the cyclic vs. non-
cyclic dichotomy: the cyclic vs. non-cyclic dichotomy is only relevant to rules,
which can apply in the two different modes, and to affixes, which can define
two different types of domain, compatible with cyclic and with non-cyclic
rules, respectively. Evidently, therefore, an alternative implementation is called
for if we are to call upon the Peripherality Condition to account for the d[e]pth
vs. dlit]pness paradox.
The first step in this alternative implementation involves formalizing the
Peripherality Condition on extraprosodicity as in (15):
(15) *...<X>Y], for Y non-null
The key aspect of the formalization in (15), which otherwise simply restates
the definition of the condition in chapter 12, is the “]” context. In particu-
lar, (15) makes it explicit that the extraprosodic element must immediately
precede a morphological right bracket. With this in mind, let us consider An extraprosodic
the morphological representations of depth and deepness: element must
immediately
(16) precede a
a. [[dep]6] b. [[dizp]nas] morphological
right bracket
You will notice that [p] indeed immediately precedes a right bracket in both
configurations. Consequently, according to (15), it should qualify for extra-
prosodicity in both cases, but it obviously does not in a.: some other factor
must therefore be at work to prevent this result.
Why are we ae 0g
tha
silGay
We suggest that the factor in question is the degree of integration of the vari-
ous morphemes in the word, as we shall now explain.
[4 Word-Internal Cohesion: The Bracket Erasure Convention
In section 4 we mentioned the fact that such cyclic suffixes as -al in arrival
occur with a fixed set of bases, rather than being productively attached to
any base at the will of the speaker. This is also the case with -th. By con-
trast, suffixes like -ness are highly productive.
The dichotomy
productive vs.
non-productive
affix correlates
well with the
dichotomy
Importantly, the dichotomy productive vs. non-productive affix (correspond- non-cyclic vs.
cyclic affix:
ing to the opposition between non-lexicalized and lexicalized constructions,
cyclic suffixes
respectively) correlates well with the dichotomy non-cyclic vs. cyclic affix: LEXICALLY
the unproductive -al and -th define a cyclic domain, whereas the productive SELECT their
-ness defines a non-cyclic domain. Technically, we say that cyclic suffixes bases
LEXICALLY SELECT their bases: the information as to what bases they attach
to is an integral part of their lexical entry. Non-cyclic affixes are, however,

not so restricted. We spell out the contrast between the two in (17) with a
small sample of (simplified) lexical entries of the two kinds:
(17) Types of lexical entry:

a. Cyclic/Unproductive:
-al. farrive .~ ., etc.
=tht (ACEP ten, tK.
b. Non-cyclic/Productive:
-ness
-less
The obvious implication of the restrictedness of -al, -th, etc., vis-a-vis -ness,
-less, etc., is that the degree of internal integration is greater in words with
the former (cyclic) affixes than in words with the latter (non-cyclic) affixes.
Indeed, the internal unity that characterizes cyclic domains manifests itself
The internal unity in several spheres. For instance, semantically, non-cyclic suffixes tend to be
that characterizes
compositional, with their meaning simply added to the meaning of the base:
cyclic domains
manifests itself in
neighbourhoodlessness means exactly what neighbour + hood + less + ness mean.
several spheres This is far less so the case with cyclic suffixes, which are often semantically
integrated with the base. For example, a word like transmission (clearly derived
from transmit morphologically) can refer to a specific part of an automobile,
as well, of course, as to the act of transmitting. Expectedly, transmissionful or
transmissionless, with the non-cyclic suffixes -ful or -less added to transmission,
mean ‘full of transmission’ and ‘without transmission’, respectively, what-
ever the meaning of transmission may be: words with non-cyclic affixes are
less prone to develop idiosyncratic meanings.
Another obvious manifestation of the close integration of cyclic affixes and

the looser amalgamation of non-cyclic affixes has to do with phonological
exceptionality: it is far more likely for words containing cyclic affixes to be
exceptions to phonological rules than for words with non-cyclic affixes. For
instance, we have already seen in chapter 15 that obesity, with the cyclic suffix
-ity, fails to undergo vowel shortening: obli:]sity, not *ob[e|sity.
The particularly close unity characterizing cyclic domains in all areas
is given formal interpretation in the following BRACKET ERASURE
CONVENTION: The BRACKET
ERASURE
CONVENTION
(18) Bracket Erasure Convention:
disposes of all
Internal morphological brackets are erased (= become invisible domain-internal
= become inaccessible) at the end of each level. brackets at the
end of each level
The implication of the Bracket Erasure Convention in (18) is that, after the
cyclic derivation is completed, a word like depth will keep no trace of its
original internal morphological structure:
(19) [dis] — [[dir]6] -— [di:6]
Clearly, the representation [di:6], with p extraprosodic, violates the

Peripherality Condition in (15), since the [p] does not immediately precede
a morphological right bracket. This situation triggers the shortening of the
vowel to allow the licensing of p by the rime, as we have explained. If p were
not thus licensed, it would have fallen foul of the rule of STRAY ERASURE,
which we must assume deletes material which is not prosodically licensed: STRAY ERASURE
deletes material
which is not
(20) Stray Erasure: prosodically
Delete material which is not prosodically licensed. licensed
Let us now turn to deepness, with the non-cyclic suffix -ness, and no vowel
shortening. At the point where Stray Erasure applies, the word deepness still
has its internal morphological structure: [[dee]ness]. Given this, p remains
licensed by the morphological right bracket that immediately follows it, and
consequently the conditions for Stray Erasure in (20) will not be met. This
explains the difference between depth and deepness, and similarly for the
parallel data we provided in section 1 above.
Non-Cyclic Processes
The proposal we are advancing accounts fora rich array of data that go beyond
vowel shortening, as we shall now briefly review.
-
In chapter 10 we came across the possibility of sonorant consonants occupy
ing the syllable nucleus. We illustrate this again in (21):
(21) cycle
centre (in rhotic accents)
rhythm
As we pointed out at the time, this unexpected result is motivated by

Sonority Sequencing, which would be violated if the sonorants were inte-
grated in the previous syllable. Ostensibly, the situation in question can only
arise word-finally, since word-internally the sonorants can readily be incor-
porated into the next onset:
(22) cyclic
central
rhythmic
Paradoxically, however, in the forms in (23) the sonorant is not (neces-

sarily) parsed in the onset:
(23) cycling
centring (in rhotic accents)
rhythmish
A considerable
number of pro-
; : ;
cesses in English
ee Scaniwenton In (23), the sonorant may still be parsed in the nucleus, despite the fact that
Hae preseneaer the obstruent + sonorant cluster is followed by a vowel, as in (22) above,
a right morpho- and therefore the cluster is parsable in the onset. The paradox is resolved if
logical bracket we make the process of sonorant nucleus formation contingent on the pres-
ence of a right morphological bracket:
(24) Sonorant Nucleus Formation:

N
[+sonorant] — [+sonorant]/__]
The rule of Sonorant Nucleus Formation is obviously non-cyclic: if it were

cyclic we would indeed obtain *rhy.thm.ic, and correspondingly.
On the other hand, if Sonorant Nucleus Formation is non-cyclic, it will not

apply in rhythmic, since when this form reaches the non-cyclic block the
relevant sonorant m is no longer followed by the required morphological
right bracket, having been disposed of by the Bracket Erasure Convention
in (18):
(25) Cyclic phase

[rhythm]
iN
o Syllabification, ,,.
[rhythm]
Oo
i
[[rhythmlic],
o Oo Syllabification, ,,
[rhythmic].
ih
oO Ke Bracket Erasure Convention
\ Ih
[rhythmic].
Non-cyclic phase
Gun.
[rhythmic]
NA Sonorant Nucleus Formation,,,
NA Bracket Erasure Convention
rhythm,
Sonorant Nucleus Formation as in (24) will, however, apply in both
required
where the m is word-final, and therefore necessarily adjacent to the
the word-
bracket, and rhythmish, where the suffix -ish is non-cyclic, and
internal bracket is therefore preserved:
(26) Cyclic phase

[rhythm] [rhythm]
o Oo Syllabification, ,.
(seme
[rhythm] [rhythm]
NA NA Bracket Erasure Convention
Non-cyclic phase
oO Oo
[rhythm] [[rhythmlish],,
Oo Oo Sonorant Nucleus Formation,
fel
[rhythm] [[rhythmlish],,.
jj jI Syllabification, ,,.
[rhythm] [[rhythmlish],,.
jj jAN Bracket Erasure Convention
[rhythm] [rhythmish],,,
Another phenomenon we have come across is the deletion of word-final

g after a nasal in many accents of English (see chapter 3 above):
(27) long — strong

This process is matched by the deletion of b under similar conditions in all

accents:
(28) bom iam
Yin
The deletion of these stops word-finally after the nasal is probably motivated
by a requirement of minimal sonority distance.
The stop does surface word-internally in such derivatives as those in (29):
(29) a. bombard b. longest
Unexpectedly at first sight, b and g are deleted word-internally in forms like

those in (30):
(30) a. bom/ing b. longing
Again, the matter is resolved if we include a right morphological bracket in

the environment of the appropriate deletion rule:
(31) Postnasal Voiced Stop Deletion:

[-continuant] > © / [+nasal]___]
Condition: the input may not contain a path to [coronal]
If Postnasal Voiced Stop Deletion as in (31) is non-cyclic it will apply,

for instance, in [[bomb]ing], which keeps the word-internal bracket in the
non-cyclic domain, but not, for instance, in [[bomb]ard], which loses it on

exit from the cyclic domain: [bombard].
Also subject to word-final deletion is n following m (again, probably

because of sonority restrictions):
(2) <a: hymy b. hymnal

condem condemnation
autumy autumnal
The lexical presence of the n is justified by the forms in b. In their base coun-
terparts in a., however, this n disappears. Puzzlingly, n also disappears in
the forms in (33), despite the fact that it is parsable as the onset of the next
syllable, just as in (32b):
(33) hymying
condemping
autumypy
Once more, we shall assume that the relevant rule deleting the n is non-cyclic
and includes a morphological right bracket in its environment (the deleting
n must obviously be preceded by m: cf. hen, kiln, barn in rhotic accents, etc.,
with no deletion):
(34) Nasal Simplification:

Aoi) i ea
As a final process affecting segments, consider the deletion of g before a

word-final nasal:
(65) #a: sign b. signature

paradigm paradigmatic
resign resignation
In a., the sequence g + nasal is indeed word-final. In this context, the syl-
labification of g would violate Sonority Sequencing, and consequently the g
deletes (NB there is no rule to parse the nasal as a nucleus here).
In (35b), g + nasal is not word-final: therefore, the nasal can be onset to the
following vowel, and the g does not delete. In this case also, there is an appar-
ent paradox, illustrated by the data in (36):
(36) signer
paradi¢gmy
resigning
In (36) g deletes despite the fact that in these forms the sequence g + nasal
is not word-final, and the nasal could in principle be syllabified in the fol-
lowing onset. By this stage, the solution is completely obvious: the non-cyclic
rule deleting g includes a right morphological bracket in its environment:
(37) Prenasal g Drop:

g 7 O/__nl
The model we are proposing also provides an explanation for the retrac-
tion of stress in forms like péregrinating, where the retracted stress is not word-
final in the source structure, peregrindting. All we have to do, in fact, is make
such non-cyclic retraction sensitive to the presence of a right bracket, rather
than to the word end as such:
(38) Final Syllable Stress Retraction:

= Line 0
zi He 2 ce Line 1
IS 7A Se ee NR fe
Clearly, this rule will enact retraction in both [peregrindtle and [[pere-
grindt]ing], since the retractable asterisk is adjacent to a right bracket in both
cases.
Eq Ordered Affixes
The model of the interaction between morphology and phonology we have

been presenting can be schematized as in (39):
(39) Interaction between morphology and phonology:

Morphology Item Phonology
g
Underlyin representation
iit LUN
8kad Hips Cyclic rule,
Intermediate representation,
Gain natiie oman yahnes as pe Cyclic rule,
|
Intermediate representation,
CY CHC ALIXaUOUM aiciurciecs

ai’hia) >
Cyclic affixed form,
Cialis: are Cyclic rule,

Cyclic affixed intermediate representation, ,,
Qintata talelalate = = oieo minisia nln Cyclic rule,
Cyclic aiiixation, ore merece >
Cyclic affixed form,,
Qo - oe ee eee eee eee eee eee
|
Cyclic rule,
Cyclic affixed intermediate representation, ,,,
Bracket Erasure Convention ------ >
Output of cyclic phase
Non-cyclic affixation, -----"------- >
Non-cyclic affixed form,
Non-cyclic affixation, -----7-7--7---7- >
Non-cyclic affixed form,,
od kt Non-cyclic rule,
Sooopenis
Non-cyclic affixed intermediate representation,
RSsnaan
ag eee Non-cyclic rule,
Output representation
The model in (39) implies that cyclic affixes are concatenated first, and non-
Cyclic affixes are cyclic affixes subsequently. The obvious implication is that no cyclic affix
concatenated will ever occupy a more outward position than a non-cyclic affix: non-cyclic
first, and non-
cyclic affixes
suffixes will occur to the right of cyclic suffixes, and non-cyclic prefixes to
subsequently the left of cyclic prefixes.
There is a considerable amount of evidence for this hypothesis. We give
a sample in (40):
(40) a. un-in-capable b. *in-un-capable

nation-al-ist *nation-ist-al
Wilson-ian-ism *Wilson-ism-ian
creat-iv-ist *creat-ist-ive
immun-ity-ite *immun-ite-ity
The forms in (40a), containing at least two affixes, are all legitimate. How-
ever, their counterparts in (40b) are not, even though they contain exactly
the same affixes. The reason is, of course, that in (40b) the order of the affixes
has been inverted, in violation of the strict order cyclic affix > non-cyclic
affix we are referring to (the arrow head “>” indicates the obvious linear
ordering).
The discovery that both affixes and phonological rules are allotted to classes
which interact in the manner we have been describing, and that the order
of affixes is fixed as between classes (the order of affixes in the same class
is supposed in principle to be free, regulated only by syntactico-semantic
considerations), is one of the most important contributions of the decade strad-
dling the 1970s and the 1980s.
Lexical Phonology: Problematic Orderings
We have already seen that each class is defined both by order (stacking
order for affixes, application order for rules) and by a set of contrastive pro-
perties (cyclic vs. non-cyclic application, early versus late internal bracket
erasure, and other dichotomies listed in (52) below). This state of affairs was
encapsulated in a model where both the morphology (that is, affixation) and
the phonology are organized in autonomous interacting blocks, as schemat-
ized in (41):
(41) Organization of the morphology and the phonology in blocks

Lexical representation
Block 1
Pee Phonology Block 2
Surface representation
The blocks of our flow chart in (41) are often referred to as CLASSES,
Blocks of rules
LEVELS or STRATA, and the model that incorporates them as LEXICAL
and affixal
PHONOLOGY or STRATAL PHONOLOGY. There have been several ver- domains are
sions of Lexical Phonology through the years, varying in a number of tech- referred to as
nical details not sufficiently central to our present concerns to warrant CLASSES, LEVELS
discussion here. In essence, (41) merely compresses our more detailed dia- or STRATA, and
gram (39) above, and must be interpreted in a similar manner. the model that
incorporates them
as LEXICAL
PHONOLOGY
Both figures (39) and (41) are of course adapted to the requirements of English,
and the exact mechanics of the blocks may differ in other languages. The
number of blocks also differs from language to language, and even from pro-
posal to proposal.
For completeness, we now summarize the properties characterizing the
rules of each of the two blocks we have been proposing for English:
(42) Rule properties:

Block 1 Block 2
Cyclic Non-cyclic
Exceptions likely Exceptions unlikely
May not refer to internal brackets May refer to internal brackets
Non-productive Productive
Common semantic opacity Usual semantic transparency
Structure-preserving Non-structure-preserving
STRUCTURE PRESERVATION refers to the fact that some phonological rules

STRUCTURE are strictly respectful of the basic structural properties of the language, such
PRESERVATION
as the identity of lexical segments or the core syllable structure (in English
refers to the fact
that some phono-
CVX, or CVXC word-finally). Non-structure-preserving rules can, by con-
logical rules are trast, violate these properties.
strictly respectful
of the basic struc-
tural properties of
the language
Lexical Phonology held considerable promise of reducing the inter-

action between morphology and phonology to a truly simple, compact
formalism. As research went on, however, significant problems concerning
As research went the ordering of the affixes came to light: many (in fact most) legitimate affix
on, significant combinations are missing, while a number of putatively illegitimate ones
problems concern-
ing the ordering
do occur.
of the affixes With respect to the non-occurrence of legitimate affix strings, it appears
came to light that, out of 1,849 logically possible combinations of the 43 most common
English suffixes, only 40-50 exist. Many of the failures can perhaps be
accounted for by syntactico-semantic factors: for instance, we cannot form
*nightness from night because the suffix -ness only attaches to adjectives (white-
ness is fine, but night is a noun). After putting such cases aside, there is still
a remnant of over 600 combinations that ought to be possible, but most of
them fail to occur.
Obviously, therefore, Lexical Phonology is underpowered: it lets too much

through.
The other side of the coin is a subset of combinations which are predicted
to be illegitimate but do occur. These have received considerable attention
in the literature under the label ORDERING PARADOXES, and can appro-
priately be illustrated by the comparative formation unhappier. In Modern
English, comparatives in -er are restricted to adjectives with at most two syl-
lables, the second of which must moreover be light, as, for instance, inha.ppy.
Therefore, the morphological constituency of unhappier must be [un[happier]l,
since a base unhappy (un.ha.ppy) would simply not qualify for -er suffixation.
The morphological bracketing [un[happier]], however, implies a meaning ‘not

happier’, as against the real meaning ‘more unhappy’. The true meaning of
unhappier presupposes the bracketing [[unhappyler], which is at odds with
the derivational facts mentioned: hence the paradox.
Bracketing para-
This and other similar cases are bracketing paradoxes precisely because of
doxes are so
the assumption we have been operating with that the domains available because of the
to phonological rules word-internally are provided by their morphological assumption that
constituency. For the data we have been presenting, this assumption has the domains
indeed served us well, but we are now seeing that it runs aground when available to
phonological rules
the data set is extended in certain directions. A proposal to circumvent this
word-internally are
problem involves the abandonment of the assumption that morphological provided by their
and phonological constituents are identical, and we will present it in section morphological
16 below. First, however, we must turn our attention to domains bigger than constituency
the word, where by definition morphology is irrelevant.
The Phonological Phrase
In chapter 11 above and at various points since we have referred to the very
common English phenomenon of stress retraction under clash: Sue Ann vs.
Sue Ann Céok, antiques vs. antique chdir, and so on.
The data in (43), however, do not exhibit retraction, in spite of the fact
that they contain stress clashes:
(43) Japanése railways and motorways

Tennessée’s politics and religion
rabbits reproduce quickly enough
Mary persevéres firmly but gently
In particular, the asterisk clashes in the constructions in (43) are to all

appearances identical to those that we have been saying all along are sub-
ject to retraction, as is the case in (44):
(44) Japanese railways

Ténnessee’s pOlitics
rabbits réproduce quickly
Mary pérseveres firmly
We will now show that the answer to this problem involves assigning
phonological rules to specifically phonological domains.
In principle, it might be reasonable to identify phrase-size domains of phono-
logical rule application with syntactic constituents, in the same way as we
have been identifying word-internal application domains with morpho-
logical constituents, quite successfully up to the end of the previous section.
Indeed, this is what we assumed when we discussed stress retraction in
previous chapters. The assumption that grammatical constituents (whether

morphological or syntactic) double up as application domains for phono-
logical rules was already made in SPE, and is of course maximally parsi-
monious and natural, providing as it does a ready-made interface between
the phonology and the grammar.
Although the identification of phonological phrasal domains with syntactic

constituents is plausible in principle, it runs into considerable empirical diffi- Although the
culties. In the case we are discussing, it is clear that the phrases in (43) share identification of
phonological
their syntactic constituency with their counterparts in (44), and yet the two
phrasal domains
sets behave differently with regard to stress retraction: retraction takes with syntactic
place in (44), but in (43) the clash persists. constituents is
plausible in princi-
ple, it runs into
Whyis thesyntac considerable
(44) identical? empirical diffi-
culties
In order to resolve this apparent paradox we have to accept that the exten-
sion of the syntactic phrase is not the same as the extension of the phono-
logical phrase, and that the domain of application of the Rhythm Rule is the The extension
of the syntactic
phonological phrase, rather than the syntactic phrase. The syntactic phrase
phrase is not
and the phonological phrase are related, but they are still distinct, in ways the same as the
that we shall now explain. extension of the
The construction of the phonological phrase is parasitic on the structure phonological
of the syntactic phrase. A typical syntactic phrase (a noun phrase, a verb phrase
phrase, an adjective phrase) has a head (a noun, a verb or an adjective, respect-
ively) and (optionally) additional material on either side of the head:
(45) A” A” =major adjective phrase
vet A’ =minor adjective phrase
joeee DW = degree word

DW =A Pr
ed ec
ele
= adjective
very fond ofphonology PP =prepositional phrase

Head
For any one language, one of the two sides of the phrase is “recursive” — in
principle it admits of an unlimited number of “complements”:
(46) very fond of phonetics, phonology, morphology, syntax, semantics,

pragmatics, ...
The other side of the phrase, by contrast, is non-recursive, in that the ele-
ments it can take (“specifiers”) are limited:
(47) fond of phonology > very fond of phonology — not very fond of
phonology — ??
elements?
The syntactic configuration exemplified in (45) provides the criteria for the
The construction construction of the PHONOLOGICAL PHRASE, which includes (obligat-
of the PHONO-
orily) the syntactic head and the elements in the non-recursive side which
LOGICAL PHRASE
is parasitic on the
are not themselves syntactic heads — if they are syntactic heads, they will
structure of the constitute nuclei of independent phonological phrases.
syntactic phrase:
it includes (obliga-
torily) the syntac- What will the phonological phrases be in.
tic head and the
elements in the
non-recursive Optionally, phonological phrases include a single word on the complement
side which are side of the head, as we show in (48) (PP = phonological phrase; NB do not
not themselves
confuse it with PP = prepositional phrase above!):
syntactic heads
(48) [rabbits]pp [reproduce]pp [quickly]pp

or
[rabbits],p [reproduce quickly]pp
an be integrated into [rep
Crucially, if the complement side contains more than one word, it cannot
be incorporated into the PP, as we illustrate in (49):
(49) [rabbits],, [reproduce]pp [quickly enough]pp

not
[rabbits], *[reproduce quickly enough]p,
We formalize the criteria for the formation of the phonological phrase in (50):
(50) Phonological phrase formation:

A phonological phrase is made up of:
(i) a syntactic head +
(ii) all the words on the non-recursive side of the head’s phrasal
domain +
(iii) optionally, a single-word constituent on the complement side of
the head
Given the definition in (50), we can account for the occurrence of retraction
in (44), but not in (43), simply by declaring the phonological phrase the domain
of the English Rhythm Rule:
(51) a. [Japanése railways]pp > [Japanese railways]pp

b. [Japanése]pp [railways and motorways]pp
You can see that there is no formal stress clash in (51b) (and correspond-
- ingly in the other phrases in (43)): the clash occurs across the boundaries of
the phonological phrase and, therefore, formally it is only apparent.
The Intonational Phrase
The phonological phrase is the smallest of our phrasal phonological domains.

The next phonological domain larger than the phonological phrase is the
INTONATIONAL PHRASE, which is traditionally identified with the domain
where the intonational melodies we studied in chapter 14 associate to the
segmental material.
As you will recall, just as intonational melodies associate to individual

words uttered in isolation, they also spread over whole (and fairly lengthy)
phrases, like the dark clouds in the sky threaten imminent rain. Clearly,
domains such as these are potentially larger than phonological phrases, and
therefore they need to be given independent formal status.
The criteria for delimiting the intonational phrase are less clear-cut than
the criteria for delimiting the phonological phrase. In particular, while the
phonological phrase appears to be definable on purely syntactic terms,
as our formalization in (50) displays, the delimitation of the intonational
The delimitation phrase can also involve semantic and phonological factors, and even be
of the intonational
influenced by performance. We list these heterogeneous criteria in (52), with
phrase involves
syntactic, semantic
the appropriate exemplification:
and phonological
factors, and is (52) Criteria for the delimitation of the intonational phrase (IP):
even influenced
Syntactic criteria:
by performance
(i) The PPs making up an IP must be sequential:
[[Picasso]pp [was a truly]pp |great]pp [painter] pphp
therefore:
(ii) “Parenthetic’” PPs (technically, PPs unattached to the highest
sentence node) make up independent IPs:
Picasso [[as you know]pp],p was a great painter

Picasso was [[as you know]pphp a great painter
and:
(iii). Each of the remaining PP sequences attached to the highest sent-
ence node makes up an IP:
[[Picasso]pp]p
[las you know]pp]p
[[was a great]pp [painter] ppl»
[[Picasso]pplwas
pp]]yp
[[as you know]pp]p
[la great]pp [painter] pplip
or, optionally:
(iv) IPs can be split up into smaller IPs
usually after a noun phrase:
[[Picasso]pp [was a painter]pp]p,

[[a lover] pphip,
[land a bon viveur] pp]p
[[My brother] ppp

[[ found] pp [several two-pound]pp [coins] pp]p
[[on the church floor]pp],p
or before a subordinate sentence:
[[my brother]pp [found] pplip

[[that phonology]pp [is a most interesting]pp [subject]
pp]p
Semantic criteria:
(v) Contrastive prominence induces the breakup of an IP:
[[Siegfried]pp [loved]pp [Briinnhilde]pplp

[[before SHE]pp]IP
[[loved]pp [HIM] pphip
SSS
(vi) IPs are not usually broken up between the verb and its follow-
ing obligatory argument (an “obligatory argument” is a type of
word the meaning of which is integrated in the meaning of the
verb in a particularly close manner):
[UT always] pel give]pplmoney|pplto the buskers]pplthat cheer me up]pp]p
cf. [[I alwayslpp [givelpp [money]pphp

[[ for the buskers]pp [that cheer me up|peli
Phonological criteria:
(vii) The greater the length, the more likely the break-up of the IP:
[[my brother] pphip
[[only recently]pp [realized] pp]p
[[that his girlfriend]pp lof twenty years] pp]yp
[[had been relentlessly pursued] pplip
[[by a crowd]pp lofwealthy] pplsuitorslpp |p
[[ for the past]pp [ten years]pp lor more! |pplip
Performance-based criteria:
(viii) Rate and style of speech affect IP breakup: the slower and the
more formal the style, the more IPs.
You can see that the criteria for the delimitation of intonational phrases
are quite flexible, although some strict guidelines, included in (52), do exist.
There are also general conditions on the structure of all phonological

domains, in particular a requirement of proper domain inclusiveness: for
example, an IP cannot begin or end in the middle of a PP. The requirement

that a phonological domain must properly include the immediately smaller
domain, with no leftover, is encapsulated in the so-called “Strict Layer
Hypothesis”, which we formalize in (58), further below.
Like other phonological domains, the intonational phrase can in prin-

ciple serve as the domain of any type of phonological rule, not just of into-
national association. In fact, there is some evidence that the intonational
phrase may not after all necessarily constitute the domain of intonational There is some
evidence that
mapping, paradoxically so if the label is taken seriously. Consider the two
the intonational
phrases in (53): phrase may not
necessarily con-
(53) a. But we are not telling John stitute the domain
of intonational
b. But we are not going John
mapping
(53b) (conventionally written with a comma between going and John) must
contain two IPs (one on each side of the comma) according to criterion (52ii).
Yet its pattern of association of the intonational melody is identical to the
pattern in (53a).
Among the non-intonational processes that occur in the domain of the into-
national phrase is optional nasal assimilation, a rule of English we introduced
and discussed in chapter 4:
(54) The trai[m] passed us as we were going up the hill

The trai[n] came in late
cf.
They came by trai[n], Peter and Jane
They came by trai[n], contrary to advice
It is clear from our discussion that the extension of the intonational

The extension of phrase need not coincide with the extension of any type of syntactic phrase.
the intonational
A dramatic example of this, made classic by SPE, is provided by the rhyme
phrase need not
coincide with the
this is the cat that killed the rat that ate the malt that ..., with the syntactic con-
extension of any stituency in (55a) but the intonational phrasing in (65b) (NP = noun phrase,
type of syntactic VP = verb phrase, S = sentence, all syntactic constituents):
phrase
(55) a. [this [is [the cat that killed [the rat that ate [the malt]yp]xplnplvels
b. [this is the cat], [that killed the rat], [that ate the malt],
The Phonological Utterance
One phonological domain even larger than the intonational phrase is the
Phonological Utterance (PU). The phonological utterance constitutes the
domain of application of such phonological rules as r-insertion (in RP, east-
ern New England and many other accents of English) and flapping (in General
American, and also in some non-rhotic accents). A useful pair illustrating
the role of this domain in r-insertion is given in (56):
(56) a. Don’t sit on that sofa!: it’s broken.

b. Don’t sit on that sofa! It’s Mary.
In the phrase in (56a), pronounced with an intrusive [1] (... sofa[ajit ...), we
give the reason for not sitting on the sofa: it is broken. By contrast, in (56b) the
two phrases are semantically unrelated: it’s Mary refers, for example, to the
knock on the door we hear as I say don’t sit on that sofa. Most curiously, there
is no [1] intrusion here, even if the two phrases are said in absolute succes-
sion, with no break between fa and it.
Indeed, each of the phrases in (56) can be pronounced with or without

r-insertion, depending again on whether or not there is a semantic link between
their two component phrases: compare Don’t sit on that sofa! It’s broken, with
it’s broken referring to an object other than the sofa, and Don’t sit on that sofa:
[1] it’s Mary, with Mary identified with the sofa (Mary had, for instance, dis-
guised herself as the sofa!).
It appears that the construction of the Phonological Utterance involves

criteria from practically every component of the grammar, phonological or The construction
of the PHONO-
non-phonological, as we specify in (57), where the superscript * indicates
LOGICAL UTTER-
optional recursion: ANCE involves
criteria from
(57) Conditions on the formation of the Phonological Utterance: practically every
Ld... ip X =a syntactic label component of
the grammar,
Condition: no deliberate internal pause
phonological or
non-phonological
Conditions on (optional) merger of Phonological Utterances:
Pragmatic: utterer and addressee identity
Phonological: shortness and no pause
Syntactic: “ellipsis” or “anaphora” relation
e.g. [You didn’t invite Martha. [1] I did]py
(invite Martha)
or Semantic: “and”, “therefore” or “because” relation
e.g. [Dont’ sit on that sofa [.]. It’s broken]
py
(because)
Properties of Phonological Domains
The three phrase-level phonological domains we have been presenting (the

phonological phrase, the intonational phrase and the phonological utterance)
are commonly labelled PROSODIC DOMAINS, and the branch of phono-
logy concerned with them PROSODIC PHONOLOGY. You must bear in mind,
however, that the term “prosodic” has through the years been applied to a
variety of phenomena, and is therefore potentially ambiguous. Remember,
for instance, that in part II we referred to such suprasegmental constituents
as the syllable and the foot as prosodic. Here as elsewhere in this book, we
are endeavouring to select terminology which is transparent and cannot lead
to misunderstanding, hence our use of the expression PHONOLOGICAL
DOMAINS in preference to “prosodic domains”.
The three phonological domains in question stand in a relationship of
inclusiveness, such that the Intonational Phrase is made up of (one or more)
The requirement Phonological Phrases, with no material left over, and the Phonological Utter-
that a phonologi-
ance is made up of one or more Intonational Phrases, again, with no material
cal domain must
properly include left over. As we mentioned above, this requirement of proper inclusiveness
the immediately is known as the STRICT LAYER HYPOTHESIS, which we now formulate
smaller domain, in (58):
with no leftover,
is encapsulated in
the STRICT LAYER (58) Strict Layer Hypothesis:
HYPOTHESIS Each phonological domain contains precisely one or more phonolog-
ical domains of the rank immediately below
The Strict Layer Hypothesis fares reasonably well across domains and
across languages, although some problematic cases are known to exist. We
will be proposing a general solution to the problem of exceptional behavi-
our in phonology in chapter 19.
In the Lexical Phonology model of morphology-phonology interaction we
introduced in section 8 above, phrasal phonological domains are directly
identified with syntactic constituency, misguidedly, as we are now seeing.
In Lexical Phonology, phonological rules that apply in phrasal domains are
Phonological rules labelled POSTLEXICAL RULES, and are assumed to apply under conditions
that apply in
even more liberal than those on the rules in the lexical non-cyclic block.
phrasal domains
are labelled
POSTLEXICAL
RULES
In (59) we provide a list of the properties originally thought to be associ-

ated with each type of rule:
(59) Lexical vs. postlexical rule properties:

Lexical Postlexical
Word-bounded Yes No
Access to word-internal structure Yes No
Cyclic Yes No
Only applies in derived environments Yes No
Structure-preserving Yes No
Only applies to lexical categories Yes No
May have exceptions Yes No
It is of course not surprising that the properties of lexical and postlexical

rules should contrast, since the elements making up postlexical domains
(= words) are even more self-sufficient than the non-cyclic affixes that make
up the non-cyclic block. This notwithstanding, the divide between the two
types of rule sketched in (59) is too sharp and neat, and was soon found to
leak in both directions, as our earlier discussion of non-cyclic rules already
hints (compare (59) with (42) above).
Subphrasal Phonological Domains
We shall now turn our attention to the phonological domains below the phrase,
namely, the domains of word-size or shorter.
In the first part of this chapter we used words and (some of) their con-
stituent morphemes as domains for phonological rules, in the general con-
text of Lexical Phonology. In a parallel tradition, however, it is the foot,
the syllable and, on occasions, the mora that are viewed as word-internal
phonological domains, structured hierarchically in accordance with the Strict
Layer Hypothesis.
It is clearly desirable to bring these two traditions together, in a synthesis

which would allow us to preserve the best of either camp. In order to attain
this goal, several inconsistencies need to be overcome. Thus, if grammat-
ical constituents do not have a direct role as phonological domains above
the word, it appears reasonable to expect them not to have such a role below
the word either, but they do in Lexical Phonology. In particular, there is an
obvious inconsistency between delimiting phrasal phonological domains prim-
arily through a mapping from syntactic structure, as we have shown we
must, and identifying subphrasal domains directly with morphological con-

stituency, without the mediation of a similar mapping.
“The word “mappin 3

sure you understand it
e have been using c
AN
On the other hand, if the word-internal phonological domains are to be

identified with the foot, the syllable and the mora, as they are in the altern-
ative tradition, the construction of these domains from scratch we adopted
in chapters 9-13 above is inconsistent with the construction of their phrasal
equivalents essentially through a mapping from the syntax.
One possible response to the difficulties just mentioned is to draw a dis-

We draw adis- tinction between phonological constituents (the mora, the syllable, the foot)
tinction between
and phonological domains. From this perspective, PHONOLOGICAL CON-
PHONOLOGICAL
CONSTITUENTS,
STITUENTS make up phonological structure, whereas PHONOLOGICAL
which make up DOMAINS simply supply the spatial bounds within which phonological
phonological rules apply.
structure, and
phonological
domains, which
supply the spatial
bounds within
which phonolog-
ical rules apply This division between phonological constituents and phonological domains
allows us to formalize the word-internal phonological domains as mappings
from morphological constituents, thus preserving the formal parallel with
the mappings from syntactic constituents for phrasal domains.
Word-internal phonological domains contrast with their phrasal counterparts

in that more often than not they are identical to morphological constituents,
the stem and the word in particular.
In the mapping approach, however, mismatches between the two domains

(morphological and phonological) are possible.
The largest subphrasal phonological domain, and thus the component

element of the Phonological Phrase, is the PHONOLOGICAL WORD (PW). The largest sub-
One of the characteristics of the Phonological Word in English we are phrasal phono-
logical domain,
already familiar with is the presence of a prominent foot, with the primary and thus the com-
stressed syllable in the word. Such prominence is, of course, maintained when ponent element of
the word is included in larger domains: the Phonological
Phrase, is the
PHONOLOGICAL
WORD
(60) uniVERsity
the provision [of uniVERsity education]pp
[the provision of uniVERsity education];p is one of the government's
fundamental duties
[the provision of uniVERsity education is one of the government's fun-
damental duties]py
You can see that the syllable ver, the head of the foot versi, is prominent in
university irrespective of the size of the phonological domain: it is not the
case that in the phonological utterance, for instance, uniVERsity becomes uni-
versity, with no prominence.
The common
situation is for
phonological
The common situation is for phonological words to have the same exten- words to have the
sion as lexical words. A revealing instantiation is provided by English com- same extension as
pounds, where the prominence contours of the component (lexical) words lexical words
are preserved:
(61) matérnity hospital

defénce department
delivery véhicle
These data suggest that the phonological word status of each of the com-
pound’s component words is maintained in English.
Now consider Modern Greek:
(62) a. kukla b. spiti c. kuklospiti

‘doll’ ‘house’ ‘doll’s house’
psychi pedi psychopédi

‘spirit’ ‘child’ ‘adopted child’
nichta filaki nichtofilaki

‘night’ ‘guard’ ‘night guard’
You can see that the stress patterns of the individual words are not pre-
served in Greek compounds, each of which has its own individual stress
pattern, idiosyncratically. The obvious inference is that Modern Greek
compounds constitute phonological words, in contrast with their English
counterparts.
The difference between Modern Greek and English compounds illustrates

the variability in size of the phonological word between languages. In
section 16 we shall see that phonological words can also be smaller than
lexical words.
Segmental Affiliation to the Phonological Word

At this point, we can make use of the phonological word to solve the
problem that is still pending with respect to the affiliation of /@/ in depth.
Remember that we decided that the final /p/ of deep is licensed by extra-
prosodicity ([di:]) and attaches to the syllable node: [di:]. A similar
analysis of the /8/ of depth is, however, not possible, for the simple reason
that the sequence /p8/ does not comply with Sonority Sequencing, a pre-
requisite for syllabic membership. As we have already mentioned, though,
coronal obstruents can appear at the end of English words with no obvious
limit: depths, ropes, raked, text, texts, and so on. Coronal obstru-
ents can appear at
the end of English
words with no
obvious limit
Extra word-final coronal obstruents have often been interpreted as an

“appendix”. The question that obviously arises is what constituent they are
appendices to, and the common answer is that they are appendices to the
phonological word.
The appending of English extra word-final coronal obstruents directly to
the phonological word obviously presupposes that the Phonological Word
is not only a phonological domain, on a par with the Phonological Phrase,
the Intonational Phrase and the Phonological Utterance, but also a phono-
logical constituent, on a par with the foot and the syllable.
From this perspective, therefore, the phonological word has a dual identity:
it is a member of both the phonological domain hierarchy and the phono-
logical constituent hierarchy.
This conclusion does not seem unreasonable, given the similar dual status
of the lexical word, of which the phonological word is the phonological cor-
relate: the lexical word is both the largest morphological unit and the small-
We affiliate
est syntactic unit.
English word-
final extra coronal
obstruents and
extrasyllabic
word-initial
s- directly to the
If we thus accept the phonological word as the highest phonological consti- phonological word
tuent, we can affiliate English word-final extra coronal obstruents directly node
to the phonological word node:
(63), fagienRWi beep
The approach can be extended to the also extrasyllabic word-initial s-, an

issue pending from chapter 10 above:
(64) PW
Seep
/\
omer
Opportunely, the configurations in (63) and (64) exemplify a violation of the

Strict Layer Hypothesis, a possibility we referred to above.
ES
The phonological Small Phonological Words

word can be
smaller than the
lexical word We said above that the phonological word can be smaller than the lexical
—————————_ word. For example, in the Australian aboriginal language Yidin’ (referred
to in chapter 12 in connection with stress), some suffixes which are an

integral part of the lexical word can be shown to form phonological words
of their own. For instance, [gumatrilpw [dagain’y]pw ‘to become red’ is one
lexical word, formed on the stem gumari ‘red’ by double affixation (-daga-
n'y). Although gumaridagan’y is one lexical word, it exhibits (twice) the
penultimate vowel lengthening typical of phonological words with an odd
number of syllables (compare [guda:ga] ‘dog-absolutive’ with [gudagagu]
‘dog-purposive’).
The possibility of having phonological words smaller in size than lexical

words provides a resolution to the bracketing paradoxes involving prefixes The possibility of
having phonologi-
that we introduced in section 9 above. Consider, for instance, the case of
cal words smaller
unhappier. in size than lexical
words provides a
resolution to the
bracketing para-
doxes involving
prefixes
In order to account for the (apparently contradictory) facts, we can assume

that the morphosemantic structure of this word is [[wnhappi],er], — remem-
ber that unhappier means ‘more unhappy’. Phonologically, however, the struc-
ture would be [un]pw [happier]pw, from a base [un]
pw [happylpw, where the second
domain [happy]pw does comply with the requirement of maximal bisyllabic-
ity on -er suffixation (NB A = adjective):
(65) A Morphological structure
préin
Prefix A Suffix
un- we
PW PW Phonological structure
Phonological word status obviously carries certain implications, which must

be fulfilled if the approach is to be substantiated. For instance, we know that
phonological word status implies independent word stress. This prediction
appears to be fulfilled in un-, which, on the one hand, clearly lies outside
the stress domain of the main body of the lexical word happier (cf. unhappy,
not *vinhappy), and, on the other, does appear to carry prominence of its own,
in contrast to its synonym in-, which does not (cf. anhdppy vs. impossible).
Another prediction of the approach is that the 1 of un- will not syllabify
with a following vowel in the stem, and this also seems correct, at least in
slow speech: un.able seems a possible spontaneous pronunciation, whereas
in.active does not.
Finally, Nasal Assimilation does not necessarily apply to the n of un-,

although it does to the n of in-: u[n]common vs. ily|competent.
Many of the
“bracketing
paradoxes” of
English and other
languages can be
resolved by the
assumption that While many of the “bracketing paradoxes” of English and other languages
phonological may be resolved by the assumption that phonological domains are distinct
domains are from morphosemantic domains, some problems remain. For instance, in forms
distinct from
like developmental, patentability, standardization, and so on, a level-1 suffix
morphosemantic
domains, but
(-al in developmental) is preceded bya level-2 suffix (-ment), contradicting the
some problems prediction of the affix ordering hypothesis, a keystone of Lexical Phono-
remain logy. Notice in particular that -ment is non-cyclic, and -al is cyclic, as we know
from the effects they have on the position of stress, among other things:
(66) [[[develop], ment],, al],
Paradoxes such as these led to the introduction of the LOOP in Lexical

Phonology. The loop allows a form to be fed back into the previous level of — The LOOP allows
derivation, in violation of the affix ordering hypothesis: Seu
back into the
previous level
of derivation, in
violation of the
affix ordering
hypothesis
(67) The loop:

develop Level-1 affixation
development-al
develop-ment Level-2 affixation
A reinterpretation of morphological constituency as a phonological domain

will necessarily have -mental as a phonological word in developmental:
(68) (develop)pw Gmental)

py
As in the case of unhappier above, this phonological phrasing will not inter-
fere with morphological constituency:
(69) A Morphological structure
Vv N A
Stem Suffix Suffix
develop -ment -al
PW
NouPW Phonological structure
In order to make -ment define a phonological word we must of course mark

it as such in the lexicon, since we have said that phonological words are
normally the result of a simple mapping from morphosyntactic words, and
-ment is not a morphosyntactic word, but a suffix. In itself, this result is
not outrageous, since we suggested a similar analysis for the prefix un-
in English, also needed for a number of affixes in other languages. Some
serious problems ensue, however. Consider first the fact that -ment system-
atically carries the main stress of developméntal. The problem is not how
to get the stress on the ment of -mental (cf. the homophonous adjective
méntal), but to have this stress as prominent over the stress in devélop: clearly,
developmental would not constitute a word domain for End Stress [Right],
since (develop)pw (mental)pw no longer makes up one phonological word. As
a consequence, we would have to assume End Stress [Right] triggered by
phrasal constituency, in the style of time flies in chapter 11 above; developmental,
however, is clearly not a phrase, but, rather, a word, and we may have to
stretch our conception of phrases quite a bit to fit it in. Even if we succeed,
a serious problem remains. In particular, if -ment is a (lexical) phonological
word, then it should also be so in development: (develop)pw (ment)pw. Now, if
(ment) is a phonological word here, then it should carry its own stress, as
it does in developmental, and this stress ought to be prominent in the con-
struction, again for the same reasons as in developmental, whatever these may
be. The resulting stress contour, development, is, of course, incorrect, and the
approach founders.
There are no doubt a number of ad hoc ways of repairing the damage
(we could, for instance, stipulate that a monosyllabic suffix cannot surface
as a phonological word, at least in English), but we shall not resort to these
here. Instead, we will briefly refer to an alternative which does not make
use of phonological constituency, but represents in effect a development of
Lexical Phonology.
This alternative is simple in the extreme. All we have to do is take fully
We have to take on board the fact that non-cyclic affixes do not necessarily occupy the outer
fully on board the
morphological layer. We repeat the structure of developmental in (70):
fact that non-
cyclic affixes do
not necessarily (70) I[[[ develop], mentl],, all,
occupy the outer
morphological Given this structure, the intermediate form development will be skipped
layer by the cyclic stress rules of English, since cyclic rules can only apply in
cyclic domains. The rules will, however, apply when the next cyclic layer,
defined by -al, is reached, and the appropriate contour, developmental, will
be generated.
While this simple alternative does resolve the problems connected with
developmental and similar forms, it obviously entails the official abandonment
of the affix ordering hypothesis, one of the foundation stones of Lexical

Phonology.
affixation of which does not.

_ tions seem to ie the
to loop back intobine coe block and besusce;

We showed that the word is nottheee d
intonational phrase, and the phonological utteranc

logical phrase are largely determi ned by a
A number of processes SUC

by such boundaries, sir
Key Questions
1 What explanation can be advanced 12 How do we identify a Phonological

for violations of the two-mora re- Phrase? How does the Phonolog-
striction on rimes and of the three- ical Phrase relate to the syntactic
syllable window? phrase?
2 What is the relationship between 13 Name a phonological process whose
domains defined by affixes and the application is confined to the Phono-
cyclic or non-cyclic status of rules? logical Phrase.
3 List some cyclic and non-cyclic 14 What are the criteria for the delim-
affixes. itation of the Intonational Phrase?
4 What does infixation in English words 15 Is the Intonational Phrase coex-
tell us about the relationship be- tensive with any type of syntactic
tween phonology and morphology? phrase?
5 Where does Bracket Erasure apply? 16 Name aprocess of which the domain
6 How do cyclic and non-cyclic af- is the Phonological Utterance.
fixes relate to their bases? 17 What does the Strict Layer Hypo-
7 How does the Bracket Erasure Con- thesis state?
vention help to explain the contra- 18 What are postlexical processes? Are
dictory behaviour of certain forms they restricted in their properties as
with regard to certain processes? are lexical processes?
8 List some non-cyclic processes. 19 How is the Strict Layer Hypothesis
9 How are affixes ordered? violated by word-final coronals and
10 List the properties characteristic of word-initial s?
rules applying in the cyclic and non- 20 Suggest ways in which bracketing
cyclic blocks. paradoxes can be resolved.
11 Explain Structure Preservation. 21 What is the loop?
Domains of Application
Purth exmcn Parca c.t.ic.e
Dutch
The choice between the two denominal adjectival suffixes -isch and -ief
in Dutch is not arbitrary. Examine the two sets of data in a. and b. and
advance a hypothesis about what determines the choice of denominal suffix:
a. psychologie ‘psychology’ psycholégisch ‘psychological’

hysterie ‘hysteria’ hystérisch ‘hysterical’
analogie ‘analogy’ analdégisch ‘analogical’
b. agréssie ‘aggression’ agressief ‘aggressive’
invéntie ‘invention’ inventief ‘inventive’
actie ‘action’ actief ‘active’
Do these data tell us anything about the relationship between morphology

and phonology?
Maltese Arabic
In Maltese Arabic stress is either antepenultimate or penultimate, but the

details are irrelevant to the present exercise. A further rule deletes an
unstressed vowel in a word-internal open syllable. We illustrate the inter-
action of these rules on the affixation of subject agreement in the paradigm
of the verb hataf ‘snatch’ (“h” stands for IPA [h], which represents a voice-
less pharyngeal fricative)
a. /hataf-t/ [htaft] ‘I snatched’

/hataf-na/ [htafna] ‘we snatched’
/hataf/ [hataf] “he snatched’
/hataf-u/ [hatfu] ‘she snatched’
/hataf-it/. [hatfet] ‘she snatched’
(i) | Show how this interaction of rules works in the examples in a.
Other morphemes, including object suffixes -na ‘us’, -ik ‘you’ (sg.), -kum ‘you’
_ (pl) and -f negative, can be added to the verb after the affixation of subject
agreement, yielding examples such as:
b. [hatafna] ‘he snatched us’

[hataff ] ‘he didn’t snatch”
[hatfitkom] ‘she snatched you’ (pl.)
[hatfek] ‘he snatched you’ (sg.)
(ii) Assuming that there are two lexical levels involved, trace the deriva-
tions of the forms in b.
Polish
Consider the consonant clusters in word-initial position in Polish (the symbol

[é] stands for a prepalatal affricate and [$] stands for a prepalatal fricative):
[fscliekly ‘furious’ [bzd]Jura ‘nonsense’

[pfflola ‘bee’ [fsplanialy ‘great’
[fstlyd ‘shame’ [g3bliet ‘back’
[Isn]i¢ ‘shine’ [gnla¢é ‘to stick’
[mdll]i¢ ‘to feel seasick’ [ms¢]i¢ sie ‘avenge’
(i) Do these clusters present a problem for the Sonority Sequencing

principle?
(ii) If so, can the problem be resolved in the light of the discussion in
section 15?
(iii) Assuming that Polish onsets (like English onsets) are maximally
binary, say how the constraints on Polish onsets differ from those on
English onsets.
Italian Raddoppiamento Sintattico
The Italian phenomenon known as Raddoppiamento Sintattico causes the

lengthening of the word-initial consonant in the second in a sequence of two
phonological words if the first word ends in a vowel bearing main stress.
The process applies in the examples in a. but not those in b.
a. Avra [t:]rovato il pescecane ‘He must have found the shark’

La gabbia é [dg:Jia [k:]aduta ‘The cage has already fallen’
Perché [k:]Jarlo non é venuto ‘Why didn’t Carlo come?’
E appena passato con tre ‘He has just passed by with
[k:Jani three dogs’
Devi comprare delle mappe “You must buy some very old
di citta [m]olto vecchie city maps’
La gabbia era dipinta di gia ‘The cage was already completely
[k]ompletamente painted’
Che c’é un perché [k]arlo lo sa ‘Carlo knows that there is a reason’
Ne aveva soltanto tré [d]i ‘He had only three dachshunds’
bassotti
Explain why RS occurs in a. but not in b.

French Liaison
The phenomenon of liaison occurring in French is illustrated in the follow-

ing examples:
il est petit [ilepati] ‘he is small’, petit mec [patimek] ‘small guy’ vs. petit enfant
[patitafa] ‘small child’
In the last sentence, an otherwise latent word-final consonant (shown

emboldened in the example) is attached to the initial vowel of the follow-
ing word.
In the examples that follow, liaison occurs in a. (as marked by an under-
scripted tie bar _ ) but not in b. (as indicated by double slashes //):
a. Ils sont_arrivés [ilsStarive] ‘They arrived’

Des beaux_italiens [debozitaljé] ‘Beautiful Italians’
Ses anciens_amis _[sezasjézamil] ‘His former friends’
Un savant_anglais [€savatagle] ‘A wise Englishman’
b. Ils sont arrivés // en retard ‘They arrived late’
Des maisons // italiennes ‘Italian houses’
Ses amis // anciens ‘His ancient friends’
Un savant // anglais ‘An English scholar’
What is the domain in which liaison occurs?
Greek s-Voicing
In Greek s is voiced when it is followed by a voiced consonant, both in-

side words and across word boundaries, as the following examples illustrate:
[kozmos] ‘people’, [6eliz na pas] ‘Do you want to go?’. Consider the following
examples and suggest what the domain for s-voicing is:
[o petroz den ine ma8imenoz na ‘Petros is not used to eating sea

troi axinuz me psomil] urchins with bread’
[o andras aftos, mu fenete, ‘This man, it seems to me,
ine poli eksipnos] is very bright’
[ekinos o andras, martiz mu oO Geos, ‘This man, God be my witness,
den 8a bi pote sto spiti mu] will never enter my house’
ASPECTS OF LEXICAL REPRESENTATION
UNDERSPECIFICATION, IMARKEDNESS AND FEATURE GEOMETRY
m “Marked” and “unmarkec

m Unmarked values not being
redundancy rules.
We now have a rich autosegmental structure of features, tones and timing

slots or moras, augmented with syllables and metrical grids. We have shown
a variety of processess applying on this structure in the appropriate phono-
logical domains. For some of these processes, Turkish vowel harmony and
Shona tone spread in particular, we assumed some of the feature specifica-
tions to be missing from the lexicon. In this chapter we explore further
and formalize the lexical underspecification of features, with particular
reference to the theories of Radical and Contrast-restricted Underspecifica-
tion. The substance of underspecification is related to universal principles
of markedness, by which certain feature values or combinations of feature
values are preferred. The features themselves are organized in a geometry of
feature dependencies, which has obvious repercussions for underspecifica-
tion. Likewise, one of the two binary values of at least some features could
in principle be redundant, and we examine some empirical evidence in this
respect.
Effects of Strict Cyclicity
Let us consider the pairs in (1):
(1) a. gymnastics b. gymnasium

Caucasus Caucasian
Malthus Malthusian
fantasy fantasia
The forms in (1a) and (1b) are obviously related, but they exhibit several
phonological differences.
Compare, for example, Caucasus with Caucasian. Caucasus has antepenultim-

ate stress. Therefore, the a in the penult must be short underlyingly, as it is
in the surface: if it were not lexically short, it would have attracted stress.
Moreover, the s that follows the a surfaces as voiceless: Cauclas]us. By con-

trast, in Caucasian the corresponding a is diphthongal in the surface, and
the s voiced: Caucleiz]ian (Cauc[er3]ian in many accents after palatalization
of [z], as we will explain in section 11 below and in chapter 18). Similar
observations apply to the other pairs in (1) (the [z] of gymnasium remains
unpalatalized in all accents).
We will discuss the mechanics of vowel lengthening in forms like those

in (1b) in the next chapter, and will at this point focus exclusively on the
alternation between voiceless and voiced s. Prerequisites for voicing in this
set are the bimoraicity of the preceding vowel and the presence of a vocalic
segment after the s, as we illustrate in (2):
500 Aspects of Lexical Representation
(2) a. V:[z]V miser, magnesium, Pusey, cosy, posey, Moses, music

b. VIs]V assembly, misogyny, pussy, russet, Russell, philosophy,
potassium
c. V:[s]C wastrel, pastry, maestro, Easter, acoustic, oestrogen
You will notice that, in order to voice, the s has to be preceded by a long
vowel or a diphthong and followed by a vowel. We formulate the rule of
s-Voicing in line with these facts in (3) (we omit the technical details of the
definition of s to keep matters simple):
‘You 8
practice.
(3) s-Voicing:
s > [+voice] / [-cons]___ [-cons]
Wow (UW)
The forms in (4) look like straightforward exceptions to s-Voicing (the
spelling difference between s and c is irrelevant in the present context):
(4) basin, mason, isolate, Isocrates, isosceles, rhesus,

Croesus, lucid, Lucy, license, recent, recess, decent
Do the forms in (4)

diff i
If you examine the forms in (4) more closely, you will notice an important
point of difference with their predecessors in (1b): in the forms in (1b), the
s occurs at the end of a morpheme, and the requisite following vowel in the
The fact that there next morpheme (the suffix); by contrast, in the forms in (4) the s occurs in
are many more the middle of a morpheme.
exceptions
morpheme-
internally than
between mor-
phemes arouses
our suspicion that
there is something We could of course dismiss this difference as irrelevant. On the other
in this distribu- hand, the fact that there are many more exceptions to s-Voicing morpheme-
tional difference internally than between morphemes arouses our suspicion that there is
something in this distributional difference. The rule can also fail to apply across
morphemes, as in basic or facial (with [f] after palatalization, as we will explain

in chapter 18), but the morpheme-internal cases are more numerous.
We will suggest that what is at work here is the Principle of Strict
Cyclicity that we introduced in chapter 15, and now repeat in (5) as a reminder:
(5) Principle of Strict Cyclicity:

Structure-changing cyclic rules only apply in environments derived in
that cycle, where “derived” = resulting from a morphological process,
or, in some cases, from a phonological change
If s-Voicing is cyclic, it will systematically fail to apply in non-derived envir-

onments, such as those in (4) above. On the other hand, its failure to apply The failure of
s-Voicing to
in some derived environments (basic, facial, etc.) can simply be attributed to
apply in many
the by now familiar fact that phonological rules (in particular, cyclic rules) non-derived
can have lexical exceptions. environments is
accounted for
by declaring the
rule cyclic
The cyclicity of s-Voicing is confirmed by the fact that it never applies in

domains defined by suffixes that we know independently to be non-cyclic, The cyclicity
of s-Voicing is
even when its environment is met (remember that the spelling contrast s vs.
confirmed by the
c is inconsequential for our concerns): fact that it never
applies in domains
(6) mousy placing _ nicish racist racism defined by suffixes
pricy splicing _loosish that we know
icy voicing sprucish independently to
be non-cyclic
juicy sprucing
saucy piecing
ntroduced in
As we showed in chapter 16, cyclic rules may not apply in domains defined
by non-cyclic affixes. Therefore, if s-Voicing is cyclic, its failure to apply in
the forms in (6) will be entirely predictable.
Lexical Underspecification
In (2a) above, s-Voicing has apparently applied morpheme-internally, in

contravention of the Principle of Strict Cyclicity in (5). We provide a few
more such cases in (7):
(7) raisin reason

season Eisenhower
poison Susan
laser quasar
Jesus _ daisy
Joseph aphasia
The ensuing paradox poses an obvious challenge to the analysis we are

proposing.
We shall get round this paradox through the introduction of underspecifica-

Apparent tion in the lexical representations of the forms in (7). We made use of under-
morpheme- specification in our discussion of Turkish vowel harmony and Shona tone
internal applica-
tions of s-Voicing
distribution, in chapters 6 and 14, respectively. In particular, we posited that
are accounted the vowels of Turkish suffixes lack lexical information about their backness,
for through the and that Shona suffixes are lexically void of tonal information. In both cases,
postulation of we obtained the surface representation by spreading the corresponding
underspecifica- features from the root:
tion in the cor-
responding lexical
representations (8) Turkish vowel harmony Shona tone spread
he ot at © yerlinbdasen teng+ ester a
[thigh] [thigh] [thigh] XXXX XX XX X
[dorsal] [dorsal] [dorsal] H
We will now extend and generalize this approach in the context of a fully
fledged THEORY OF UNDERSPECIFICATION, which we shall introduce.
Aspects of Lexical Representation 50:
Suppose that in forms like raisin in (7) the s contains no lexical informa-
tion about [+voice], whereas in forms like basin in (4) it is marked [—voice].
What follows from this? Quite simply, our cyclic s-Voicing rule in (3) will
be prevented from applying in basin and the like, as we already explained:
if it did apply, it would infringe the Principle of Strict Cyclicity in (5), by
changing structure in an environment which has not been derived in that
cycle, indeed has not been derived at all.
By contrast, the application of s-Voicing in raisin and the like will not incur
any such infringement of Strict Cyclicity, because, in the absence of lexical
information for [+voice], s-Voicing will be creating structure rather than chang-
ing it. Creating structure is, of course, quite compatible with the Principle
of Strict Cyclicity (PSC). Creating structure
is quite compatible
with the Principle
of Strict Cyclicity
We illustrate the difference between basin and raisin in (9), where we ital-
icize the lexical representation of s as an informal graphic reminder that its
feature [tvoice] is autosegmentalized:
(9) basin raisin Lexical
[-voice]
Blocked by PSC pia s-Voicing
bal[s]in rai[z]in Surface
You can see that the value of [+voice] is missing from the lexical repres-
entation of raisin. This value is not supplied by spreading a neighbouring
feature, as was the case in Turkish and Shona. Instead, it is supplied by the
independently motivated rule of s-Voicing in (3) above. This approach thus
achieves maximal simplicity: a single rule (naturally encapsulating a single
process) accounts both for such alternations as in Cauca[s]us ~ Cauca[z]ian
in (1), and for the phonotactic skewness of the two segments in the given
environment, manifested in the numerical imbalance between forms like ba[s]in
in (4), incompatible with s-Voicing, and forms like rai[z]in in (7), compatible
with it.
Clearly, the simplest situation is the one where the same rule is active both
to change structure in derived environments and to supply structure in forms
lexically empty for the relevant feature(s). Exceptions to rule application are
Exceptions to rule possible, as we know, although by their very nature we expect them to be few.
application are
On the other hand, it is not unnatural for lexical items to be fully specified.
possible, although
by their very
Therefore, we shall not be unduly surprised at the existence of a substantial
nature we expect number of lexical forms contradicting a rule or process morpheme-internally,
them to be few. even though the forms compatible with it are of course still preferred, as they
On the other are formally simpler.
hand, it is not
unnatural for
lexical items to be
fully specified
Feature Transparency as Underspecification
Lexical underspecification has interesting consequences that go beyond

phonotactics. We have already brought underspecification into our account
of Turkish and Shona assimilation (see (8) above). However, in both these
cases we could have assumed full lexical representations and a structure-
changing assimilation rule: while the approach would have been more com-
plex, it is technically feasible, because the assimilations in question take place
across morphemes, and therefore they cannot be blocked by the Principle
of Strict Cyclicity.
The case from Russian we will present next does require lexical under-
specification.
In Russian, adjacent obstruents have the same value for voice, whether
the cluster is word-internal or straddles two words. The occurrence of the
alternations on the edge of morphemes leads to a construal of the phenomenon
as right-to-left assimilation, as we illustrate in (10) and (11):
(10) a. gorod-a ‘town’ (gen.) b. gorolt-k]-a ‘little town’

no3-a ‘knife’ (gen.) nolf-k]-a ‘little knife’
gotov-a ‘ready’ (fem.) gotol[f-k]-a ‘preparation’
arab-a_ = ‘Arab’ (gen.) ara[p-k]-a ‘Arab’ (fem. nom.)
(11) a. ot ozera ‘from a lake’ bs

ot strasti § ‘from passion’ old]banka ‘from a bank’
ot Pragi ‘from Prague’ old]grexa ‘froma sin’
ot ptits ‘from birds’ old]bdenija ‘from a vigil’
ae bez ozera ‘without a lake’

be[s]strasti ‘without passion’ bez banka = ‘without a bank’
be[s]Pragi ‘without Prague’ bez grexa ‘without a sin’
be[s]ptits ‘without birds’ bez bdenija ‘without a vigil’
The data in (12) show that sonorant consonants neither trigger (cf. (12a)),
undergo (cf. (12b)) nor block (cf. (12c)) voice assimilation:
(12) a. pesna ‘song’ (NB not *pezna)

tri ‘three’ (NB not *dri)
ot nravov ‘from morals’ (NB not *od nravov)
b. on pojet ‘he sings’ (*o[n] pojet]

Mtsensk ‘Mcensk’ (*[m]tsensk)
mstitel’nost ‘vindictiveness’ (*[m]stitel’nost)
c. old]mgli ‘from fog’ (*ot mgli: cf. ot ozera in (11))

o[d]lguni ‘from the liar’ (“ot Iguni)
The data considered lead to the formulation of the Voice Assimilation rule
in (13):
(13) Russian Voice Assimilation:

[-son] wittsson]
[avoi] [Bvoil]
Condition: ”.. .” does not include a syllable peak
In chapter 3 we said that sonorant consonants are normally voiced, and

we would expect this also to be the case in Russian. If sonorant consonants
are voiced, the Voice Assimilation rule in (13) ought to be blocked by inter-
vening sonorant consonants.
In particular, the application of Voice Assimilation across sonorant con-

sonants ought to cause association lines to be crossed, in contravention of
No-Crossing of Lines, which we know is one of the basic principles of
autosegmental phonology (see (17.3) in chapter 14 above):
oN
(14) =—s +5 -s (“s” = sonorant) i/z/ Mtsenska
+v 4+v -Vv (“v” = voice) ils] Mtsenska ‘out of Mcensk’
In order for lines not to cross and the fricative in iz ‘out of’ still to devoice,
we would obviously have to devoice the sonorant first, and only then spread
[-voice] to the preceding obstruent:
G5)" =s° =" = -s +s -s i/z/ Mtsenska
iV eV -v _ ifs] *[m]tsenska
The problem with the output of (15) is that it contains a voiceless sonorant,
contrary to the universal tendency of sonorants to be voiced. The solution
we shall now propose involves lexical underspecification.
All we have to do in order to get both the facts and the representation right
is to assume that sonorants do not have a value for voice specified in the The transparency
lexicon. The TRANSPARENCY of Russian sonorants with respect to assimila- of Russian sonor-
ants with respect
tion (that is, the fact that they behave as if they were not there) now follows
to voice assimila-
automatically — they are indeed not there as far as [+voice] is concerned: tion is explained
if we assume that
(16) PerEsSe=s i/z/ Mtsenska sonorants do not
have a value for
voice specified in
+V ~y ils] Mtsenska the lexicon
The representation in (16) makes clear why Russian sonorants cannot play
any role in voicing assimilation: they are lexically unspecified for [+voicel,
and remain so throughout the derivation.
Bp Underspecification and Markedness
We have just shown that Russian sonorants need to be underspecified for

voice in the lexicon and throughout the derivation. In surface representa-
tion, however, we want them to be voiced, in line with most of the world’s
languages. The discussion in the previous section shows clearly that the sur-
face voicing of sonorants cannot have originated in assimilation: so, where
does it come from?
In chapter 14 we saw that, when there is no underlying tone to associate
with a particular vowel in Shona and Tiv, a DEFAULT tone is supplied to
fill in the gap.
Similarly, it will be reasonable to provide [+voice] to lexically unspecified

sonorants, as formulated in (17):
(17) Sonorant Voicing Default (provisional formulation):

[+sonorant] > [+voice]
(17) does capture the essence of the process of sonorant voicing, but the form-
alization is ambiguous as between a REDUNDANCY RULE, which can only
fill in a gap in the input representation, and an ordinary rule, which changes
something in the input representation. The alternative formalism in (18) con-
veniently keeps redundancy rules visually separate from rules that change
structure:
(18) Sonorant Voicing Default (final formulation):

[ ] > [4voice] /
ae
The empty input matrix in (18) appropriately points to the lack of specifica-
tion for [+voice], and therefore to the structure-building nature of the rule.
Rule (18) gives obvious expression to the universal tendency of sonorants

to be voiced. There are many other universal tendencies affecting many of
the sounds of language. For instance, it was observed by Roman Jakobson,
whom we have already cited in chapters 8 and 9, that the first vowel most
children produce is [a]. Significantly, [a] is also the most common vowel in the
inventories of the world’s languages, and one of the most frequent vowels
in lexical items. These facts suggest that [a] is the most UNMARKED vowel
in language (equivalently, the least MARKED). For consonants, [p] and [Et]
are most unmarked, hence the frequency of such expressions as papa and
tata in child language the world over (also baba and dada, with voicing, and
mama and nana, with nasalization).
The observation
that the sounds
of language
are ranked on
grounds of
The observation that the sounds of language are ranked on grounds of
naturalness has
given rise to the
naturalness has given rise to the THEORY OF MARKEDNESS. The term
THEORY OF “markedness” means that some feature combinations, whether paradigmatic
MARKEDNESS (that is, within the same segment) or syntagmatic (that is, across segments),
are less natural than others, and therefore less likely to crop up in the world’s
languages. The combination [+sonorant, —voice], for instance, is marked para-

digmatically, and the sequences [coronal] + [labial] or [coronal] + [dorsal]
are marked syntagmatically.
The set of all natural, or unmarked, combinations of features makes up

the NATURAL PHONOLOGY of language, that is, the phonology that is The set of all
natural, or
spontaneously present in children in the absence of marked input from the
UNMARKED,
language of their environment.
combinations of
features makes
up the NATURAL
PHONOLOGY
of language
We now provide a selection of the main universal MARKEDNESS

STATEMENTS affecting segments paradigmatically. We formulate these
statements as FILTERS or CONSTRAINTS prohibiting a particular feature
or combination of features:
(19) Some markedness statements:

I Vowels:
a. *[+high, tlow] (= Vowels cannot be high and low at
the same time)
*[-high, —low] (= Vowels cannot be non-high and non-
low at the same time)
*[+low, —back] (= Vowels cannot be low and front at
the same time)
*[+low, tround] (= Vowels cannot be low and round at
the same time)
*[around, —aback, —low] (= non-low vowels cannot have oppos-
ite values for roundness and backness)
*[ao ATR, alow] (= Vowels cannot have the same value
for lowness and ATRness)
II Consonants:
g. *[dorsal]p (= Consonants cannot be dorsal)
h. *[labial], (= Consonants cannot be labial)
i. *[-anterior] (= Consonants cannot be non-anterior)
j. *L+distributed] (= Consonants cannot be distributed)
k. *[+lateral] (= Consonants cannot be lateral)
l. *[+round] (= Consonants cannot be round)
m. *[+continuant] (= Consonants cannot be continuant)
n. *[acontinuant, —astrident] (= Consonants cannot have opposite
values for continuancy and stridency)
III_ Common to vowels and consonants:

0. *[aconsonantal, asonorant] (= The values for consonantality and
sonorancy cannot agree)
p. “*lasonorant, —avoice] (= The values for sonorancy and
voice cannot disagree)
q. *[+nasal] (= Segments cannot be nasal)
r. *[+nasal, —sonorant] (= Nasals cannot be obstruent)
Each of the constraints in (19) expresses a universal prohibition on feature

Each markedness occurrence or co-occurrence.
constraint
expresses a
(usually relative)
universal prohibi-
tion on feature
occurrence or
co-occurrence
Most of the prohibitions in (19) are, however, relative: they only express a
preferred tendency which individual languages respect to a greater or lesser
extent, depending on their degree of markedness. Indeed, only the prohibi-
tions in (19a) and (19r) are assumed to be inviolable, as a matter of sheer
physics.
The rule of Sonorant Voicing Default in (18), which we said above is

ultimately responsible for the underspecification of the feature [+voice] in
the underlying representation of Russian sonorants, is obviously related to
the markedness statement (19p), according to which the values for sonorancy
and voice cannot disagree: the implicational relation [+sonorant] —> [+voice]
that drives the default rule in (18) follows directly from constraint (19p).
Indeed, each of the constraints in (19) automatically gives rise to one or more
implicational statements, in accordance with the laws of logic. In particular, Each markedness
constraint auto-
the logical expression ~[A & B] (“not ‘A and B’”) can readily be developed
matically gives rise
into A > ~B (“if A, then not B”) and B > ~A (“if B, then not A”).
to one or more
implicational state-
ments, in accord-
ance with the
laws of logic
To paraphrase from ordinary life, if we say that it cannot be day and night
(simultaneously), we are effectively saying that if it is day, then it is not night,
and if it is night, then it is not day. The same relations exist in the more
abstract realm of distinctive features, as we have already seen in connection
with sonorant voicing default, and as we will have the occasion to explore
further in the next section.
The Theory of Radical Underspecification
While the idea of underspecification and the motivation behind it are fairly
straightforward, its actual formal implementation has been an object of debate
in the literature, and to a large extent the matter remains unsettled. Here
we shall limit ourselves to a brief survey of the two main positions.
Suppose we take seriously the idea that economy is paramount in the
evaluation of alternative phonological descriptions. This will mean that the
simpler the description (that is, the fewer elements it contains), the more
highly valued it will be. This perspective, therefore, validates our early pro-
posal to leave English [z] lexically underspecified for voice in environments
that match the s-Voicing rule: this saves on [+voice] entries, even allow-
ing for “exceptions”, which will be lexically encoded as [-voice], as we
explained.
If we take the idea of economy to its limit, we will want to save as many
lexical features as is compatible with the differentiation of lexical entries.
Obviously, we will have to draw the economy line at a point before the dif-
ferences between lexemes are lost: in the extreme case, a lexical inventory
with no feature specifications at all will fail to do the very job for which it
is intended, namely, the differentiation of lexical entries.
The theory of RADICAL UNDERSPECIFICATION takes this programme fully

The theory of on board. We will present this theory through a hypothetical example:
RADICAL UNDER-
hypothetical examples have the advantage of simplicity relative to their real-
SPECIFICATION
takes the idea of
world counterparts.
lexical economy to Suppose a language has the vowel inventory /i, e, a, 0, u/, which we know
its limit: we will is not at all unusual among the languages of the world. The full specifica-
want to save as tion of these vowels requires information on the values they take for the fea-
many lexical features [thigh], [+low], [+back], [tround] and [+ATRI, as represented in the
tures as is com-
table in (20):
patible with the
differentiation of
lexical entries (20) Cm An Ome
High + - —- —- +
Low - - + - =
Back Se
Round -—- —- — + +
CR ete et
The universal markedness statements in (19) above allow us to cut down

the amount of information in this table, as in (21):
(21) PREM Ao TH
High + - Op
Low - + -
Back - = + +
Round
ATR
The reason the simplifications in (21) are possible is that the empty cells can
eventually be filled in by the effect of the implicational statements in (22),
which are directly derived from the corresponding constraints in (19):
(22) A set of default rules for vowels:
a. [] > Lhigh] | iMesdiaaws ‘| |(em1985

+low thigh
QT 259 acid /| | (from (19¢))

+low
or Ts round (| (from (19d))

: +low
e. []— [around] / (from (19e))

aback
—low
f. []— [aATR] / =] (from (19f) )

—alow
Constraint (19b), which disfavours mid vowels, is obviously not operative

in this language, which does have /e/ and /o/. Notice, importantly, that the
default rules in (22) incur no cost, since we are assuming that the markedness
statements they interpret are part of universal grammar. The end result is,
therefore, a considerable simplification of the lexical representation of all the
vowels, as (21) attests.
Radical Underspecification does not stop here in its endeavour to rid lex-
ical representations of unnecessary clutter. In order to meet this goal fully,
an imaginative step is taken next. Suppose that we take all feature values
out of one of the vowels, say, /i/, for the sake of argument. The new, fur-
ther simplified table will be as in (23):
(23) L260 a) EO a
High ee eat
Low - + -
Back — + +
Round
ATR
Following the logic of the strategy, we will have to admit that table (23) still
contains too much information. Specifically, if the values [+high], [-low],
[-back], [round], and [-ATR] have been completely eliminated from the
/i/ column on grounds of redundancy, logic dictates that they are also redund-
ant everywhere else. We give the totally redundancy-free representation in
(24), where, crucially, all the segments are still kept apart, since each pair is
differentiated by the value of at least one of the features:
(24) ire yao,

High ~ -
Low +
Back + +
Round
ATR
Each feature value

that is missing
from the maxim-
ally underspecified
lexical segment The underspecified entries in (24) correspond of course to lexical repres-
automatically gives entation. At the surface, however, we are assuming that all the features need
rise to a COM-
to be specified to enable segments to be realized phonetically. How can this
PLEMENT RULE
that eventually
be achieved if (24) contains a number of blanks? The answer is that each
supplies the miss- feature value that is missing from the maximally underspecified segment
ing feature value (here /i/) automatically gives rise to a COMPLEMENT RULE that eventually
supplies the missing feature:
(25) Complement rules for maximally underspecified /i/:

[] [thigh]
[] [-low]
[] [-back]
[| [-round]
[] bel
ildy[+ATR]
The application of the complement rules in (25) and the default rules in
(22) will ensure that all the feature values are represented in the surface:
the surface values will in fact be identical to the values in the table in (20)
above.
Two questions remain outstanding. One concerns the timing of the applica-
tion of redundancy rules, and we will provide the answer in the next section.
The second question concerns the criterion for the selection of one segment
as fully underspecified in the lexicon. In our present example, we simply
chose /i/ at random, but clearly a principled criterion is required.
The solution provided by Radical Underspecification is grounded in one
specific empirical fact: in many (perhaps all) languages, one of the segments
in each major class behaves asymmetrically, in that it alone systematically In many (perhaps
appears in contexts of epenthesis, or in that it alone triggers, fails to trigger, all) languages, one
of the segments in
or is pervasive in, certain rules. For example, in a vowel system like the
each major class
one we have been discussing (similar in fact to the system of Japanese), [i] behaves asymmet-
will systematically turn up as the epenthetic vowel, will be transparent to rically: its selection
harmony, and so on. as maximally
underspecified fol-
lows automatically
Among consonants, coronals have been found to exhibit such asymmetries

in many languages, confirming the universal unmarked status given to them
in (19).
Once one segment in the relevant class singles itself out by such skewed
behaviour, its selection as maximally underspecified follows automatically,
both as a matter of theory-internal congruence and on account of the accur-
acy of the empirical predictions made by the model, in particular with regard
to transparency or invisibility.
Explain in your ow
segment as undersp
the theory ofRadical UI
[4 Problems for Radical Underspecification
The data in (26), from the Australian aboriginal language Gooniyandi,

There are anum- —_ prove problematic for the Radical Underspecification approach:
ber of problems
with Radical :
Underspecification (26) a. duwu ~ duwu cave
langija ~ langija ‘midday’
b. diripindi (*ditipindi) ‘he entered’

dili (*dili) ‘flame, light’
What is notable about
(26a) shows that word-initial coronals are freely realized as [+anterior]

([d], [1]) or [-anterior] [qd], [l}).
(26b) shows that such free variation is suspended when the word-initial cor-
onal is followed by another coronal in the same word — this second coronal
transmits its value for [+anterior] to the word-initial coronal by assimilation:
Aspects of Lexical Representation SAL,
7) tindata Miglissniadins APelwi
[cor] ali [cor] [cor]
[-ant] [tant]
In (27) you can see that word-initial coronals are indeed lexically unspeci-
fied for [+anterior]. In the absence of a [anterior] segment medially in the
same word, such lack of specification gives rise to the phonetic fluctuation
illustrated in (26a) — this obviously means that the markedness statement
(19i) does not apply in Gooniyandi, and therefore that Gooniyandi lacks a
default rule supplying a value for the feature [+anterior].
in this case.
Word-medially, no such fluctuation between anterior and non-anterior

coronals exists. Indeed, we are seeing that the lexical [tanterior] value of
Gooniyandi word-medial coronals is transmitted to their unspecified word-
initial counterparts. The conclusion is obvious: Gooniyandi word-medial coro-
nals are fully specified in the lexicon.
fully specified underlying i Gooniyandi
This result contradicts the prediction of Radical Underspecification that one

of the two complementary values ([+anterior] or [—anterior]) will be lexically
empty, and subsequently filled by the corresponding complement rule, or
by a default rule along the lines of (28), which interprets the markedness
statement in (191):
(28) []-— [+anterior]/

coronal
How does (28) interpret (19i)?Why should onesofthe values

of [anterior] be absent from the =
There are other problems with Radical Underspecification. The default rule
in (18), repeated now as (29), was pivotal in the analysis of the Russian voice
assimilation process in section 3:
(29) []- [+voice] |

+sonorant
We saw then that the existence of this rule allows sonorants to be lexically
unspecified for [+voice], thereby failing to playa role in the process of regres-
sive voice assimilation that affects obstruents. Crucially, however, in order
for this to be so, we need to stipulate that the Russian rule of Voice Assim-
ilation applies before the default rule in (29).
Remind yourself v
S SSSA ~
(29).
This result is unsatisfactory for two reasons. First, it is conceptually undesir-

able to subject default rules, inspired in universal markedness statements,
to language-specific ordering.
Second, ordering Voice Assimilation before (29) is incompatible with the

The Redundancy REDUNDANCY RULE ORDERING CONSTRAINT in (30). The Redundancy
Rule Ordering
Rule Ordering Constraint is a universal principle of Radical Underspecifica-
Constraint predicts
the ordering of a tion that predicts the ordering of redundancy rules, default rules included:
redundancy rule
immediately (30) Redundancy Rule Ordering Constraint:
before the first A redundancy rule applies immediately before the first (ordinary) rule
(ordinary) rule that refers to the output of the redundancy rule
that refers to the
output of the
redundancy rule
The ordering of Russian Voice Assimilation before (29) is incompatible

with the Redundancy Rule Ordering Constraint in (30), because the Voice
Assimilation rule in (13) above makes reference to [+voice], and therefore
by the Redundancy Rule Ordering Constraint the redundancy rule in (29)

must apply first. However, if (29) applies before Voice Assimilation (13), the
account of assimilation we proposed above collapses:
(31) -s +s -s i/z/ Mtsenska
+v
ea
+v -v i*[z] Mtsenska
As shown in (31), the value [+voice] of /m/ now intervenes between the
source and the putative target of Voice Assimilation, and consequently the
process gets blocked (cf. also (15) above).
Note that the motivation for the Redundancy Rule Ordering Constraint is
overall very strong in the general context of Radical Underspecification Theory.
In particular, the Redundancy Rule Ordering Constraint is needed to pre-
serve feature binarity at the point where structure-changing rules apply: in —_The Redundancy
the absence of the Redundancy Rule Ordering Constraint, feature binarity Rule Ordering
would be jeopardized by the use of underspecification, which effectively makes Siege te
available a third value for features, [OF], beyond the two binary values [+F] ceive feature Bina.
and [-F] (the full demonstration of this point is rather technical, and best tity at the point
side-stepped in the present context). where structure-
changing rules
apply
A further problem area for Radical Underspecification concerns the

identification of the maximally underspecified segment with the epenthetic
segment. First, non-epenthetic null segments do exist: segments present in
lexical representation (and thus not epenthetic) but still underspecified.
Explain ttfhedifference be
Indeed, we have seen that complete underspecification of one segment per

class is basic to the Radical Underspecification machinery.
The mutual independence of the pattern of underlying underspecification
and the choice of epenthetic segment is not only logical, but also backed by
fact.
For instance, in Basque (a non-Indo-European language of the corner of

the Bay of Biscay) the genitive indefinite plural suffix is made up of an
unspecified vowel followed by n: -Vn. We can reasonably assume that
the quality of this vowel is unspecified because it is filled in by spreading
from the adjacent vowel: mendi ‘mountain’ gives mendiin ‘of mountains’,
asto ‘donkey’ gives astoon, and so on. At the same time, Basque has an
epenthetic vowel e, manifested in such forms as giSonek ‘a man’, from gison
‘man’ + k (indefinite suffix): compare mendik ‘a mountain’, astok ‘a donkey’,
etc., with no vowel. Clearly, this epenthetic e is distinct from the maximally
underspecified vowel, the unspecified vowel of -Vn. Similarly, in Mohawk
(a native language of North America), the epenthetic vowel i is distinct
from its maximally unspecified vowel [A], manifestly asymmetric with the
remainder of the inventory /i, e, a, 0, u/. And so on.
Next, some languages have more than one epenthetic segment in any
one sound class: Hindi, for instance, the main language of India, has two
epenthetic vowels: i before s + obstruent clusters, and schwa elsewhere.
Finally, in some languages one and the same surface segment can origin-
ate as lexically specified or as lexically underspecified, as attested by the
behaviour of the segment. This is the case, for instance, with the Basque e:
the e in gisonek is epenthetic, as we have just seen, but the e of the genitive
plural definite -en is lexical. In particular, the e of -en triggers raising in a
preceding vowel (asto + en = astuen ‘of the donkeys’), as do other vowels

(asto + a = aStua ‘donkey’ (abs. sg.); asto + ok = aStuok ‘donkey’ (abs. pl. prox.)):
if so, the e of -en cannot be underspecified.
Contrast-Restricted Underspecification
In the face of all the difficulties we have just mentioned, it is obviously well
worth exploring an alternative formalization of underspecification. One such
alternative is based on the tenet that only features that implement lexical
contrasts have both their values lexically specified in the relevant environ- CONTRAST-
RESTRICTED
ment. For instance, in the Gooniyandi case discussed above, [tanterior] is
UNDERSPECIFICA-
lexically contrastive word-medially, and consequently it would need to TION is based
be fully specified in this position. This prediction agrees with the facts, as on the tenet that
we saw. only features that
implement lexical
contrasts have
both their values
lexically specified
in the relevant
environment
Similarly, [+voice] is lexically contrastive in Russian obstruents, but not
in sonorants. Therefore, Russian sonorants can be left underspecified in the
lexicon, but not so obstruents, again in agreement with the facts.
The specification of both values of the same feature in the same environ-
ment undermines the basic tenet of Radical Underspecification that attain-
ment of lexical economy is paramount.
In particular, while still favouring lexical economy (the values of features

which are not contractive in the lexicon are still left underspecified, and filled
in by the familiar default rules derived from the markedness statements),
the alternative approach we are presenting gives priority to the explicit ex-
pression of lexical contrast over the attainment of radical lexical economy,
supposedly on empirical grounds. This alternative approach to underspe-
cification can transparently be referred to as CONTRAST-RESTRICTED
UNDERSPECIFICATION, where the expression “contrast-restricted” must
be interpreted as “restricted by contrast” — crucially, not “restricted to con-
trast”, as the more common label CONTRASTIVE UNDERSPECIFICATION
paradoxically appears to suggest. We spell out the basic principle of Contrast-
restricted Underspecification in (32):
(32) Contrast-Restricted Underspecification:

In Contrast- Feature values are left unspecified in the lexicon if they are predict-
Restricted able from the pattern of distributional neutralization in the language
Underspecification
feature values are
left unspecified in
the lexicon if they
are predictable
from the pattern
of distributional
neutralization in
the language We will see below that Contrast-restricted Underspecification is not empir-
ically problem-free either. Before going into this matter, however, we need
to be fully conversant with the way distinctive features are organized, and
accordingly we turn to this matter in the next section.
FE] Feature Dependencies
In chapter 4 we saw that the features [tdistributed] and [tanterior] sub-

classify sounds which are [coronal] (that is, articulated with the blade of
the tongue). It follows as a matter of logic that the two features in ques-
tion will simply be absent from sounds which are exclusively [labial] or
[dorsal].
Similarly, the feature [tround] is restricted to [labial] sounds, and does not
occur with sounds which are exclusively [coronal] or [dorsal]. Finally,
[thigh], [4low] and [+back] are restricted to [dorsal] segments, and will not
turn up with sounds which are exclusively [labial] or [coronal].
As we have been hinting, all these restrictions derive from the fact that
the binary features in question introduce finer divisions in the class defined
by the feature to which they are circumscribed: they define subsets within
a set. Dependent
features define
subsets within
a set
In chapters 4 and 6, we formalized the situation we are describing by means

of feature dependencies:
(33) a. [labial] b. [coronal] ce [dorsal]
[+round] [+distr] [tant] [thigh] [tlow] [+back]
The linking lines in (33) are genuine autosegmental lines, denoting the
simultaneous timing of the individual distinctive features.
As we pointed out in chapter 4 and more than once since, this autosegmental
formalism allows a very simple formalization of assimilation and dissimila-
tion processes.
[Feature Geometry
There are other dependencies among features beyond the ones we have men-
tioned. Indeed, the full set of features makes up a web of dependencies, which
we formalize in (34). This web of feature dependencies is conventionally
The full set of referred to as FEATURE GEOMETRY:
features makes
up a web of
dependencies (34) [+cont]
referred to [+strid]
as FEATURE
[+nasal]
GEOMETRY
[+lat]
[+voice],
[round] —————— [labial],
[+dist]
[tant] SS [coronal],
tcons
[thigh] +tson
[+low] > [dorsal],
[+back]
be OR acca
You can see in (34) that the root of the feature tree (rotated by 90 degrees
from the more usual display to make it more manageable graphically) is
made up of the conjunction of the features [+cons] and [+son]. These are the
most fundamental of all features.
The substantive claim behind the inclusion of [+cons] and [+son] in the tree
root is thatthese two features do not exhibit autosegmental behaviour inde-
pendently of the other features: when [+cons] and [+son] spread, for example,
they drag along the remainder of the features (= total assimilation). This hap-
pens, for instance, with the English prefix in- before sonorants: in-legal >
illegal, in-regular > irregular, but in-ability, in-capable, im-pertinent, and so on:
(35) English total assimilation of nasals:

o o
{
+cons] [+cons
+son} | +son
[+nas] [cor]
For convenience and for historical reasons, the root of the feature tree is some-
times simply referred to as the ROOT, often abbreviated to R.
Among the features dependent on the root, some are direct dependents,
and others indirect dependents, in that further dependents intervene between
them and the root.
Once more you must bear in mind that the only motivating factor for the
specific geometry in (34) is the autosegmental behaviour of each feature. The motivating
factor for the spe-
cific geometry is
the autosegmental
behaviour of each
feature
So, the claim behind the direct dependency of [+cont], [+strid], [+nasal] and
[tat] on the root is literally that the root features [+cons, tson] are the only
features that are superordinate to [tcont], [tstrid], [nasal] and [+lat], which
are otherwise independent.
The situation is similar for [tvoice] (in fact a shorthand for a set of other
features, to be introduced in the next section), and for the place features [labial],
[coronal], [dorsal] and [radical], each of them with its respective dependents,
already commented on.
~ Again, say what each of [lab coronal], [dorsa>l] and [radic:

oe
of.
is independen t
Notice the subscripts L (on [tvoice]) and P (on the place features), which
allow us to refer to the whole class thus designated: [ .. . ]p for the class of
place features, and [... ], for the class of laryngeal features, at the moment
only with one member. The label LARYNGEAL obviously means ‘of the
larynx’, the space where the corresponding sound(s) are articulated.
You will remember that the subscript formalism signals membership of

a common class: features sharing a given subscript exhibit parallel phono-
Features sharing a logical behaviour. For instance, we saw in chapter 4 that English consonants
given subscript
transmit their place of articulation feature, whichever this is, to a preceding
exhibit parallel
phonological
coronal nasal.
behaviour
Class Nodes
For historical reasons, our subscripting practice is not common in the liter-
ature, which instead gives autosegmental representation to our subscripts
in the form of CLASS NODES, as in (36), where we have emboldened the
two class nodes to make them stand out visually:
(36) [+cont]
[+strid]
[+nasal]
[+lat]
[+voice] —_———————————_ Laryngeal
[+round] [labial]
[+dist]
[tant] = [coronal]
ees
Place +son
[thigh]
[+low] => [dorsal]
[+back]
[+ATR] ———— [radical]
You will notice that we have not enclosed “Laryngeal” and “Place” in
square brackets. This omission is quite deliberate, to signal that these labels
do not correspond to distinctive features.
The intercalation of class node labels in what is otherwise a network of

autosegmentalized features is conceptually a step of questionable legitimacy.
In particular, the practice enshrined in (36) grants autosegmental status to

“Laryngeal” and “Place”. These, labels, however, have no individual refer-
ence, as we know. Instead, they define classes of individual features, a func- “Laryngeal” and
“Place” have no
tion which is arguably better reflected in the subscript notation.
individual refer-
ence: they define
classes of indi-
vidual features
Be that as it may, the class node notation entails no loss of expressive power:
nasal assimilation in (37a), for instance, is in all equivalent to our rule with
subscript notation of chapter 4 above, repeated in (37b) to make comparison
easy:
CA a +cons +cons b. +cons tial

+son —son +son —son
[+nas] Place Place [+nas] [coronal], [...]
|
[coronal]
You can see that both notations capture the generalization that coronal nasals
assimilate in place of articulation to the following obstruent irrespective of
the place of articulation of this obstruent.
Throughout the book so far, we have been using the feature [tvoice] to
formalize the voiced or voiceless nature of segments, and indeed this is the
only feature we included in the laryngeal class in (34) and (36) above. Besides
voice, however, the larynx is responsible for the production of aspiration and
glottalization. We give the features responsible for these various activities
in (38), with their respective values:
4
|
(38) Aspiration Glottalization Voice

[spread glottis] + =
[constricted glottis] =
[stiff vocal folds] *
[slack vocal folds] +
The phonetic content of each of the four laryngeal features is expressed with
reasonable transparency by the respective labels — for instance, SPREAD
GLOTTIS means that the glottis is spread open, and so on. The full tech-
nicalities associated with these features are rather complex, and best left out
of the present discussion. Just as a for instance, the exact implementation of
voice is contingent on the specific type of segment, the values included in

(38) only being valid for obstruents. There is also a connection between voic-
ing and low tone, and between voicelessness and high tone. To keep mat-
ters simple, however, many authors, including ourselves, refer to voice
activity with the relatively informal feature [+voice], instead of relying on There are four
the more complex set of features in (38). laryngeal features,
but many authors
refer to voice
activity with the
relatively informal
hend the general riftof the larynges
eal features. feature [+voice]
Relations between Vowels and Consonants
One important issue in feature geometry concerns the relationships

between features for vowels and features for consonants. In particular, it is
not uncommon in the languages of the world for the place of articulation
of a consonant to be affected by the place of articulation of the following
vowel. For instance, front high vowels often palatalize a preceding coronal
consonant, as illustrated in (39) for English:
(39) impress — impre[flion

race — ralflial
right — righ[tf]eous
Christ — Chris[ffJian
It would obviously be desirable to express a process such as the one in

(39) as assimilation. This goal is, however, unattainable in the context of It is desirable to
the feature geometry in (36) above, where front high vowels are defined as express the influ-
ence of the place
[dorsal, +high, —back], while palatal consonants are [coronal, —anterior]: clearly,
of articulation of
there is no formal connection between these two sets of features. a vowel on the
place of articula-
tion of the pre-
ceding consonant
as assimilation.
This goal is
achieved if vowels
and consonants
The situation we just described could be interpreted as an indictment of share the relevant
the feature geometry in (36), and indeed an alternative is available in the set of features
literature, the relevant aspects of which are displayed in (40). As usual, dif-
ferent line heights are intended to suggest different tiers, each obviously defin-
ing a different plane:
(40) Unified place features for consonants and vowels:

Consonants Vowels
+sonorant +sonorant
tapproximant +approximant
—vocoid ee
Oral cavity Oral cavity
C-place a
Vocalic
Aperture
[labial] ee
-pl papa
[dorsal] [labial]
[coronal] [dorsal]
teantl [coronal]
[+distr] [+ant]
[+distr]
The key difference between the feature geometry in (40) and its predecessor
in (34)/(36) lies in the fact that in (40) vowels and consonants share most of
their features, assumed to receive a slightly different phonetic interpretation
depending on whether they are dominated by the C-place or the V-place
class node.
The system in (40) includes a new feature [+open], to express degree of

tongue height in vowels. This feature is assumed to be recursive, reflecting
the open-ended nature of vowel height.
The various
manifestations of
[open] are
gathered under
the class node
Aperture The various manifestations of [topen] are gathered under the also new class
node Aperture:
(41) Vowel height under the class node Aperture:

/i, u/ /e,0/ /a/
Aperture Aperture Aperture
Pope TANG apEGNIOY Bape Re
[-open] [+open] [+open]
Notice the similarities between this proposal and the one based on the three
quantum vowels that we discussed in chapter 8 in the context of the English
Vowel Shift. In particular, [+open] in (41) could be claimed to correspond
to <a> in the quantum vowel proposal, and [—-open] to .
The existence of various alternative proposals for feature geometry is clear

even from our brief exposé here. The question now is which of these altern-
atives is correct. The answer, unfortunately, is that the matter is still un-
resolved, since each of the models presented (and others we have ignored)
has advantages and disadvantages. The issue of distinctive features is of course
extremely complex, not least because it constitutes the interface of phono-
logy with phonetics, and much more research is needed before the matter
is settled.
Redundancies between Features
The existence of a feature geometry tree (in whatever incarnation) introduces

an additional source of redundancy beyond those discussed earlier in the There are redund-
ancies in the
chapter.
feature geometry
tree as a result
of the feature
dependencies
Think of the contrast between retroflex and non-retroflex coronals in

Gooniyandi we came across in section 6: compare, for example, [d] vs. [d]
in diripindi ‘he entered’. We know that the two types of segment are dis-
tinguished by their opposed specification for [+anterior]: retroflexes are
[—-anterior] and non-retroflexes [+anterior]. These values need therefore to
be specified in the lexicon for each of the respective segments. The question
is, do we have to specify the rest of the feature geometry?
The answer is contained in the feature geometry itself. You know, in par-
ticular, that the geometry contains lines linking some features, but not all
of them. For instance, in the case of [d] and [d] there is no need to specify
[coronal] in the lexicon, for the simple reason that if a segment has a value
for [+anterior] then it must of necessity be [coronal]. On the other hand, the
presence of a value for [tanterior] says nothing about the possible value of,
say, [tvoice] or [tcontinuant].
The redundancies contained in the feature geometry tree in (36) above as

a result of feature dependencies are now displayed in (42) (“>” = ‘implies’,
and is thus equivalent to the “>” we have been using in default rules):
(42) Feature geometry implicational redundancies:

[+voice] > Laryngeal > [tconsonant, +sonorant]
[+spread glottis] > Laryngeal > [+consonant, +sonorant]
[+constricted glottis] > Laryngeal > [+consonant, tsonorant]
[+stiff vocal folds] > Laryngeal > [+consonant, tsonorant]
[+slack vocal folds] > Laryngeal > [+consonant, tsonorant]
[+round] > [labial] > Place > [+consonant, +sonorant]
[+dist] > [coronal] > Place > [+consonant, +sonorant]
[tant] > [coronal] > Place > [+consonant, +sonorant]
[thigh] > [dorsal] > Place > [+consonant, +sonorant]
[+low] > [dorsal] > Place > [+consonant, +sonorant]
[+back] > [dorsal] > Place > [+consonant, +tsonorant]
[tATR] > [radical] > Place > [+consonant, +sonorant]
[+lat] > [tconsonant, +sonorant]
[+nasal] > [+consonant, tsonorant]
[+continuant] > [+consonant, tsonorant]
[+strident] > [tconsonant, +sonorant]
The implications in (42) are of course simple reformulations of the tree in (36).
Notice, in particular, that the set of implications in (42) includes class nodes
(Laryngeal, Place) as well as features. From a more substantive perspective,
however, class nodes are not implied, but, rather, contained in the relevant
features, in a manner best expressed by our earlier subscript notation.
The removal of class nodes from (42) would obviously reduce the implica-
tional set to cases involving place features and to the across-the-board
implication of the root features.
The implications in (42) (with or without the class nodes) are clearly dif-
ferent from the implications of the markedness relations discussed earlier.
In particular, the implications in (42) are implicit in the feature geometry

tree, and therefore need no separate statement: we have only provided the The implications in
statements in (42) to make the exposition clearer. By contrast, markedness the feature geo-
metry tree need
statements must be specifically included in the grammar (ideally, in universal
no separate state-
grammar). ment. By contrast,
markedness state-
ments must be
specifically
included in uni-
versal grammar
In addition, while feature geometry implications are inviolable by their very

nature, most markedness implications are violable, as we know (the excep- While feature
geometry implica-
tion are cases considered of physical necessity, such as *[+high, +low] or
tions are inviolable
*[+nasal, -sonorant]). All this points to a dichotomy between “hard” uni- by their very
versals (inviolable) and “soft” universals (violable), the latter better formal- nature, most
ized as parameters. markedness
implications are
violable
Privative Features
A different type of implication (and thus of redundancy) is contained in the

binary notation itself: [wF] automatically implies not [-oF], and conversely.
Mathematically, if [«F] = ~[-oF], and [-aF] = ~[aF] (“~” is the logical sym-
bol for “not”), it may be possible to dispense with o altogether and reduce
the contrast to the presence of the unary feature [F] versus its absence.
Empirically, the strongest case against binarity would come from the demon-
Empirically, the stration that one of the values of a feature plays no role in the phonological
strongest case
system. This seems to be the case to different degrees of likelihood for the
against binarity
would come from
settings [—nasal], [-round], and others.
the demonstration Evidence against the existence of [—round] in the context of Contrast-
that one of the restricted underspecification is available from the round harmony system
values of a feature of Khalkha Mongolian, a language already cited in connection with stress
plays no role in
in chapter 13 above. Khalkha Mongolian reputedly possesses the relatively
the phonological
system
simple vowel system in (43) (there is some dispute as to the real phonetic
substance of [y] and [g] in Khalkha Mongolian, notwithstanding the phon-
etic symbols, which we will have to take at face value here):
(43)
Oe
Rounding harmony in Khalkha Mongolian only affects non-high vowels: it

does not affect [i, y, u]. The process involves rightward spread of the fea-
ture [labial] attached to [+round] from the initial vowel. In keeping with the
phonetic symbols /y/, /a/, we shall assume there is also [tback] harmony
in this language, in a manner similar to Turkish:
(44) sons-ogd-ox ‘to be heard’

erg-egd-ex ‘to be raised’
cf. nee-gd-ex ‘to be opened’

Intervening high round vowels prevent the spread of [labial, +round] past
them - more technically, high round vowels act as harmony BLOCKERS:
(45) boogd-uul-ax ‘to hinder’
This state of affairs receives a straightforward formalization in our model

of features:
U6) ib we ed sur Pa x
[-high] [+high] [-high]
\
Place
fous
Place Place
faecrvienfi
[labial] [labial]
[+round]
a
[+round]
The presence of the [thigh] /u:/ between the source /o:/ and the possible
target /a/ blocks the rightward spread of [labial, tround].
Crucially, /i/ does not behave as a blocker:
(47) oril-ox ‘tO weep’

oril-ogd-ox ‘to be wept’
This outcome only makes sense if /i/ is unspecified for [labial], and hence
for [-round], since otherwise [labial, -round] would interfere in the propaga-
tion of [labial, tround] in precisely the same way as [labial, tround] did
in (46):
(48) ojsicrmill
ii eels Of fx
[-high] [+high]__ [-high]

mhtypeofapts
Place
thd anclhy
Place Place
[labial]
nN
[+round]
In the context of the theory of Contrast-restricted Underspecification,

however, the underspecification of Khalkha Mongolian /i/ for [tround] is
problematic. In particular, in the Khalkha Mongolian system, /i/ contrasts
minimally with /y/ with regard to [tround], and therefore it ought to be
fully specified for this feature in the lexicon.
ically specified?
In turn, if /i/ is lexically [round], it is by implication also [labial], since

[+round] is a dependent of [labial]:
(49) O it i ] oOwex«
[-high] [thigh] [-high]
A ae
nesonyhata
Place Place Place
fie
[labial] [labial]
[t+round] [-round]
However, if /i/ is [labial, -round], [labial, +round] will not be able to spread
from the o of the root to the underspecified o of the suffix, again for the
simple reason that such spreading will violate the no-crossing constraint:
(50) aa ET 3h 5X
[-high] [+high] —_[-high]
[dorsal] [dorsal] ao
Ne
Place
Place __ Place
ere
[+round] _[-round]
From all this we could of course conclude that Contrast-restricted Underspe-

cification needs to be abandoned. However, we have seen above that Radical
Underspecification, its main alternative, does not necessarily fare much better.
The Khalkha Mongolian problem is overcome if [round] is formalized
as monovalent, that is, if the absence of rounding does not have a formal
representation:
(51) oO r i 1 OEE
[-high] __[+high] [-high]

\
Prd.Place o-anh
ipieee
Place. Place
V o
fiabiall
\
[round]
Arguments are
being put forward
in the literature
for the mono-
valency of other
Similar arguments are being put forward in the literature for the mono-
distinctive fea-
valency of other distinctive features, although more work is necessary to tures, although
dispel all doubts and allow wider acceptance across the field. For the time more work is nec-
being, the question of how many features (if any) are monovalent must be essary to dispel all
added to the not inconsiderable list of issues in the area of distinctive fea- doubts
tures that still await a definitive solution.
and then in chapter 14 in the on

explanation for the application o
voicing to underived forms as wel
that the lexically unspecified vai .
can be predicted by universal mi ednes
straints. We presented two

t
specification and Contrast-r
| methrown up by the theories of unde

r deal ith by a theory that recognizes
Key Questions
How can underspecification help to 9 What is Contrast-restricted Under-

explain the fact that some cyclic specification? How does it differ
rules apparently apply in underived from Radical Underspecification?
contexts? 10 What motivates Feature Geometry?
What is the difference between 11. “Laryngeal” and “Place” are class
structure-changing and _ structure- nodes: what does this signify?
building rules? 12 How is a type of configuration in
Why are some sounds transparent which vowels and consonants share
with respect to processes such as a feature specification for place of
assimilation? articulation more revealing in some
What is a “default rule"? contexts?
What is meant by marked and un- 13. How do the redundancies inher-
marked feature combinations? ent in the Feature Geometry tree
What observation lies behind the result?
Theory of Radical Underspecifica- 14 On what grounds is it suggested
tion? that some features apart from
What is the role of a complement place of articulation are in fact
rule? What does is complement? monovalent?
What is the Redundancy Rule Order-
ing Constraint? Does it always apply
unproblematically?
Further Practice
Bahasa Melayu/Indonesia Nasal Spread
The Austronesian language spoken in the states of Indonesia, Malaysia,

Singapore and Brunei, and in some neighbouring areas, is known as Bahasa
Indonesia in Indonesia, and as Bahasa Melayu elsewhere. In this language,
the appearance of nasal vowels is totally predictable and non-distinctive.
Nasalization spreads both within morphemes and across morpheme
boundaries.
[makan] ‘to eat’ [kasama?an] ‘similarity’

[malam] ‘night’ [kaméwahhan] ‘prosperity’
[taman] ‘garden’ [kematijan] ‘death’
[naé?] ‘to ascend’ [kasonijan] ‘art’
[madt] ‘died’ [mdnaja?] ‘sift’ (active)
[mawat] ‘fit’ [penaja?an] ‘enriching’
(i) | State the formal mechanics of the spread of the nasal feature.
(ii) Where and why does this spreading stop?
(iii) Provide the nasalization patterns for the following forms:
penawallan ‘guarding’ mahal ‘expensive’ manikot ‘follow’ (active)

majan ‘stalk’ (palm) mewah ‘luxury’ rambottan ‘rambutan’
kenallan ‘friend’ tanam ‘to bury’ makannan ‘food’
(iv) What light do these data shed on the characterization of consonants

for place of articulation?
Markedness
Each of the sets of sounds below would be an “unnatural” system for lan-
guage. State which of the markedness statements are contravened in each
case:
(i) Vowel system (iv) Fricatives (vii) Consonant system

fend te Tay: @- Pi Se tx
ze Gar. Bis ve ga e
Coats ]
(ii) Vowel system (v) Vowel system (viii) Vowel system
feat y w 2.40
D Clas 2050
omen a
(iii) Stop consonants (vi) Stop consonants
DFek nm y
b 8 n m y
Yoruba Vowel Assimilation
The oral vowel system of standard Yoruba is as listed below (the language
also has three nasal vowels which we shall ignore for the purposes of this
exercise):
Work out the full feature specification for these vowels.
Vowel assimilation in Yoruba is illustrated in a. (tones are omitted):
a. owotade owa ade

‘Ade’s money’
Owo+0jo OWO Ojo
‘Ojo’s money’
owo+omd ow) om3d
‘child’s money’
owo+emu owe emu
“wine money’
owo+epu owe epu
‘oil money’
awo+ejo ‘colour of a snake’
awe ejo
ara+oke aro oke
‘northern Yoruba’
ebat+odo ebo odo
‘near the river’
atd+oogun ato oogun ‘medicine dispenser’
ile+tayo ila ay ‘Ayo’s house’
arat+ebun are ebun ‘Ebun’s body’
aratejide tee
+Peat
eee
are ejide ‘Edjide’s body’
(i) Formalize the assimilation process.
Now consider the examples in b.:
b. aratilu *ariilu ‘townsman’

erutigi *erl igi ‘bundle of wood’
ile+ise *ili ise ‘office’
(ii) Can you suggest any reason why assimilation does not occur in the
cases in b.? What is there in the forms in b. that prevents them from
undergoing assimilation?
(iii) How can a theory of underspecification help to account for the data
above?
Japanese
In the native Yamato vocabulary of Japanese a combination of voiced

obstruents within a root is avoided. Thus while such words as buta ‘pig’,
futa ‘lid’ fuda ‘sign’ are acceptable, *buda is not possible. This constraint, known
as “Lyman’s law”, has the effect of undoing the effect of the pervasive rule
of “Rendaku”, which systematically voices the initial obstruent in the second

component of a compound. The examples below illustrate the combined effects
of these two processes:
a. iro ‘coloured’ + kami ‘paper’ — irogami ‘coloured paper’

take ‘bamboo’ + sao ‘pole’ — takezao ‘bamboo pole’
eda ‘branch’ + ke ‘hair’ — edage ‘split hair’
zjuzu ‘rosary’ + tama ‘beads’ — zjuzudama ‘(prayer) beads’
b. kita ‘north’ + kaze ‘wind’ — kitakaze (*kitagaze)

‘freezing north wind’
firo ‘white’ + tabi ‘tabi’ — firotabi (*firodabi) ‘white tabi’
zjuzu ‘rosary’ + +> zjuzutunagi @zyuzudunagi)
tunagi ‘sequence’ ‘roping together’
taikutsu ‘boredom’ + — taikutsufinogi (*taiuksuzinogi)
Jinogi ‘avoiding’ ‘time-killer’
Lyman’s Law is clearly a dissimilation process. In the compounded forms

in b. Lyman’s Law has reversed any voicing effect that Rendaku might have
had. Remember, however, that all sonorant consonants are inherently voiced.
Notice that in a. there are examples where a voiced consonant does occur
in the compounded form followed by a sonorant, but where the constraint
of Lyman’s Law is not applicable (irogami, zjuzudama).
(i) | Show how Lyman’s Law works.

(ii) Why does Lyman’s Law not apply in irogami and zjuzudama?
(iii) Explain why it applies in zjuzutunagi and taikutsufinogi.
RULES AND DERIVATIONS
In the model we have been presenting throughout the book, alternants are
related by means of rules which act on the lexical form, or on a repres-
entation ultimately derived from the lexical form. In particular, lexical forms
are mapped onto surface forms ina series of steps, each defined bya rule:
the sequence of representations thus obtained constitutes a “derivation”. The
action of a rule on a representation can have decisive effects on the applic-
ability of some other rule, either by creating the input required for this rule
or by destroying it. Rules can therefore interact with each other, and we need
to monitor such interaction by ordering the rules precisely in the manner
that will yield the desired output. In this chapter we study the mechanics
of rule interaction and rule ordering making use of a substantial portion of
the segmental rules of English, which accordingly we also survey. We pay
close attention to the evidence that bears on the cyclic or non-cyclic status
of the rules, and to the ensuing organization of the rules into blocks.
Rule Ordering: Feeding and Counterfeeding
In the previous chapter we considered alternations like those in (1), and

accounted for them with the s-Voicing rule in (2) (as we said, in some
accents some of the [z]s in (1) are only intermediate, and subsequently undergo
palatalization):
544 Rules and Derivations
(1) a. gymnastics b. gymnasium

Caucasus Caucasian
Malthus Malthusian
fantasy fantasia
(2) s—[+voice] / [-cons] [—-cons]
Wo (UW)
We commented at the time that the long vowel that triggers s-Voicing in
the set in (1b) cannot be lexical, for, if it were, the stress pattern of some of
the forms in (la) would have been different.
Further examples of vowel lengthening, this time independent of s-Voicing,

are provided in (3a). Notice once more that the corresponding vowels in the
bases in (3b) are short:
(3) a. Canadian b. Canada

Jordanian Jordan
Mongolia Mongol
Arabia Arab
Babylonian Babylon
regalia regal
collegiate college
courageous courage
felonious felon
colonial colony
comedian comedy
Gregorian Gregory
In (4) we formulate the rule responsible for the lengthening of the (crucially,
non-high) vowel in (1b) and (3a). We shall refer to this rule as “CiV-
Lengthening”, for the simple reason that the lengthening in question takes
Rules and Derivations 545
place precisely in front of the sequence consonant + i + vowel (cf. Bosnian,

satin, with no lengthening):
7
YS
|
(4) CiV-Lengthening:
Oo Oo
Da seetpaceaies macro
UW ee aL 1 [+cons] [—cons] [—cons]
[—cons] [-cons] __[-high] [—back]

Condition: the contextual i is metrically weak
Before we proceed, we must say a word about the way we formalize dis-
tinctive features in this and other rules throughout this chapter. Features
are of course organized autosegmentally, as we reiterated in chapter 17. To
keep the representations free from unnecessary clutter, however, we do not
include the feature geometry in our rules, except in cases where such geo-
metry is directly relevant to the operation of the rule. Whenever possible,
therefore, we adopt the practice of simply stacking up the features, with no
indication of their position in the geometry.
Not unexpectedly, there are a number of exceptions to the rule of CiV-

Lengthening in (4). Some are idiosyncratic:
(5) Italian centennial rebellious Maxwellian
Other exceptions are more systematic:
(6) companion battalion medallion rebellion scullion
Quite simply, CiV-Lengthening does not take place in front of the suffix
-10N.
Returning at this point to the data in (1), you will notice that s-Voicing
requires the prior application of CiV-Lengthening in the relevant forms.
In particular, in order to undergo s-Voicing, the /s/ needs to be preceded

by a long vowel, but there is no such vowel in the underlying repres-
entation of the forms in (1), as we have pointed out: Cauc/z/sus, etc.
Consequently, such forms would not be able to undergo s-Voicing if CiV-
Lengthening had not applied first, as is substantiated by the forms in (7):
(7) hessian massive classic

missile passive impressive
In (8) we show that the application of CiV-Lengthening to forms like

Cauc/z/sian creates the appropriate environment for s-Voicing:
(8) Cauc/e s/ian

eet CiV-FASO
Z, s-Voicing
The type of ordering illustrated in (8), where the first rule creates the input
The type of order- required by the second rule, is known as FEEDING ORDER, and the first
ing where the first
rule is said to FEED the second.
of two rules cre-
ates the input
required by the
second rule is
known as FEED-
ING ORDER (the
first rule FEEDS Clearly, the opposite ordering between the two rules we are discussing would
the second) not give the desired result:
(9) Cauc/e s/ian

NA s-Voicing
Eo CiV-Lengthening
In (9), s-Voicing is prevented from applying because it has (mistakenly) been

ordered before CiV-Lengthening: at the time s-Voicing is available in (9), the
relevant contextual vowel is still short, and consequently s-Voicing cannot
apply. This type of ordering, where the second of two rules would feed the
first rule if the ordering were reversed, is referred to as COUNTERFEED- The type of order-
ing where the sec-
ING ORDER (the first rule COUNTERFEEDS the second).
ond of two rules
would feed the
y Uy
first rule if the
Stop reading until you are sure you see the difference en | ordering were
the feeding and pouctericedine orderings. exp i reversed is known
as COUNTER-
FEEDING ORDER
We will come across many instances of RULE ORDERING in the re- (the first rule
COUNTERFEEDS
mainder of the chapter. Rule ordering is indeed one of the formal tools that
the second)
characterize the model of phonology associated with SPE. The device of
stipulating the order in which phonological rules apply is not, however, to
the taste of all phonologists, and there have been through the years a num-
ber of attempts to remove it from the theory. Many rule orderings can indeed
be removed from individual grammars, because they are either INTRINSIC Many rule order-
ings are either
ORDERINGS, that is, orderings that obtain even if the rules are allowed
INTRINSIC
to apply unfettered, or UNIVERSAL ORDERINGS, that is, orderings that ORDERINGS
follow from some universal principle of rule interaction, such as the Else- or UNIVERSAL
where Condition. ORDERINGS, but
a core of recalcit-
rant cases seems
to require stipulat-
ive EXTRINSIC
ORDERING
The previous considerations notwithstanding, a core of recalcitrant cases

remains which seems to require stipulative EXTRINSIC ORDERING, in
English and in other languages.
between Civ-Lengthening andsS-\
A recent move from a novel, radical perspective to do away with rule

ordering, indeed with rules and derivations themselves, will be presented
in chapter 19.
The output of the derivation in (8) obviously needs some further process-
ing: compare *Cauclzet|sian with the correct Cauc[et|sian. Such processing is
r 8 above,
carried out by the rules of Vowel Shift, which we discussed in chapte
and Diphthongization, which we also mentioned then but did not formu-
late, in order to keep the presentation simple.
We formalize Diphthongization in (10):
(10) Diphthongization:
oO Oo
MW ow U Wu
A RO, a eh
[—cons] [—cons] [—cons]
[aback] [aback] [aback]
[around]
[+high]
The general effects of Diphthongization for the front vowels are specified
in (11):
(11) it srt
These outputs essentially match the phonetic realization (see chapter 7

above for more details), Cauc[et|sian in the case at hand, after the laxing of
the high vowel at the end of the diphthong.
Bleeding and Counterbleeding
At this point, we must elucidate the ordering of the two additional rules we
have just introduced, both with respect to each other and with respect to
CiV-lengthening and s-Voicing.
Consider first the mutual ordering of Vowel Shift and Diphthongization.
The order must be Vowel Shift > Diphthongization (the symbol “>” separ-
ates rules that are ordered; unordered rules are separated by a comma).
The reason is that the application of Vowel Shift is restricted to (stressed)
long vowels, as we specify in (12), which updates the formulation in (38) of
chapter 8 above (notice that we have split the environment, to make the pre-
sentation clearer):
(12) Vowel Shift:

[-ohigh] /[___] Z
[ohigh] 2
[-low] (
[-cons] > / —
plow / 42 | /\
[Blow] PnP
[-high ] \/
Oo
Now, as you can see in (10) above, Diphthongization destroys the bimoraicity
of the input vowel, by robbing it of its second mora, which it associates with
a (newly created) high vowel.
Obviously, if Diphthongization were allowed to apply before Vowel Shift,

Vowel Shift would no longer be able to apply, because the vowel mentioned
in the rule would not be long any more. We illustrate the situation in (13)
with the /ze1/ of Caucasian:
550 ~=Rules and Derivations
(13) Cauc/ee:/sian
el Diphthongization
NA Vowel Shift
*Cauclet|sian
app
The situation where a rule removes material that would be necessary for
The situation the application of a subsequently ordered rule is referred to as BLEEDING
where a rule ORDER, and the first rule is said to BLEED the second rule. Therefore,
removes material
that would be
Diphthongization bleeds Vowel Shift in (13).
necessary for the
application of a
subsequently Whaatexactly is the difference betwe
ordered rule is
referred to as
BLEEDING ORDER
(the first rule The correct order of application between Diphthongization and Vowel Shift
BLEEDS the is obviously the opposite of the one in (13): by giving the rules the oppos-
second) ite order, we prevent the unwanted bleeding from taking place.
Reverse the order
Reversing the order of two rules to prevent bleeding results in a COUN-

Reversing the TERBLEEDING ORDER (the first rule now COUNTERBLEEDS the second).
order of two rules
to prevent bleed-
ing results in a
COUNTERBLEED-
ING ORDER (the
first rule now
COUNTERBLEEDS
the second)
The (correct) counterbleeding order of Vowel Shift and Diphthongization is

illustrated in (14):
(14) Cauc/ez:/sian
e! Vowel Shift
el Diphthongization
Caucl[et]sian
It may be worth pointing out at this juncture that all the four possible rule
ordering relations we have introduced (feeding, counterfeeding, bleeding, All four possible
counterbleeding) are attested in the languages of the world, most often within rule ordering rela-
tions (feeding,
the rule system of one and the same language. counterfeeding,
bleeding, coun-
terbleeding) are
attested in the
languages of the
world
Consequently, you must not interpret the feeding and counterbleeding

interactions between the English rules we have discussed as automatically
correct, and the opposite orderings (counterfeeding and bleeding) as auto-
matically incorrect. The orderings in question are indeed correct and incor-
rect, respectively, for the English data we have discussed, but the mutual
ordering of any two rules in any one language must be established empir- The mutual order-
ing of any two
ically, on the basis of the facts of the language, as indeed will be our prac-
rules in any one
tice throughout the chapter. language must
be established
empirically
In fact, s-Voicing and Vowel Shift are NON-INTERACTING, since the envir-
onments of both rules are met whatever the order of application:
(15) a. Cauc/ers/ian b. Cauc/e:s/ian

ex Vowel Shift Z s-Voicing
Zi s-Voicing er Vowel Shift
You can see that we obtain the correct output irrespective of the order in
which the two rules are applied.
Transitivity
Sometimes, two or more rules have to be left unordered because there is no

evidence on the way they interact. More often than not, however, indirect
evidence on their ordering can be gathered. In particular, if the orderings
If the orderings A > Band B>C can be established empirically, then it follows by TRANS-
A>BandB>C
can be established
ITIVITY that A > C. We will now see that an argument of this kind can be
empirically, then it used to order s-Voicing before Vowel Shift, as in (15b) above.
follows by TRANS- The argument comes from the existence of the rule of Velar Softening in
ITIVITY that English. Consider the alternations in (16), found in the “Latin” part of the
A>C English vocabulary:
(16) a. kl [s] b. [gl [a5]

critical criticism analogous analogical
classical _ classicist pedagogue pedagogic
public _ publicity prodigal prodigy
medic medicine regal regicide
Catholic Catholicism fungus fungivorous
State precisely whatthese alternatior
In (16a), [k] alternates with [s], and in (16b) [g] alternates with [ds]. Phon-
etically, each of these alternations is somewhat idiosyncratic: there is not much
in common between [k] and [s], or between [g] and [ds].
Show this by breaking the se;

components. _
What interests us here, however, is the environment. This is much more

straightforward: /k/ and /g/ “soften” to [s] and [ds], respectively, before
a front non-low vowel (also high in the data in (16), but this is irrelevant:
magnificent, maleficent, etc.). The rule in (17) provides an adequate formal-
ization of the situation. Note that angled brackets indicate a two-way implica-
Angled brackets tion, that is, here either [-voice] and [+continuant, +anterior], or neither:
indicate a two-
way implication
Explain the function of angled brackets in. your own words.

(17) Velar Softening:

[—cont] [+strid] [—cons]
[+back] — [coronal] /_ [-back]
<—voice> <+cont> [-low]
<+ant>
The restriction of the environment of Velar Softening to non-low vowels

prevents the application of the rule in the set in (16a): /z/ (cf. criticality) is
a low vowel, and consequently it does not trigger Velar Softening.
The data in (18) present a challenge to our account:
(18) a. criticize b. medicate

publicize rusticate
anglicize implicate
On the face of it, Velar Softening has overapplied in (18a) (cf. the low vowel
[ze] in critic[eet]se), while in the forms in (18b) it has underapplied ([e] in
mediclet]te is not low).
The puzzle disappears if Velar Softening is ordered before Vowel Shift, so

that we have Velar Softening counterbleeding Vowel Shift in the forms in
(18a), and counterfeeding it in those in (18b): the contextual vowel for Velar
Softening will be /i:/ and /z:/, respectively.
Having established that Velar Softening precedes Vowel Shift, we shall now
make a case for ordering Velar Softening after s-Voicing. If so, s-Voicing will
precede Vowel Shift by transitivity.
Consider forms like those in (19):
(19) a. Isl b. [z]

recite resign
recede resist
receive design
decide reside
We shall analyse the forms in both sets in (19) as composed of a prefix

(re-, de-) and a stem: -cede (cf. recede), -sist (cf. resist), and so on.
The fact that the forms in (19b) have undergone s-Voicing (re-[z]ist vs. con-
[s]ist) suggests that the e of the prefix must be long, at least underlyingly.
The obvious problem with the analysis as it stands concerns the forms in
(19a), which still fail to undergo s-Voicing, and will therefore have to be treated
as exceptions.
“Check that recite,

point.
While phonological rules often do have exceptions, we naturally want to min-

imize their number and integrate the corresponding data into the system as
much as possible. In the present case, the stem of the forms in (19a) is spelled
with an initial c (re-cede), whereas the stem of the forms in (19b) is spelled
with s (re-sist). While spelling is, most emphatically, irrelevant to phono-
logy, as we have been repeating ad nauseam throughout the book, in the
present case the difference in spelling contains a useful clue: indeed, most
of the literature assumes that forms like recede have an underlying /k/
(re/k/ede), and forms like resist an underlying /s/ (re/s/ist).
The /s/ of resist will of course be voiced by s-Voicing (again, on the

assumption that the vowel in the prefix rein question is long at that stage).
Velar Softening turns /k/s into [s]s, and therefore the /k/ of re/k/ede would
also end up as [z] if the ordering were Velar Softening > s-Voicing, with
Velar Softening feeding s-Voicing.
checks
> s-Voicing gi ses the
If we adopt the opposite, counterfeeding, order, however, the desired result

will be achieved:
(20) re/k/ede re/s/ist

NA Z s-Voicing
s NA _ Velar Softening
re[slede _ re|zlist
By this point, we have established the following orderings on empirical

grounds:
(21) Paired rule orderings:

CiV-Lengthening > s-Voicing (Caucasian vs. hessian)
s-Voicing > Velar Softening (re[slede vs. re[z]ist)
Velar Softening > Vowel Shift (criticise vs. critical)
Vowel Shift > Diphthongization (Cauclet|sian vs. *Cauclzt]sian)
By the (logical) principle of transitivity, we derive the overall ordering rela-

tions in (22):
(22) Overall rule orderings:

CiV-Lengthening (4)
s-Voicing (2)
Velar Softening (17)
Vowel Shift (12)
Diphthongization (10)
You can see that we have now established that Vowel Shift must follow s-
Voicing, even though there is no direct interaction between the two rules.
E¥ Palatalization
We mentioned at the beginning of the chapter (and in chapter 17) that the
[z] that is the output of s-Voicing is subject to palatalization in many words
(in most accents). We illustrate such data again in (23):
(23) Caucasian Malthusian fantasia
Palatalization is in fact an extremely productive process in English. We pro-

vide further exemplification of the phenomenon in (24), this time independ-
ently of s-Voicing:
(24) a. [f] b:, -[s]

impression impress
obsession obsess
digression digress
racial race
official office
ferocious ferocity
efficacious efficacy
[3] [z]
diffusion diffuse
confusion confuse
_ infusion infuse
erasure erase
incision incisor
vision visor
enclosure enclose
Observation of the spelling of the alternants with the palatal reveals the
sequence i + vowel after the segment that undergoes the palatalization.
This orthographic i does not, however, have any correlate in the pronunci-
ation (imprelfa]n, not *imprelfialn, etc.).
558 ‘Rules and Derivations
We could conclude from this fact that the i is purely a spelling matter, and
therefore that there is no/1/ present in the respective lexical representations.
However, it is obvious that the suffix in the forms in (24a) does include /1/
in non-palatalizing environments. Note that we are postulating /1/, rather
than /i/, on the grounds that the segment in question is short (although not
necessarily lax in the cases where it does go through to the surface: this mat-
ter will be clarified below):
(25) rebellion rebellious familiar curious

menial _ Italian centurion imperial
Also, the palatalization of coronals before /1/ is very natural, because both
Palatalization of coronals and /1/ are pronounced at the front of the mouth: indeed, /1/ (like
coronals before /1/
the other front vowels) is treated as [coronal] in the model of feature geo-
is very natural
metry we discussed in section 11 of the previous chapter.
Pause and
_ coronal cons
Let us thus assume that there is indeed an /1/ in the lexical representation
of the suffixes in question, acting as the trigger of the palatalization process,
which we formalize in (26) (we provide a choice between [—back] and [—ante-
rior], a dependent of [coronal], with regard to the specification of /1/, in line with
the two alternative models of feature geometry we presented in chapter 17):
(26) Palatalization:
[-son] ~ [-ant] _____ [--cons] [—cons]

[cor] [+strid] [+high]
[—back OR —anterior]
Condition: 6 immediately follows the syllable with the word’s primary

stress
Notice the direct affiliation of [-cons, +high, —back OR —anterior] to the

6 node in (26), with no mora intervening. The non-moraicity of this vowel
is therefore a precondition for the application of the rule, and corresponds Non-moraicity of
a vowel corres-
to what much of the literature refers to by the label “glide”. We have, how-
ponds to what
ever, raised problems with this notion in chapter 10, and we are now see- much of the liter-
ing that there is actually no need for it, since the same results can be ature refers to by
achieved by reference to non-moraicity. In chapter 13 we suggested that the label “glide”
vowels generally projected moras in the lexicon. What seems to be happening
here is that the vowel /1/ demorifies in the context stressed syllable __
vowel. The application of this rule is, however, subject to idiolectal idiosyn-
crasies, and, in the absence of solid primary research, we shall not go any
further into the matter. To maximize the clarity of the exposition, we shall
transcribe such non-moraic [1] as [j], following the practice we introduced
in chapter 10. You must of course bear in mind that we are in no way imply-
ing a different segmental identity for [j]: it is identical to [1] or [i], but it has
no mora attached to it.
After Palatalization takes place (turning, for instance, regre[sjlon into
regre| fjlon) the /j/ gets absorbed into the palatal.
The rule in (27) formalizes the process:
(27) [j]-Deletion:
o
U
[—cons]
[+high] — @/ [-anterior] ___ [-cons]
[—back OR —anterior]
The rule of [j]-Deletion in (27) must obviously be ordered after Palataliza-

tion, a feeding relationship.
Where does Palatalization fit in the rule set in (22) above?

In fact, Palatalization must be ordered after Velar Softening, because of the

progression logi/k/ian — logilslian — logi[flian. The /k/ of the initial form
logi/k/ian is obviously justified by the base logi[k]. In order to end up as
[f], by Palatalization, /k/ must first be turned into [s] by Velar Softening,
which thus feeds Palatalization:
(28) logi/k/ian
S Velar Softening
Hf Palatalization
Clearly, the opposite, counterfeeding ordering would not deliver the desired
result.
Turning now to Vowel Shift and Diphthongization, there is no evidence

of their ordering with respect to Palatalization: these two rules and
Palatalization seem to be non-interacting.
0S
Diphthongization inte
The state of affairs we have established is set out in the augmented list of
rule orderings in (29):

CiV-Lengthening (4)
|
s-Voicing (2)
Velar Softening (17)
Vowel Shift (12) Palatalization (26)
Diphthongization (10) [j]-Deletion (27)

As you can see, the ordering branches after Velar Softening, one branch con-
taining Vowel Shift > Diphthongization, and the other Palatalization > [j]-
Deletion.
Further Twists
The forms in (30a) obviously do not undergo [j]-Deletion:
(30) a. artifi[fi]ality b. artifi[fJal

presiden[filality presiden|[f
Jal
confiden[fiJality confiden|fJal
You will notice that /j/ has been deleted in the b. set in accordance with
the rule of [j]-Deletion in (27). However, /j/ survives as [i] in the closely
related a. set, which contains an additional cycle. The failure of [j]-Deletion
to apply in the a. set is in fact systematic, and is caused by the presence of
stress on the vowel that follows it. Accordingly, we formulate the corres-
ponding rule of [j]-Morification in (31), where we use arrow heads to indic-
ate prosodic projection:
(31) [j]-Morification:
oO Oo
ce
.
[—anterior] [+high] [—cons]
[-back OR —anterior] (*
2
(31) needs to be ordered between Palatalization (26) and [j]-deletion (27).

In the examples in (32), Palatalization applies as predicted, as does, sub-
sequently, [j]-Deletion:
(32) Christian question digestion (in most accents)

bastion celestial _bestial (in some accents)
In particular, Palatalization turns the underlying /t/ (cf. Christ, etc.) into [ff]
— Canadian, comedian and the like obviously have /i/, rather than /j/.
Consider, however, the data in (33):
(33) Egyptian exemption deletion extinction

invention torrential partial contortion
decision exclusion delusion division
allusion explosion erosion invasion
Here, the palatal becomes fricative, rather than affricate, as it should on the
basis of Egypt, decide, etc.
Simple application of Palatalization to the underlying stops would have pro-

duced the affricates [t{], [d3], respectively: compare the forms in (32) above.
Clearly, then, there must be a further rule fricativizing the coronal stops
in (33) before Palatalization comes into effect: Egyp/t/ > Egyplslien >
Egyplflian.
Sy exactlywhy:we need:his fu
intoAG es
The rule that turns coronal stops into fricatives is known as Spirantization,
and we formulate it in (34):
(34) Spirantization:
[-son] [+cont] [+son] [-cons]
[+cor] — [+strid] / ah thigh]
[-cont] [—back]
[-cons]
Condition: [+high] does not support a stress bearer
[—back]
The condition on (34) limits the scope of the rule to cases where the high
front vowel is directly attached to the syllable node (Egyptian), or where it
is idiosyncratically outside metrical structure, in the suffix -y we discussed
in chapter 15 (presidency).
Just as desired, the [+son] or [-cont] disjunction in the environment

excludes the cases in (32) above, where the preceding contextual segment
is precisely [—son, +cont]: remember that we do not get Chris[flian, etc.
The data in (35) appear to contradict the rule of Spirantization in (34):
(35) a. [ff] b. [ds]

conceptual gradual
eventual residual
habitual individual
virtuous deciduous
adventurous arduous
impetuous incredulous
These and similar forms exhibit an affricate ({f] or [dg]), even though the
environment of Spirantization appears to be met, since the stop is followed
by /i/ and is not preceded byafricative.
The solution to this puzzle is found in the idiosyncrasy of the cluster [iu],
to which we have already referred in chapter 10 above.
The [i] of [iu] is
either not present
in underlying rep-
resentation or not
incorporated into
syllable structure
until well into the
What seems to be happening is that, at the point where Spirantization derivation
becomes operative, the [i] of [iu] is still not available, either because it is not
present in underlying representation, being inserted subsequently by rule,

or because, if it is present underlyingly, it has not yet been incorporated into
syllable structure, and therefore is still not prosodically licensed.
Either way, forms like those in (35) only have /u/ visible at the point when
Spirantization applies, and therefore the environment of the rule is not met.
Subsequently, [i] becomes available as [j] and triggers Palatalization. The order
of these three rules is, therefore, Spirantization > [iu]-Formation (in some
format we will not decide on) > Palatalization.
Say why we need
We now incorporate these rules, and their orderings, into the overall rule
diagram:

CiV-Lengthening (4)
s-Voicing (2)
Velar-Softening (17) Spirantization (34)
Vowel Shift (12) [iu]-Formation
|
Diphthongization (10) Palatalization (26)
[j]-Morification (31)
[j]-Deletion (27)
You can see that Spirantization does not interact with Velar Softening. It must,
however, follow s-Voicing, which counterfeeds it: the product of Spir-
antization is not subject to s-Voicing (cf. promotion). In turn, Palataliza-
tion is fed by [iu]-Formation (and indirectly by Spirantization) and by Velar
Softening.
[4 Vowel Length Alternations. Tensing
In chapter 15 we came across the shortening of vowels in forms like depth

and divinity (compare deep and divine, respectively). We attributed this
shortening to a tendency to favour a maximum of two moras per foot, which
we formalized as the shortening rule repeated here as (37) (2 = foot):
(37) Vowel Shortening:

[-cons] _ [cons]
Li etl ee aa
The interaction of Vowel Shortening with CiV-Lengthening is a matter of

some interest. Consider again the data in (3a) above, repeated here as (38):
(38) Canadian collegiate

Jordanian courageous
Mongolia _ felonious
Arabia colonial
Babylonian comedian
regalia Gregorian
These forms meet the conditions for both CiV-Lengthening (cf. /aente/
in Jordanian) and Vowel Shortening (cf. the trimoraic foot (deI.nI) also in
Jordanian). The fact that CiV-Lengthening wins suggests the order Vowel
CiV-
Shortening > CiV-Lengthening, with Vowel Shortening counterfeeding
Lengthening:
(39) Jord/ent/ an
(Fa) <> Stress
+
NA Vowel Shortening
ei Vowel Lengthening
CiV-Lengthening is not the only rule that lengthens vowels in English.

Consider the contrasts in (40):
(40) a. various b. variety

social society
notorious notoriety
simultaneous simultaneity
maniac maniacal
In the forms in column a., the i of the suffix must be underlyingly short,
because it does not attract stress (vdrious, not *various, etc.), thus justifying
our decision to represent this vowel as /1/. In b., however, the same under-
lying vowel turns up as long and shifted (and, of course, stressed).
The feature
LEATRI is lexically In order to explain these facts, we must consider the role of the feature
unspecified in [+ATR] in the English vowel system. Briefly, all lexical long non-low vowels
English vowels end up as [+ATR], that is to say, they are all pronounced tense. The redund-
ancy rule in (41) encapsulates this state of affairs:
(41) []>[+ATR]/_
[—-cons]
[-low]
oe
Wee
By contrast, lexical short vowels are in principle [-ATR]. This suggests the
redundancy rule in (42), which complements (41):
(42) []—[-ATR]/_
[-cons]
(41) applies across the board: there simply are no lax (= [-ATR]) phonolog-
ically long vowels in standard English.
By contrast, (42) is violated in the surface in several contexts. One such con-
text is prevocalic position, since in most accents of English short vowels sys-
tematically surface as tense when they immediately precede another vowel: — English short vow-
els systematically
s ‘ ae surface as tense
(43) menial various’ affiliate toreador when they tne:
manual graduate tortuous sensual diately precede
another vowel
The situation illustrated in (43) is expressed in the rule of Prevocalic

Tensing in (44):
(44) Prevocalic Tensing:

[-cons] > [+ATR] / [-cons]
As the result of this rule, the /1/ of both various and variety will be tensed
into [i]. We now want the [i] of variety (but not the [i] of various) to undergo
Vowel Shift and Diphthongization. How can this happen? The answer is in
fact quite simple. First, Prevocalic Tensing must precede Vowel Shift, so as
to feed it. Notice, however, that our formulation of Vowel Shift in (12) above
requires length in the input vowel, not tenseness. This means that the feed-
ing of Vowel Shift by Prevocalic Tensing is indirect: the vowel tensed by
Prevocalic Tensing must first lengthen in variety, but, crucially, not in vari-
ous (cf. *var[at]ous).
Saywhy th
various.
Fortunately, the formal differentiation between these two vowels is quite

straightforward: [i] carries stress in variety, but not in various.
Following these observations, we formulate the rule of Stressed Tense

Vowel Lengthening as in (45):
(45) Stressed Tense Vowel Lengthening:

Z ;
-
&
pe Pre cepa te ofrres
[—cons} [—cons]
[+ATR] [+ATR]
The rule of Stressed Tense Vowel Lengthening in (45) feeds Vowel Shift, fol-
lowed by Diphthongization, and the desired output var[aility is obtained:
(46) var/t/i ty
Sone a> Stress
i Prevocalic tensing
it Stressed Tense Vowel Lengthening
et Vowel Shift
ei Diphthongization
varleetjity
English non-low vowels also tense in word-final position, as demon-

strated by the data in (47) (partly a selection from (25) in chapter 13 above): — English non-low
vowels also tense
in word-final
(47) buffalo pity position
mosquito jelly
window yeti
jujitsu wacky
tofu vanity
Kikuyu _ hockey
Of these, the front high vowel [i] remains short in many (albeit not all) accents,
while its counterparts lengthen and/or undergo Diphthongization (but not
Vowel Shift, because the vowel remains stressless):
(48) buffal[ou] pitli]

mosquit[ou] _ jell[i]
wind[ou] yetli]
jujits[u:] wack[i]
tof[u:] vanit[i]
Kikuy[u:] hock{i]
The desired results can be obtained either by subjecting non-low final vowels
to a tensing rule and then lengthening them all but 7, or by lengthening them
all but i in the first place, and then tensing them. We leave it to the reader
to work out the details of these alternative analyses, and to evaluate their
respective merit.
We wind up the section with an update of our overall list of rule orderings:

Vowel-Shortening (37)
CiV-Lengthening (4)
s-Voicing (2)
Velar Softening (17) Spirantization (34)
Prevocalic Tensing (44)
Stressed Tense
Vowel Lengthening (45)
Vowel Shift (12) {iu]-Formation
[j]-Deletion (27)
Cyclic Rules
By now, we have accumulated a sizeable number of rules related to the seg-

mental patterns of English (we said at the beginning that we are deliber-
ately keeping the focus of the chapter on segmental phonology). As we have
been showing, these rules enter into specific ordering relations with each
other.
In chapter 16 we explained that phonological rules affecting words are
organized into (at least) two blocks, each with its own specific properties,
and the most important of these properties is whether or not cyclic rules
can apply.
aeyour memory on the empirical inotivation roa he ‘diveion

into rule —
In turn, cyclic rules obey the Principle of Strict Cyclicity, which you know
blocks their application to any input that has not undergone some previous The Principle of
change in the same cycle. Sn Sate
restricts the appli-
cation of cyclic
; “i So | rules to derived
Remind yourself of the function and the mech. environments
cycle.
We will now set about establishing the cyclic or non-cyclic status of each of
the rules in (49) above. We will conclude that the dividing line between the
two blocks falls just after s-Voicing.
We shall first establish that s-Voicing is cyclic. In chapter 17 we argued
that s-Voicing obeys the Principle of Strict Cyclicity.
In particular, a fairly substantial number of monomorphemic forms con-

tradict the results of s-Voicing: basin, mason, and many others (cf. (4) in
chapter 17). This is an awkward situation for a non-cyclic rule, which ought
also to have voiced the s here, but is unproblematic for a cyclic rule, which
has no power to change morpheme-internal structure.
As you know, there are also some forms where s-Voicing fails to apply across
morphemes (facial and a few others), but their number is much smaller.
Explain how
imbalance. |
As we explained at the time, this situation is easily accountable for if we

leave the voice value of the anterior coronal strident (/s/, /z/) under-
specified morpheme-internally, to allow s-Voicing to fill in the + value in a
structure-building mode, thus bypassing the Principle of Strict Cyclicity.
On this analysis, the numerous monomorphemic exceptions are simply

the result of a lexical specification [—voice].
On the other hand, the Principle of Strict Cyclicity does not block the rule
across morphemes. As expected, the output [z] is present in all forms with
cyclic suffixes but a handful of exceptions.
s-Voicing, however, systematically fails to apply in forms with non-cyclic

suffixes. As we mentioned in the previous chapter, this is evidence that s-
Voicing is cyclic.
Forms like resist in (19b) above present a problem for the cyclicity of s-
Voicing. In particular, if the Principle of Strict Cyclicity restricts the applica-
tion of cyclic rules to derived environments, the environment in which the
/s/ becomes [z] should be derived in (19b) also. At first sight it is, since we
have argued that such forms are made up of a prefix and a stem (re-sist,
etc.). While this is true, there is evidence across languages that cyclic rules
Cyclic rules only only apply within word domains, and sist and the other stems in (19b) are
apply within a
clearly not words.
word domain
Say exactly what the Dr
The solution to this conundrum involves leaving the relevant s in all such
stems unspecified for [voice] in the lexicon: the rule of s-Voicing will now
apply without violating strict cyclicity.
Why will s-Voicing not violate strict cyclicity now? — .
On the grounds discussed, we will assume the cyclic status of s-Voicing, with
most of the literature.
Rules
and Derivations 573
_ Ifs-Voicing iscyclic, thenallthe rules ordered before itmust also be cyclic,

given the assumption we are operating under that cyclic rules and non-cyclic
rules are organi in coherent
zed blocks.
Thus, in particular, CiV-Lengthening feeds s-Voicing, therefore preceding it

in the ordering. Consequently, CiV-Lengthening must also be cyclic.
—s
ia
We are not aware of any evidence against the cyclicity of CiV-Lengthening.

Next, we have shown that Vowel Shortening must precede CiV-Leng-
If Vowel Shortening precedes CiV-Lengthening, then Vowel Shortening

must also be cyclic.
}Infact,inchapter 15 we presented independent evidence forthecyclicity of

Vowel Shortening: its operation is restricted to cyclic domains, as attested
_by the contrast div[i)nity vs. div{at\nishy, a possible word.
A similar diagnostic ought to be applicable to CiV-Lengthening, but the test

is made difficult by the shortage of suitable suffixes.
We have now established that s-Voicing and the two rules that precede
it (Vowel Shortening > CiV-Lengthening) are cyclic. The next question is
whether the two rules that follow s-Voicing in the ordering (Velar Softening
and Spirantization, in the respective branches) are also cyclic.
If cyclic rules obey the Principle of Strict Cyclicity, as we are maintaining

they do, then Velar Softening cannot be cyclic. The reason is that in forms
like re/k/ede, etc., there has not been any previous change in the input, since
the stems cede, etc., do not constitute cyclic domains, for the same reasons
we argued for sist in resist above.
QUE
YA
at are ther a
One obvious move forward is to declare Velar Softening non-cyclic, as

indeed does most of the literature. An alternative would be to represent the
forms in question with a lexical /s/, thus specified as [—voice]: a cyclic rule
of Velar Softening would now be powerless to apply and voice the /s/, on
account of the Principle of Strict Cyclicity. Be that as it may, we will assume
with the bulk of the literature that Velar Softening is non-cyclic.
Turning now to Spirantization, it is clearly ordered after s-Voicing: vaca-
tion has [f], rather than [3].
i the elev.
On the other hand, Spirantization must precede [iu]-formation, which it coun-

terfeeds: habitual has [f], not [f].
It appears that [iu]-Formation must be non-cyclic, since it applies morpheme-

internally with no need for a previous change: cute, dispute, etc., have [iu:],
not [u:].
Rules and Derivations BID
If Spirantization precedes non-cyclic [iu]-Formation and follows cyclic s-

Voicing, it could in principle be the last cyclic rule or the first non-cyclic
rule.
We are not aware of the existence of a clear argument either way. On the
one hand, a number of forms, both derived and underived, fail to undergo
Spirantization:
(50) a. Nadia, Tatiana, Katya, radio, patio, sodium, Arcadia

b. Canadian, comedian, Ovidian, guardian, custodian
These forms could be taken as proof that Spirantization is cyclic: if it

were non-cyclic, its application would not be constrained by strict cyclicity,
and all the forms in (50) would be expected to undergo it. If Spirantization
were cyclic, however, only the forms in (50b) would have to be marked
as exceptions.
On the other hand, we could and shall assume that such forms as those in
(50) simply fail to meet the environment of Spirantization, on account of the
i of the environment being moraic, at least at the point where Spirantization
applies.
In the absence of a clear empirical argument for the non-cyclic status of

Spirantization, assumed in most of the literature, we shall fall back on the
universal Principle of Late Block Rule Assignment:
(51) Late Block Rule Assignment:

Rules are assigned Rules are assigned to the latest possible block in the absence of con-
to the latest possi- trary evidence
ble block in the
absence of con-
trary evidence Following this principle, we shall assume with the rest of the literature
that Spirantization is ordered in the non-cyclic block, and therefore that it
is non-cyclic.
Having established that the dividing line between the cyclic and the
non-cyclic blocks falls immediately after s-Voicing, we make the identity of
the cyclic rules and their ordering explicit in (52):
(52) Cyclic Block:

Vowel Shortening (37)
CiV-Lengthening (4)
s-Voicing (2)
All the rules still to be discussed follow either Velar Softening (17) or
Spirantization (34), and therefore they must be non-cyclic, given the organ-
ization of the rules into coherent blocks.
Remind yourself of whic

ordered.
El Non-Cyclic Rules
In (49) Velar Softening is followed by Prevocalic Tensing > Stressed Tense

Vowel Lengthening > Vowel Shift > Diphthongization in one branch. Vowel
Shift must obviously follow Velar Softening, which counterbleeds or counter-
feeds it: criticize and medicate, respectively.
If Velar Softening is non-cyclic, as we have suggested is the case, then by

transitivity Vowel Shift must also be non-cyclic.
Vowel Shift can be fed by the pair Prevocalic Tensing > Stressed Tense Vowel
Lengthening, which must consequently precede it.
Prevocalic Tensing applies in defiance of the Principle of Strict Cyclicity

(cf. radio, patio, etc.), and therefore it must be non-cyclic. If so, Stressed
Tense Vowel Lengthening, which follows Prevocalic Tensing, must also be
non-cyclic.
Vowel Shift counterbleeds Diphthongization, which must therefore be

ordered after it. Consequently, Diphthongization is also non-cyclic — there is
indeed specific evidence that Diphthongization applies within morphemes,
as also does Vowel Shift.
In the bottom branch of the non-cyclic block, Palatalization is fed by

[iu]-Formation, which therefore precedes it. Accordingly, Palatalization is
non-cyclic.
Finally, Palatalization feeds [1]-Deletion, which must consequently follow it,

and thus also be non-cyclic.
Notice that Palatalization must also be part of the postlexical block, because
it is applicable across words: dildglou?, goltf{Jou!, I milflyou, do I plea[3lou?,
etc. We have already encountered rules assigned to both the cyclic and the
non-cyclic blocks, and we must now extend the possibility of multiple block
membership to the postlexical block. Multiple block membership is standardly
constrained by the Stratum Contiguity Hypothesis of (53):
(53) Stratum Contiguity Hypothesis:

The strata a rule is The strata a rule is assigned to must be contiguous
assigned to must
be contiguous
We have now checked the status of all the rules in our list in (49) above
with regard to cyclicity. In (53) we provide the full list of non-cyclic rules
with their orderings, to complement the list of cyclic rules in (51):
(53) Non-cyclic Block:

Velar Softening _—_ (17) Spirantization (34)
Prevocalic Tensing (44) [iu]-Formation
Stressed Tense
Vowel Lengthening (45)
Vowel Shift (12)
[j]-Deletion (27)
There are a number of segmental rules mentioned in previous chapters

which we have omitted from the present discussion. Among these rules are
Aspiration, Flapping, Glottalization, Sonorant Devoicing and /-Velarization.
These rules must be assumed to be postlexical (NB not just non-cyclic),

for the simple reason that they possess all the properties that characterize
postlexical rules: they are not structure-preserving (that is, they create
segments which are not underlying), they have no exceptions, they apply
in all types of word (not just in words which constitute “lexical categories”),
they are not word-bounded, and so on (see chapter 16 above for a full
list of properties of postlexical rules). We shall not explore the mutual
ordering of these rules here, in order not to lengthen the chapter any fur-
ther, but we warmly invite you to work out this ordering as an addi-
tional exercise.
discussed in chapter 17, we she

Caucasus the rule of Ci V-Lengthening hasto
be 0
ing ordering is reversed, the tw Fules sad in a “counter

We used the rules of Vowel Shift andDiphthongization discusse
e Vowel Shift, bowelShift woe

known as a “bleeding order”. In
, Se Although
Key Questions
What is meant by “intrinsic” and 6 What do angled brackets in a rule

“extrinsic” ordering of phonological indicate?
rules? 7 What restriction does the Principle
2 Rules which stand in an Elsewhere of Strict Cyclicity impose on the ap-
relationship are said to be universally plication of cyclic rules?
ordered: why? 8 How does the Principle of Late Block
3 What is meant by a feeding order Rule Assignment help to determine
and a counterfeeding order? Give an the dividing line between the cyclic
example of each. and non-cyclic blocks of rules?
4 What is meant by a bleeding order 9 What are the terms of the Stratum
and a counterbleeding order? Give Contiguity Hypothesis?
examples.
5 How does transitivity determine the
overall order of rule application?
Further Po r-atert’pPcze
Rule Ordering in Serbo-Croatian
Consider the following examples of adjectival and past participle

paradigms from the South Slavic language Serbo-Croatian, spoken in much
of the former Yugoslavia:
an Masco: en. Netta. Pls

mlad mladaé mladé6 mladi ‘young’
pust pusta pust6 pusti ‘empty’
zelén zelenaé zelené zeleni ‘green’
Yést festa fests testi ‘frequent’
G) In the context of this evidence, how is main stress assigned in Serbo-

Croatian?
Now consider the forms in b.:
b. Masc. Fem. Neut. Ris

debéo debelé debel6 debeli ‘fat’
vidéo videla videl6 videli ‘seen’
béo bela bel6 beli ‘white’
hodao hodala hodalé hodali ‘walked’
(ii) Account for the stem alternations in b. by means of a rule.

(iii) How does this rule interact with the stress assignment procedure?
The forms in c. illustrate a further variation:
c. Mase. Fem. -Neut?Ph

dobar dobra dobré dobri ‘good’
jasan jasna jasnd jasni ‘clear’
sitan sitna sitnd sitni ‘tiny’
Oftar oftra oftrd oftri ‘sharp’
(iv) Account for the stem alternations in c. by means of a further rule.

(v) How does this rule interact with either of those you have proposed
so far?
Now consider the set of forms in d.:
d. Masc. Fem. Neut. a

podmukao podmuklé podmukl6 podmukli. ‘treacherous’
okrugao okrugla okrugl6 okrugli ‘round’
6bao obla obl6 obli ‘plump’
podao podla podlé podli ‘pase’
(vi) What do the forms in d. tell us about the interaction of the rules you
have proposed?
Rule Ordering in Tangale
Consider the following nominal paradigms from the Chadic language

Tangale, spoken in Nigeria (tones omitted):
a. Noun Definite Possessive Possessive Possessive

Isg. 28g. 3sg.fem.
loo loo-i loo-no loo-go loo-do ‘meat’
bugat bugat-i bugad-no bugat-ko bugat-to ‘window’
tugat tugad-i tugad-no tugad-go tugad-do_ ‘berry’
aduk aduk-i adug-no aduk-ko aduk-to ‘load’
kuluk kulug-i kulug-no_ kulug-go kulug-do ‘harp’
(i) | What are the underlying forms of the nouns listed?

(ii) Write rules to account for the stem-final voicing alternations.
(iii) Write a rule to account for the alternation in the 2sg. and 3sg.fem. pos-
sessive suffixes.
Now consider the noun paradigms in b.:
b. Noun Definite Possessive Possessive Possessive

Isg. 28g. 3sg.fem.
wudo wud-i wud-no wud-go wud-do ‘tooth’
litiee wiertst lut-no lut-ko lut-go ‘bag’
taga _—tag-i tag-no tag-go tag-do ‘shoe’
duka duk-i duk-no duk-ko duk-to ‘salt’
yuli yul-i nul-no nul-go nul-do ‘truth’ -
(iv) What are the underlying forms of the nouns listed in b.?
(v) What further rule operates on the noun paradigm shown in b.?
(vi) How does this rule interact with those you proposed for a.?
The Elsewhere Condition
Western Finnish
In Western Finnish, word-final /k/ (synchronically justified in the grammar)

is either assimilated to the word-initial consonant of a following word, or
deleted before a pause or if the following word is vowel initial, as we can
see in the examples below:
menek pois — menep pois ‘go away’

menek alas — menealas ‘go down’
menek ~ 1. mene ‘g0’
(i) Write two rules to derive both these facts.

(ii) Can the Elsewhere Condition prevent the incorrect deletion of the
word final /k/ from menek kotiin ({menek kotiin], not *[mene kotiin])?
Lardil
In the Australian language Lardil the inflection marking future agreement

on nouns in the accusative case is -ur [ur] (e.g. [tunal-ur] ‘tree’). If the
uninflected form is vowel-final then the [u] is deleted, as we show in col-
umn a., except if that final vowel is [i], in which case [w] is inserted
between [i] and [u], as we show in column b.
a. wite-t ‘inside’ b. kenti-wur ‘wife’

mela-t ‘sea’ nini-wut ‘skin’
nuku-t ‘water’ tjimpi-wur ‘tail’
Demonstrate how an Elsewhere relationship between the rule of vowel dele-

tion and [w] insertion can derive the forms above.
Chapt
In this chapter you will learn abo
The model of phonology we have presented throughout the book consists

of several levels of representation related by context-sensitive rules. In par-
ticular, forms stored in the lexicon are mapped onto surface forms in a series
of steps, each defined by a rule. The complete set of steps mapping a lexical
form onto a surface form constitutes a derivation. The effects of the applica-
tion of one rule can affect the subsequent application of some other rule,
either by inducing it or by blocking it. Interaction between rules is regulated
through extrinsic rule ordering, although some general principles that pro-
mote intrinsic rule ordering also exist. The two basic elements of this model
can thus be summarized as follows: (i) there are two levels of representa-
tion, namely, the lexical level and the surface level; (ii) the mapping from
the lexical level to the surface level is guided, and restrained, by rules. In
the present chapter we will explore an alternative, though closely related,
model. In particular, we will allow rules a completely free hand: it is in some
way as if all rules of all languages applied in each and every possible con-
text. The obviously wild output of this procedure is then filtered through a
set of constraints, geared to excluding all the undesired outputs. Clearly, the
target outputs vary from language to language: if they didn’t, only one lan-
guage would exist. Such language-specific targets are attained through the
Constraints 585
language-specific ranking of the crucially violable constraints, the substance

of which is ideally conceived of as universal. This alternative model, still very
much part of generative phonology, is known as OPTIMALITY THEORY,
and it can be extended to morphology and even to syntax.
Naturalness of Phonological Inventory: Markedness
In chapter 17 we formalized markedness through the negative statements,

or constraints, in (1):
(1) Some markedness statements:

I Vowels:
a. *[+high, t+low] (= Vowels cannot be high and low at the
same time)
b. *[-high, —low] (= Vowels cannot be non-high and non-low
at the same time)
c. *[+low, —back] (= Vowels cannot be low and front at the
same time)
d. *[+low, +round] (= Vowels cannot be low and round at the
same time)
e. *[around, —aback, (= Non-low vowels cannot have opposite
—low] values for roundness and backness)
fi *[x ATR, alow] (= Vowels cannot have the same value for
lowness and ATRness)
II Consonants:
g. *[dorsall]p (= Consonants cannot be dorsal)
h. *[labial], (= Consonants cannot be labial)
i. *[-anterior] (= Consonants cannot be non-anterior)
j. *[+distributed] (= Consonants cannot be distributed)
k. *[+lateral] (= Consonants cannot be lateral)
1. *[+round] (= Consonants cannot be round)
m. *[+continuant] (= Consonants cannot be continuant)
n. *[acontinuant, (= Consonants cannot have opposite values
—astrident] for continuancy and stridency)
Ill Common to consonants and vowels:

o. *[aconsonantal, (= The values for consonantality and sonor-
asonorant] ancy cannot agree)
p- *[asonorant, —avoice] (= The values for sonorancy and voice
cannot disagree)
q- *[+nasal] (= Segments cannot be nasal)
*[+nasal, —-sonorant] (= Nasals cannot be obstruent)
586 Constraints
Taken literally, the constraints in (1) prohibit the occurrence of segments that
contain any of the features or feature combinations they mention.
The following is a selection of prohibited segments:
(2) I Vowels:
Empty set
*e, *o
*2e
“Cheb
*Oe, +7, ty, “WH, + A
paos
mo AT Tepe at sO
Consonants:
Ele *o
Sp; tb; *f
mes
FOaX0)
*
ah
Efe [OF ty tf *h
5.8
2S
oe
See *9
II1_ Common to consonants and vowels:

ai Se *m, any *n
*b, ty *) +7 a hy *a
my ih, *n, wee O)
0s2°5
Empty set
It is obvious that, if all these predictions were fulfilled, the segmental

inventory of natural languages would reduce to the vowels /a, i, u/ and
the consonant /t/.
This is not much material with which to build the tens of thousands of words
usually present in any language.
Constraints 587
As we pointed out in chapter 17, the constraints in (1a) and (Ir) are taken
to embody physical contradictions. All the other constraints in (1), however,
rule out segments which do exist in at least some natural languages (some
of them in many), at the underlying or surface levels. This means that these
constraints are not to be interpreted literally, as absolute prohibitions on the
segments they define.
Constraints (1b) to (1q) need instead to be interpreted as formal encapsula-

tions of partial deviations from the simplest ideal, to which the various Most markedness
constraints are
languages adhere to varying degrees. For instance, we saw that the most
formal encapsula-
common vocalic system in the world’s languages is /i, e, a, 0, u/, and this
tions of partial
set contains two violations of (1b). For consonants, practically all languages deviations from
have bilabial stops (in violation of (1h)), most also have velar stops (in viola- the simplest ideal
tion of (1g)), a majority has fricatives (in violation of (1m)) and nasals (in
violation of (1q)), and so on.
The real function of the constraints in question, therefore, is that of pro-
viding a formal measure of the “markedness” inherent in a system.
We achieved this goal in chapter 17 through a set of implicational statements,

or REDUNDANCY RULES, directly derived from the negative statements or We achieve a
formal measure of
constraints in (1) in accordance with the general principles of logic (see (22)
the markedness
in chapter 17). These redundancy rules allow us to save on the correspond- inherent in a sys-
ing lexical information: the less marked the system, the less information it will tem through a set
need in the lexicon. For instance, the low vowel /a/ will only be lexically of REDUNDANCY
marked as [+low], and the high vowels /i, u/ as [+high] and [aback] (— and RULES logically
+, respectively): in this way, there will be a direct correspondence between derived from
the markedness
economy of lexical features and naturalness (= less markedness, formally).
constraints
588 Constraints
At this point we must ask why languages have segments which are
unnatural, or, in less strong and more formal terms, marked. The answer is
that there is an obvious correlation between segment markedness and the
There is an size of the system: the more marked the system, the more segments it will
obvious correlation
contain, assuming of course that the presence of a marked segment pre-
between segment
markedness and
supposes the presence of its unmarked counterpart(s). In turn, there is an
the size of the obvious correlation between the size of the system and its encoding power:
system the more segments the system contains, the more lexical contrasts we can
build without unduly lengthening the words.
One plausible assumption in this connection is that, the younger we are,

The younger we the greater the power the markedness constraints hold upon us. As we grow
are, the greater
up, however, we become more capable of producing sounds which are less
the power the
natural, hence the existence of many such sounds in most of the world’s
markedness con-
straints hold upon languages. Clearly, thus, the tension between the child’s tendency towards
us naturalness and the almost certain presence of markedness in the language
of the environment eventually resolves itself in favour of the language of the
environment: sooner or later the child has to accept and learn the sounds
contained in that language.
Constraint Ranking: Faithfulness
We shall now propose a formalization for the situation we have been

describing. You will soon realise that this formalization is considerably
simpler than the one embodied in the rule-and-derivation model we have
been presenting up to this point. Indeed, the formalization we are about to
introduce may well be the simplest one compatible with the phonology of
natural languages.
Constraints 589
Suppose that we adopt the constraints in (1) above, and other similar
ones, as the only formal devices responsible for the shape of surface repres-
entations. In particular, suppose that we do away with rules and derivations
altogether. This is the essence of OPTIMALITY THEORY (OT), a novel Optimality
Theory does away
development in generative phonology which we will be presenting in the
with rules and
remainder of the chapter. In the model we have sketched, the least marked derivations
vowel system /a, i, u/ will simply follow from the presence in the corres-
ponding grammar of the constraints in (1b) to (1f) above, which we repeat
here for your convenience:
~~
(3) b. *[-high, -low] (= Vowels cannot be non-high and non-

low at the same time)
c. *[+low, —back] (= Vowels cannot be low and front at the
same time)
d. *[+low, +round] (= Vowels cannot be low and round at
the same time)
e. *[around, —aback, -low] (= Non-low vowels cannot have oppos-
ite values for roundness and backness)
f. *[aATR, alow] (= Vowels cannot have the same value
for lowness and ATRness)
How do these constraints bring about the desired state of affairs?
Quite simply, any vowel that is not one of /i, a, u/ will violate one of
the constraints in (3), and therefore it will be ruled out by the system.
Specifically, as we indicated in (2), /e, o/ violate (3b), /z/ violates (3c),
/v/ violates (3d), /y,.../ as well as /w,.../ violate Ge), andi f.G.8a,
violate (3f).
So, as we said then, the system as it stands will only allow /i, a, u/ through.
Notice, crucially, that we are achieving this result exclusively through
590 ~=Constraints
the medium of constraints, without any need for implicational statements,

redundancy rules, derivations and the like.
The question that springs to mind immediately is: given this framework,
how come languages do include (some of) the marked segments in their
inventories? For instance, how can a language have /e, 0/, in violation of
(3b), /ze/, in violation of (3c), and so on? Obviously, unless the constraints
are neutralized somehow, they will simply block the realization of any under-
lying form which is not compatible with them. If so, the surface representa-
tion will be as impoverished as it was at the child’s initial stage.
The way Optimality Theory deals with this situation is as follows. Besides
the markedness constraints in (1)/(3) above, the phonology is assumed to
include the requirement, also formalized as a set of “constraints”, that sur-
Surface forms
face forms must be FAITHFUL to the corresponding lexical forms, that is,
must be
FAITHFUL to
that surface forms must simply reproduce their lexical counterparts, with
the corresponding no change. Given this drive for faithfulness, a pronunciation of an under-
lexical forms lying form /pen/ as [pin] (or [pan]), induced by the markedness constraint
(1b)/(3b), will obviously be disallowed.
Or will it? Why isn’t the conflict between the faithfulness constraint and
the markedness constraint resolved in favour of the markedness constraint?
This question constitutes in fact the nub of Optimality Theory. What we are
witnessing is a tug of war between two constraints: if one wins, the other
will automatically lose. This outcome contradicts our natural expectation that
both constraints will be obeyed, the way rules are when their turn comes.
Constraints 591
The answer provided by Optimality Theory is, realistically, that the world
is not totally harmonious: more often than not, in order to obey one con-
straint, another constraint needs to be disobeyed. Formally, the constraints
are given a RANKING, the OT equivalent of the ordering of rules that we
explored in depth in the preceding chapter. The consequence of mutually
ranking two constraints is, of course, that the higher-ranked constraint will A higher-ranked
be complied with in preference to the lower-ranked constraint: constraint is
complied with
in preference to
(4) Ranking metaprinciple: a lower-ranked
A higher-ranked constraint is complied with in preference to a lower- constraint
ranked constraint
In our /pen/ —/— [pin] example, the ranking must therefore be

IO-IDENT,,,; » MARKEDNESS (1b)/(3b). The constraint I[NpuT]O[urruT]-
IDENT[ITY],:x{urs] CMbodies the specific requirement that surface features
be faithful (= identical) to their underlying correspondents. In turn, “»”
signals ranking between constraints, just as “>” signals ordering between
rules. At the time the child still says [pin] (if that is indeed what the child
says in this instance) we can assume that the ranking IO-IDENT,,,, »
MARKEDNESS (1b)/(3b) still has not been established.
The difference
between gram-
mars is a function
of differences in
At this time the markedness constraints (1b)/(3b) (and, undoubtedly, others)
constraint rankings
are still undominated, hence MARKEDNESS (1b)/(3b) » IO-IDENT,,,,.. What
follows from this simple model is far-reaching: the difference between
grammars, whether between child and adult grammar or between grammars
of different languages, will be a function of differences in constraint rank- Constraints are
ings, the constraints themselves being universal. This is Optimality Theory universal
in a nutshell.
| ay diffe
erences in the
Constraint summary:
MarKeDNeEss: Surface forms must be phonetically natural
IO-IDENT,«;: Surface features must be faithful to their underlying correspondents
ea SS
592 Constraints
Structural Constraints: Syllables
Let us now add to the edifice of Optimality Theory we are building. By now,
we have encountered constraints of two kinds: faithfulness constraints, such
as IO-IDENT,,,,, and markedness constraints, inventoried in (1)/(3) above.
Constraints therefore come in families (we will come across other faithful-
Constraints come ness constraints besides IO-IDENT,,,,; as we go along), each constraint family
in families, each
corresponding to one simple basic concept. For instance, each faithfulness
constraint family
corresponding to
constraint expresses one aspect of the requirement that surface forms must
one simple basic be faithful to the underlying form, all markedness constraints are embodi-
concept ments of the tendency of language towards phonetic naturalness, and so on.
The next constraint family we shall look at is made up of structural con-

straints. Structural constraints are the constraints behind such prosodic
Structural structures as syllables and metrical grids.
constraints are
We know from chapter 9 that the most basic syllable is made up of a con-
responsible for
prosodic structure
sonant and a vowel: CV. Of these two segments, the consonant makes up
the syllable onset, and the vowel the nucleus. How can we obtain this state
of affairs in our present, constraint-only system? The repertoire of constraints
in Optimality Theory includes the set of structural constraints in (5):
(5) Syllable-structure constraints:

OnseT = All syllables must have an onset
Nucteus = All syllables must have a nucleus
No-Copa = Syllables must not have codas
“COMPLEX = Constituents must not be complex
Some of the constraints in (5) are stated positively, that is, as requirements,
rather than as prohibitions. It is clear that the constraints in (5) will license
precisely the structure we are aiming for: a simple onset followed by a simple
nucleus. The table in (6) spells out how this is achieved, on the reasonable
assumption that consonants go in the syllable margin, and vowels in the syl-
lable nucleus, an assumption we formalize directly below. In (6) a tick (V)
indicates compliance with the corresponding constraint, and an asterisk (*)
indicates a violation:
Constraints 593
(6) CVVECVINEVE CVV ile

ONSET Jv v Jv v rey
NUcLEuS v Jv v v Sf ott
No-copA ¥ Jv 3 Jv vow.
*COMPLEX J = v : Cees
It is clear from (6) that, given the proposed set of unranked constraints, the
structure CV, made up of a simple onset and a simple nucleus, will emerge
victorious.
Before testing out this model on the actual syllable structure of English,
we must make formally explicit the reason why consonants go in the syl-
lable margins, and vowels in the nucleus. You are already aware that this
distribution is motivated by sonority, in particular by the fact that syllable
peaks aim for maximum sonority: vowels of course have more sonority than
consonants. In Optimality Theory, this situation is formalized by means of
the two constraints in (7):
(7) Sonority constraints:

*M/V = Vowels must not occupy the syllable margin
*N/C = Consonants must not occupy the syllable nucleus
We illustrate the action of these two constraints in (8), where we revert to

the non-moraic syllable formalism to keep in with the OT labelling of the
relevant constraints (cf. ONSET, NucLEus, No-CopA, etc.):
(8)
594 Constraints
Out of the three possibilities, only CV does not incur a violation of *M/V
or *N/C, and therefore CV is the best syllable.
aS
Constraint summary: :
ONSET = All syllables must have an onset
Nucteus = All syllables must have a nucleus
No-copa = Syllables must not have codas
*COMPLEX = Constituents must not be complex
*M/V = Vowels must not occupy the syllable margin
*N/C = Consonants must not occupy the syllable nucleus
9 The Generator. Tableaux
Before we proceed any further, we need to bring out an issue which is as

yet only implicit: if the grammar only contains constraints, where do the dif-
ferent CANDIDATES that we are evaluating originate?
This is of course a central question, because without candidates there would be

no evaluation, and without evaluation the constraints would remain unenforced.
The answer is that the model includes a further component, named GEN,
GEN randomly for “generator”, which randomly generates surface forms from each lexical
oes representation, using only the most elementary principles of linguistic logic.
each lexical
representation
GEN will therefore produce a variety of syllabic parsings, a variety of asso-

ciations between elements in different tiers (including features), and so on.
For instance, given the sequence /hzepi/, we can assume that GEN will offer
us the candidates in (9), at least:
(9) Oo Oo o Oo G6 Oo
R
aa R R R R
N
ie peg
N NN N
aseptic
linFees ep AL hve p 1
Constraints 595
The job of the constraints, specifically ranked for each grammar, is the
EVALUATION of these candidates, that is, the selection of one candidate The job of
as the most HARMONIC with respect to the grammar in question. eee Si
EVALUATION of
candidates, one of
which is selected
as the most
HARMONIC
The candidate thus selected is the WINNER of the evaluation procedure.

Accordingly, this candidate emerges as the surface representative of the
corresponding underlying form.
The interaction between constraints and candidates is conventionally
represented in tables with a special format, known as “tableaux” (singular:
TABLEAU). In (10) we give the tableau for the syllabic structure of happy:
(10) ONSET NucLEUS 3No-coDA 7*COMPLEX ;

*M/V |
*N/C
596 Constraints
You will notice in (10) that constraint violations are signalled with an asterisk
in the appropriate box, while constraint compliances are simply left blank.
You can also see that a tableau is simply a plotting of a set of candidates
A tableau is a (usually listed in the left column) against the set of constraints (usually listed
plotting of a set in the top row), with boxes in every intersection, in the style of a bar of choc-
of candidates
olate. Unranked constraints are separated by a dotted line, the tableau equi-
against the set of
constraints valent of a comma in the text. The superiority of candidate (10a) (marked with
a pointing hand as the winner) is clear at a glance: it is the only candidate
which does not incur any violation of any constraint.
The visual advantages of tableaux are obvious. We could of course have
included other candidates in (10) (and, correspondingly, in (9)), since, as we
mentioned earlier, GEN knows few limits, if any, on the number of candidates
it can produce. However, the common practice is, sensibly, to limit the selec-
tion of candidates to plausible forms. Similarly, only the constraints that are con-
ceivably relevant to the phenomenon being examined are usually considered.
“relevant constral
For your convenience, we end this section with a schematic summing up

of the machinery of Optimality Theory we have introduced so far:
(11) Optimality Theory machinery:

Device Function
GEN(erator) Generates candidates
Constraints Reject candidates
Constraint rankings Define specific grammars
Tableaux Provide a visual representation of the process
of evaluation of specific lexical forms by the
ranked constraints
Ponder over this lis
Basic English Syllables

We will now put the machinery we have introduced to use to analyse English
syllables. We shall enrich and refine this machinery as the need arises.
Constraints 597
Consider first a simple and ordinary word like egg. This form violates ONSET
and No-Copa.
The violation of Onset could in principle be remedied by the insertion of a

consonant in the onset position. Such insertion, or EPENTHESIS, is indeed the
norm in some languages: for instance, Arabic epenthesizes [?] when there is
no lexical segment to occupy the onset position, as does German foot-initially.
In English, however, the pronunciation [eg] prevails over its epenthesized
alternative [?eg], at least in ordinary speech.
The prevalence of the pronunciation [eg] means that, in the case we are
discussing, faithfulness to the lexical form takes priority over filling the
onset. The (faithfulness) constraint that blocks epenthesis is known as
Dep[ENDENCY] (FILL in early versions of Optimality Theory). DEP requires
every surface segment to have a lexical correspondent, on which it there-
fore “depends”.
Clearly, in order to get [eg] rather than *[?eg] we need to rank Dep higher
than ONSET, as in (12). Higher ranking is represented in tableaux by means
of a solid vertical line between the columns of the two relevant constraints:
You can see that the consequence of ranking Dep higher than ONsET is the
selection of [eg] over *[?eg] as the surface form: [eg] obviously fares better
than *[?eg] with regard to the higher-ranked constraint DEP. Notice in par-
ticular that [?eg] incurs a violation of Dep, while [eg] does not. The violation
598 Constraints
of the higher-ranked Dep by [?eg] is therefore a FATAL VIOLATION, and

An end-game situ- the evaluation need not be carried any further. This end-game situation is
ation is signalled
signalled by the exclamation mark after the asterisk in the Dep column and
by an exclamation
mark after the the shading of all the subsequent boxes, which are now irrelevant to the
asterisk and the evaluation procedure: the violation of the lower-ranked constraint ONSET
shading of all the by the winning candidate [eg] in (12) is of no consequence.
subsequent boxes
Next, let us examine the violation of No-Copa by [eg]. There are two obvi-
ous ways of circumventing this violation. First, we can epenthetize a final
vowel, say, [a], to create an additional syllable of which /g/ will be the onset:
[e.ga]. This result obviously does not match the data, and therefore we con-
clude that No-Copa is also ranked lower than Dep (ONsET and No-Copa
are of course equally ranked, as indicated by the dotted line separating their
respective columns):
Number of viola-
tions is irrelevant
In (13) it becomes clear that number of violations is irrelevant to the evalu-
to the evaluation ation procedure: the two candidates in (13) tie in number of violations, but
procedure [eg] still wins. The decisive criterion is, obviously, compliance with the higher-
ranked constraints, in line with the Ranking Metaprinciple in (4) above.
An alternative strategy to circumvent the violation of No-Copa by /eg/
involves deleting /g/, to obtain the output form [e]. This form may violate
other constraints of English (in particular, it violates the constraint that words
must contain at least one binary foot), but it manifestly does not violate
No-Copa. However, faithfulness prevails over structural perfection here too:
you will recall from chapters 9, 10 and 16 that English does allow for codas.
Formally, therefore, No-Copa needs to be ranked lower than Max{[IMALITy]
(PaRsE in early Optimality Theory), the constraint responsible for the maxi-
mal mapping of underlying segments onto the surface:
Constraints 599
OnsET : No-Copa
You will notice that in (14) Dep and Max are fatally violated by the losing
candidates ([ega] and [e], respectively). Note that, because these two con-
straints are mutually unranked, the Max box is not shaded, even though
the preceding box contains a fatal violation.
Tableaux (12) to (14) obviously leave out many constraints.
es
Indeed, tableaux are routinely simplified in order to keep them within man-
ageable spatial limits: the total number of constraints is very high, probably _Tableaux are rou-
running into the hundreds, and the number of candidates is also very high, "ely simplified in
order keep them
indeed infinite if epenthesis is given a free rein.
within manage-
able spatial limits
Granted all this, we now combine in (15) the tableaux in (12) and (14)
(which subsumes (13)), for a more complete evaluation of [eg] and its
competitors:
(15)
Tableau (15) makes clear why [eg] is selected over all its competitors: it fares
best with regard to the highest-ranked constraints, Dep and Max.
600 Constraints
Constraint summary:
ONSET
No-Coba
Dep (formerly Fit) = Every surface segment must have a lexical correspondent
Max (formerly Parse) = Underlying segments must be mapped onto the surface
[4 Syllable Complexities
We have now analysed egg successfully. You know, however, that English
syllables may include complex onsets and complex nuclei. We capture this
legitimate complexity of English onsets and nuclei by ranking the structural
constraint *COMPLEX lower than the faithfulness constraints DEP and Max,
as we illustrate with clay in (16):
OnsET :No-Copa } *COMPLEX
Now, while clay is a legitimate English word, */kay obviously is not, even
though it contains exactly the same segments.
Constraints 601
The reason for the difference between clay and *Ikay lies, of course, in the
fact that *Ikay violates Sonority Sequencing, which we incorporate into our
repertoire of constraints under the label Son[ority].
/\ker/ ONSET :No-Copa 7*COMPLEX
Notice that the table in (17) does not constitute a tableau as such, because
it carries out no evaluation. We have included it here simply to illustrate
the high ranking of Son.
A real-life example of a Son violation is provided by the word sky. We
obviously also want Son to prevent /sk/ from making up a complex onset.
In chapter 16 we suggested that such a word-initial /s/ is salvaged from
phonetic obliteration by direct affiliation to the Phonological Word (PW),
skipping the o node.
Sonority Sequen-
cing and the Strict
Layer Hypothesis
are encapsulated
in the constraints
The affiliation of the word-initial /s/ to PW violates the Strict Layer son and SLH,
Hypothesis in (58) of chapter 16, which we now incorporate into the OT model __ respectively
as the constraint SLH.
602 Constraints
(18) ONSET /No-Copa *COMPLEX :SLH_-
PW - a
The inevitable violation of SLH by word-initial s + obstruent in English occurs

because, as we mentioned in chapter 16, PW edges are endowed with spe-
cial licensing powers, at least in English. Word-edge licensing also accounts
for English extra word-final coronals, which we suggested in chapter 16 above
also attach directly to the PW node (cf. si[ks@s]).
In chapters 10 and 16 we also saw that English allows one extra word-
final consonant to be directly licensed by 6, provided Sonority Sequencing
is complied with: lame, lamp.
At present, the final p of lamp would be parsable in the coda, together with
m, since we have ranked *CoMPLEx lower than Dep and Max, to allow for
complex onsets. However, given the arguments in chapters 10 and 16, this
parsing is undesirable.
Constraints 603
We shall resolve the tension between the need for a low ranking of *Com-
PLEX for onsets, and the need for a high ranking for codas by the inclusion
of a specific constraint “COMPLEX”, over and above the general constraint
*COMPLEX. *COMPLEX®”* must obviously be ranked higher than Dep and
Max. By transitivity, it will also be ranked higher than *ComPLex, indeed
as follows independently from the Elsewhere Condition, renamed PANINI’S
THEOREM in Optimality Theory to honour its original discoverer: In OT the Else-
where Condition is
renamed PANINI'S
THEOREM, to
honour its original
discoverer
(19)
One last issue needs addressing before we close the discussion on syl-
lables. In particular, the low ranking of *ComPLex is consistent with the selec-
tion of both onsets and rimes with more than two elements, as in *[blju:]
and [lerm], respectively (complex codas are of course totally ruled out by
*COMPLEX™™*).
604 Constraints
We clearly need an undominated constraint to rule out such configurations.

Crucially, this constraint must outrank Max, which would force any number
of lexical segments through to the surface.
The constraint we are looking for is *COMPLEX&COMPLEX. This type of
constraint differs from all our previous ones in that it contains a self-
A self-conjoined conjunction. Clearly, the self-conjunction of COMPLEX will rule out any con-
constraint rules stituent which includes more complexity than the minimal two-element one.
out forms which
A rime with a complex nucleus and a coda, for instance, will be complex
incur more
violations than twice over:
the ones barred
by the simple
constraint (20) R
tay
/\ eleey
ae Complex
—.~—=Gomplex
The configuration in (20) violates *CompLex&CompLex, and consequently it

will be disallowed if this constraint is sufficiently high in the ranking: it is
in fact undominated in English.
Constraint summary:
* COMPLEX
Dep
Max
* COMPLEX" = No complex codas
*COMPLEX&COMPLEX = No joint complexity
SLH = Strict Layer Hypothesis
SON = Sonority Sequencing
—_—_—_—_—_—eeeee————————————————
Constraints 605
Basic Metrical Structure
We will now propose an interpretation of stress in terms of Optimality Theory.

You know from chapter 13 above that in English verbs and primary adject-
ives stress is maximally penultimate.
In Optimality Theory, this pattern is formalized by means of the constraints

V/u, Foor Binarity, FOOT TYPE oan and ALIGN-RIGHT, as we shall now explain.
V/u demands that rime vowels be moraic. F[oo]T Brn[ARITy] requires feet
to be strictly binary, either in terms of moras (in quantity-sensitive systems) FTBIN requires feet
or in terms of syllables (in quantity-insensitive systems). English is of course . Perey
quantity-sensitive, and therefore all English feet will have to be bimoraic. pa
F[OO]T TYPE,,ocfucey licenses left-headed feet (“trochees”), as against right-
headed feet (“iambs”), which require the opposite setting, F[oo]T TYPE,,.,.-
We illustrate the action of these three constraints in (21), before proceeding
to discuss the fourth constraint we listed, ALIGN-RIGHT:
(21) | /develop/ 3 FrBIN Fr-TyPe TROC
ii
develop
bas*
(W pL)
develop
*+
(Hf)
develop
oars
(uw)
develop
606 Constraints
You can see that candidates b. and c. are rejected, on account of their viola-
tion of FrBIN and FT-TyPE,,,<, respectively. By contrast, both candidate a. and
candidate d. emerge unscathed. We know, however, that only candidate a. is
legitimate. The necessary further selection is carried out by the constraint ALIGN-
RIGHT, a member of the ALIGN family. As suggested by the label, ALIGN con-
ALIGN constraints straints regulate the spatial relation between two elements. In particular, ALIGN
implement either implements either strict linearization, by requiring that two elements be
strict linearization
sequentially abutting, or structural alignment, by requiring that the edges of
(by requiring that
two elements
two constituents in different structural levels be superimposed. In the case
be sequentially we are discussing, we want the right edge of the prosodic word to be super-
abutting) or struc- imposed on the right edge of a foot. We illustrate the situation in (22):
tural alignment
(by requiring
that the edges of V/u FrBIN 7Fr-TyPE TROC
7ALIGN-RIGHT
two constituents
in different struc-
tural levels be
superimposed)
(W Wu
develop
The desired result, devélop, has now been attained, as you can see.
————
Constraint summary:
ALIGN-RIGHT = The right edge of the prosodic word must be superimposed on the right edge of
a foot
FTBIN = Feet must be strictly binary
FT-TYPE.y, = Feet must be right-headed (“iambs”)
FT-TYPE;eo¢ = Feet must be left-headed (“trochees”)
V/u = Rime vowels must be moraic
rr
Constraints 607
El Extrametricality
You will recall that the stress pattern of English nouns is like the stress
Right extrametri-
pattern of English verbs, with right extrametricality added. In Optimality
cality is encoded
Theory, right extrametricality is encoded in the constraint NoN-FIN[ALITY], in the constraint
which keeps metrical structure away from the final edge of the phonological NON-FINALITY
word:
(23)
a +
(Uw)
Americ a
The foot of candidate b. includes the mora on the right edge of the phono-
logical word, and therefore it incurs a violation of NoN-FIN. Candidate a. of
course incurs a violation of ALIGN-RIGHT, which you know requires that the
right edge of the phonological word be superimposed on the right edge of
a foot. Indeed, the requirements of ALIGN-RicHT and Non-FIN are mutu-
ally at odds, since the satisfaction of one of these constraints inevitably results
in the violation of its counterpart.
The existence of a conflict between the two constraints obviously means that
for the set of English nouns and derived adjectives NON-FIN must outrank
ALIGN-RIGHT:
608 Constraints
For verbs and underived adjectives we need the opposite ranking —

remember that we are assuming that all constraints are present in all gram-
mars, although their ranking does of course vary from grammar to grammar:
(25)
*
a.
L(t HL)
develop
The tableau in (26) incorporates the full set of constraints directly relev-
ant to stress in nouns:
Constraints 609
(26) V/u 7FTBIN :Fr-TyPE., 3Non-FIN ALIGN-RIGHT
a. es
(UL
ma America
b. *
ML iyWW)
Amer ica
Cr
(4 WHE
America
dots
Me (WpBW)
America
+
We (Wp WL
Americ a
You can see that the most harmonic candidate is, inevitably, a., which only
incurs one violation of ALIGN-RicuT. Notice in particular that candidate c.
incurs two such violations, and, consequently, it is discarded.
Constraint summary:
ALIGN-RIGHT
FTBIN
FT-TYPE
Roc
V/u
NON-FIN = No metrical structure on the final edge of the phonological word
EJ Quantity-Sensitivity
In chapter 13 we saw that a heavy syllable in the vicinity of the right edge
attracts the stress: compare agénda with América, and collapse with devélop.
610 Constraints
The OT constraint responsible for this attraction is called the WEIGHT-TO-

In OT, accenting Srress PRINCIPLE (WTS), and simply stipulates that a heavy syllable must
of heavy syllables carry stress. The application of WTS to agenda or collapse obviously presup-
is encapsulated in
poses the assessment of gen and lapse as heavy. We achieve this assessment
a constraint
labelled WeIGHT- through the OT equivalent of the Weight-by-Position principle projecting
TO-STRESS PRINCIPLE moras from coda consonants. Remember, however, that the word-final con-
sonant does not project a mora: compare colldpse with devélop, not *develdp,
because it has only one consonant on the right edge. This fact prompted the
restriction of WEIGHT BY PosiTIOoN (WBP) to coda consonants immediately
followed by another consonant (and preceded by a vowel) —in all other cases,
mora provision is suppressed by the constraint *u, a manifestation of the
more general *STRUCTURE, which simply disfavours structure:
(27) | /develop/ |V/u :FrBIN 7Fr-TyPE soc, ALIGN-| WTS| WBP

iRIGHT |
Ce
(WW Wy)
develop
d. ‘i
p(y)
deve lop
e. He
L(U ML)
ei
develop
WTS|WBP|Non- *p
ALIGN-,,./
/collapse/ |V/ FrBin |Fr-Tyrr,
7
(28)
iRIGHT | | FIN
a. .
LL (LL)
collapse
bot
(HW Wp
colla pse
c +
(Wu Hy)
collapse
d.#
(u MW)
collapse
e +
(up)
collapse
The evaluation of agenda runs along similar lines to that of collapse, even
though the consonant cluster is now word-internal. The ranking of ALIGN-
Ricut and Non-FIn is, of course, reversed, as we know is the case with all
nouns:
612 Constraints
/agenda/ | V/ W.FrBrn |
Fr-TyPe 'TROC _Non-/
} WTS WBP ALIGN- | *U
FIN
Constraint summary:
ALIGN-RIGHT
FTBIN
FT-TYPE,po¢
NON-FIN
V/p
WBP = Weight by Position
WTS = Heavy syllables must carry stress
*u = No mora
*StTRUCTURE = No structure
Secondary Footing
In chapter 15 we saw that in words like prosopopéia the word-initial secondary

stress is obtained by means of left-to-right (non-cyclic) footing. In OT, this
requires the addition of a constraint ALIGN-LEFT, yielding the opposite effect
to ALIGN-RicuT. In words like these, ALIGN-Ricut therefore needs to be re-
stricted to the head foot, and thus we recast it as ALIGN-RicutT"™”"", ALIGN-
Ricut’™*’™ outranks the more general ALIGN-LErt by Panini’s Theorem, as
we show in (30). To keep the tableau simple we omit those constraints which
must be obvious by now:
Constraints 613
(30) / prosopopeia/ Non-FIN |WTS |WBP ALIGN- ALIGN-LEFT

: : RicuTr™>Ft
(MB) M(H)
ZS
prosopopei a
(WW) WC)
prosopopei a
wu W)(up)
prosopopei a
Words like Epaminéndas, with secondary stress in the second syllable,

ostensibly require a constraint NoNn-INITIAL, mirroring NON-FIN, to ensure
that the initial syllable is skipped.
Constraint summary:
NON-FIN
WBP
WTS
ALIGN-LEFT = Feet must be aligned with the word's left edge
ALIGN-RiGHT"“°"" = The head foot must be aligned with the word's right edge
NON-INITIAL = The word-initial syllable must not be metrified
Correspondence Constraints Ordering, whether

of rules or of
strata, is at odds
In chapter 16 we presented evidence for the stratification of phonological with the common
rules. The resulting strata, or levels, are by definition mutually ordered. tenet of OT that
Ordering, whether of rules or of strata, is at odds with the common tenet of constraints apply
OT that constraints apply in parallel, with no intermediate stages between in parallel
the lexical and the surface representations. We shall resolve this conflict by
614 Constraints
introducing transderivational CORRESPONDENCE CONSTRAINTS, that is,

CORRESPON- constraints that impose correspondences between surface forms.
pee In the English of both New York and Philadelphia the low front lax vowel
STRAINTS impose [ze] tenses to [45] in closed syllables, subject to additional requirements on
correspondences ; ; :
ce aartace the identity of the coda consoriant, which are irrelevant for us here ([AE] is
forms our ad hoc phonetic symbol for a tense [z]):
(31) a. [4] b. [el]

man. ma.nage
can.did ca.nnibal
mas.ter Ma.ssachusetts
jam. janitor
plant. pla.net
This alternation permeates some derivational paradigms, as expected:
(32) a. [A] b. [e]

class. cla.ssic
mass. ma.ssive
pass. pa.ssive
In other forms, however, [a] tenses even in ostensibly open syllables:
(23) teary LA DL Ae
class. cla.ssy
mass. ma.ssable
pass. pa.ssing
In (33b), the principle of Minimal Onset Satisfaction, now recast as the OT

constraint ONsET, will parse the intervocalic consonant in the second syl-
lable: cla.ssy, etc. Despite this, the [ze] tenses to [AE], as shown.
By this stage you should have no difficulty in detecting a crucial differ-
ence between the forms in (32b) and those in (33b), over and above the dif-
ference in the quality of the vowel in question.
In particular, the derivatives in (32b) have suffixes that in chapter 16 we

allotted to level 1, whereas the derivatives in (33b) have level-2 suffixes.
Intuitively, this difference is responsible for the different fate of the vowel
in the two sets. We will now propose an OT formalization for this intuition.
Constraints 615
We will encapsulate the tensing of /z/ in a constraint prohibiting the occur-

rence of [ze] in closed syllables: *[aec.]. We will assume that other constraints,
which we will not include in the evaluation to keep the exposition simple,
rule out undesired solutions to the problem input /zC./ — for instance, the
deletion of the C (ruled out by a high ranking of Max), the epenthesis of
a vowel (ruled out by a high ranking of Dep), or the realization of the
underlying /a/ as a vowel other than [A] (ruled out by a high ranking of
the relevant IO-IDENT,,,, constraints).
The other constraints directly involved in the evaluation are *TENSE,,,, Which
prohibits tense low vowels, and IO-IDENT,,,,,, Which requires the surface
realization of the underlying value for tense. We illustrate with class [kl4s]
in (34):
klees
b.
You can see that the candidate with [A] is selected irrespective of whether
we postulate /Ai/ or /z/ underlyingly. For classic, with an open first sy]l-
lable, the result will be the opposite one, correctly so:
(35). a:
a
616 Constraints
If we apply the same procedure to classy, we will obviously obtain cl[z]ssy,

which is incorrect for the accents in question. As a consequence, we shall
adopt a constraint BD-IDENT,,,,, imposing identity in tenseness between the
BASE class and its DERIVATIVE classy. Like the IO-IDENT constraints
before, BD-IDENT,,5, presupposes a CORRESPONDENCE relation between two
Constraints that terms. Such constraints are accordingly categorized as CORRESPONDENCE
presuppose a cor-
CONSTRAINTS, and the branch of Optimality Theory that encompasses
respondence rela-
tion between two
them is referred to as CORRESPONDENCE THEORY. Clearly, BD-
terms are categor- IDENT,,xsz Must outrank all the other relevant conflicting constraints if the
ized as CORRES- desired output is to be obtained. Against this background, the evaluation
PONDENCE of cl[A]ssy goes as follows:
CONSTRAINTS
The branch of
Optimality Theory
that encompasses
correspondence
constraints is
referred to as
CORRESPON-
DENCE THEORY
Because of the high ranking of BD-IDENT,,,.,, the output cl[A]ssy prevails,

irrespective of the requirements of the other constraints.
The high ranking of BD-IDENT,,,,, must obviously be confined to forms

with level-2 suffixes, such as the -y of classy. In forms with level-1 suffixes,
BD-IDENT,,,., must be outranked by *[zc.], TENSE,,,, and IO-IDENT yxxy557 aS
you can verify by re-examining the tableau in (35) above. This means that
the difference between strata needs to be maintained under OT, as it surely
does under any theory, given the facts.
What will vary is the mode of implementation — by rule and affixation

ordering in rule-and-derivation phonology, and by _transderivational
Constraints 617
correspondence constraints and constraint ranking in OT. Notice that there

is a close connection between both approaches, despite apparent disparities.
Indeed, it is possible to mimic the correspondence formalism of OT in rule-
and-derivation formalism by stipulating that in level 2 phonological segmental
processes take place before affixation:
(37) /kles/ /kles/ /klzes/

Level 1:
-Ik suffixation — --------- klees-Ik —---------
Syllabification -kles. _.kle.sik. .kles.
Level 2:
ze-Tensing oe NA A
-Isuffixation = ---------__--------- kKlAs.1.
Resyllabification NA NA KIA. sI.
[kl4s] [klestk] [kl4s1]
You can see that z-Tensing needs to be ordered in stratum 2. In particular,

if it were ordered in stratum 1 it would be cyclic, and it would wrongly apply
in forms with stratum-1 suffixes like classic. By contrast, by being ordered
before affixation in stratum 2 it has precisely the desired effects.
Constraint summary:
*[ec.] = No lax [e] in closed syllables
BD-IDENT,.;. = Base-derivative faithfulness for tenseness
IO-IDENTyeyse. = Input—output faithfulness for tenseness
*TENSE,., = No tense low vowels
Cyclic Effects
In chapter 15 we presented some rather spectacular results of the cyclic

application of rules in the tonal phonology of Tiv and in the metrical phono-
logy of English. Correspondence constraints can also replicate the cycle of
derivational phonology. Consider the words in (38), which supplement the
ones we used in chapter 15 to justify the English stress cycle:
618 Constraints
(38) a. medicinal b. medicinality

divisible divisibility
Napédleon napolednic
personify personification
assimilable ' assimilability
You will notice once more that the secondary stress of the derivatives in b.
falls on the syllable that bears the primary stress in their bases in a., rather
than on the initial syllable predicted by the secondary stress procedure: com-
pare prodsopopéia in section 10 above. We have of course come across other
instances of secondary stress in the second syllable (cf. Epaminondas), but,
clearly, none of them is derivative the way those in (38) are.
The answer lies again in the correspondence constraint BD-IDENT, in this

case referring to stress: BD-IDENT,:_:s;. Clearly, this constraint must be
ranked higher than the ALIcN-LerT that yields initial stress in prosopopéia:
(39)
* *
LW(w)(LLd)
TS
napo le onic
(HW HW) (Up)
napo le onic
The confrontation between napdleénic and *napolednic is decided in favour

of the former by BD-IDENT,;2:5, Which is fatally violated by *napoleonic.
Constraint summary:
ALIGN-LEFT
ALIGN-RIGHT"*°""
BD-IDENT. +s; = The stress of the base must be preserved in the derivative
SS
Constraints 619
Word Formation through Truncation
Correspondence constraints can also govern the phonology of word forma-

tion. An interesting example is provided by such word truncations as those —_Correspondence
in (40): constraints can
also govern the
phonology of
Janice Jan
cafeteria caf
Massachusetts Mass
The relevant question is whether the vowel [z] of the truncates will
undergo tensing to [A] in New York and Philadelphia.
It obviously should on purely phonological grounds, since the constraint

*[zec.] rules out the non-tense vowel [e] in closed syllables. In (40) and
similar cases, however, the vowel surfaces as lax: P[a]m, etc.
The a of the base Pamela is of course non-tense, given the constraint

*TENSE,,,, and the obvious irrelevance of *[zc.] here:
620 Constraints
/pAEmela/ | BD-IDENT,,,,,. | *eeC. IO-IDENT

pyr
/
p4.mela
Piaee oneal
You can see that the predominance of *TENSE,,w enforces a non-tense [a] even
when we hypothesize an underlying tense /A:/. In turn, the [z] of the base
Pamela is transmitted to the truncate Pam by the ranking of the correspond-
ence constraint BD-IDENT,,,,, above the phonological constraint *[zec.].
Constraint summary:
*[ec.]
BD-IDENTyexce
IO-IDENTyexse
* TENSE,ow
OT Morphology: English Plurals
OT is also relevant to morphology. In particular, we shall now show that

The ALIGN con- the ALIGN constraint family can be used to get morphemes concatenated in
straint family can
the correct order. Moreover, these constraints interact with phonological con-
be used to get
morphemes constraints in the selection of the correct candidate. It is currently being sug-
catenated in the gested that OT is also relevant to syntax, but this matter obviously falls beyond
correct order the scope of this book.
Orthographically, regular English plurals are formed from the singular by
the addition of -(e)s. In particular, the forms in (43b) are well-formed plurals
of the forms in (43a), but those in (43c) are not:
Morphological
constraints interact
(43) a. map b. maps c. smap
with phonological
constraints in the pod pods spods
selection of the day days day
correct candidate ash ashes sash
coat coats coast
Clearly, the reason for the ill-formedness of the pseudoplurals in (43c) is not
phonological. Rather, it is a function of the suffixal status of the regular
plural morpheme, orthographically -(e)s. Indeed, not only is -(e)s a suffix,
but also it must be the right-most one:
Constraints 621
(44) a. action b. actions c. *actsion

printer printers *printser
cartoonist cartoonists *cartoonsist
institution institutions *institutestion
Our formal arsenal already includes the resources we need to obtain the
correct result, namely, the ALIGN family of constraints. In particular, the
ALIGN family makes available the constraint in (45), among many others:
(45) ALIGN LEFT-RIGHT,, yeat-stem:

The left edge of the plural morpheme /z/ must be aligned with the
right edge of the stem
The inclusion of this constraint in the morphological grammar of English

accounts for the well-formedness of the forms in the b. columns of (43) and
(44), and the ill-formedness of their counterparts in the c. columns. This is
thus a successful application of OT to morphology, if a particularly simple
one. More complex cases of course exist, many of them in languages “exotic”
to the English speaker: we shall not go into these here in order not to lengthen
the exposition unduly (some such cases are included in the companion work-
book, and you can check them out there).
To ease our way in, we have been using the orthographic representation
-(e)s for the English plural morpheme, but by now you are fully aware that
spelling is ultimately irrelevant to phonology. Consequently, we must now
supply the proper phonological details of English plural formation.
The regular English plural morpheme has three allomorphs, as follows:
(46) English regular plural allomorphs:

[z]: | day-s, pie-s, shampoo-s, piano-s, zebra-s
road-s, leg-s, parcel-s, foundation-s, nerve-s
[s]: | map-s, lot-s, obelisk-s, puff-s, birth-s
[az]: ash-es, fortress-es, surprise-s, age-s, approach-es
622 Constraints
We shall rationalize these allomorphs by adopting /z/ as the underlying

representation, from which the two other forms will be derived by voice assim-
ilation (-z — -s) and schwa epenthesis (-z — -az), respectively. These two
changes are phonetically grounded: voice assimilation comes about as a result
of inertia of vocal fold activity (there is related discussion in chapter 2 above).
In turn, the avoidance of a sibilant cluster, here resolved by the insertion of
the maximally underspecified vowel [a], follows from the difficulties asso-
ciated with the articulation and perception of such clusters.
bilant”
isan infor
ill the sibila
In rule-and-derivation phonology, we would need the two rules in (47) to

obtain the surface forms. In (47a) we have abbreviated to z all the features
of [z] other than [+voice], in order to keep the representation simple:
(47) a. Plural devoicing:
[+cons] [+cons]
[- son] Stem [- son] Plural
Laryngeal Laryngeal
[—voice] [+voice]
b. Plural epenthesis:
OX /X X
“
[—cons]
.
[+cons]
i
[+cons]
ents aan
[coronal] [coronal]
Constraints 623
In (47b) the underspecification of the inserted vowel eventually leads to schwa,

correctly so for many accents, although in some the plural vowel is [1], an
irrelevant matter here.
The OT constraints equivalent to the rules in (47) are as in (48):
(48) a. VorckE Harmony: *[avoice] [-avoice]

b. OCP on SIBILANTs: *[cor] [cor]
[+strid] [+strid]
Obviously, additional constraints are needed if precisely the desired out-

put is to be selected, rather than an output involving deletion of one of
the two clashing segments, and so on. In particular, it is necessary to get
violations of VorcE HARMony in (48a) resolved by devoicing, rather than
by epenthesis, which is only fallen back on to as alast resort. Accordingly,
we shall introduce the faithfulness constraint ANCHOR-LEFT,,y,,,, Which
requires the left edge of the lexical plural morpheme to remain such in the
surface. The remainder of the constraints and their rankings will simply be
assumed:
624 Constraints
ANCHOR-LEFT,,
year
Constraint summary:
ALIGN LEFT—RIGHTpiyeat-srem= Left edge of plural morpheme aligns with right edge of stem
ANCHOR-LEFTo: yea: = Surface left edge of plural morpheme corresponds to its lexical left
edge
OCP ON SIBILANTS = No adjacent sibilants
Voice HARMONY = Uniform voice in consonant clusters
English Possessives and Correspondence Theory
The mechanics of the English possessive (’s) would seem to be similar to

those of the plural:
(50) a. [z]: Ted’s, John’s, Tony’s, Gayle’s

b. [s]: Pat's, Dick’s, Ralph’s, Math’s
c. [az]: Rose’s, Reg’s, Maurice’s, Thomas’s
Indeed, the conditions on the distribution of the possessive allomorphs seem

to be identical to those of their counterparts in the plural.
There is an interesting twist with the possessive of the plural, however.

In particular, the possessive of a form with the plural morpheme /-z/ is iden-
The possessive of tical to the non-possessive plural:
a form with the
plural morpheme
(51) a. - the cats house here
/-z/ is identical to
the non-possessive b. the cats’ house is here
plural
This situation is indeed peculiar on the strength of our constraints above,
which would make us expect *catses.
Constraints 625
(52) the men who run’s yellow jumper
Crucially, there is no incompatibility between the possessive and the plural:
(53) children’s, oxen’s, mice’s
If possessives could not be added to plurals, the forms in (53) would be

ungrammatical, but clearly they are not.
All the facts considered reveal that the possessive marker -z cannot be added
to the plural marker -z. In a rule-based approach, we would first have to
allow the concatenation of both markers and then delete one. Alternatively,
we could block the concatenation of one of the markers. Either way, the ques-
tion arises as to which marker gets left out, and the answer has to be arbit-
rary, undesirably so.
The OT framework allows a more principled answer. All we have to do is

adapt the ALIGN LEFT-RIGHT,, yar-srem Of (45) above to possessives, as in (54):
(54) ALIGN LEFT—RIGHT possassive-stem:

The left edge of the possessive morpheme -s must be aligned with the
right edge of the stem
The crucial aspect of this constraint is that, like its counterpart for the plural,
it enforces alignment of the possessive morpheme with the stem.
626 Constraints
The obvious consequence of this formulation is that both the plural and the
It is unusual for possessive morphemes will be competing for the same position. What is un-
two morphemes
usual about this case is that they both get their way. Indeed, the machinery
competing for the
same position of Correspondence Theory is remarkably well suited to the attainment of
both to get their this goal, as we now showfor the form cats’. Note that the identity of each
way segment is made explicit in the representation as a subscript:
(55) |/k, ity24,25/|VorCe-HARM OCP-Sin ALIGN-RicHT ANCHOR-LEFT POSSESSIVE
The machinery of
Correspondence
Theory is remark-
ably well suited to
the attainment of
this goal
You can see that in the winning candidate the noun’s stem is aligned with
both the plural morpheme (z,) and the possessive morpheme (z;). Therefore,
the requirements of the two respective ALIGN-RIGHT constraints are com-
plied with in this form.
Constraint summary:
OCP ON SIBILANTS
Voice HARMONY
ALIGN LEFT-RIGHTyossessvesrem = The left edge of the possessive morpheme -s must be aligned with
the right edge of the stem
ANCHOR-LEFT, POSSESSIVE = Surface left edge of possessive morpheme corresponds to its lex-
ical left edge
In this chapter we have introduced an alt

tionship between lexical forms and |
Optimality Theory. logic.
Constraints 627
eratesa list of execs N

aps being the on
harmonic, form, while ae ran

straint violations incurred by the
asa “tableau”. In the course of this
Key Ques tions
1 What is the effect of faithfulness 5 What is the effect of the Max

constraints outranking markedness constraint?
ones? 6 What is the effect of the Dep
2 List the constraints which exist on constraint?
syllable markedness. 7 Under what name have you pre-
3 What is GEN? viously encountered Panini’s The-
4 The ranked constraints evaluate orem? What is its relevance in
candidates: what for? OT?
628 Constraints
8 Outline the constraints relevant to 42 How do correspondence constraints

basic metrical structure. replicate cyclic effects?
9 How is extrametricality incorporated 13 How can the Aticn family of con-
into OT? straints ensure the correct order of
10 How is quantity-sensitivity incor- morpheme concatenation?
porated into OT?
11 What are correspondence con-
straints? What function do they
perform?
Fou rt hier. 2Po franc. Iicae
Syllable Structure in Urban Hijazi Arabic
Alternative candidates have been provided in the following data set from
Urban Hijazi Arabic, spoken in Saudi Arabia:
mak.ta.bi *ma.kV.ta.bi ‘my office’

mak.ta.ba.ti *mak.tab.at.i ‘my library’
dah.ra.da.ti *da.hra.da.ti ‘my rolling’
fa.@a.ra.ttuhum *fa.dja.ra.ttu.khu.mV ‘their tree’
(i) | Work out the correct rankings for the following constraints, in order
to produce the correct forms:
No-Copa
ONSET
*COMPLEX
DEP
(ii) Show the effect of your rankings on maktabati and dahradgati.

(iii) Can your constraint ranking account for:
faanuus *fa.Ca.nu.Cus *fa.nu ‘a lantern’
(iv) Suggest further constraints to accommodate the additional data.

(v) How does your ranking need to be adjusted?
Stress in Latin
We repeat below the data from Latin which we provided earlier. Basing your
analysis on the set of constraints outlined in the text, work out the constraint
ranking which will yield the correct metrical structure for Latin:
Constraints 629
amiicus ‘friend’ refé:cit ‘set over’ (3sg.perf)

agricolatrum ‘farmer’ (gen.pl.) agricola ‘farmer’ (nom.sg.)
volucres ‘winged’ (nom.1.) — pepérci— ‘spare’ (1sg.perf)
simula: ‘snub nosed’ (fem.) magister ‘master’
Icelandic Deverbal Nouns
Syllable structure considerations dictate the choice of nominative singular

allomorph in Icelandic, as we can see from the following alternations:
a. Nom.sg. Acc.sg.
dagur dag ‘day’
beer bee ‘farmhouse’
leknir lekni ‘physician’
Nom.sg. Dat.sg.
lifur lifri ‘liver’
akur agri ‘field’
hestur hesti ‘horse’
b. Nom.sg. Acc.sg. Dat.pl.

lyfur lyf lyfjum ‘medicine’
bylur byl byljum ‘snowstorm’
(i) | What are the two allomorphs?

(ii) | Give the constraint rankings for the constraints Max, Son and Dep.
Deverbal action nouns in Icelandic are exceptions to the final cluster con-
straints, as we show in c::
c. Infinitive Deverbal action noun

klifta ‘climb’ klifr ‘climbing’
kumra_ ‘bleat’ kumr_ ‘bleating’
grenja_ ‘cry’ grenj ‘crying’
(iii) What further constraint interacts with those already proposed?

(iv) Show the new rankings for the forms in c.
LOOKING BACK AND MOVING ON
mw An introduction to the history of distinctive feat et

mental representation. — _ -
So far in the text of the book, we have endeavoured not to burden the reader
with contradictory or disparate information, and therefore we have tried to
present a homogeneous picture of phonology as clear and free from debate
as possible: quite obviously, it is more productive to direct all efforts to the
acquisition of the basic tools of the phonological trade, not an easy task in
itself, than to waste energy on internal disputes, arguably sterile at this pre-
liminary stage. In aiming for this mode of presentation, we have deliber-
ately modelled ourselves on the standard textbooks of the natural sciences,
where concepts and formal apparatus have priority over the historic details
of their elaboration. Of necessity, however, this approach involves a degree
of oversimplification: behind the smooth, shiny surface lies the chaos of stag-
gered discovery and intellectual debate, or even straight disagreement. The
forceful arrival of Optimality Theory on the scene provides a pointed ex-
ample of this internal strife. The dust will have to settle before we know with
certainty the outcome of this particular contest, and in the meantime we have
opted for the cautious strategy of presenting the body of phonology in as
neutral, non-doctrinaire a manner as possible, with the aim of providing keen
learners with the tool kit they need to achieve full participation in the sub-
ject. In this closing chapter, however, we turn the stone over and reveal some
Looking Back and Moving On 631
of the life teeming underneath. We also offer pointers to enable the reader
to proceed beyond the necessarily limited confines of this book. For addi-
tional practice of the points raised throughout we recommend the exercises
in the companion volume A Workbook in Phonology.
Phonetics
In chapters 1, 3, 5 and 7 we presented the phonetic matter we consider

a necessary prerequisite to the study of phonology. In particular, we
reviewed in some detail the mechanics of the articulation of obstruent and
sonorant consonants, and of vowels. We paid special attention to the
description of the sounds of English, in the belief that a good understand-
ing of the mechanics of sound production needs to be rooted in one’s own
personal experience of the sound — we naturally assume that the reader of
this book carries along a wealth of experience with the sounds of English. We
have deliberately ignored the two other aspects of linguistic sound, acoustics
and auditory phonetics, on account of the fact that it is perfectly possible
to get into phonology without any previous grounding in these areas.
Briefly here, ACOUSTIC PHONETICS is the branch of phonetics (and of
physics) that investigates the patterns of vibration of molecules in the air
caused by articulation. In turn, AUDITORY PHONETICS deals with the effects
of these patterns on the human ear and with their subsequent reception and
perception in the brain. You can find reasonably accessible information about
acoustic phonetics in Denes and Pinson (1993), Fry (1979), and Kent and Read
(1992), and more concise summaries in chapter 4 of Catford (1977), chapter
8 of Ladefoged (1993), chapter 3 of Borden et al. (1993), chapter 4 of
Kenstowicz (1994a), and chapter 7 of J. Clark and Yallop (1995). Auditory
phonetics is less well developed than its two counterparts, perhaps on
account of the evasiveness of its subject matter. For an introduction, you can
consult chapter 6 of Borden et al. (1993), chapter 4 of Kenstowicz (1994a),
or chapter 7 of J. Clark and Yallop (1995).
There are a number of useful books on articulatory phonetics, and we
have based our phonetic chapters on them: here and elsewhere, we could
paraphrase Sir Isaac Newton and openly declare that we have stood on the
shoulders of giants! The most recent such book is Ladefoged and Mad-
dieson (1996). This is a thorough, rigorous survey of the sounds of the world’s
languages, including some likely to feel exotic to the exclusively English-
speaking reader. The obvious predecessors are Ladefoged (1993) and
Maddieson (1984). Ladefoged (1993) has been for some time, and arguably
still is, the standard coursebook for phonetics (for articulatory phonetics, in
particular), while Maddieson (1984) usefully surveys the sound inventories
of 317 languages, carefully selected as representative of the approximately
five thousand languages still spoken on the earth.
632 Looking Back and Moving On
The most recent textbook on articulatory phonetics is Laver (1994). This

is an extremely thorough survey of the production of speech sound, from
respiration through phonation to the various articulations. It also deals with
the temporal organization of speech, including its prosodic and metrical
aspects. Unusually, but usefully, it frames the discussion in the wider con-
text of human communication and linguistic variation. Also useful are
Daniloff et al. (1980) and Catford (1977), the latter a bit of a classic. Catford
(1988) presents the subject matter in the form of practical tasks. The physi-
ology of speech production is surveyed in great detail in Hardcastle (1973).
Hardcastle and Laver (1997) brings together a number of specialized papers
on the central aspects of phonetics.
As we shall systematically do throughout this chapter, we will now pro-
vide specific references for the topics of the phonetics chapters in part I.
Importantly, these sources will best be understood, and thus most pro-
fitably read, at this point in the exposition: it is most unlikely that any one
of the sources we refer to keeps pace with the deliberately graded structure
of our presentation in the corresponding chapters of this book. Also import-
antly, we do not attempt to provide an exhaustive bibliography on any
topic, or even a complete list of major works on that area: our aim, as ever,
is to help readers increase their knowledge, and accordingly we have been
pedagogically selective in our choice of literature.
We start off with chapter 1. For useful discussion of the relationship between
phonetics and phonology, and of the differences between the two disciplines,
you can go to Pierrehumbert (1990, 1991), Diehl (1991b), chapter 2 of Laver
(1994) and Ohala (1997). An overview of the physiology of articulation and
the anatomy of the relevant organs, a bit technical, appears in Hardcastle
(1973). The articulation of consonants, with special reference to English, is
examined in chapter 3 of Ladefoged (1993). The articulation of fricatives
is discussed in chapter 9 of Laver (1994). For a survey of actual fricative sounds,
see chapter 5 of Ladefoged and Maddieson (1996). Chapter 7 of Ladefoged
(1993) covers both place and manner of articulation. Place of articulation
alone is reviewed in chapter 2 of Ladefoged and Maddieson (1996) and in
chapter 8 of Catford (1977), who discusses manner of articulation in chap-
ter 7. The principles and problems of phonetic transcription are examined
in chapter 2 of Ladefoged (1993) and chapter 18 of Laver (1994). The phon-
etic symbols (standard and less standard) are reviewed and discussed
individually in Pullum and Ladusaw (1986). The official guide to the IPA
alphabet, International Phonetic Association (1949), is now out of print. A
considerably revised edition is being prepared, and Nolan (1995) offers a
comprehensive preview. Ladefoged (1990) comments on the then latest
version of the IPA alphabet and reflects on the IPA transcription sytem in
general. Ladefoged and Halle (1988) reassess the foundations of phonetic
transcription and offers some suggestions for its improvement.
Voice production and phonation types are dealt with in chapter 6 of

Catford (1977), chapter 6 of Ladefoged (1993), and chapter 7 of Laver (1994).
Phonation types are typologically surveyed in chapter 3.1 of Ladefoged and
Maddieson (1996). Stops are examined in chapter 8 of Laver (1994), and sur-
veyed in chapter 3 of Ladefoged and Maddieson (1996). Finally, affricates
and other multiple articulations are covered in detail in chapter 11 of Laver
(1994), and surveyed in chapter 3.3 of Ladefoged and Maddieson (1996).
In chapter 3 we dealt with sonorant consonants: nasals, laterals and
rhotics. A general discussion of sonorants, under the label “resonants”, can
be found in chapter 10 of Laver (1994). Ladefoged and Maddieson (1996)
devotes three chapters to the typology of the specific sonorant classes:
nasals in chapter 4, laterals in chapter 6, and rhotics in chapter 7.
In chapter 5 we discussed the articulation of vowels in the framework
of Daniel Jones’s cardinal vowels. Besides chapter 8 of D. Jones (1967),
cardinal vowels are discussed in chapter 9 of Ladefoged (1993). They
are usefully presented through exercises in chapter 8 of Catford (1988). A
recording of the cardinal vowels by Daniel Jones is available from
University College London. For a preliminary overview of vowel articula-
tion you can go to chapter 4 of Ladefoged (1993). There are general discus-
sions of vowels in the first half of chapter 10 of Laver (1994) and chapter 9
of Catford (1977), both with substantial reference to cardinal vowels.
Chapter 9 of Ladefoged and Maddieson (1996) reviews the basic parameters
relevant to vowels, with the appropriate cross-linguistic exemplification.
For vowel typologies you can consult Crothers (1978) and Lass (1984). The
topic of quantum vowels is a little technical, but a number of useful papers
are collected in Journal of Phonetics 17, 1989, including an update of the the-
ory by its originator, Kenneth Stevens: the quantum vowels were introduced
in Stevens (1972).
In chapter 7 we reviewed the realization of English vowels in some detail.
Wells (1982) gives.a wealth of information on the English sounds of both
standard and regional accents, with the emphasis on vowels: the three vol-
umes respectively contain an introduction and general survey of English
sounds, a discussion of the accents of the British Islands, and a discussion
of English accents in the rest of the world. Some similar information can be
found much abridged in Kreidler (1989, 1997). Specialized works on the vowel
and consonant sounds of British English, RP in particular, are D. Jones’s two
manuals (1966, 1967) and the follow-up by Gimson (1994). There is a sum-
mary of RP sounds, aimed at foreign learners but useful generally as an intro-
duction, in Roach (1991). Ramsaran (1990b) reviews the present state of RP.
Classics on American English pronunciation, both general and regional, are
Bronstein (1960), Kenyon (1935), Krapp (1969) and Thomas (1958). Labov (1994)
comments on the pronunciation of many vowels in the US, but the context
he is writing in (chain shifts and language change) does not make for easy
identification of each individual vowel. For a handy comparison of some

American vowel systems, including GA, you can check Moulton (1990).
Canadian raising is specifically discussed in Chambers (1973). Burchfield (1994)
contains specialized papers on the English of Australia (by G. Turner), New
Zealand (by L. Bauer) and South Africa (by W. Branford). For the accents
of New Zealand and South Africa, you can also consult Bauer (1986) and
Lass (1990), respectively.
Foundations of Phonology
In the even-numbered chapters of part I (chapters 2, 4, 6 and 8) we presented
the foundations of phonological theory, which the rest of the book elabor-
ates on.
In Chapter 2 we introduced phonology, making use of some assimilation
processes of English which are beyond the threshold of consciousness in most
speakers. Processes like these are referred to in varying degree of detail
in Wells (1982), Nespor and Vogel (1986), Kreidler (1989), chapter 8 of
Giegerich (1992), Hawkins (1992), Ladefoged (1993) and Gimson (1994), and
are conveniently brought together in chapter 6 of Spencer (1996). The exist-
ence of these assimilation processes led us to draw a distinction between
the phonetic level of actual sound, and the phonological level of sound struc-
ture. The phonological level is rooted in Saussure’s conception of language
as a system of oppositions, and is kept in check by a principle of economy
that we related to Occam's razor. Useful summaries of Saussure’s doctrine
from the perspective of phonology can be found in chapter 5 of Jakobson
(1990) and in chapter 2 of S. Anderson (1985), the latter an excellent source
of historical information on phonology. A recent edited and annotated
English translation of Saussure’s Course is R. Harris (1987). You will find a
summary of Occam’s philosophy, including his famous razor, in chapter 14
of B. Russell (1996).
In classical generative phonology the phonological and phonetic levels
(relabelled lexical or underlying level and surface level, respectively) are
related by means of phonological rules. For an early discussion of the role
of rules in phonology, you can go to Halle (1962), although you should
be warned that some of the features are now obsolete. Phonological rules
include a focus, made up of the input sound to be changed and of its changed
output, and a context or environment, made up of the constant surround-
ing sounds that condition the change. Each phonological rule effects one
change in an input representation: a set of rules therefore carries out a sequence
of changes, collectively known as a derivation. A particularly clear account
of the mechanics and properties of rules and derivations appears in S.
Anderson (1974). You can find an inventory of the formal machinery of
early generative phonology in a ten-page appendix to chapter 8 of SPE.
In chapter 4 we broke down the phoneme into its primitive building blocks,
the distinctive features. The concept of the phoneme is discussed in chap-
ter 15 of Jakobson (1990), and its history is traced in S. Anderson (1985), chap-
ters 3 and 4 in particular. For general discussion of distinctive features, you
can consult chapter 5 of S. Anderson (1985), Clements (1992a), Halle (1983)
and chapter 16 of Jakobson (1990) (also chapter 17, but this is more tech-
nical, and many of the features discussed in it have now been superseded).
Halle and Clements (1983) and Halle (1991) are reader-friendly sources for
a more or less contemporary inventory of features — references to more tech-
nical, up-to-date accounts framed in the theory of feature geometry are pro-
vided in section 7 below. Each distinctive feature captures one key aspect
of the articulation of sound. It is normally related to the articulator actively
involved in its production, and is usually given one of two values: positive
if the aspect in question is activated, and negative if it isn’t. This binarism
permeates Jakobson’s work (see the entry binarism in the index of Jakobson
1990), and is succintly defended in Halle (1957). In the SPE system all fea-
tures are binary: features are discussed in chapter 7 of SPE. Subsequently,
however, binarism came under attack from various directions. First, sev-
eral alternative theories adopt feature unarism as one of their tenets:
Dependency Phonology (J. Anderson and Ewen 1987, den Dicken and van
der Hulst 1988, van der Hulst 1988, 1995), Government Phonology (Kaye et
al. 1985, J. Harris and Lindsey 1995), Particle Phonology (Schane 1984). Second,
major place of articulation features are now generally considered unary,
on the grounds that the range of places of articulation is defined by the
set of these features, rather than each individual place feature defining two
existing opposites (Sagey 1986a, McCarthy 1988, Yip 1989a, Halle 1991,
1995). Last, a number of mainstream individual features are being argued
by some to be “privative”, that is, to have only the positive value, with
particular reference to underspecification (Steriade 1995 offers a convenient
summary). For the standard position on the matter of feature valency you
can check Pulleyblank (1995). Place of articulation features are cumulative
as a consequence of their monovalency, which originated in Sagey (1986a)
(see also Halle 1991), and is forcefully argued for in McCarthy (1988).
Trubetzkoy’s distinction between binary and multiple oppositions is an
obvious antecedent. Trubetzkoy’s (1939) classic has been translated into
English (1969), and is usefully discussed in Fischer-Jorgensen (1975). For
specific arguments for binarity both in place features and in other features,
see Lombardi (1996).
Distinctive features provide a formal characterization of the notion “nat-
ural class”, that is, of the fact that phonological processes universally tend
to affect specific families of segments, rather than random sets: formally, the
segments in question share one or more distinctive features. The mutual
interaction of features in rules is constrained by the autosegmental mode of
representation: each feature occupies its own autonomous layer, or “tier”,

and is linked to other features by means of association lines. Autosegmental
phonology was first fully articulated in Goldsmith (1976a). It is summar-
ized in Goldsmith (1976b, 1979), and updated in Pulleyblank (1986a) and
Goldsmith (1990). There is also a useful, if slightly dated, overview in the
introduction of van der Hulst and Smith (1982b). Features that express
multiple settings of one abstract parameter (for instance, place of articula-
tion features) enjoy a high degree of functional unity, which we formalized
by assigning a common subscript to the features in question, a notation akin
to the one in Hayes (1990a) (commented on in Bird 1991), and, by exten-
sion, Halle (1995). The autonomy of some features is curtailed by their sys-
tematic formal attachment to some other specific feature, on which they are
said to be dependent. Feature dependencies were officially introduced in main-
stream generative phonology in Clements (1985), and have been adopted
by most practitioners since (Sagey 1986a, McCarthy 1988, Odden 1991,
Halle 1991, 1995, Clements and Hume 1995, Padgett 1995).
In chapter 6 we proposed aset of distinctive features for vowels and
explored their empirical impact through the phenomena of back harmony,
in Turkish, and umlaut, in German and also in English, where it is now
reduced to the status of historical relic. The phonetic foundations of vowel
features are presented in Lindau (1978) and Keating (1987). The standard
generative account of Turkish vowel harmony is Clements and Sezer (1982).
For German umlaut, see Lodge (1989) and the discussion in chapter 7.2 of
Wiese (1996a). Aoki (1968), Ringen (1975), Halle and Vergnaud (1981) and
van der Hulst and van de Weijer (1995) include useful typologies of vowel
harmony. Clements (1977) pioneers the autosegmental treatment of the phe-
nomenon. For vowel harmony in specific languages you can also consult
Ringen (1975), Goldsmith (1985), Pulleyblank (1986b) and the papers in the
second volume of van der Hulst and Smith (1988). Ringen and Vago (1995)
scrutinize Hungarian vowel harmony in the context of Optimality Theory.
Consonant harmony is relatively common in child speech (Vihman 1978),
but rare in adult language (see Shaw 1991 for an overview).
In chapter 8 we applied the distinctive feature model to the analysis of
the English Great Vowel Shift. The synchronic analysis of the Great Vowel
Shift originated in SPE, and was updated in Halle and Mohanan (1985).
Reference to history sheds useful light on the opaque set of contemporary
vowel shift alternations, and you can go to Wolfe (1972) for a good his-
torical account. Summary references to the phenomenon appear in C. Jones
(1989), Lass (1984), Strang (1986) and Pyles and Algeo (1992). Wells (1982)
also gives the essentials.
Our analysis of the Great Vowel Shift in chapter 8 was carried out in the
context of the “timing tier” or “skeleton”, an additional autosegmental
tier formalizing the relative timing of segments: universally, vowels can be
short or long, and consonants simple, geminate or affricate. The timing tier,
or CV-tier, was introduced in Clements and Keyser (1983), and streamlined
to an X-tier in Levin (1983), an unpublished but influential paper. The
timing tier, or “skeleton”, has been applied in the areas of compensatory
lengthening and template-based morphology, the latter typical of, but
not exclusive to, Semitic languages (see McCarthy 1981, 1984, McCarthy and
Prince 1990, 1995). For premoraic treatments of compensatory lengthening,
you can go to Ingria (1980) and Wetzels and Sezer (1986). Hayward (1988)
puts up a defence of the timing tier in the context of four Ethiopian lan-
guages. We saw that the timing tier also allows a satisfactory analysis
of affricates and of complex segments in general. The issue of the auto-
segmental representation of complex segments is examined in Sagey (1986b).
For the specific case of affricates, see Hualde (1988) and Lombardi (1990).
Yip (1989b) purposely relates contour tones to affricates.
Syllables
In chapter 9 we introduced the notion of the “syllable”, the architecture of

which we discussed both from a general perspective and in the context of
English. In chapter 10 we specifically explored the syllable structure of English,
which we saw exhibits a considerable amount of complexity.
We started chapter 9 with the observation that young children systemat-
ically simplify the sound input they receive, aiming for the core syllable CV
in their own output. The child data we provided at the beginning of chap-
ter 9 are from Vihman (1996 [“Timmy”, on pp. 258-60]), a valuable survey
of child phonology. A CV output also pervades the adaptation of English
loans by Japanese adult speakers, as explained in detail in Lovins (1975)
— both Shibatani (1990) and Tsujimura (1996) refer to the matter briefly,
summarized usefully in Kimura (1996). It6 and Mester (1995, 1996a) explore
Japanese loan phonology in the context of the overall structure of the
Japanese vocabulary. The omnipresence of the core syllable was noticed
by Jakobson (see Jakobson 1941, and chapter 18 of Jakobson 1990). The
alternation of a consonant (with low sonority) with a vowel (with high
sonority) of course best embodies the regular alternation of sonority that
constitutes speech: Ohman (1966). Other phonetic arguments for the core
syllable are provided in Ohala and Kawasaki (1984). The core syllable can
be altered minimally by deleting the onset or adding a coda, the latter mak-
ing up a rime with the nucleus: much of the syllable typology of the world’s
languages can be reduced to this basic inventory. Syllable typologies are to
be found in Clements and Keyser (1983) and Blevins (1995). Pike and Pike
(1947), Hockett (1956), Hooper (1972, 1976) and Vennemann (1972) have fairly
pretheoretical discussions of the structure of the syllable and its motivation.
The anchoring of syllable structure on the skeleton is justified in Clements

and Keyser (1983) and in Levin (1985). For the X-bar notation in syntax you
can check Kornai and Pullum (1990), or Cook and Newson (1996) for a less
technical summary. Borowsky (1986) argues for a two-timing-unit maximum
in English onsets and nuclei. In essence, syllables are mountains of sonor-
ity. The sonority hierarchy is discussed in various guises in Saussure (1916),
Hankamer and Aissen (1974), Hooper (1976: chapters 10 and 11), Selkirk
(1984a), Clements (1991) and Rice (1992), among many others. The sonor-
ity relations between the segments of any one syllable are governed by the
principle of Sonority Sequencing (Selkirk 1984a) and the Minimal Sonority
Distance parameter (Steriade 1982).
In chapter 10 we gave data hinting at the existence of complex codas in
English. The matter, however, is less clear cut than for onsets and nuclei:
for discussion, see Selkirk (1982), Borowsky (1986) and J. Harris (1994). In
chapters 15 and 16 we essentially followed the analyses of Myers (1987b)
and Borowksy (1989): we return to this issue in section 7 below. The
identification of syllable nuclei with vowels and of syllable margins with
consonants, while generally sound, was found to leak in both directions. First,
sonorant consonants can be compelled by sonority sequencing to become
nuclei. Bell (1978) provides a universal typology of syllabic consonants, and
Rubach (1990) gives a particularly enlightening discussion of the German
case. For English, reference to the matter is made in Mohanan (1985) and
Borowsky (1993). Second, high vowels are allowed in the onset in English
(see again Mohanan 1985), and quite generally cross-linguistically. In com-
plex onsets, the onset parsing of high vowels is substantiated by OCP
effects (see J. Harris 1994 for a brief discussion). However, Selkirk (1982)
and Steriade (1994) prefer an analysis in terms of single segments. The OCP
disfavours the inclusion of similar segments in the same subsyllabic con-
stituent: for discussion of the OCP, see McCarthy (1986), Odden (1986) and
Yip (1988). The syllabification of intervocalic consonants is governed by the
principle of Minimal Onset Satisfaction (see Roca 1994) — known variously
as the Maximal Onset Satisfaction Principle (Selkirk 1982), the CV rule
(Steriade 1982), etc. - and by the Onset Maximization Parameter (Roca
1994). Some French data discussed in the text demonstrate the separate
identity of the Maximal Onset Satisfaction Principle and the Onset Maxim-
ization Parameter, not always explicit in the literature: the Maximal Onset
Satisfaction Principle is stronger than the Onset Maximization Parameter
(for French data, see Tranel 1995). Allophony provides support for many of
the syllabification principles we have presented, which we showed interact
in interesting ways. We have already referred to Wells (1982), Nespor and
Vogel (1986), Kreidler (1989), chapter 8 of Giegerich (1992), Hawkins (1992),
Ladefoged (1993), Gimson (1994), and Spencer’s (1996) summary, in connec-
tion with English allophonic processes. Kahn (1976) offers a particularly
thorough review of English stop allophones, and Gussenhoven (1986) is

also worth consulting. The idiosyncratic behaviour of s in English is dis-
cussed in Selkirk (1982) and, in a different theoretical framework, in Kaye
(1992).
EZ Stress
In chapters 11 to 13 we dealt with the phenomenon of stress. The first

complete generative treatment of stress is found in SPE. Hyman (1977a)
gathers an interesting collection of papers on the topic. Van der Hulst (in
press a) focuses on the stress and tone patterns of European languages, but
also contains useful general papers, among them an overview of stress by
van der Hulst (chapter 1) and a survey of the word prosodic systems of
European languages by van der Hulst, Hendriks and van de Weijer (chap-
ter 7). Kager (1995) is an excellent, very clear overview of the vicissitudes
of metrical theory from its inception to the date of publication.
While the intuition that certain syllables are “stressed” is probably avail-
able to all speakers of stress languages, the nature of stress itself is much
less obvious. A particularly useful window on stress is the phenomenon of
stress movement, which we surveyed in chapter 11 for English. Differences
in stress placement are often reducible to differences in syntactic category
or syntactic constituency. For noun vs. verb stress, you can check sections
1.1 and 1.2 of chapter 7 in Halle and Vergnaud (1987a), and for compound
vs. phrase stress section 1 in chapter 2 of SPE (in a now superseded for-
malism), and sections 9.1 and 9.3 in chapter 7 of Halle and Vergnaud
(1987a). For a discussion of English compounds, you can go to Lieber
(1989), and Downing (1977), and to Selkirk (1984b) for a summary. Meaning
has no part in regular stress assignment, besides cases of marked focus: dis-
cussion on the role of focus in stress can be found in Selkirk (1984b) and
Ladd (1997). Stress is a manifestation of rhythm, more basic than poetic
rhythm: for rhythm in verse you can refer to Kiparsky (1977), Attridge (1982,
1989) and the papers in Kiparsky and Youmans (1989). The idea of relating
stress to rhythm is developed in Liberman (1975) and Liberman and Prince
(1977).
We adopted the metrical grid as a formal device for the representation of
stress: Prince (1983) is probably the best source for acquiring familiarity with
the grid. The metrical grid allows a simple account of apparent anomalies
in the distribution of stress. In particular, English disfavours stress clash, a
clashing asterisk retracting to the position marked by the nearest asterisk
in the line below: Prince (1983), 7.9.2 in Halle and Vergnaud (1987a) and
Hayes (1995). Gussenhoven (1991) offers a deletion alternative to asterisk
movement. Retraction takes place both across and within words. Some
retraction failures are accountable for by the ban imposed by the Con-
tinuous Column Constraint on gaps in metrical columns (Hayes 1995). In
German, asterisks can move forward, as well as retract, to avoid clash: see
the summary in Wiese (1996a). Besides asterisk movement, rhythm can also
induce asterisk insertion (Selkirk 1984b). Nespor and Vogel (1989) review
the phenomenon in a number of languages, including English. General dis-
cussion of English linguistic rhythm appears in Abercrombie (1967) and in
Liberman and Prince (1977). Nespor (1990a, 1990b) provides useful discus-
sion of the dichotomy stress rhythm vs. syllable rhythm. The repercussions
of metrical structure on such segmental phenomena as vowel reduction
and consonant allophony contribute additional evidence for stress and its
distribution: see Sainz (1992) for vowel reduction, and Gimson (1994), Kahn
(1976) and Gussenhoven (1986) for English stop allophony.
Metrical structure can be minimally reduced by systematically disre-
garding the metrical element on the edge, a device known as “extramet-
ricality” (Liberman and Prince 1977, Hayes 1979, 1980, 1982, 1995, Halle and
Vergnaud 1987a, Archangeli 1988a, Roca 1988, 1992). The edge requirement
is encapsulated in the Peripherality Condition. Prince and Smolensky (1993)
provide a very useful critical overview of the properties of extrametricality.
The interaction of overlapping rules is subject to the Elsewhere Condition:
Kiparsky (1973), Koutsoudas et al. (1974) and Iverson and Wheeler (1988).
For a brief discussion of French stress, see Tranel (1987). The metrical pat-
terns of English are developed in Hayes (1980, 1982), Prince (1983) and Halle
and Vergnaud (1987a, 1987b), among other sources. Kager (1989) provides
a valuable critical evaluation. The metrical foot as the basic unit of rhythm
is elaborated in Liberman and Prince (1977), Selkirk (1980), Hayes (1980, 1995)
and Halle and Vergnaud (1987a). For iteration in English stress, see Prince
(1983) and Selkirk (1980, 1984b). Metrical parameters are examined in Hayes
(1980), Prince (1983), Halle and Vergnaud (1987a) and Dresher and Kaye (1990).
For the stress patterns of Polish and Macedonian, see Franks (1985 and 1987,
respectively), and also Dogil (1998) for Polish. Aklan, Maranungku and YidinY
are discussed in Hayes (1980), and Winnebago in K. Hale and White Eagle
(1980) and Halle and Idsardi (1995). Line conflation is specifically justified
in Halle (1990), while the opposing iterativeness parameter is defended in
Blevins (1990). Hayes (1995) formulates the Faithfulness Condition, build-
ing on Halle and Vergnaud (1987a). For the so-called “Duke of York
Gambit”, see Pullum (1976).
The effects of syllable weight on stress are taken account of in SPE, but
their distinct formalization in terms of moras begins with Hyman (1985). The
Free Element Condition, formulated in Prince (1985), captures the idea that
stress rules are structure-building, in the sense of Kiparsky (1982, 1985).
Degenerate feet are paid special attention in Hayes (1995). The effects of final
consonants on English stress are carefully reviewed in Ross (1972) in an SPE
framework. Final vowel lengthening was introduced in SPE. Hyman’s (1985)

moraic formalization of syllable structure is modified in Hayes (1989a), who
also formulates Weight by Position. For a comparison between moras and
skeletal slots see Broselow (1995), and Rubach (1993) with specific reference
to Slovak. The model in Hayes (1995) is grounded in the empirical asym-
metries in the distribution of foot types, underpinned by the perceptual
findings reported in Woodrow (1951). Kager (1993) favours a No-Lapse
Constraint over the Iambic-Trochaic Law. The unbounded feet of Khalkha
Mongolian are discussed in Hayes (1980), Halle and Vergnaud (1987a) and
Halle and Idsardi (1995), and the Selkup data are taken from Halle and
Clements (1983). Irregular stress is formalized as accent in Halle and Verg-
naud (1987a) and Halle (1990). Sanskrit stress is discussed in Kiparsky and
Halle (1977), and summarized in Halle and Mohanan (1985) and Halle and
Vergnaud (1987a, 1987b). Halle and Idsardi (1995) develop the theory of Halle
and Vergnaud (1987a) in suggestive ways.
Tone
In chapter 14 we presented three linguistic functions of pitch: intonation,

pitch accent and tone. Variations in pitch result from variations in the
frequency of vibration of the vocal folds. Vocal fold vibration is an aero-
dynamic effect induced by the outgoing air on a given vocal fold configura-
tion. Phonologically, high pitch can contrast with low pitch, and mid pitch
with both (Ladefoged 1993). Such units of pitch contrast are referred to as
“tones”. The difference between intonation, pitch accent and tone languages
hinges on the pattern of association of the tones to the segmental material,
and on the function of the tones in the system. In particular, in intonation
strings of tones making up “tunes” associate to certain segments in a cer-
tain domain (probably the “intonational phrase”, but see section 7 below)
to express functional or attitudinal meanings: whether the utterance is a state-
ment or a question, whether the speaker is enthusiastic or reserved about
it, and so on. Bolinger (1989) provides an accessible overview. In pitch accent
languages, a constant, meaningless tune (or a restricted set of tunes) asso-
ciates to the segmental material algorithmically. Finally, in tone languages
tone differentiates words or expresses morphological functions: in the for-
mer case, different words (with a different meaning, for instance) which are
segmentally identical are kept apart by their tonal shape; in the latter, mor-
phological functions like tense in verbs or plurality in nouns are expressed
by a change in tone, rather than by affixation. The lexical function of tone
prevails throughout South East Asian languages, while the morphological
function is common in African languages. Yip (1995) and Odden (1995) pro-
vide excellent overviews of Asian and African tones, respectively.
We illustrated intonation with English. We saw that each intonational

tune is composed of a string of tones, divided into boundary tones, word
tones (one of them dominant) and a phrasal tone (Pierrehumbert 1980,
Ladd 1997). For pitch accent we focused on Japanese. Japanese words can
be accented (with a mark on one of the syllables) or unaccented, idiosyn-
cratically. Accented syllables attract the dominant tone of a fixed, meaningless
tonal melody, the remainder of the tones being associated to the remain-
ing syllables in a predetermined manner (Haraguchi 1977). In words with
no accented syllable the dominant tone associates to the word-final syllable.
Chinese provides the prototypical example of a lexical tonal language: seg-
mentally homophonous words can have different lexical tones. In the last
two sections of the chapter we illustrated the mechanics of tone in African
languages. The autosegmental approach to phonology was in fact developed
in connection with African tones, and we scrutinized the basic principles of
autosegmental association in this context. Many of the principles originally
proposed (Goldsmith 1976a) have subsequently been found not to be uni-
versal (Pulleyblank 1986a). We showed that the most resilient of these prin-
ciples, the No-Crossing of Lines Constraint, is in fact derivable from the
logic of the geometry: see Sagey (1988), followed up by Hammond (1988)
and, in a different context, Bagemihl (1989). Coleman and Local (1991) offer
criticism of the constraint.
Bolinger (1986) provides a painless introduction to the phenomenon of
intonation — also, more concisely, the introduction and chapter 1 of Bolinger
(1972). Chapter 1 of Ladd (1997) is very useful, althoughalittle less beginner-
friendly — in particular, it offers a critical overview of the main models of
intonation available. For a basic intonational typology, see Bolinger (1978).
For the mechanics of voice production, see chapter 7.4 of Laver (1994),
and for the use of pitch in intonation, chapter 15 of Ladefoged (1993). The
autosegmental analysis of intonation started with Liberman (1975) and
came of age with Pierrehumbert (1980), otherwise a little technical. Useful
summaries appear in Ladd (1992), Pierrehumbert and Hirschberg (1990)
(commented on in Hobbs 1990), and, of course, Ladd (1997). The connec-
tion between intonation and primary stress in English is spelled out in Hayes
(1995). Cruttenden (1986) is a good source for pre-Pierrehumbert intonation
from the particular perspective of the so-called “British school”. For the
tones of English, you can also look at Leben (1976) and Goldsmith (1981).
M. Beckman and Pierrehumbert (1986) is a ground-breaking comparison
of English and Japanese intonation, in the footsteps of M. Beckman’s (1986)
investigation of stress and accent, but again most of it is rather technical.
An oft-cited study of tone is Pike (1948). Fromkin (1978) is a most useful
early generative collection. It includes, among others, chapters by Ohala on
the phonetic production of tone, Leben on the representation of tone, S.
Anderson on tone features, McCawley on what constitutes a tone language,
and Schuh on the types of rule that affect tone.
For an early autosegmental investigation of Chinese tone you must go

to Yip (1980). More recent discussion appears in Duanmu (1990) and Bao
(1990), the latter from a general theoretical perspective. The specific issue of
contour tones is addressed in Yip (1989b) and Duanmu (1994). The intro-
duction of van der Hulst and Smith (1988) reviews the typology of pitch
accent systems. The autosegmental analysis of Japanese pitch accent began
with McCawley (1977) and Haraguchi (1977). The next important study
was Poser (1984), followed by Pierrehumbert and Beckman (1988). There are
convenient summaries in Shibatani (1990) and Tsujimura (1996). M. Clark
(1987) attempts an analysis of Japanese as a tone language. Kubozono
(1993) includes several experimentally based studies of various aspects of
Tokyo Japanese pitch. In the European area, the classical study of Swedish
pitch accent is Bruce (1977). For Serbo-Croatian you can look at Inkelas and
Zec (1988), and for Lithuanian at Blevins (1993). Hualde (1991a, 1998) exam-
ines several aspects of Basque pitch-accent. Clements and Goldsmith (1984)
is an important early collection on African tones. The Mende data are dis-
cussed in Leben (1978). The original principles of autosegmental association
are presented and justified in Goldsmith (1976a). For Shona tonology, you
can see Odden (1981) and Myers (1987a). Pulleyblank (1986a) updates the
general theory and discusses such central phenomena as floating tones
and downstep, much of it in the context of Tiv tonology. For the non-
universality of tonal association conventions see also Hyman and Ngunga
(1994), and for Meeussen’s law Goldsmith (1984). Tone features are discussed
in Yip (1980, 1989b), Pulleyblank (1986a), Snider (1988, 1990), Bao (1990)
and Duanmu (1990), among others. For the role of accent in African tonal
systems, see Goldsmith (1987) and Hyman (1978, 1989).
[4 The Interaction between Morphology and Phonology
In chapters 15 and 16 we carried out a systematic investigation of the

influence of grammar, morphology in particular, on phonology. Chapter 15
focused on the cycle and chapter 16 on the layering of phonology. The com-
mon thread is the delimitation of the domains in which any one phono-
logical rule applies. Our first approximation was to identify these domains
with grammatical constituents: syntactic at the phrasal level and morpho-
logical inside the word. However, we eventually saw that this approach
runs into problems, and in the second part of chapter 16 we explored the
construction of specifically phonological domains on grammatical con-
stituents, to allow for mismatches.
In our investigation of stress in chapter 11 we came across the phe-
nomenon of stress retraction in English phrases under conditions of clash.
In chapter 15 we saw that the correct results follow if we build the grid in
tandem with the grammatical constituency of the phrase, a cyclic mode of
construction (Prince 1983, Selkirk 1984b, Halle and Vergnaud 1987a, Hayes
1984, 1995). Cyclicity also governs tone association in many languages, and
we illustrated the particular case of Tiv (Pulleyblank 1986a). More recently,
however, there has been a turn away from the cycle: see Cole and Coleman
(1992), for instance, in a Declarative Phonology framework. Hyman (1994)
presents data supportive of the cycle from the Bantu language Cibemba.
For an overall discussion of these and other issues related to the cycle, you
can go to Cole (1995).
In chapter 15 we completed the picture of English word stress that we
started putting together in chapter 12, and investigated the mechanics
of English secondary stress in some detail. First, we showed the need to
reapply the footing procedure in a non-cyclic mode, that is to say, in the
maximal word domain (Halle and Vergnaud 1987a, 1987b). We suggested
that this non-cyclic reapplication of footing builds feet from left to right,
the opposite direction of its cyclic counterpart (Halle and Kenstowicz 1991),
hence our dubbing non-cyclic footing “refooting”. Refooting is followed by
destressing, a specifically non-cyclic procedure which deletes a mono-
moraic degenerate foot that immediately precedes a foot head: cf. Halle
and Vergnaud’s (1987a) construct “stress well”. Next comes a non-cyclic
reapplication of accenting, rendered necessary by the destructive action
of conflation at the end of each cycle: heavy syllables tend to have some
degree of surface stress, whether or not they carry the word’s main stress
(Halle and Kenstowicz 1991). Non-cyclic accenting is subject to a fair bit
of lexical exceptionality, in direct contrast to cyclic accenting, which is prac-
tically exceptionless — this skewedness runs in the wrong direction (it is
non-cyclic rules that tend to be exceptionless), possibly revealing a weak-
ness in the analysis. One further rule retracts primary stress from the
word’s last syllable. We provisionally assimilated this rule to the Rhythm
Rule that effects retraction under clash (Halle and Vergnaud 1987a), and
ordered it last in the sequence of non-cyclic stress rules. We went on to demon-
strate the cyclicity of the rules responsible for the assignment of word-
primary stress: the primary stress of words embedded in larger words tends
to surface (as secondary) irrespective of rhythm (SPE, Halle and Vergnaud
1987a, Hammond 1989, Halle and Kenstowicz 1991). In order to account
for this situation, we preserved internal primary stresses on independent
planes, eventually copying them onto the plane of the largest domain to
provide accentual sites for secondary stresses, over and above the rhythmic
secondary stresses assigned by the non-cyclic stress algorithm. This pro-
cedure is justified and explained in Halle and Vergnaud (1987a), and sum-
marized in Kager (1995). An early discussion of the stress cycle in languages
other than English is to be found in Brame (1974). Sainz (1992) is an attempt
to dispense with the English stress cycle altogether, while both Hargus
(1993) and Inkelas (1993) argue for maintainting the relationship between
phonological rule application and morphological formation.
We investigated English vowel shortening also in chapter 15. The Great

Vowel Shift alternations we discussed in chapter 8 are anchored in a length
contrast: the lexical long vowel only shortens in the longer alternant, and in
chapter 15 we attributed this shortening to a pressure for bimoraic feet,
which we formalized as a rule of Vowel Shortening (Borowsky 1986, 1989,
Myers 1987b, Halle and Vergnaud 1987a, Rubach 1996). We observed that
Vowel Shortening fails to apply in a number of forms. Some of these forms
are straightforward exceptions, but others simply fall under the remit of
the Principle of Strict Cyclity, which limits the application of cyclic rules
to domains derived in their cycle: lexical material can therefore remain
unaffected by cyclic rules. The idea of the strict cycle was introduced in phono-
logy in Kean (1974) and was further developed in Mascar6 (1976). For addi-
tional discussion of the strict cycle you can go to Kiparsky (1982, 1985, 1993),
Hualde (1989), and Poser (1993). Cole (1995) offers a convenient summary.
Vowel Shortening is also active in the word-final syllable, for the same
reason as word-internally: preference for bimoraic feet. Rubach (1984a) is
one of the earliest global applications of the theory of the cycle to one lan-
guage (Polish), and Rubach (1984b) accounts for the segmental rules of English
in a similar mould.
In chapter 16 we accounted for the imperviousness of some affixes to cyclic
rules by dividing affixes idiosyncratically into two sets, namely, the set of
affixes which define a domain where cyclic rules apply and the set of affixes
which do not define such a domain: cyclic rules can therefore skip some
morphological structure, typically the outer layers. The observation that
the linear order of affixes determines their behaviour with respect to cyclic
rules is akin to Siegel’s (1974) “Ordering Hypothesis”. Affixation processes
can be sensitive to non-lexical phonological properties of the input: cf.
McCarthy’s (1982) discussion of English expletive infixation, for instance. If
the properties in question are non-lexical, then they must be derived; if they
are derived, then some phonological rules must precede some affixation:
in a nutshell, morphology and phonology are interleaved, not segregated.
The model of Lexical Phonology integrates this interleaving with the
Ordering Hypothesis. Lexical Phonology developed in the early 1980s in
such work as Kiparsky (1982, 1983, 1985) and Mohanan (1986), following on
from Siegel (1974), Allen (1978) and Pesetzky (1979), among others. A use-
ful collection of articles on the Lexical Phonology of the 1980s appeared in
Phonology Yearbook 2 (1985), Kaisse and Shaw’s opening paper conveniently
summarizing the model. Hargus and Kaisse (1993) took further stock at a
time when Lexical Phonology was no longer in the limelight. Of particu-
lar theoretical interest in this new collection are the papers by Kiparsky,
on the relationship between Strict Cyclicity and the derived status of the
environment, by Borowsky, on the ordering of morphology and phonology
in level 2, and by Odden, on the general interaction between morphology
and phonology. The introduction by Kaisse and Hargus evaluates the
evolution of Lexical Phonology and maps out its present and its possible
future. Inkelas and Zec (1990), a sister collection, investigates the relation-
ship between phonology and syntax. It contains, among others, papers by
Hayes on precompiled phonology, by Kaisse on the properties of post-
lexical rules, by Nespor on the separation of prosody and rhythm, by Rice on
the prediction of rule domains in the phrasal phonology, by Vogel and Kenesei
on the influence of syntax and semantics on phonology, and by Zec and Inkelas
on prosodically constrained syntax. Szpyra (1992) is a critical evaluation of
the theory of cyclic and lexical phonology in the context of Polish and English,
and Mohanan (1995) takes a fresh look at the overall organization of the gram-
mar. Pulleyblank (1986a) uses Lexical Phonology to analyse tone associ-
ation in several African languages. The most complete treatment of English
phonology in a Lexical Phonology framework is probably Borowsky (1986),
akin to Rubach (1984a) for Polish. Rubach (1985) focuses on the distinction
between lexical and postlexical rules, and Rubach (1990) applies the theory
to a set of puzzling data concerning the voice value of word-final obstru-
ents in German. For a recent evaluation of the theoretical apparatus of Lexical
Phonology, you can go to Odden (1993), and for a state-of-the-art overview
to Booij (1994).
Phonological Domains
In chapter 16 we argued for licensing the English word-final consonant directly

from the syllable node, to account for the fact that the maximally bimoraic
rime can be followed by an extra consonant word-finally, only subject to
Sonority Sequencing (Selkirk 1982, 1984a, Myers 1987b, Borowsky 1989). On
addition of a consonantal suffix (-[6] to deep, for instance), the formerly word-
final consonant is pushed into the rime, where it takes over one of the moras
of the vowel, thus shortened: dep[6] (cf. Myers 1987b). This shortening does
not take place before some suffixes (consider deepness, for instance), and we
resolved this paradox through a three-fold strategy: (i) we allotted the -[6]
suffix to level 1, and both the -ness suffix and the rule of Stray Erasure to
level 2; (ii) we made the licensing of the word-final consonant contingent
on adjacency to a morphological right bracket (the analysis in Borowsky 1993
is, in essence, equivalent); (iii) we adopted the convention of deleting word-
internal morphological brackets at the end of each level — Mohanan (1986)
discusses the rationale for bracket erasure in Lexical Phonology. Following
this procedure, the | at the end of deep will license the p in level 2 even after
the non-cyclic -ness is suffixed, also in level 2. Not so, however, for the p in
dep[8], which will get deleted by the Bracket Erasure Convention at the end
of level 1. Therefore, in level 2 the two forms will be represented [[deep]ness]
and [dep[8]], respectively. As a consequence, level-2 Stray Erasure will have
no effect on the p of deepness; by contrast, it would have deleted the p of

depth if this p had not become licensed by one of the vowel’s two moras.
Many English processes similarly exhibiting paradoxical behaviour are
brought to rule once the proposed machinery is in place, the retraction of
primary stress from the word-final syllable among them: Borowsky (1993)
provides a good survey. While most of the data can indeed be accounted
for in this way, a non-negligible set of bracketing paradoxes remains where
the bracketing provided by the morphology and the semantics is at odds
with the bracketing required by the phonology, as usefully summarized in
Spencer (1990). Marantz (1988) and Sproat (1988) propose a mapping of mor-
phological structure onto phonological structure as a solution. Also prob-
lematic is the fact that most of the affix combinations predicted by the model
fail to materialize (Fabb 1988). All these difficulties obviously weaken the
approach, pointing at the illegitimacy of identifying phonological domains
with grammatical constituency (Inkelas 1989, 1993).
There is strong evidence that in phrasal domains phonological rules
do not act directly on the domains defined by the syntax, but, rather, on
phonological domains built parasitically on syntactic constituency: Selkirk
(1981, 1986, 1990), Nespor and Vogel (1986) and Hayes (1989b); a direct-
syntax approach is advocated in Kaisse (1985). One such phonological
domain, the Phonological Phrase, constrains the application of clash-driven
asterisk retraction in English (Nespor and Vogel 1986). The next higher domain,
the Intonational Phrase, is usually construed as the domain of intonational
association (Selkirk 1984b, Nespor and Vogel 1986), albeit not uncontrover-
sially (Gussenhoven 1990, Gussenhoven and Rietveld 1992). The Intonational
phrase can also constrain the application of phonological rules (Selkirk
1984b, Nespor and Vogel 1986, Rice 1987, Vogel and Kenesei 1990), and
we mentioned the rule of place of assimilation of nasals in English in this
connection. The largest phonological domain, the Phonological Utterance, does
not just correspond to a speech utterance, but is also defined on specific,
if varied, criteria. Phonological domains (often referred to as “prosodic
domains” in the literature) are expected to comply with the Strict Layer
Hypothesis (SLH), which conceives of each phonological domain as made
up of one or more occurrences of the phonological domain immediately
lower in the hierarchy, with no material left out (Selkirk 1984b, 1990).
Selkirk (1996) decomposes the SLH into a set of four more basic constraints.
Phrasal phonological domains provide the bounds for the application of the
so-called postlexical rules of Lexical Phonology. The postlexical component
is specifically investigated in Booij and Rubach (1987) and Iverson (1993),
among others. The Phonological Word normally corresponds to the morpho-
syntactic word, but can be larger or smaller. The Clitic Group, made up of
a Phonological Word and one or more clitics attached to it (Nespor and Vogel
1986), is not universally accepted (see Booij 1996).
Below the word, we drew a distinction between phonological domains and

phonological constituents. In particular, we argued that prosodic elements
such as syllables and feet constitute phonological constituents, while phono-
logical domains are the result of a (frequently vacuous) mapping from
morphological structure, parallelling the mapping of syntactic structure
onto phrasal phonological domains (Inkelas 1989, 1993). From this per-
spective, the Phonological Word must be deemed to enjoy dual status, as a
phonological constituent (dominating the foot and the syllable) and as a
phonological domain (dominated by the Phonological Phrase, or maybe the
Clitic Group). The availability of the Phonological Word (PW) as a phono-
logical constituent makes possible the direct licensing from the PW node of
such word-peripheral extra segmental material as the word-initial s- or (any
number of) word-final coronal obstruents in English (see Borowsky 1986; also
Rubach and Booij 1990a for Polish).
The model we have outlined achieves a welcome integration between Lexical
Phonology and the “Prosodic Phonology” of most of the literature. Prosodic
Phonology (our “phonological domains”) must of course not be confused
with the Prosodic Morphology developed in the work of McCarthy and Prince,
and of which McCarthy and Prince (1995) provides a convenient summary.
For the relationship between Lexical Phonology and Prosodic Phonology you
can check, Booij and Rubach (1984), Booij (1988) and Booij and Lieber (1993).
One pending problem concerns the bracketing paradoxes we mentioned above.
In Lexical Phonology, Mohanan’s (1986) looping back into a previous level
(see also Halle and Mohanan 1985) looks suspiciously like an admission
of defeat. An alternative made available by the autonomy of phonological
domains involves the elevation of some affixes to the rank of phonological
words (Booij and Lieber 1993), but some forms seemingly remain intract-
able. One solution involves interspersing non-cyclic affixes among the cyclic
affixes, then allowing cyclic rules to skip non-cyclic domains (Halle and
Vergnaud 1987a). However, this strategy entails the abandonment of the
Ordering Hypothesis, perhaps the very heart of Lexical Phonology. As we
said above, Marantz (1988) and Sproat (1988) advocate a rebracketing strat-
egy to overcome this.
El Aspects of Lexical Representation
The Principle of Strict Cyclicity that we presented in chapter 15 restricts

the application of the cyclic rule of s-Voicing to derived environments. In
chapter 17 we came across exceptions to this prediction, but this is of course
to be expected with any phonological rule. More challenging is the fact
that s-Voicing appears to have applied morpheme-internally in a majority
of monomorphemic forms, contrary to the ruling of the Principle of Strict
Cyclicity. The answer is that the lexical forms in question do not have a
lexical value for [+voice], and therefore s-Voicing can fill in the gap with-
out contravening Strict Cyclicity. Monomorphemic forms where the s does
not voice include the lexical specification [—voice], which blocks s-Voicing
by Strict Cyclicity (Kiparsky 1982, 1985). This model automatically accounts
for the transparency of some segments to some assimilation processes: these
segments are in effect not there when the assimilation takes place, since the
relevant feature is unspecified. A case in point is the regressive assimilation
of voice in Russian obstruents, in which sonorants play no role (Halle and
Vergnaud 1981, Hayes 1984, Kiparsky 1985).
Underspecification interprets natural values as literally “unmarked” in the
lexicon: they are simply left out of lexical representation. Jakobson (1941)
is the obvious precursor of this view. The natural values of the distinctive
features are inventoried in a set of universal markedness statements, which
give rise to a set of implicational relationships deriving some feature values
from others (SPE, Kean 1975, Calabrese 1995). Stampe’s (1969) Natural Phono-
logy addresses the acquisition of phonology from a markedness perspective:
see Donegan and Stampe (1979) for a convenient summary.
The formal implementation of underspecification has given rise to several
theories, of which in chapter 17 we presented Radical Underspecification and
Contrast-restricted Underspecification. Radical Underspecification aims for
maximal (“radical”) economy of lexical inventory (Archangeli 1984b, 1988).
To achieve this aim, one of the segments in each class (for instance, one
of the vowels) is assumed not to have feature values in the lexicon: such a
segment is maximally underspecified. The identity of the maximally under-
specified segment must be determined language by language: it is based on
such language-specific facts as the asymmetric behaviour of the segment
with respect to some phonological rules, its status as the epenthetic vowel
of the language, and so on. The feature values omitted from the maximally
underspecified segment are also omitted from all the other segments in the
class. This language-specific underspecification adds to the general under-
specification derivable from markedness, and obviously results in a drastic
simplification of lexical entries, the overt goal of Radical Underspecification.
Each feature value missing from the maximally underspecified segment gives
rise to a “complement rule” that eventually supplies the value in question.
Whether supplied by a complement rule or by a markedness statement,
underspecified values must be filled in before they are appealed to in a
feature-changing rule, an ordering convention known as the “Redundancy
Rule Ordering Constraint”. However, ongoing research on actual phonetic
implementation is casting doubt on the adequacy of feature filling by
phonological redundancy rule, since the phonetic value of underspecified
segments often seems to originate in interpolation: see Keating (1984, 1988,
1990) and Pierrehumbert and Beckman (1988), for instance.
For successful applications of the theory of Radical Underspecification

you can consult Pulleyblank (1986b, 1988a, 1988b), Abaglo and Archangeli
(1989), and Archangeli and Pulleyblank (1989), among others. The predic-
tions of this theory do not always come true, however (Steriade 1995,
Hualde 1991a, 1991b), and the theory of Contrast-restricted Underspeci-
fication has been brought in (Steriade 1987, 1995, Christdas 1988, Clements
1988). Contrast-restricted Underspecification restricts underspecification to
those values that do not contrast in a particular environment. Contrast-
restricted underspecification is not trouble-free either. One move to preserve
the validity of the theory involves the formalization of (some) distinctive
features as unary: the standard negative value would simply correspond
to the absence of the feature. Indeed, in chapter 17 we showed that a
principled analysis of Khalkha Mongolian backness harmony in the context
of Contrast-restricted Underspecification points to a unary feature [round]
(Steriade 1987, 1995). For evaluation of the two theories of underspecifica-
tion, you can go to Archangeli (1988b) and Steriade (1995), while Mohanan
(1991) and Myers (1991) take a general stand against underspecification
as such. Archangeli and Pulleyblank (1994) develops Radical Under-
specification into Combinatorial Specification.
Another major theme in chapter 17 concerned the organization of features:
see Broe (1992) for a particularly clear overview. First, we reminded our-
selves of the fact that many features are linked in a relationship of depend-
ency, extended by many to class nodes (Clements 1985, Sagey 1986a,
McCarthy 1988, Halle 1991, Padgett 1995), a formal equivalent of our sub-
script notation (cf. Hayes 1990a and Halle 1995). The global network of
features is commonly referred to as “Feature Geometry”. There are several
versions of the feature geometry, with a shared basic structure, but differ-
ing in some of the details. The root of the feature geometry tree is assumed
by most to be composed of the features [tconsonantal] and [+tsonorant]
(McCarthy 1988), although Kaisse (1992) argues for the independence of
[tconsonantal], and Hume and Odden (1996) against the very existence of
the feature. The idea underlying feature geometry is that associations
between features are restricted in the ways encoded in the geometry.
Clements (1991) and Clements and Hume (1995) advance a feature geo-
metry common to both consonants and vowels, simply augmented for vowels
with the class node “Aperture” and its dependent feature [topen]. Halle
and Stevens (1971) and Ladefoged (1973) scrutinize the laryngeal features
responsible for voicing, aspiration, glottalization, and so on. The feature
geometry tree naturally implies a specific set of redundancies between fea-
tures: a feature lower in the tree automatically presupposes all the higher
features (and class nodes) along its path to the root. Stevens et al. (1987) and
Stevens and Keyser (1989) put forward the notion of “enhancement” to
justify the frequent co-occurrence of some feature values — [+round] with

[+back], for instance.
EE] Derivational Theory
In chapter 18 we reviewed rule ordering, one of the trademarks of SPE-based

phonology, anticipated in Bloomfield (1939). In particular, the lexical and
surface levels of representation are assumed to be mediated by an open num-
ber of sequential levels, each mapped onto the next by means ofa rule. This
sequential set of levels mapping the lexical representation of any one form
onto its surface representation constitutes a derivation: see S. Anderson (1974)
for a particularly lucid presentation. The ordering of the rules, that is, the
fact that rule X applies before rule Y, can be automatic: given an input A,
rule X (but not rule Y) can and will apply, to yield B; in turn, Y (but not X)
can and will apply to B, to yield C; and so on. The situation just described,
where rule application is only driven by the substance of the input, is
known as “intrinsic ordering”. Very often, however, an intrinsic order of
application yields incorrect results, and outside intervention (by the linguist
or, more to the point, by the learner) is necessary. In this case, we talk about
“extrinsic ordering”, and we say that the rules in question are extrinsic-
ally ordered. Extrinsic ordering by definition disrupts intrinsic ordering: if
it didn’t, it would be redundant. Rule ordering is discussed in chapter 8.3
of SPE. Chafe (1968) defends persistent rule application, as does Myers (1991)
in a more modern context. For specialized treatment of the issues of rule
ordering see Iverson (1974), and for useful general discussion Booij (1981).
Opposition to extrinsic rule ordering has been a recurring theme through
the years, and several proposals have been made to dispense with it. Some
ordering can indeed be attributed to universal principles: SPE’s disjunc-
tive ordering, Kiparsky’s (1973) Elsewhere Condition, Archangeli’s (1984)
Redundancy Rule Ordering Constraint, and so on. Other authors have
attempted to extend free ordering beyond areas regulated by such prin-
ciples: see Koutsoudas et al. (1974), Hooper (1976), Koutsoudas (1980). The
main aspects of this debate and other issues concerning rule ordering are
summarized in Iverson (1995).
The application of a rule X to an input A can affect the potential of some
other rule, Y, to apply. Thus, it is possible that, while the conditions for the
application of Y are not met in A (simply because A does not include the
environment required for the application of Y), such conditions are met after
X has changed A into B: B does include the environment required by Y. In
a case like this we say that X and Y stand in a feeding relation, and that X
feeds Y (in reality, though, it is the application of X to A that feeds Y). If we
invert the order of the two rules, the relation becomes one of counterfeed-
ing, the reverse of feeding: X would feed Y if it were ordered prior to it, as
we have just seen. The establishment of counterfeeding orderings obviously
requires outside intervention: if the rules were left to their own devices, X
would apply to A (because it can), but not to Y (because it can’t); in turn,
after X has produced B from A, Y will apply to B (again, simply because it
can); and so on. Actual examples of these ordering relations in chapter 18
are the English rules of CiV-Lengthening and s-Voicing for feeding (CiV-
Lengthening > s-Voicing), and Spirantization and [iu]-Formation for coun-
terfeeding (Spirantization > [iu]-Formation). Defenders of free ordering
face the challenge of reinterpreting counterfeeding orderings, to achieve
the same result without extrinsic ordering. One obvious strategy involves
enriching the environment of the rule we want to prevent from applying to
the initial input (Koutsoudas et al. 1974): if we enrich the environment of
rule Y in our abstract example, so that it is no longer met in A, there will
be no need to order Y after X.
Counterfeeding must not be confused with bleeding - all the concepts
we are discussing, and their labelling, originate in Kiparsky (1968): see S.
Anderson (1974) and chapter 5 of Hooper (1976) for particular, clear illus-
trations. In counterfeeding, an ordering Y > X prevents X from feeding Y,
that is, it blocks the creation of forms to which Y would be applicable: at
the time Y is called up (crucially, before X) such forms are simply not
available. They will of course become available after the application of X,
but by this time Y will no longer be applicable, because of the ordering: the
turn of Y has gone, so to speak. Bleeding is different. In bleeding, the applica-
tion of a rule X to an input A actually removes some material required
for the application of another rule Y, ordered after X. In this case, we
say that X bleeds Y. Clearly, if we invert the ordering (Y > X, rather than
X > Y), bleeding will no longer take place, and Y will be able to apply to
A (whatever X does next is immaterial). In cases like these we talk about
a counterbleeding order, and say that Y counterbleeds X. Counterbleeding,
therefore, simply means ‘the opposite of bleeding’, in the way that coun-
terfeeding means ‘the opposite of feeding’. You can now see that bleeding
is not the same as counterfeeding: counterfeeding does not remove any-
thing from the input (the input simply fails to meet the environment of the
next rule), but bleeding does, by definition. If bleeding is not the same as
counterfeeding, then counterbleeding cannot be the same as feeding, and
indeed it isn’t — feeding creates the environment required by the next rule,
but counterbleeding does not create anything: it simply prevents destruc-
tion. The rules in chapter 18 also provide examples of these additional rela-
tionships: [j]-Morification > [j]-Deletion for bleeding, and Palatalization >
[j]-Morification for counterbleeding. As with counterfeeding, a bleeding
relation can only be established on the basis of extrinsic ordering. Without
this outside intervention, rules will apply whenever their environment is

met: if rule Y has its environment met in a form A (although it wouldn’t if
rule X had applied first), it will simply apply to it, in a counterbleeding order.
Rules in a system do not necessarily affect each other: rules can be
non-interacting. However, the effects of transitivity often impose a specific
ordering even between non-interacting rules. Thus, suppose that we can estab-
lish empirically that rule X precedes rule Y in the ordering, and that rule Y
precedes rule Z. If so, by the principles of logic X must also precede Z. For
notice what would happen if it did not. In this case, we would have the
following ordering relations: on the one hand, Z > X (ex hypothesi), and on
the other X > Y > Z (empirically). But these two orderings are contradictory,
because if X > Y > Z, then we cannot in the same breath hypothesize that Z
> X, the opposite order. This is precisely what we mean by “transitivity”: if
X > Y, and Y > Z, then of necessity X > Z. Transitivity is therefore a matter
of logic, once we assume, as practically everybody else does, that the order
of rules is fixed once and for all for each system: each two rules can only
have one ordering (in S. Anderson's 1974 “local ordering”, however, the order-
ing of rules need not be constant across derivations). In chapter 18 we showed
that Vowel Shift follows s-Voicing (s-Voicing > Vowel Shift), even though
there is no direct interaction between these two rules: s-Voicing must pre-
cede Velar Softening (s-Voicing > Velar Softening), and Velar Softening must
precede Vowel Shift (Velar Softening > Vowel Shift); therefore, by trans-
itivity, s-Voicing precedes Vowel Shift (s-Voicing >... > Vowel Shift). The
interaction between these and other rules of English is examined in detail
in Rubach (1984b), Halle and Mohanan (1985) and Borowsky (1986), among
others. Jensen (1993) provides a convenient summary of these and other
matters of English phonology. As we said above, rule ordering is part and
parcel of the model of phonology that stemmed from SPE, even though there
have been several attempts through the years to remove it from the theory.
Indeed, extrinsic rule ordering has proved one of the most resilient aspects
of SPE-inspired phonology: see Bromberger and Halle (1989) for a recent
statement, and Kaye (1990), Klausenberger (1990) and Hyman (1993) for
alternative views. However, recent developments departing from SPE in
significant ways can, and do, dispense with rule ordering. Of course, this
outcome presupposes the abandonment of derivations, which in turn
entails narrowing the distance between the lexical and the surface representa-
tion, either because these two representations are related directly, as is the
case in Optimality Theory (Prince and Smolenky 1993), or because there are
no such two representations in the first place: only surface representation,
as in Declarative Phonology (see Scobbie 1993, Bird 1995 and Coleman
1998) and in some versions of Correspondence Theory, a development in
Optimality Theory which we will present in the next section (see Burzio 1996,
1997a, for an insightful defence of this position).
In chapter 18 we exemplified and justified rule ordering with a range of

English segmental rules: Vowel Shortening, CiV-Lengthening, s-Voicing, Velar
Softening, Spirantization, [iu]-Formation, Palatalization, [j]-Morification, [j]-
Deletion, Prevocalic Tensing, Stressed Tense Vowel Lengthening, Vowel
Shift and Diphthongization (cf. Rubach 1984b, Borowsky 1986, Halle and
Mohanan 1985, Jensen 1993). Of these, Vowel Shortening, CiV-Lengthening
and s-Voicing are cyclic, that is, apply stepwise in each morphological
constituent, from smallest to largest. If they are cyclic, then they must
obey the Principle of Strict Cyclicity, and therefore they can only apply in
derived environments (Kean 1974, Mascar6 1976, Kiparsky 1982, 1985, 1993;
see Cole 1995 for a convenient summary). Indeed, arguments advanced for
the cyclicity of a given rule often rely on its behaviour with regard to Strict
Cyclicity. In the absence of empirical arguments for the cyclicity or other-
wise of a rule (either from direct empirical considerations, or indirectly
via Strict Cyclicity), the Principle of Late Block Rule Assignment (Halle
and Mohanan 1985) assigns the rule to the latest possible stratum. All the
other English rules just mentioned have been assigned to the non-cyclic
lexical stratum, mostly on empirical considerations, but on the strength of
the Principle of Late Block Rule Assignment in the case of Spirantization.
In addition to lexical rules, languages have postlexical rules, subject to a
number of properties that we listed in (59) in chapter 16. At the end of chap-
ter 18 we mentioned a few postlexical rules of English, but did not go into
details (for discussion of English postlexical rules see Rubach 1985, Booij and
Rubach 1987). Rules can be assigned to more than one stratum (Halle and
Mohanan 1985, Kiparsky 1985). When this happens, the strata in question
must be contiguous, in keeping with the Stratum Contiguity Hypothesis,
referred to in the literature as the Stratum Continuity Hypothesis, perhaps
a less transparent label (see Mohanan 1986).
Optimality Theory
In chapter 19 we explained how the markedness constraints introduced in

chapter 17 can be brought to bear on the simplifications typical of the lan-
guage of young children. From this perspective, the task of the language-
learning child would be to achieve faithfulness to the adult output by
suppressing the effect of the markedness constraints (cf. Stampe 1972). One
simple way of modelling this situation consists in also formalizing faith-
fulness as a constraint (a set of constraints, in fact), ranked vis-a-vis the
markedness constraints. Assuming that, in case of conflict between two con-
straints, the higher-ranked constraint prevails, we can naturally construe
the child’s phonology as involving a higher ranking of the markedness
constraints, and the adult’s phonology as involving a higher ranking of the
faithfulness constraints (Gnanadesikan 1995, Smolensky 1996, Tesar and

Smolensky 1998; but see M. Hale and Reiss 1997). Similarly, differences
between the phonologies of different languages can be attributed to differ-
ences in the respective ranking of the constraints: the constraints themselves
are conceived of as universal, simply the incarnation of natural tendencies,
as proposed in Prince and Smolensky (1993) and elaborated on in Hayes (1996)
and Myers (1997), among others (Archangeli and Pulleyblank 1994 is an obvi-
ous antecedent). The model just outlined is known as Optimality Theory
(OT). One major innovation of Optimality Theory is, obviously, the reduc-
tion of the derivation to a direct mapping of the underlying form onto the
surface form. The main body of OT is laid out in Prince and Smolenky (1993).
For (updated) summaries you can consult Archangeli (1997) and Sherrard
(1997), and for a textbook presentation Kager (in press).
The technical apparatus of Optimality Theory is conceptually very simple.
As in more traditional forms of generative phonology, surface alternations
are regarded as arising from a shared lexical representation. OT lexical forms
and surface forms are in effect related by rules, again as in traditional gen-
erative phonology. The difference is that in traditional generative phonology
the rules are tailor-made to the requirements of the particular language, as
we have seen throughout the book. In Optimality Theory, by contrast, the
rules are universal and act freely on the lexical form to change it into an
infinite number of surface forms, not just one as in traditional generative
phonology. Moreover, because of their universal and automatic nature, rules
have remained unformalized in the OT literature: they are tucked away
in the GEN component of the model, effectively a black box closed to
the observer (see Bromberger and Halle’s 1997 assessment of Tesar 1995a,
1995b; Norton 1998 argues that GEN is redundant). The wild output of GEN
obviously requires drastic cleaning up if the correct output form is to be
obtained: indeed, all the surface forms generated by GEN but one need to
be thrown out. This result is achieved by the filtering action of the ranked
constraints. In the body of the book we have also encountered constraints
as part of the traditional theory, where they are needed for exactly the same
reason as they are needed in OT: to correct mismatches between rule out-
puts and actual forms. There are two main differences between the two
theories with regard to the role of constraints, however. First, in traditional
theory the constraints are few and far between, because the rules themselves
are shaped in such a way that they usually generate the correct output. By
contrast, in OT, rules (= GEN) act wildly and generate incalculable garbage:
in principle, an infinite number of candidates. Consequently, in OT the main
burden of selecting the appropriate surface form rests on the constraints. The
second difference is that in traditional theory constraints are inviolable, in
a manner similar to the general laws of physics. By contrast, in Optimality
Theory constraints are freely violable, and indeed for any particular form
most constraints will be violated. The reason constraints can be (and often
are) violated is that, as in real everyday life, obedience to superiors is more
important than obedience to inferiors: indeed, obedience to inferiors is
rather a contradiction in terms! This means that the guiding principle of the
procedure evaluating the adequacy of candidates will be the maximization
of obedience to the higher constraints: first the constraints ranked highest,
then the constraints ranked next highest, and so on in ordered steps. The
evaluation procedure (EVAL) is made visually explicit in tableaux where
the ranked constraints are plotted against a list of plausible candidates: con-
straint violations are marked with an asterisk in the corresponding box of
the tableau. The optimal candidate will be the candidate which best com-
plies with the constraints, in the manner just explained. Notice, importantly,
that the optimal candidate (which must obviously correspond to the surface
form) need not be the ideal candidate in the sense of complying with all the
constraints: in evaluating candidates, again as in real life, best does not
necessarily mean perfect. This point is forcefully made in McCarthy and Prince
(1994), who label this misconception “the fallacy of perfection”. McCarthy
and Prince (1994) also demonstrate that the effect of lower-ranked constraints
is not necessarily lost for ever: lower constraints become relevant whenever
their higher-ranked counterparts are inactive, a phenomenon McCarthy
and Prince refer to as “the emergence of the unmarked”.
Constraints come in families, and in chapter 19 we examined a number
of these, illustrating their operation with data usually drawn from English.
We have already mentioned two such families: the family of markedness
constraints and the family of faithfulness constraints. In essence, marked-
ness constraints derive the most natural phonetic forms, if left unimpeded:
Golston (1996) indeed identifies (non-natural) lexical representations with
markedness violations. Faithfulness constraints account for the fact that the
phonologies of natural languages need not coincide with the most natural
phonology: less natural forms can make their way into the language, and
then be respected as a result of pressure from the faithfulness constraints.
We mentioned Max and Der as the two central faithfulness constraints, for-
bidding the alteration of lexical forms by deletion and insertion, respectively
(McCarthy and Prince 1996). Identity of segmental material between the
two levels of representation is captured by a set of IDENT constraints, in the
obvious way (McCarthy and Prince 1996: McCarthy and Prince (in press) is
a more accessible version of this paper). Another constraint family contains
the constraints responsible for the prosodic structures we discussed in part
Il of the book: syllabic and metrical structure in particular (Kenstowicz 1994b,
1995, are reasonable introductions to the application of OT in the respective
areas). The syllable constraints are geared to the selection of the optimal
syllable, CV, by imposing a nucleus (Nuc) and an onset (ONsET), ruling out
a coda (No-Copa) and any type of complex constituent (*CoMPLEx), allotting
vowels to the nucleus (*M/V) and consonants to the margins (*N/C), and
enforcing the correct sonority profile (Son). Prince and Smolensky (1993) give
a good rundown of the OT syllabification procedure. Many of the constraints
(but not all) are of course violated in the phonologies of real languages,
among them English, as we would expect they would be. Clements (1997)
applies the procedure to Berber, a language notorious for its liberality with
regard to the sonority threshold of the nucleus, and Rubach (1997) to Polish,
famous for its consonant clusters. The constraints responsible for metrical
structure include FTBIN (to enforce foot binarity), Ft-TyPE (to select between
trochees and iambs), and Non-FIn (to implement extrametricality). Quantity-
sensitivity results from the action of the constraints WTS (Weight-to-Stress
Principle) and WBP (Weight by Position). For surveys of metrical constraints,
you can consult Hammond (1997a) and Roca and Al-Ageli (in press), the
latter cross-comparing OT constraints with the parameters of classical met-
rical theory.
Particularly active in OT analyses is the ALIGN family of constraints,
developed in McCarthy and Prince (1993a) (see also McCarthy and Prince
1993b). As we explained in chapter 19, ALIGN is brought in to enforce either
abuttedness between two constituents in the same level or superimposition
of the edges of two constituents, possibly of different grammatical types, in
different levels. Among the effects derived from ALIGN are the enhancement
of the first or the last of the word’s feet (similarly to Prince and Smolensky’s
1993 EpcEmost), the directionality of iteration (Mester and Padgett 1994; cf.
It6 1986, 1989), and the formal reduction of some of the syllabification con-
straints (Its and Mester 1994). Functionally related to ALIGN are CONTIGUITY
and ANcHoR. ConTIicuITy (Kenstowicz 1994b) requires abuttedness of elements
(segments, syllables, feet), thus blocking epenthesis and/or deletion, to this
extent replicating the effect of Dep and Max string-internally. ANCHOR
establishes correspondences between peripheral elements, and effectively sub-
sumes ALIGN (McCarthy and Prince 1996). At a more general level, the Strict
Layer Hypothesis translates into the constraint SLH (Rubach 1997), decom-
posed in Selkirk (1996) into four more specific constraints, each dealing with
a specific aspect of the SLH, as we have already mentioned. The Elsewhere
Condition is recast under the label “Panini’s Theorem”, interpreted as a prin-
ciple governing constraint ranking (Prince and Smolensky 1993).
The version of Optimality Theory currently predominant, “Correspond-
ence Theory” (McCarthy and Prince 1996), increases the scope and number
of correspondence constraints. Paramount among these are surface-to-
surface corresponding constraints, that simply relate surface forms, in a man-
ner unacceptable to standard generative phonology (the via rules in Hooper
1976 can perhaps be considered a precedent). In chapter 19 we showed how
these correspondence constraints can replicate the effects of both the cycle
and level ordering in traditional theory: see Kenstowicz (1995), McCarthy
(1995), Benua (1996, 1997) and Burzio (1997b). While level ordering also implies
constraint reranking, this not an outrageous outcome, given the very nature
of the strata. More worrying is perhaps the use of constraint reranking in
cases like English noun and verb stress. Constraint reranking is problem-
atic because of learnability. In particular, a model where the substance of
constraints is universal and the ranking between constraints is learnt once
and for all for each language meets a reasonable level of plausibility (Tesar
and Smolensky 1993, 1998; Pulleyblank and Turkel 1997). However, if we
allow the ranking of the constraints to be specific to word classes within a
language (Cohn and McCarthy 1994), perhaps even to individual words them-
selves (Hammond 1997b), the learnability of the system obviously becomes
problematic.
At the end of chapter 19 we examined the applicability of OT to mor-
phology (cf. McCarthy and Prince 1993b, 1996): to the processes of regular
plural formation and possessive suffixation in English, to be more precise.
We saw that the model is indeed successful in predicting both the correct
order of concatenation of the morpheme and the selection of its correct
allophonic shape (see K. Russell 1997 for specific discussion). Particularly
interesting is the interaction between the plural -s and the possessive -’s, which
Correspondence Theory allows us to construe as simple allophone overlap.
Some puzzling facts of the phonology of English truncates (the apparent
overapplication of the rule of [z]-Tensing, in particular) are also resolved
satisfactorily when correspondence constraints are brought into the picture,
provided that we also allow for constraint reranking (Benua 1996).
Two areas are clearly problematic to Optimality Theory. The first such
area concerns the possible need for rules of a traditional kind outside GEN
(McCarthy 1993). Halle and Idsardi (1997) view this development as a
direct indictment of OT, but Blevins (1997) argues that rules can be integrated
into OT without damaging it. The second problematic area concerns the treat-
ment of opaque surface forms. Idsardi (1997), Noyer (1997) and Paradis (1997)
interpret the existence of such forms as support for traditional derivations.
Booij (1997) and Rubach (1997) accommodate such data into OT theory
by appealing to level ordering and by augmenting Optimality Theory into
Derivational Optimality Theory, respectively. McCarthy’s (1998) “Sympathy
Theory” proposes to solve this problem by designating a (possibly) domin-
ated constraint as the “selector”. The selector selects a specific candidate
as the “object of sympathy”: it is the best of the candidates that comply with
the selector. The overall winner (which must of course match the surface
form) is then the candidate that best corresponds with the object of sym-
pathy. For further discussion of Sympathy Theory, you can go to It6é and
Mester (1996b).
In addition to the regular publication outlets, there are several collec-
tions where you can find out more about Optimality Theory, chiefly among
them J. Beckman et al. (1996) and Roca (1997), the latter deliberately aimed
at comparing OT with derivational theory. Most usefully also, many OT practi-

tioners make their unpublished papers or dissertations available on the World
Wide Web, in the so-called Rutger’s Optimality Archive (ROA). At the time
of writing there are 286 such papers available, and the number is growing
steadily. You can download the files freely at the following address:
http: //ruccs.rutgers.edu/roa.html
We have now reached the end of this book. Assuming you have read the
text carefully and thoughtfully, and worked through the various exercises
we have proposed, you are now in a position to tackle the primary literat-
ure of generative phonology. However, some of this literature is not imme-
diately transparent, even to the connoisseur, and you may wish to ease your
way in by first looking into such advanced distillations of the subject as Roca
(1994) and Kenstowicz (1994a), most profitably in this order (Kenstowicz 1994a
is reviewed in Lingua 96, 1995, 189-95 [J. Szpyra] and Phonology 12, 1995,
131-4 [D.A. Dinnsen] and Roca 1994 in Journal of Linguistics 32, 1996, 534-6
[J. Durand] and Phonology 13, 1996, 433-8 [T.A. Hall]).
The purpose ofthis final chhapte

covered in the book ana toe ec
cerned with distinctive featur

allows suchfeaturess freedo
itleaves its essential meanin

Intonation alsoprovides additio
Key Questions
What is meant by “articulation”? List the different ways in which

How does articulation relate to the modulation of pitch known as
acoustics and to perception? “tone” is used in the languages
Explain why some distinctive features of the world.
have binary values, whereas others How does the Principle of Strict
are unary. Cyclicity regulate the interaction be-
Describe the role of sonority tween phonology and morphology?
syllabification. The application of phonological
List the parameters available to processes is frequently restricted
languages in the determination of to particular domains, apparently
stress patterns. defined by morphological and
syntactic criteria. List the domains 9 What types of relationship can

discussed and say how they are exist between pairs of extrinsically
defined. ordered rules?
8 Underlying forms do not need to be 10 In OT, what is an “optimal” candi-
marked for some feature values. date? How is it chosen?
How is the realization of these val-
ues in the surface forms eventually
arrived at?
Be Ghats
ty ee iP Liraacc deere
In chapter 19 we discussed transderivational correspondence constraints relat-

ing a base form with its derivative. Correspondence Theory was initially
conceived as an answer to the problems characterizing the relationship
between base and reduplicant in cases of reduplication. Such a relationship
gives rise to three’ different patterns, known as “overapplication”, “under-
application” and “normal application”:
(i) | overapplication involves a situation where, in the interest of preserv-

ing a phonological match between the base and the reduplicant, some
phonological processes apply to both base and reduplicant, although
the triggering condition is only found in one of them;
(ii) | underapplication obtains when one or other of the reduplicative part-
ners fails to undergo some phonological process, again in the interest
of a phonological match, even though the triggering condition is pre-
sent in that partner;
(iii) under normal application the phonology takes precedence over the iden-
tity relationship, which is sacrificed in the interests of the phonolo-
gical process(es).
The family of IDENT correspondence constraints we introduced in chapter

19 will therefore need to be extended to include BR-IpEenT [base—redupli-
cant identity].
Bahasa Melayu/Indonesia (1)
In Bahasa Melayu/Indonesia, the prefixation of a stem with an initial voice-

less obstruent by a nasal-final prefix causes the nasal to coalesce with the
obstruent, which leaves its place behind (e.g. maN+poton — mamoton, for
“N” = nasal consonant unspecified for place). In reduplication cases the
active voice morpheme maN may be preposed or it may be interposed, as
we show in a. and b. below:
a. Preposed prefix /maN+base+reduplicant]/

poton mamoton-monon ‘cut’ (intensive, repetitive)
tulis | meanulis-nulis ‘write’ (intensive, repetitive)
kira manira-nira ‘guess’ (intensive, repetitive)
b. Interposed prefix /base+maN+reduplicant]/

pukul pukul-mamukul ‘hit’ (reciprocal)
tari tari-monari ‘dance’ (reciprocal)
(i) The two data sets above show examples of two types of application we
referred to above: which are they?
(ii) Provide a suitable set of candidates for evaluation for each of the cases
illustrated, and show how different rankings of the same constraints
can yield the two types of application in a. and b. (For convenience,
the nasal coalescence process can be covered by a general constraint
labelled PHoN-Cons [phonotactic constraints].)
Bahasa Melayu/Indonesia (2)
In Bahasa Melayu/Indonesia there is a prohibition against final /ri The effects

of this are the following:
Absolute root final /r/ is deleted and the preceding vowel is lengthened
(/kotor/ — [koto:] ‘dirty’)
Where the root is followed by a vowel initial suffix, /r/ geminates
(/kotortan/ — [kotorran])
Where an /r/-final prefix precedes a vowel initial root, /r/ is deleted
(/bartkerdga/ — [bakardga] ‘work’).
Consider the following set of forms which show underapplication in

reduplication:
Root Bare root Affixed root

/basar/ ‘big’ [basa:-basa:] [basar-basarran]
/saior/ ‘vegetable’ _[sajoz-sajo:] _[sajor-sajorran]
/tabor/ ‘spread’ [tabo:-tabo:] [tabor-taborran]
Taking into account the fact that /r/-gemination is prompted by a con-

straint against onsetless syllables, work out a suitable constraint ranking
for the affixed root set. (Hint: your constraint set may need to include an
ANCHOR constraint (see chapter 19, section 14) in order to exclude some
apparently well-motivated candidates.)
Japanese
The consequence of the rule of Voiced Velar Nasalization (VVN) in conser-

vative Tokyo speech is that the voiced dorsal stop [g] and the dorsal nasal
[n] appear in complementary distribution: [g] only occurs word-initially
and [n] only word-internally ([geta] *[neta] ‘clogs’ vs. *[kagi] [kani] ‘key’).
This distribution extends to compounds formed with “bound” roots (roots
which cannot occur independently), yielding such alternations as gai+dsiN
‘foreigner’ vs. koku+yai ‘abroad’. We can therefore posit the following con-
straints, which we offer in a shorthand (but we hope transparent) form:
*[n No dorsal nasal in prosodic word-initial position

*e = No voiced dorsal
In ordinary compounds VVN is optional:
a. geta ‘clogs’ niwa+g/yeta ‘garden clogs’

gara ‘pattern’ Sima+g/nara ‘striped pattern’
gei ‘craft, art’ firooto+g/nei ‘amateur’s skill’
A pervasive process in Japanese compounding is Rendaku (see the exercise

in chapter 17). The effect of Rendaku is to voice the initial obstruent in the
second element of a compound (e.g. ama ‘nun’ + tera ‘temple’ — amadera
‘nunnery’). Unlike the situation with compounds of the type shown in
a., when [g] would result from the operation of Rendaku it is obligatorily
pronounced as [n]:
b. kuti ‘mouth’ doku+nuffi ‘abusive language’

kuni ‘country’ jukitnuni = ‘snow country’
kami ‘paper’ ori+yami ‘origami paper’
In order to account for these facts we need to add some additional constraints
to our pair above:
IDENTs, = Surface-to-surface identity (segmental identity between a bound

stem and its free form)
IDENT,, = Lexical-to-surface identity (segmental identity between a lexical
and surface form for [nasal])
RENDAKU
Show the appropriate rankings for niwa-geta, niwa-yeta and doku-yufi, bear-
ing in mind that optionality of surface forms must be the result of free con-
straint rankings.
Axininca Campa
In the Arawakan language Axininca Campa, spoken in Peru, a hiatus at a

V + V juncture is disallowed, leading to the epenthesis of [t]:
/i-N-koma-i/ in.ko.ma.ti ‘he will paddle’

/i-N-koma-aa-i/ —_in.ko.ma.taa.ti ‘he will paddle again’
/i-N-koma-ako-i/ in.ko.ma.ta.ko.ti “he will paddle for’
/i-N-t{*ik-i/ in.tf"i.ki ‘he will cut’
/i-N-t{"ik-aa-i/ in.t{*i.kaa.ti ‘he will cut again’
/i-N-t{"ik-ako-i/ —_in.tf"i.ka.ko.ti ‘he will cut for’
Notice that although epenthesis seems to be prompted by an ONSET con-

straint, this does not affect initial vowels, nor is there coalescence between
adjacent vowels of stem and affix. Notice also that there is a phonotactic con-
straint on codas: in this position the language only permits nasals homor-
ganic to a following stop or affricate. This constraint can be labelled Copa
CoNnD[ITION].
Show how the constraints indicated above interact with the two follow-
ing alignment constraints:
ALIGN-LEFT = Align(Stem, L, PrWd, L)

ALIGN-RicuT = Align(Stem, R, 0, R)
REFERENCES
(ROA = Rutgers Optimality Archive; see page 659.)
Abaglo, P. and Archangeli, D. 1989: Language-particular underspecification:

Gengbe /e/ and Yoruba /i/, Linguistic Inquiry 20, 457-80.
Abercrombie, D. 1967: Elements of General Phonetics. Edinburgh: Edinburgh Uni-
versity Press.
Allen M. 1978: Morphological investigations. PhD, University of Conneticut.
Anderson, J. and Ewen, C. 1987: Principles of Dependency Phonology. Cambridge:
Cambridge University Press.
Anderson, S. 1974: The Organization of Phonology. New York: Academic Press.
Anderson, S. 1978: Tone features. In Fromkin (1978), 133-75.
Anderson, S. 1985: Phonology in the Twentieth Century. Chicago: University of
Chicago Press.
Aoki, H. 1968: Toward a typology of vowel harmony. Internalional Journal of
American Linguistics 34, 142-5.
Archangeli, D. 1984: Underspecification in Yawelmani phonology and morphology.
PhD, MIT. Published by Garland, New York, 1988.
Archangeli, D. 1988a: Extrametricality in Yawelmani. The Linguistic Review 4, 101-20.
Archangeli, D. 1988b: Aspects of underspecification theory. Phonology 5, 183-207.
Archangeli, D. 1997: Optimality theory: an introduction to linguistics in the 1990s.
In Archangeli and Langendoen (1997), 1-32.
Archangeli, D. and Langendoen, T. (eds) 1997: Optimality Theory: An Overview.
Oxford: Blackwell.
Archangeli, D. and Pulleyblank, D. 1989: Yoruba vowel harmony. Linguistic Inquiry
20, 173-217.
Archangeli, D. and Pulleyblank, D. 1994: Grounded Phonology. Cambridge, MA: MIT
Press.
Aronoff, M. and Oehrle, R. (eds) 1984: Language Sound Structure. Cambridge MA: MIT
Press.
Attridge, D. 1982: The Rhythms of English Poetry. London: Longman.
Attridge, D. 1989: Linguistic theory and literary criticism: The Rhythms of English Poetry
revisited. In Kiparsky and Youmans (1989), 183-99.
Bagemihl, B. 1989: The Crossing Constraint and ‘backwards “languages” ’. Natural
Language and Linguistic Theory 7, 481-549.
Bao, Z. 1990: On the nature of tone. PhD, MIT.
Bauer, L. 1986: Notes on New Zealand English phonetics and phonology. English World
Wide 7, 225-58.
666 References
Bauer, L. 1994: English in New Zealand. In Burchfield (1994), 382-429.

Beckman, J., Walsh Dickey, L. and Urbanczyk, S. (eds) 1996: Papers in Optimality Theory.
University of Massachusetts Occasional Papers in Linguistics 18. Graduate Linguistic
Student Association.
Beckman, M. 1986: Stress and Non-Stress Accent. Dordrecht: Foris.
Beckman, M. and Pierrehumbert, J. 1986: Intonational structure in English and
Japanese. Phonology Yearbook 3, 255-309.
Bell, A. 1978: Syllabic consonants. In Greenberg (1978), 153-201.
Benua, L. 1996: Identity effects in morphological truncation. In Beckman et al.
(1996), 77-136.
Benua, L. 1997: Transderivational identity: phonological relations between words. PhD,
University of Massachusetts at Amherst.
Bird, S. 1991: Feature structures and indices. Phonology 8, 137-44.
Bird, S. 1995: Computational Phonology: A Constraint-based Approach. Cambridge:
Cambridge University Press.
Blevins, J. 1990: Alternatives to exhaustivity and conflation in metrical theory. MS,
University of Texas at Austin.
Blevins, J. 1993: A tonal analysis of Lithuanian nominal accent. Language 69, 237-73.
Blevins, J. 1995: The syllable in phonological theory. In Goldsmith (1995), 206-44.
Blevins, J. 1997: Rules in Optimality Theory: two case studies. In Roca (1997),
227-60.
Bloomfield, L. 1939: Menomini morphophonemics. Travaux du Cercle Linguistique de
Prague 8, 105-15.
Bolinger, D. (ed.) 1972: Intonation. Harmondsworth: Penguin.
Bolinger, D. 1978: Intonation across languages. In Greenberg (1978), 471-524.
Bolinger, D. 1986: Intonation and its Parts: Melody in Spoken English. London: Edward
Arnold.
Bolinger, D. 1989: Intonation and its Uses. London: Edward Arnold.
Booij, G. 1981: Rule ordering, rule application, and the organization of grammars.
In W.U. Dressler (ed.), Phonologica 1980, Innsbruck: Institut fiir Sprachwissens-
chaft, 45-56.
Booij, G. 1988: On the relation between lexical and prosodic phonology. In P.M.
Bertinetto and M. Loporcaro (eds), Certamen Phonologicum: Papers from the 1987 Cortona
Phonology Meeting, Turin: Rosenberg and Sellier, 63-76.
Booij, G. 1994: Lexical Phonology: a review. Lingua e Stile 29, 525-55.
Booij, G. 1996: Cliticisation as prosodic integration. The Linguistic Review 13, 219-42.
Booij, G. 1997: Non-derivational phonology meets lexical phonology. In Roca (1997),
261-88.
Booij, G. and Lieber, R. 1993: On the simultaneity of morphological and prosodic
structure. In Hargus and Kaisse (1993), 23-44.
Booij, G. and Rubach, J. 1984: Morphological and prosodic domains in Lexical
Phonology. Phonology Yearbook 1, 1-27.
Booij, G. and Rubach, J. 1987: Postcyclic versus postlexical rules in Lexical Phono-
logy. Linguistic Inquiry 18, 1-44.
Borden, G., Harris, K. and Raphael, L. 1993 [1980]: Speech Science Primer. Baltimore
and London: Williams and Wilkins.
References 667
Borowsky, T. 1986: Topics in the lexical phonology of English. PhD, University of

Massachusetts at Amherst. Circulated by the Graduate Linguistic Student Asso-
ciation. Published by Garland, New York, 1990.
Borowsky, T. 1987: Antigemination in English phonology. Linguistic Inquiry 16,
671-8.
Borowsky, T. 1989: Structure preservation and the syllable coda in English. Natural
Borowsky, T. 1993: On the word level. In Hargus and Kaisse (1993), 199-234.
Brame, M. 1974: The cycle in phonology: stress in Palestinian, Maltese and Spanish.
Linguistic Inquiry 5, 39-60.
Branford, W. 1994: English in South Africa. In Burchfield (1994), 430-96.
Broe, M. 1992: An introduction to feature geometry. In Docherty and Ladd (1992),
149-65.
Bromberger, S. and Halle, M. 1989: Why phonology is different. Linguistic Inquiry
20, 51-70.
Bromberger, S. and Halle, M. 1997: The contents of phonological signs: a compar-
ison between their use in derivational theories and in optimality theories. In Roca
(1997), 93-123.
Bronstein, A.J. 1960: The Pronunciation of American English. New York: Appelton-
Century-Crofts.
Broselow, E. 1995: Skeletal positions and moras. In Goldsmith (1995), 175-205.
Bruce, G. 1977: Swedish Word Accents in Sentence Perspective. Lund: Gleerup.
Burchfield, R. (ed.) 1994: The Cambridge History of the English Language, vol. 5: English
in Britain and Overseas: Origins and Development. Cambridge: Cambridge Univer-
sity Press.
Burzio, L. 1996: Surface constraints versus underlying representations. In Durand and
Katamba (1996), 123-41.
Burzio, L. 1997a:. Strength in numbers. University of Maryland Working Papers in
Linguistics 5, 29-52.
Burzio, L. 1997b: Cycles, non-derived-environment blocking, and correspondence.
MS, Department of Cognitive Science, Johns Hopkins University.
Calabrese, A. 1995: A constraint-based theory of phonological markedness and sim-
plification procedures. Linguistic Inquiry 26, 373-463.
Catford, J.C. 1977: Fundamental Problems in Phonetics. Edinburgh: Edinburgh Uni-
versity Press.
Catford, J.C. 1988: A Practical Introduction to Phonetics. Oxford: Oxford University Press.
Chafe, W. 1968: The ordering of phonological rules. International Journal of American
Chambers, J. 1973: Canadian raising. Canadian Journal of Linguistics 18, 113-35.
Chomsky, N. and Halle, M. 1968: The Sound Pattern of English. New York: Harper
and Row.
Christdas, P. 1988: The phonology and morphology of Tamil. PhD, Cornell
University.
Clark, J. and Yallop, C. 1995 [1990]: An Introduction to Phonetics and Phonology.
Oxford: Blackwell.
Clark, M. 1987: Japanese as a Tone Language. Dordrecht: Foris.
668 References
Clements, G.N. 1977: The autosegmental treatment of vowel harmony. In W.

Dressler and O. Pfeiffer (eds), Phonologica 1976, Innsbruck: Innsbrucker Beitrager
zur Sprachwissenschaft, 111-19.
Clements, G.N. 1983: The hierarchical representation of tone. In LR. Dihoff (ed.), Current
Approaches to African Linguistics, vol. 1, Dordrecht: Foris, 145-76.
Clements, G.N. 1985: The geometry of phonological features. Phonology Yearbook
2, 225-52.
Clements, G.N. 1988: Toward a substantive theory of feature specification. North Eastern
Linguistic Society 18, 79-93.
Clements, G.N. 1991: Place of articulation in consonants and vowels: a unified
theory. Working Papers of the Cornell Phonetics Laboratory 5, 77-123.
Clements, G.N. 1992a: Phonological primes: features or gestures? Phonetica 49,
181-93.
Clements, G.N. 1992b: The sonority cycle and syllable organization. In W. Dressler,
H. Luschiitzky, O. Pfeiffer and J. Rennison (eds), Phonologica 1988, Cambridge:
Cambridge University Press, 63-76.
Clements, G.N. 1997: Berber syllabification: derivations or constraints? In Roca
(1997), 289-330.
Clements, G.N. and Goldsmith, J. 1984: Autosegmental Studies in Bantu Tone.
Dordrecht: Foris.
Clements, G.N. and Hume, E. 1995: The internal organisation of speech sounds.
In Goldsmith (1995), 245-306.
Clements, G.N. and Keyser, S.J. 1983: CV Phonology: A Generative Theory of the
Syllable. Cambridge, MA: MIT Press.
Clements, G.N. and Sezer, E. 1982: Vowel consonant and disharmony in Turkish.
In van der Hulst and Smith (1982), part 2, 213-55.
Cohen, P.R., Morgan, J. and Pollack, M.E. (eds), 1990: Intentions in Communication.
Cambridge, MA: MIT Press.
Cohn, A. and McCarthy, J. 1994: Alignment and parallelism in Indonesian phono-
logy. MS, Cornell University and University of Massachusetts at Amherst.
ROA-25.
Cole, J. 1995: The cycle in phonology. In Goldsmith (1995), 70-113.
Cole, J. and Coleman, J. 1992: No need for cyclicity in generative grammar. Chicago
Linguistic Society 28, vol. 2: The Parasession on the Cycle in Linguistic Theory, 36—50.
Cole, J. and Kisseberth, C. 1994: Perspectives in Phonology. Stanford: CSLI
Publications.
Coleman, J. 1998: Phonological Representations. Cambridge: Cambridge University Press.
Coleman, J. and Local, J. 1991: The “No crossing constraint” in autosegmental
phonology. Linguistics and Philosophy 14, 295-338.
Cook, V. and Newson, M. 1996: Chomsky’s Universal Grammar. Oxford: Blackwell.
Crothers, J. 1978: Typology and universals in vowel systems. In Greenberg (1978),
93=152:-
Cruttenden, A. 1986: Intonation. Cambridge: Cambridge University Press.
Daniloff, R., Schuckers, G. and Feth, L. 1980: The Physiology of Speech and Hearing. An
Introduction. Englewood Cliffs, NJ: Prentice-Hall.
Denes, P. and Pinson, E. 1993 [1973]: The Speech Chain. Garden City, NY: Freeman.
References 669
Diehl, R. (ed.) 1991a: On the Relationship between Phonetics and Phonology. Special issue
of Phonetica 48.2-4.
Diehl, R. 1991b: The role of phonetics within the study of language. In Diehl (1991a),
120-34.
Dinnsen, D. 1979: Current Approaches to Phonological Theory. Bloomington, IN:
Indiana University Press.
Docherty, G. and Ladd, R. (eds) 1992: Papers in Laboratory Phonology, II: Gesture, Segment,
Prosody. Cambridge: Cambridge University Press.
Dogil, G. 1998: Slavic languages. In van der Hulst [in press a], ch. 11.1.
Donegan, P. and Stampe, D. 1979: The study of natural phonology. In Dinnsen (1979),
126-73.
Downing, P. 1977: On the creation and use of English compound nouns. Language
53, 810-42.
Dresher, E. and Kaye, J. 1990: A computational learning model for metrical phono-
logy. Cognition 34, 137-95.
Duanmu, S. 1990: A formal study of syllable, tone, stress and domain in Chinese
languages. PhD, MIT.
Duanmu, S. 1994: Against contour tone units. Linguistic Inquiry 25, 555-608.
Durand, J. and Katamba, F. (eds) 1995: Frontiers of Phonology. London and New York:
Longman.
Durand, J. and Laks, B. (eds) 1996: Current Trends in Phonology: Models and Methods.
Salford: University of Salford Press.
Fabb, N. 1988: English suffixation is constrained only by selectional restrictions. Natural
Fischer-Jorgensen, E. 1975: Trends in Phonological Theory. A Historical Introduction.
Copenhagen: Akademisk Forlag.
Franks, S. 1985: Extrametricality and stress in Polish. Linguistic Inquiry 16, 144-51.
Franks, S. 1987: Regular and irregular stress in Macedonian. International Journal of
Slavic Linguistics and Poetics 1, 227-40.
Fromkin, V. (ed.) 1978: Tone: A Linguistic Survey. New York: Academic Press.
Fry, D. 1979: The Physics of Speech. Cambridge: Cambridge University Press.
Giegerich, H. 1992: English Phonology. Cambridge: Cambridge University Press.
Gimson, A. 1994 [1970]: An Introduction to the Pronunciation of English [revised by
A. Cruttenden]. London: Edward Arnold.
Gnanadesikan, A. 1995: Markedness and faithfulness conditions in child phonology.
MS, University of Massachusetts at Amherst. ROA-106.
Goldsmith, J. 1976a: Autosegmental phonology. PhD, MIT. Circulated by the
Indiana University Linguistics Club. Published by Garland, New York, 1979.
Goldsmith, J. 1976b: An overview of autosegmental phonology. Linguistic Analysis
2, 23-68.
Goldsmith, J. 1979: The aims of autosegmental phonology. In Dinnsen (1979), 202-22.
Goldsmith, J. 1981: English as a tone language. In D. Goyvaerts (1981), Phonology in
the 1980's, Ghent: Story-Scientia, 287-308.
Goldsmith, J. 1984: Meeussen’s rule. In Aronoff and Oehrle (1984), 245-59.
Goldsmith, J. 1985: Vowel harmony in Khalkha Mongolian, Yata, Finnish and
Hungarian. Phonology Yearbook 2, 251-75.
670 References
Goldsmith, J. 1987: Tone and accent and getting the two together. Berkeley Linguistic
Society 13, 88-104.
Goldsmith, J. 1990: Autosegmental and Metrical Phonology. Oxford: Blackwell.
Goldsmith, J. (ed.) 1995: The Handbook of Phonological Theory. Oxford: Blackwell.
Golston, C. 1996: Direct Optimality Theory: representation as pure markedness.
Language 72, 713-48.
Greenberg, J. (ed.) 1978: Universals of Human Language, vol. 2: Phonology. Stanford:
Stanford University Press.
Gussenhoven, C. 1986: English plosive allophones and ambisyllabicity. Gramma 10,
119-41.
Gussenhoven, C. 1990: Tonal association domains and the prosodic hierarchy in English.
In Ramsaran (1990a), 27-37.
Gussenhoven 1991: The English rhythm rule as an accent deletion rule. Phonology 8,
1-35.
Gussenhoven, C. and Rietveld, A. 1992: Intonation contours, prosodic structure and
preboundary lengthening. Journal of Phonetics 20, 283-303.
Hale, K. and White Eagle, J. 1980: A preliminary metrical account of Winnebago accent.
International Journal of American Linguistics 46, 117-32.
Hale, M. and Reiss, C. 1997: Formal and empirical arguments concerning phono-
logical acquisition. MS, Concordia University, Montreal. ROA-233. [To appear in
Linguistic Inquiry.]
Halle, M. 1957: In defense of the number two. In E. Pulgram (ed.), Studies Presented
to Joshua Whatmough, The Hague: Mouton, 65-72.
Halle, M. 1962: Phonology in generative grammar. Word 18, 54-72.
Halle, M. 1983: On distinctive features and their articulatory implementation.
Natural Language and Linguistic Theory1, 91-105.
Halle, M. 1990: Respecting metrical structure. Natural Language and Linguistic Theory
8, 149-76.
Halle, M. 1991: Phonological features. In W. Bright (ed.), International Encyclopedia of
Linguistics, Oxford: Oxford University Press, 207-12.
Halle, M. 1995: Feature geometry and feature spreading. Linguistic Inquiry 26,
1-46.
Halle, M. and Clements, G.N. 1983: Problem Book in Phonology. Cambridge, MA:
MIT Press.
Halle, M. and Idsardi, W. 1995: General properties of stress and metrical structure.
In Goldsmith (1995), 403-43.
Halle, M. and Idsardi, W. 1997: r, hypercorrection, and the Elsewhere Condition.
In Roca (1997), 331-48.
Halle, M. and Kenstowicz, M. 1991: The Free Element Condition and cyclic versus
noncyclic stress. Linguistic Inquiry 22, 457-501.
Halle, M. and Mohanan, K.P. 1985: Segmental phonology of Modern English.
Halle, M. and Stevens, K. 1971: A note on laryngeal features. MIT Quarterly Progress
Report 11, Research Laboratory of Electronics, MIT, 198-213.
Halle, M. and Vergnaud, J.-R. 1981: Harmony processes. In W. Klein and W. Levelt
(eds), Crossing the Boundaries in Linguistics, Dordrecht: Reidel, 1-22.
Halle, M. and Vergnaud, J.-R. 1987a: An Essay on Stress. Cambridge, MA: MIT Press.
References 671
Halle, M. and Vergnaud, J.-R. 1987b: Stress and the cycle. Linguistic Inquiry 18, 45-84.
Hammond, M. 1988: On deriving the Well-Formedness Condition. Linguistic Inquiry
19, 467-99.
Hammond, M. 1989: Cyclic stress and accent in English. West Coast Conference on Formal
Hammond, M. 1997a: Optimality Theory and prosody. In Archangeli and
Langendoen (1997), 33-58.
Hammond, M. 1997b: Underlying representations in Optimality Theory. In Roca (1997),
349-65.
Hammond, M. and Noonan M. (eds) 1988: Theoretical Morphology. San Diego, CA:
Academic Press.
Hankamer, J. and Aissen, J. 1974: The sonority hierarchy. Chicago Linguistic Society
10: Parasession on Natural Phonology, 131-45.
Haraguchi, S. 1977: The Tone Pattern of Japanese: An Autosegmental Theory of Tone. Tokyo:
Kaitakusha.
Hardcastle, W. 1973: Physiology of Speech Production: An Introduction for Speech
Scientists. London: Academic Press.
Hardcastle, W. and Laver, J. (eds) 1997: The Handbook of the Phonetic Sciences. Oxford:
Blackwell.
Hargus, S. 1993: Modelling the phonology—morphology interface. In Hargus and Kaisse
(1993), 45-74.
Hargus, S. and Kaisse, E. (eds) 1993: Studies in Lexical Phonology. San Diego, CA:
Academic Press.
Harris, J. 1994: English Sound Structure. Oxford: Blackwell.
Harris, J. and Lindsey, G. 1995: The elements of phonological representation. In Durand
and Katamba (1995), 34-79.
Harris, R. 1987: Reading Saussure. La Salle, IL: Open Court.
Hawkins, P. 1992: Introducing Phonology. London: Routledge.
Hayes, B. 1979: Extrametricality. MIT Working Papers in Linguistics 1, 77-86.
Hayes, B. 1980: A metrical theory of stress rules. PhD, MIT. Circulated by the
Indiana University Linguistics Club, 1981. Published by Garland, New York, 1985.
Hayes, B. 1982: Extrametricality and English stress. Linguistic Inquiry 13, 227-76.
Hayes, B. 1984: The phonology of rhythm in English. Linguistic Inquiry 15, 33-74.
Hayes, B. 1989a: Compensatory lengthening in moraic phonology. Linguistic Inquiry
20, 253-306.
Hayes, B. 1989b: The prosodic hierarchy in meter. In Kiparsky and Youmans
(1989), 201-60.
Hayes, B. 1990a: Diphthongization and coindexing. Phonology 7, 31-71.
Hayes, B. 1990b: Precompiled phrasal phonology. In Inkelas and Zec (1990), 85-108.
Hayes, B. 1995: Metrical Stress Theory: Principles and Case Studies. Chicago: University
of Chicago Press.
Hayes, B. 1996: Phonetically-driven phonology: the role of Optimality Theory and
inductive grounding. MS, University of California, Los Angeles. ROA-158.
Hayward, R. 1988: In defence of the skeletal tier. Studies in African Linguistics 19, 131-72.
Hobbs, J. 1990: The Pierrehumbert-Hirschberg theory of intonational meaning
made simple: comments on Pierrehumbert and Hirschberg. In Cohen et al. (1990),
313=23.
672 References
Hockett, C. 1956: A Manual of Phonology. Baltimore, MD: Waverly Press.

Hooper, J. 1972: The syllable in phonological theory. Language 48, 525-40.
Hooper, J. 1976: An Introduction to Natural Generative Phonology. New York:
Academic Press.
Hualde, J.I. 1988: Affricates are not contour segments. West Coast Conference on
Formal Linguistics 7, 143-57.
Hualde, J.I. 1989: The strict cycle condition and noncyclic rules. Linguistic Inquiry
20, 675-80.
Hualde, J.I. 1991a: Basque Phonology. London: Routledge.
Hualde, J.I. 1991b: Unspecified and unmarked vowels. Linguistic Inquiry 22, 205-9.
Hualde, J.I. 1998: Basque accentuation. In van der Hulst [in press a], ch. 14.
Hulst, H. van der 1988: Atoms of segmental structure: components, gestures, and
dependency. Phonology 6, 253-84.
Hulst, H. van der 1995: Radical CV Phonology: the categorial gesture. In Durand
and Katamba (1995), 80-116.
Hulst, H. van der (ed.) [in press a]: Word Prosodic Systems in the Languages of Europe.
Berlin: Mouton de Gruyter.
Hulst, H. van der [in press b]: Word accent. In van der Hulst [in press a], ch. 1.
Hulst, H. van der and Smith, N. 1982a: Prosodic domains and opaque segments
in autosegmental theory. In van der Hulst and Smith (1982b), part 2, 311-36.
Hulst, H. van der and Smith, N. (eds) 1982b: The Structure of Phonological
Representations. Dordrecht: Foris.
Hulst, H. van der and Smith, N. (eds) 1988: Features, Segmental Structure, and
Harmony Processes. Dordrecht: Foris.
Hulst, H. van der and Weijer, J. van de 1995: Vowel harmony. In Goldsmith (1995),
495-534.
Hulst, H. van der, Hendriks, B. and Weijer, J. van de [in press]: A survey of word
prosodic systems of European languages. In van der Hulst [in press al, ch. 7.
Hume, E. and Odden, D. 1996: Reconsidering [consonantal]. Phonology 13, 345-76.
Hyman, L. (ed.) 1977a: Studies in Stress and Accent. Special issue of Southern California
Occasional Papers in Linguistics 4, Department of Linguistics, University of
Southern California.
Hyman, L. 1977b: On the nature of linguistics stress. In Hyman (1977a), 37-82.
Hyman, L. 1978: Tone and/or accent. In D.J. Napoli (ed.), Elements of Tone, Stress,
and Intonation, Washington, DC: Georgetown University Press, 1-20.
Hyman, L. 1985: A Theory of Phonological Weight. Dordrecht: Foris.
Hyman, L. 1989: Accent in Bantu: a reappraisal. Studies in the Linguistic Sciences 19,
111-28.
Hyman, L. 1993: Problems for rule ordering in phonology: two Bantu cases. In
J. Goldsmith (ed.), The Last Phonological Rule, Chicago: University of Chicago
Press, 195-222.
Hyman, L. 1994: Cyclic phonology and morphology in Cibemba. In Cole and
Kisserberth (1994), 81-112.
Hyman, L. and Ngunga, A. 1994: On the non-universality of tonal association ‘con-
ventions’: evidence from Ciyao. Phonology 11, 25-68.
Idsardi, W. 1997: Phonological derivations and historical changes in Hebrew spir-
antisations. In Roca (1997), 367-92.
References 673
Ingria, R. 1980: Compensatory lengthening as a metrical phenomenon. Linguistic Inquiry

11, 465-95.
Inkelas, S. 1989: Prosodic constituency in the lexicon. PhD, Stanford University.
Published by Garland, New York, 1990.
Inkelas, S. 1993: Deriving cyclicity. In Hargus and Kaisse (1993), 75-110.
Inkelas, S. and Zec, D. 1988: Serbo-Croatian pitch accent: the interactions of tone,
stress, and intonation. Language 64, 227-48.
Inkelas, S. and Zec, D. 1990: The Phonology-Syntax Connection. Chicago: University
of Chicago Press.
International Phonetic Association. 1949: The Principles of the International Phonetic
Association. International Phonetic Association.
It6, J. 1986: Syllable theory in prosodic phonology. PhD, University of Massachu-
setts at Amherst. Circulated by the Graduate Linguistic Student Association.
It6, J. 1989: A prosodic theory of epenthesis. Natural Language and Linguistic Theory
7, 217-60.
It6, J. and Mester, A. 1994: Reflections on CodaCon and Alignment. Phonology at Santa
Cruz, iii, 27—46.
It6, J. and Mester, A. 1995: Japanese phonology. In Goldsmith (1995), 817-38.
It6, J. and Mester, A. 1996a: The core-periphery structure of the lexicon and con-
straints on reranking. In J. Beckman et al. (1996), 181-210.
It6, J. and Mester, A. 1996b: Sympathy theory and German truncations. University
of Maryland Working Papers in Linguistics 5, 117-38. ROA-211.
Iverson, G. 1974: Ordering constraints in phonology. PhD, University of Minnesota.
Iverson, G. 1993: (Post)lexical rule application. In Hargus and Kaisse (1993), 255-75.
Iverson, G. 1995: Rule ordering. In Goldsmith (1995), 609-14.
Iverson, G. and Wheeler, D. 1988: Blocking and the Elsewhere Condition. In
Hammond and Noonan (1998), 325-38.
Jakobson, R. 1941: Kindersprache, Aphasie und allgemeine Lautgesetze. Translation: A.R.
Keiler, Child Language, Aphasia and Phonological Universals, The Hague: Mouton, 1968.
Also in Selected Writings, vol. 1. The Hague: Mouton, 328-401.
Jakobson, R. 1990: On Language. Cambridge, MA: Harvard University Press.
Jensen, J. 1993: English Phonology. Amsterdam: Benjamins.
Jones, C. 1989: A History of English Phonology. London: Longman.
Jones, D. 1966 [1909]: The Pronunciation of English. Cambridge: Cambridge Univer-
sity Press.
Jones, D. 1967 [1918]: An Outline of English Phonetics. Cambridge: Heffer.
Kager, R. 1989: A metrical theory of stress and destressing in English and Dutch.
PhD, University of Utrecht. Published by ICG Printing, Dordrecht, 1989.
Kager, R. 1993: Alternatives to the iambic-trochaic law. Natural Language and
Linguistic Theory 11, 381-432.
Kager, R. 1995: The metrical theory of word stress. In Goldsmith (1995), 367—402.
Kager, R. [in press]: Optimality Theory: A Textbook. Cambridge: Cambridge Univer-
sity Press.
Kahn, D. 1976: Syllable-based generalizations in English phonology. PhD, MIT.
Circulated by the Indiana University Linguistics Club. Published by Garland, New
York, 1980.
674 References
Kaisse, E. 1985: Connected Speech: The Interaction of Syntax and Phonology. New York:
Academic Press.
Kaisse, E. 1990: Toward a typology of postlexical rules. In Inkelas and Zec (1990),
127-43.
Kaisse, E. 1992: Can [consonantal] spread? Language 68, 313-32.
Kaisse, E. and Hargus, S. 1993: Introduction. In Hargus and Kaisse (1993), 1-19.
Kaisse, E. and Shaw, P. 1985: On the theory of lexical phonology. Phonology Yearbook
2, 1-30.
Kaye, J. 1990: What ever happened to dialect B? In Mascar6 and Nespor (1990), 259-63.
Kaye, J. 1992: Do you believe in magic? The story of s + C sequences. SOAS Working
Papers in Linguistics and Phonetics 2, 293-313.
Kaye, J., Lowenstamm, J. and Vergnaud, J.-R. 1985: The internal structure of phono-
logical elements: a theory of charm and government. Phonology Yearbook 2, 305-28.
Kean, M.-L. 1974: The strict cycle in phonology. Linguistic Inquiry 5, 179-203.
Kean, M.-L. 1975: The theory of markedness in generative grammar. PhD, MIT.
Circulated by the Indiana University Linguistics Club.
Keating, P. 1984: Phonetic and phonological representation of stop consonant voicing.
Language 60, 286-319.
Keating, P. 1987: Survey of phonological features. UCLA Working Papers in Phonetics
66, 124-50. Circulated by the Indiana University Linguistics Club.
Keating, P. 1988: Underspecification in phonetics. Phonology 5, 275-92.
Keating, P. 1990: Phonetic representation in a generative grammar. Journal of
Phonetics 18, 321-34.
Kenstowicz, M. 1994a: Phonology in Generative Grammar. Oxford: Blackwell.
Kenstowicz, M. 1994b: Syllabification in Chuckchee: a constraint based analysis.
Formal Linguistics Society of the Midwest 4, 160-81. ROA-30.
Kenstowicz, M. 1995: Cyclic vs. non-cyclic constraint evaluation. Phonology 12, 397-436.
Kenstowicz, M. 1996: Base-identity and uniform exponence: alternatives to cyclicity.
In Durand and Laks (1996), vol. 1, 365-95.
Kent, R. and Read, C. 1992: The Acoustic Analysis of Speech. San Diego, CA: Singular
Publishing Group.
Kenyon, J.S. 1935 [1924]: American Pronunciation. Ann Arbor, MI: George Wahr.
Kimura, A. 1996: A phonological analysis of English loanwords in Japanese.
Masters, University of Essex.
Kingston, J. and Beckman, M. (eds) 1990: Papers in Laboratory Phonology I: Between
the Grammar and Physics of Speech. Cambridge: Cambridge University Press.
Kiparsky, P. 1968: Linguistic universals and linguistic change. In E. Bach and R. Harms
(eds), Universals in Linguistic Theory, New York: Holt, Rinehart and Winston,
170-202.
Kiparsky, P. 1973: “Elsewhere” in phonology. In S. Anderson and P. Kiparsky (eds),
A Festschrift for Morris Halle, New York: Holt, Rinehart and Winston, 93-106.
Kiparsky, P. 1977: The rhythmic structure of English verse. Linguistic Inquiry 8, 189-247.
Kiparsky, P. 1982: Lexical morphology and phonology. In I.-S. Yang (ed.), Linguistics
in the Morning Calm, Seoul: Hanshin Publishing, 3-91.
Kiparsky, P. 1983: Word formation and the lexicon. In Proceedings of the 1982 Mid-
America Linguistics Conference, 3-29.
References 675
Kiparsky, P. 1985: Some consequences of lexical phonology. Phonology Yearbook

2, 85-138.
Kiparsky, P. 1993: Blocking in non-derived environments. In Hargus and Kaisse (1993),
277-313.
Kiparsky, P. and Halle, M. 1977: Towards a reconstruction of the Indo-European
accent. In Hyman (1977a), 209-38.
Kiparsky, P. and Youmans, G. (eds) 1989: Rhythm and Meter. San Diego: Academic
Press. [Review: Hammond, M. 1992: Phonology 9, 358-62.]
Klausenberger, J. 1990: Topic... Comment. Natural Language and Linguistic Theory
8, 621-3.
Kornai A. and Pullum, G. 1990: The X-bar theory of phrase structure. Language 66,
24-50.
Koutsoudas, A. 1980: The question of rule ordering: some common fallacies. Journal
of Linguistics 16, 19-35.
Koutsoudas, A., Sanders, G. and Noll, C. 1974: On the application of phonological
rules. Language 50, 1-28.
Krapp, G. 1969 [1919]: The Pronunciation of Standard English in America. New York:
Ams Press.
Kreidler, C. 1989: The Pronunciation of English. Oxford: Blackwell.
Kreidler, C. 1997: Describing Spoken English. London: Routledge.
Kubozono, H. 1993: The Organization of Japanese Prosody. Tokyo: Kurosio.
Labov, W. 1994: Principles of Linguistic Change: Internal Factors. Oxford: Blackwell.
Ladd, R. 1992: An introduction to intonational phonology. In Docherty and Ladd
(1992), 321-34.
Ladd, R. 1997: Intonational Phonology. Cambridge: Cambridge University Press.
Ladefoged, P. 1973: The features of the larynx. Journal of Phonetics 1, 73-83.
Ladefoged, P. 1990: Some reflections on the IPA. Journal of Phonetics 18, 335-46.
Ladefoged, P. 1993 [1975]: A Course in Phonetics. Fort Worth, TX: Harcourt Brace
Jovanovitch.
Ladefoged, P. and Halle, M. 1988: Some major features of the International Phonetic
Alphabet. Language 64, 577-82.
Ladefoged, P. and Maddieson, I. 1996: The Sounds of the World’s Languages. Oxford:
Blackwell.
Lass, R. 1984: Phonology: An Introduction to Basic Concepts. Cambridge: Cambridge
University Press.
Lass, R. 1990: A ‘standard’ South African vowel system. In Ramsaran (1990a),
272-85.
Laver, J. 1994: Principles of Phonetics. Cambridge: Cambridge University Press.
Leben, W. 1976: The tones in English intonation. Linguistic Analysis 2, 69-107.
Leben, W. 1978: The representation of tone. In Fromkin (1978), 177-219.
Levin, J. 1983: Reduplication and prosodic structure. MS, MIT.
Levin, J. 1985: A metrical theory of syllabicity. PhD, MIT.
Liberman, M. 1975: The intonational system of English. PhD, MIT. Circulated by the
Indiana University Linguistics Club. Published by Garland, New York, 1980.
Liberman, M. and Prince, A. 1977: On stress and linguistic rhythm. Linguistic Inquiry
8, 249-336.
676 References
Lieber, R. 1989: Argument linking and compounds in English. Linguistic Inquiry 14,
251-85.
Lindau, M. 1978: Vowel features. Language 54, 541-63.
Lodge, K. 1989: A non-segmental account of German umlaut: diachronic and syn-
chronic perspectives. Linguistische Berichte 124, 470-91.
Lombardi, L. 1990: The nonlinear organization of the affricate. Natural Language and
Linguistic Theory 8, 375-425.
Lombardi, L. 1996: Postlexical rules and the status of privative features. Phonology
13) 1=38:
Lovins, J. 1975: Loanwords and the phonological structure of Japanese. PhD,
University of Chicago. Circulated by the Indiana University Linguistics Club.
McCarthy, J. 1981: A prosodic theory of non-concatenative phonology. Linguistic Inquiry
12, 443-66.
McCarthy, J. 1982: Prosodic structure and expletive infixation. Language 58, 574-90.
McCarthy, J. 1984: Prosodic structure in morphology. In Aronoff and Oehrle (1984),
299-317.
McCarthy, J. 1986: OCP effects: gemination and antigemination. Linguistic Inquiry 17,
207-63.
McCarthy, J. 1988: Feature geometry and dependency: a review. Phonetica 43,
84-108.
McCarthy, J. 1993: A surface case of constraint violation. In Paradis and LaCharité
(1993),,169=95:
McCarthy, J. 1995: Extensions of faithfulness: Rotuman revisited. MS, University of
Massachusetts at Amherst. ROA-110. ,
McCarthy, J. 1998: Sympathy and phonological opacity. ROA-252.
McCarthy, J. and Prince, A. 1990: Foot and word in prosodic morphology: the
Arabic broken plural. Natural Language and Linguistic Theory 8, 209-83.
McCarthy, J. and Prince, A. 1993a: Generalized alignment. Yearbook in Morphology
1993. Dordrecht: Kluwer, 79-153.
McCarthy, J. and Prince, A. 1993b: Prosodic morphology I. Constraint interaction and
satisfaction. MS, University of Massachusetts at Amherst and Rutgers University.
McCarthy, J. and Prince, A. 1994: The emergence of the unmarked: optimality in
prosodic morphology. North Eastern Linguistic Society 24, 333-79.
McCarthy, J. and Prince, A. 1995: Prosodic morphology. In Goldsmith (1995),
318-66.
McCarthy, J. and Prince, A. 1996: Faithfulness and reduplicative identity. In
J. Beckman et al. (1996), 249-384.
McCarthy, J. and Prince, A. [in press]: Faithfulness and identity in Prosodic
Morphology. In R. Kager, H. van der Hulst and W. Zonneveld [in press]: The
Prosody—Morphology Interface. Cambridge: Cambridge University Press. ROA-216.
McCawley, J. 1977: Accent in Japanese. In Hyman (1977a), 261-302.
Maddieson, I. 1984: Patterns of Sounds. Cambridge: Cambridge University Press.
Marantz, A. 1988: Clitics, morphological merger, and the mapping of phonological
structure. In Hammond and Noonan (1988), 253-70.
Mascar6, J. 1976: Catalan phonology and the phonological cycle. PhD, MIT.
Circulated by the Indiana University Linguistics Club.
References 677
Mascaré, J. and Nespor, M. (eds) 1990: Grammar in Progress. Glow Essays for Henk van
Riemsdijk. Dordrecht: Foris.
Mester, A. and Padgett, J. 1994: Directional syllabification in generalized alignment.
Phonology at Santa Cruz, iii, 79-87. ROA-1.
Mohanan, K.P. 1985: Syllable structure and lexical structure in English. Phonology
Yearbook 2, 139-55.
Mohanan, K.P. 1986: The Theory of Lexical Phonology. Dordrecht: Foris.
Mohanan, K.P. 1991: On the basis of Radical Underspecification. Natural Language
and Linguistic Theory 9, 285-325.
Mohanan, K.P. 1995: The organization of grammar. In Goldsmith (1995), 24-69.
Moulton, W. 1990: Some vowel systems in American English. In Ramsaran (1990a),
119-36.
Myers, S. 1987a: Tone and the structure of words in Shona. PhD, University of
Massachusetts at Amherst. Circulated by the Graduate Linguistic Student
Association.
Myers, S. 1987b: Vowel shortening in English. Natural Language and Linguistic Theory
5, 485-518.
Myers, S. 1991: Persistent rules. Linguistic Inquiry 22, 315-44.
Myers, S. 1997: Expressing phonetic naturalness in phonology. In Roca (1997),
123=52.
Nespor, M. 1990a: On the rhythm parameter in phonology. In I. Roca (ed.), Logical
Issues in Language Acquisition, Dordrecht: Foris, 157-75.
Nespor, M. 1990b: On the separation of prosodic and rhythmic phonology. In
Inkelas and Zec (1990), 243-58.
Nespor, M. and Vogel, I. 1986: Prosodic Phonology. Dordrecht: Foris.
Nespor, M. and Vogel, I. 1989: On clashes and lapses. Phonology 6, 69-116.
Nolan, F. 1995: Preview of the IPA handbook. Journal of the International Phonetic
Association 25, 1-33.
Norton, R. 1998: Optimality Theory without Gen. Essex Graduate Student Papers in
Language and Linguistics 2, 19-30.
Noyer, R. 1997: Attic Greek accentuation and intermediate derivational repres-
entations. In Roca (1997), 501-27.
Odden, D. 1981: Problems in tone assignment in Shona. PhD, University of Illinois.
Odden, D. 1986: On the Obligatory Contour Principle. Language 62, 353-83.
Odden, D. 1991: Vowel geometry. Phonology 8, 261-89.
Odden, D. 1993: Interaction between modules in Lexical Phonology. In Hargus and
Kaisse (1993), 259-77.
Odden, D. 1995: Tone: African languages. In Goldsmith (1995), 444-75.
Ohala, J. 1978: Production of tone. In Fromkin (1978), 5-39.
Ohala, J. 1997: The relation between phonetics and phonology. In Hardcastle and
Laver (1997), 674-94.
Ohala, J. and Kawasaki, H. 1984: Prosodic phonology and phonetics. Phonology
Yearbook 1, 113-27.
Ohman, S. 1966: Coarticulation in CVC utterances: spectrographic measurements.
Journal of the Acoustic Society of America 66, 1691-1702.
Padgett, J. 1995: Stricture in Feature Geometry. Stanford, CA: CSLI Publications.
678 References
Paradis, C. 1988: On constraints and repair strategies. The Linguistic Review 6, 71-97.
Paradis, C. 1997: Non-transparent constraint effects in Gere: from cycles to deriva-
tions. In Roca (1997), 529-50.
Paradis, C. and LaCharité, D. (eds) 1993: Constraint Based Theories in Multilinear
Phonology. Special issue of the Canadian Journal of Linguistics 38 (2).
Pesetzky, D. 1979: Russian morphology and lexical theory. MS, MIT.
Pierrehumbert, J. 1980: The phonetics and phonology of English intonation. PhD, MIT.
Pierrehumbert, J. 1990: Phonological and phonetic representation. Journal of Phon-
etics 18, 375-94.
Pierrehumbert, J. 1991: The whole theory of sound structure. In Diehl (1991a), 223-32.
Pierrehumbert, J. and Beckman, M. 1988: Japanese Tone Structure. Cambridge, MA:
MIT Press.
Pierrehumbert, J. and Hirschberg, J. 1990: The meaning of intonational contours in
the interpretation of discourse. In Cohen et al. (1990), 271-311.
Pike, K. 1948: Tone Languages. Ann Arbor, MI: University of Michigan Press.
Pike, K. and Pike, E. 1947: Immediate constituents of Mazatec syllables. International
Journal of American Linguistics 13, 78-91.
Poser, W. 1984: The phonetics and phonology of tone and intonation in Japanese.
PhD, MIT.
Poser, W. 1993: Are Strict Cycle effects derivable? In Hargus and Keisse (1993), 315-21.
Prince, A. 1983: Relating to the grid. Linguistic Inquiry 14, 19-100.
Prince, A. 1985: Improving tree theory. Berkeley Linguistic Society 11, 471-90.
Prince, A. and Smolensky, P. 1993: Optimality Theory: Constraint Interaction in gen-
erative grammar. MS, Rutgers University and University of Colorado.
Pulleyblank, D. 1986a: Tone in Lexical Phonology. Dordrecht: Reidel.
Pulleyblank, D. 1986b: Underspecification and low vowel harmony in Yoruba.
Studies in African Linguistics 17, 119-53.
Pulleyblank, D. 1988a: Underspecification, the feature hierarchy and Tiv vowels.
Phonology 5, 299-326.
Pulleyblank, D. 1988b: Vocalic underspecification in Yoruba. Linguistic Inquiry 19,
233-70.
Pulleyblank, D. 1996: Feature geometry and underspecification. In Durand and
Katamba (1996), 3-33.
Pulleyblank, D. 1997: Optimality Theory and features. In Archangeli and Langen-
doen (1997), 59-101.
Pulleyblank, D. and Turkel, W. 1997: Gradient retreat. In Roca (1997), 153-93.
Pullum, G. 1976: The Duke of York gambit. Journal of Linguistics 12, 83-103.
Pullum, G. and Ladusaw, W. 1986: Phonetic Symbol Guide. Chicago: University of
Chicago Press.
Pyles, T. and Algeo, J. 1992 [1978]: The Origins and Development of the English
Language. New York: Harcourt Brace Jovanovich.
Ramsaran, S. (ed.) 1990a: Studies in the Pronunciation of English. London and New
York: Routledge.
Ramsaran, S. 1990b: RP: fact and fiction. In Ramsaran (1990a), 178-90.
Rice, K. 1987: The function of structure preservation: derived environments.
North
Eastern Linguistic Society 17, 501-20.
References 679
Rice, K. 1990: Predicting rule domains in the phrasal phonology. In Inkelas and Zec
(1990), 289-312.
Rice, K. 1992: On deriving sonority: a structural account of sonority relationships.
Phonology 9, 61-99.
Ringen, C. 1975: Vowel Harmony: theoretical implications, PhD, Indiana University.
Ringen, C. and Vago, R. 1995: A constraint-based analysis of Hungarian vowel har-
mony. In I. Kenesei (ed.), Approaches to Hungarian. vol. 5: Levels and structures, Szeged:
JATE, 309-19.
Roach, P. 1991 [1983]: English Phonetics and Phonology. Cambridge: Cambridge
University Press.
Roca, I. 1988: Theoretical implications of Spanish word stress. Linguistic inquiry 19,
393-423.
Roca, I. 1992: Constraining extrametricality. In W. Dressler, H. Luschitzky, O.
Pfeiffer and J. Rennison (eds), Phonologica 1988, Cambridge: Cambridge Univer-
sity Press, 239-48.
Roca, I. 1994: Generative Phonology. London: Routledge.
Roca, I. (ed.) 1997: Derivations and Constraints in Phonology. Oxford: Clarendon Press.
Roca, I. and Al-Ageli, H. [in press]: Optimal metrics. In J.S. Hannah and M.
Davenport (eds), Proceedings of the 1994 Durham Phonology Conference.
Ross, J. 1972: A reanalysis of English word stress. In M. Brame (ed.), Contributions
to Generative Phonology, Austin and London: University of Texas Press, 229-323.
Rubach, J. 1984a: Cyclic and Lexical Phonology: The Structure of Polish. Dordrecht: Foris.
Rubach, J. 1984b: Segmental rules of English and cyclic phonology. Language 60, 21-54.
Rubach, J. 1985: Lexical Phonology: lexical and postlexical derivations. Phonology
Yearbook 2, 157-72.
Rubach, J. 1990: Final devoicing and lexical syllabification in German. Linguistic Inquiry
21, 79-94.
Rubach, J. 1993: Skeletal versus moraic representations in Slovak. Natural Language
Rubach, J. 1996: Shortening and ambisyllabicity in English. Phonology 13, 197-237.
Rubach, J. 1997: Extrasyllabic consonants in Polish: Derivational Optimality Theory.
In Roca (1997), 551-81.
Rubach, J. and Booij, G. 1990a: Edge of constituents effect in Polish. Natural Language
Rubach, J. and Booij, G. 1990b: Syllable structure assignment in Polish. Phonology
(4421358,
Russell, B. 1996 [1946]: History of Western Philosophy. London: Routledge.
Russell, K. 1997: Optimality Theory and morphology. In Archangeli and Langen-
doen (1997), 102-33.
Sagey, E. 1986a: The representation of features and relations in nonlinear phono-
logy. PhD, MIT.
Sagey, E. 1986b: On the representation of complex segments in Kinyarwanda. In Wetzels
and Sezer (1986), 251-95.
Sagey, E. 1988: On the ill-formedness of crossing association lines. Linguistic Inquiry
19, 109-18.
680 References
Sainz, S. 1992: A noncyclic approach to English word stress. Chicago Linguistic

Society 28, vol. 2: The Parasession on the Cycle in Linguistic Theory, 263-76.
Saussure, F. de 1916: Cours de Linguistique Générale. Paris: Payot. Translation: W. Baskin,
Course in General Linguistics, New York: Philosophical Library, 1959. See also
Harris, R. (1987).
Schane, S. 1984: The fundamentals of Particle Phonology. Phonology 1, 129-55.
Schuh, R.G. 1978: Tone rules. In Fromkin (1978), 221-56.
Scobbie, J.M. 1993: Constraint violation from the viewpoint of Declarative Phono-
logy. In Paradis and LaCharité (1993), 155-67.
Selkirk, E. 1980: The role of prosodic categories in English word stress. Linguistic Inquiry
11, 563-605.
Selkirk, E. 1981: On prosodic structure and its relationship to syntactic structure. In
T. Fretheim (ed.), Nordic Prosody, II, Trondheim: Tapir, 111-40.
Selkirk, E. 1982: Syllables. In van der Hulst and Smith (1982b), part 2, 337-83.
Selkirk, E. 1984a: On the major class features and syllable theory. In Aronoff and
Oehrle (1984), 107-36.
Selkirk, E. 1984b: Phonology and Syntax: The Relation between Sound and Structure.
Cambride, MA: MIT Press.
Selkirk, E. 1986: On derived domains in sentence phonology. Phonology Yearbook 3,
371-405.
Selkirk, E. 1990: On the nature of prosodic constituency. In Kingston and Beckman
(1990), 179-200.
Selkirk, E. 1996: The prosodic structure of function words. In J. Beckman et al. (1996),
439-70.
Shaw, P. 1991: Consonant harmony systems: the special status of Coronal Harmony.
In C. Paradis and J.-F. Prunet (eds), The Special Status of Coronals, San Diego, CA:
Academic Press, 125-57.
Sherrard, N. 1997: Questions of priorities: an introductory overview of Optimality
Theory in phonology. In Roca (1997), 43-89.
Shibatani, M. 1990: The Languages of Japan. Cambridge: Cambridge University Press.
Siegel, D. 1974: Topics in English morphology. PhD, MIT. Published by Garland, New
York, 1979.
Smolensky, P. 1996: On the comprehension/ production dilemma in child language.
Snider, K. 1988: Towards the representation of tone. In van der Hulst and Smith (1988b),
part 1, 237-67.
Snider, K. 1990: Tonal upstep in Krachi: evidence for a register tier. Language 66, 453-74.
SPE, see Chomsky and Halle (1968).
Spencer, A. 1990: Morphological Theory. Oxford: Blackwell.
Spencer, A. 1996: Phonology: Theory and Description. Oxford: Blackwell.
Sproat, R. 1988: Bracketing paradoxes, cliticization, and other topics. In M. Everaert,
A. Evers, R. Huybregts and M. Trommelen (eds), Morphology and Modularity,
Dordrecht: Foris, 339-60.
Stampe, D. 1969: The acquisition of phonetic representation. Chicago Linguistic
Society 5, 443-54.
Stampe, D. 1972: How I spent my summer vacation. A dissertation on natural
phonology. PhD, University of Chicago. Published by Garland, New York, 1980.
References 681
Steriade, D. 1982: Greek prosodies and the nature of syllabification. PhD, MIT.
Steriade, D. 1987: Redundant values. Chicago Linguistic Society 23, 339-62.
Steriade, D. 1994: Complex onsets as simple segments. In Cole and Kisseberth
(1994), 203-91.
Steriade, D. 1995: Underspecification and markedness. In Goldsmith (1995), 114-74.
Stevens, K. 1972: The quantal nature of speech: evidence from articulatory-acoustic
data. In E.E. David and P.B. Denes (eds), Human Communication: A Unified View,
London: Academic Press, 51-66.
Stevens, K. 1989: On the quantal nature of speech. Journal of Phonetics 17, 3-46.
Stevens, K. and Keyser, S.J. 1989: Primary features and their enhancement in con-
sonants. Language 65, 81-106.
Stevens, K., Keyser, S.J. and Kawasaki, H. 1987: Toward a phonetic and phonolo-
gical theory of redundant features. In J. Perkell and D.H. Klatt (eds), Symposium
on Invariance and Variability, Hillsdale, NJ: Lawrence Erlbaum, 426-49.
Strang, B. 1986 [1970]: A History of English. London: Routledge.
Szpyra, J. 1992: The Phonology—Morphology Interface: Cycles, Levels and Words. London:
Croom Helm.
Tesar, B. 1995a: Computing optimal forms in Optimality Theory. Technical Report
CU-CS-763-95, Feb. 1995, Department of Computer Science, University of Colo-
rado at Boulder. ROA-52.
Tesar, B. 1995b: Computational Optimality Theory. PhD, University of Colorado.
ROA-90.
Tesar, B. and Smolensky, P. 1993: The learnability of Optimality Theory. An algo-
rithm and some basic complexity results. MS, University of Colorado at Boulder.
ROA-2.
Tesar, B. and Smolensky, P. 1998: Learnability in Optimality Theory. Linguistic
Inquiry 29, 229-68. [Longer version in ROA-156.]
Thomas, C.K. 1958 [1947]: The Phonetics of American English. New York: Ronald.
Tranel, B. 1987: The Sounds of French. Cambridge: Cambridge University Press.
Tranel, B. 1995: Current issues in French phonology: liaison and position theories.
In Goldsmith (1995), 798-816.
Trubetzkoy, N.S. 1939: Grundziige der Phonologie. Guttingen: Vandenhoeck and
Ruprecht. Translation: C.A.M. Baltaxe, Principles of Phonology, Berkeley, CA: Uni-
versity of California Press, 1969.
Tsujimura, N. 1996: An Introduction to Japanese Linguistics. Oxford: Blackwell.
Turner, G. 1994: English in Australia. In Burchfield (1994), 277-327.
Vennemann, T. 1972: On the theory of syllabic phonology. Linguistische Berichte 18,
1-18.
Vihman, M. 1978: Consonant harmony: its scope and function in child language. In
Greenberg (1978), 281-334.
Vihman, M. 1996: Phonological Development. The Origins of Language in the Child. Oxford:
Blackwell.
Vogel, I. and Kenesei, I. 1990: Syntax and semantics in phonology. In Inkelas and
Zec (1990), 339-63.
Wells, J. 1982: Accents of English. 3 volumes. Cambridge: Cambridge University
Press.
682 References
Wetzels, L. and Sezer, E. (eds) 1986: Studies in Compensatory Lengthening. Dordrecht:

Foris.
Wiese, R. 1996a: The Phonology of German. Oxford: Clarendon Press.
Wiese, R. 1996b: Phonological vs. morphological rules: on German umlaut and
ablaut. Journal of Linguistics 32, 113-35.
Wolfe, P. 1972: Linguistic Change and the Great Vowel Shift in English. Berkeley, CA:
University of California Press.
Woodrow, H. 1951: Time perception. In S. Stevens (ed.), Handbook of Experimental
Psychology, New York: Wiley, 1234-6.
Yip, M. 1980: The tonal phonology of Chinese. PhD, MIT. Circulated by the Indiana
University Linguistics Club.
Yip, M. 1988: The Obligatory Contour Principle and phonological rules: a loss of iden-
tity. Linguistic Inquiry 19, 65-100.
Yip, M. 1989a: Feature geometry and cooccurrence restrictions. Phonology 6, 349-74.
Yip, M. 1989b: Contour tones. Phonology 6, 149-74.
Yip, M. 1995: Tone in East Asian languages. In Goldsmith (1995), 476-94.
Zec, D. and Inkelas, S. 1990: Prosodically constrained syntax. In Inkelas and Zec (1990),
365-78.
GLOSSARY
The purpose of this glossary is to provide the reader with a quick look-up facil-
ity. As might be expected, it must be interpreted in conjunction with the text. (For
a more general list of phonological and phonetic terms, we recommend R.L. Trask,
A Dictionary of Phonetics and Phonology, London: Routledge, 1996.)
ACCENT: (1) Each of the ways of pronouncing a language, characteristic of a geo-

graphical area, a social group, or even a single individual. (2) In metrical theory,
the projection of the baseline asterisks of heavy rimes onto line 1 in the grid. See
also Pitch Accent, Phrase Accent, Standard Accent.
ACCENTED TONE: See Dominant Tone.
ACOUSTIC PHONETICS: The branch of phonetics (and of physics) that investigates
the patterns of vibration of molecules in the air produced as a result of the articu-
lation. See Articulatory Phonetics, Auditory Phonetics.
ACTIVE ARTICULATOR: The articulator that moves as opposed to the static, or “pas-
sive”, articulator. See Passive Articulator.
AD HOC: An expression applied to an unmotivated analysis, driven only by the need
to get the right result. Similar to brute force.
ADVANCED ARTICULATION: An articulation made slightly more towards the front
of the mouth than the reference point. Cf. Retracted.
ADVANCED TONGUE ROOT (ATR): A distinctive feature ([EATR]) dependent on
[radical] and implemented by drawing the root of the tongue forward, to enlarge
the pharyngeal cavity, often also resulting in a raising of the body of the tongue.
AFFIX: A morpheme that needs to be attached to a base.
AFFIX ORDERING HYPOTHESIS: The hypothesis inherent to Lexical Phonology that
affixes are attached to the base in a specific order, which is fixed for each affix and
has phonological consequences.
AFFRICATE: A complex consonant made up of a stop gesture and africative ges-
ture made in rapid succession, without changing the position of the articulators.
ALGORITHM: A word designating a self-contained procedure, such that each step
automatically leads to the next, ultimately reaching the output.
ALIGN: In OT, a family of constraints which regulate the spatial relation between
two elements, either by requiring that they be sequentially abutting, or by requiring
that the edges of two constituents in different structural levels be superimposed.
ALLOMORPH: A contextual variant of a morpheme.
ALLOMORPHY: The situation in which a morpheme has several allomorphs.
ALLOPHONE: A contextual variant of a phoneme.
684 Glossary
ALLOPHONY: The situation in which a phoneme has several allophones.

ALTERNANT: A form that partakes in an alternation.
ALTERNATION: A phenomenon involving a systematic difference in sound between
two or more allomorphs.
ALVEOLAR: A sound articulated on the (upper) tooth ridge.
ALVEOLAR TRILL: A consonantal sound that involves the vibration of the tip of
the tongue against the upper tooth ridge.
ANCHOR: In OT, a constraint that enforces correspondence between edge elements.
ANTERIOR: A binary distinctive feature which is dependent on [coronal] and dis-
tinguishes an articulation in the alveolar or dental areas ([+anterior]) from an articu-
lation in the postalveolar or palatal areas ([—anterior]).
APERTURE: A class node pertinent to vowels in a model of feature geometry aim-
ing for a uniform place of articulation specification for consonants and vowels.
APICAL: An articulatory gesture made with just the tip of the tongue.
APPENDIX: Said of consonants occurring on the edge of the word which do not fit
into the canonical structure: the assumption is that such consonants are affiliated
to a higher constituent.
APPROXIMANT: A term applied to sounds that are continuant and frictionless.
ARTICULATORS: The organs that make contact to produce linguistic sound.
ARTICULATORY PHONETICS: The branch of phonetics which is concerned with
the production of sound.
ASH: The name conventionally given to the symbol [ze], generally assigned to a raised
version of cardinal vowel no. 4, but identified with cardinal vowel no. 4 itself in
this book.
ASPIRATION: The period between the release of the closure of a consonant and
the start of vocal fold activity for the vowel that follows it. Aspiration can be felt
physically as a puff of air.
ASSIMILATION: A term used in phonology to designate the “contamination” of a
sound by another (usually adjacent) sound, formalized as feature spreading in
autosegmental phonology.
ASSOCIATION CONVENTION: A principle of autosegmental phonology by which
autosegments (particularly tones) associate to their bearers one-to-one from left to
right.
ATR: See Advanced Tongue Root.
AUDITORY PHONETICS: The branch of phonetics that deals with the effects of
acoustic patterns on the human ear and with their subsequent reception and per-
ception in the brain.
AUTOSEGMENT: A unit of phonological structure in autosegmental phonology.
AUTOSEGMENTAL PHONOLOGY: An approach to phonology where each feature
or structural element is granted autonomy of action.
BABBLING: A stage in developmental phonology in which the child combines con-

sonants and vowels into recognizable words, at around six or nine months of age.
Cf. Cooing, Vocal Play.
BACK: A binary distinctive feature which is dependent on [dorsal] and whose posit-
ive value is characterized bya retraction of the body of the tongue.
Glossary 685
BASE: (1) The core, irreducible form of a word, to which affixes are added. (2) In
reduplication, the word which is reproduced (wholly or partly) in the reduplicant.
BBC ENGLISH: The standard form of English traditionally used in the British
media, in particular the BBC, and characterized by an RP-type accent.
BILABIAL: A sound articulated with both lips.
BILABIAL TRILL: A trill made by the vibration of both lips, used more or less uncon-
sciously by English speakers to indicate cold.
BINARISM: A type of formalism which involves the characterization of a feature as
+ or —, corresponding respectively to the presence or the absence of the property
encapsulated in the feature.
BINARY FOOT: A foot with only two elements. Cf. Unbounded Foot.
BLADE: The most mobile and versatile part of the tongue, that sticks out most eas-
ily, located behind the tip. Cf. Body, Root, Tip.
BLEEDING ORDER: Asituation in which a rule removes material that would be
necessary for the application of a subsequently ordered rule.
BLOCKER: In harmony systems, an element which stands in the way of the propa-
gation of the harmonizing feature.
BODY: The section of the tongue behind the blade, more massive and less mobile
than the blade. Cf. Blade, Root, Tip.
BOUNDARY TONE: In an intonational melody, the tones associated to elements on
the edge of the intonational domain. Cf. Phrase Tone, Word Tone.
BRACKET ERASURE CONVENTION: A convention of Lexical Phonology accord-
ing to which domain-internal morphological brackets are erased (= become invis-
ible = become inaccessible) at the end of each block or level.
BREAKING: A label which is sometimes used to indicate the unfolding of a vowel
into a diphthong.
BROAD TRANSCRIPTION: A phonetic transcription which is phonemically oriented.
Cf. Narrow Transcription.
BRUTE FORCE: An expression which suggests forcing an analysis to fit the data.
Similar to ad hoc.
C: A symbol that informally stands for a consonant, that is, a segment defined as
[+consonantal]. Cf. V.
C-PLACE: A class node for consonants in a theory of feature geometry favouring a
uniform set of place features for consonants and vowels. Cf. V-Place.
CANDIDATE: In OT, each of the forms competing for victory in the evaluation by
the ranked constraints.
CARDINAL VOWELS: A set of vowels proposed by Daniel Jones as reference
points for the description of the vowels of the world’s languages.
CASE: The form of a word expressing a specific syntactic or semantic function.
CATEGORY: The lexical class in which a word belongs according to morphological,
syntactic or semantic criteria.
CENTRALIZATION: The articulation of a usually front or back vowel sound in the
centre of the vowel space, or near it.
CENTRING DIPHTHONG: A diphthong the second phase of which is articulated
in the central area of the vowel space, usually in the area of the schwa.
686 Glossary
CLASS: See Level.

CLASS NODE: A node in the tree of the feature geometry which defines a functional
class of features, rather than a distinctive feature as such, therefore an alternative
to the subscript notation.
CLITIC: A small, stressless function word which prosodically cannot stand by itself,
but must lean on a host.
CLOSED SYLLABLE: Said of a syllable with a coda. Cf. Open Syllable.
COARTICULATION: A term by which phoneticians express essentially what phono-
logists refer to as “assimilation”. More precisely, it refers to the fact that articulat-
ory planning is not segment-sized, but may include a whole string of segments, which
are accordingly brought closer in their articulation.
CODA: The consonant that follows the nucleus, and that joins it to make up the rime.
COMBINATORIAL SPECIFICATION: A development of Radical Underspecifica-
tion that builds on the notion of “F-element” (a valued feature or a class node)
and its status with regard to association (associated or unassociated).
COMPENSATORY LENGTHENING: A phenomenon involving the lengthening of
a segment as the result of the deletion of its neighbouring segment.
COMPETENCE: The language system that underlies actual performance, and
which must be assumed to be permanently present in the brain, in some form. Cf.
Performance.
COMPLEMENT: In syntax, the material that follows the head of a phrase in languages
like English, to make up a constituent with the head. Cf. Specifier.
COMPLEMENT RULE: In Radical Underspecification, the redundancy rules which
by convention fill in the values of the distinctive features that define the maxim-
ally underspecified segment.
COMPLEMENTARITY: The property characterizing binary features, such that when
one of the values of the feature is present, the other value must of necessity be absent.
COMPLEMENTARY DISTRIBUTION: The non-overlapping distribution of two
allophones of one phoneme.
COMPOUND: A type of word created by combining other words, such as the
English nouns greenhorn and greenhouse, which contrast with the phrases green horn
and green house, and with the non-compound words greenish and greenness.
COMPOUND STRESS: The stress pattern assigned to compounds, which in English
usually involves greater prominence in the first member of the pair.
CONSONANT: A sound produced with an obstruction to the airflow. Cf. Vowel.
CONSONANTAL: The distinctive feature differentiating consonants characterized
by a drastic constriction somewhere along the centre of the oral passage ([+con-
sonantal]), from sounds which do not have such a constriction ([-consonantal]).
CONSTITUENT: An element or string of elements which fulfil a particular structural
role together, as is the case with the consonant(s) that make up the onset, or the
feet that make up a phonological word.
CONSTRAINT: An injunction in the grammar regulating the shape of a form.
CONSTRAINT FAMILY: In OT, a group of constraints that express one single basic con-
cept; for instance, faithfulness to the underlying form, or alignment of constituents.
CONSTRAINT RANKING: In OT, the hierarchization of the power of constraints,
somewhat reminiscent of the ordering of the rules in rule-based phonology.
Glossary 687
CONSTRAINT VIOLATION: In OT, the disobedience of a constraint by a candidate,

usually signalled by an asterisk (*) in the corresponding box.
CONSTRICTED GLOTTIS: The gesture of bringing the vocal folds together, respons-
ible for glottalization of sounds, and expressed by a positive specification of the
distinctive feature thus named.
CONTENT WORD: See Lexical Word.
CONTEXT: See Environment.
CONTINUANT: A binary distinctive feature expressing the state of the oral channel
during the production of the sound: for [+continuant] sounds the channel remains
unobstructed and allows the air to escape freely.
CONTINUOUS COLUMN CONSTRAINT: A constraint in metrical theory to the effect
that the columns of the metrical grid must be continuous, without lines being
skipped.
CONTRAST-RESTRICTED UNDERSPECIFICATION: A theory of underspecifica-
tion in which only non-contrastive segments (in a particular context) can be left
unspecified (in that context). Cf. Radical Underspecification.
CONTRASTIVE STRESS: The prominence assigned to a syllable or a word to make
it stand in opposition to some other syllable or word, either syntagmatically or
paradigmatically.
CONTRASTIVE UNDERSPECIFICATION: Another name for Contrast-restricted
Underspecification, used by its proponents.
COOING: An early stage in the development of phonology characterized by the pro-
duction of contented vocalizations by the infant. Cf. Babbling, Vocal Play.
CORE SYLLABLE: The most basic and most common syllable, made up of a simple
nucleus preceded by a simple onset.
CORONAL: A unary distinctive feature expressing movement of the blade of the
tongue.
CORRESPONDENCE CONSTRAINTS: In OT, a type of constraint that imposes some
form of identity between two representations.
CORRESPONDENCE THEORY: In OT, the subtheory that countenances corres-
pondence constraints.
COUNTERBLEEDING ORDER: An ordering relation between two rules such that,
if the order were reversed, a bleeding relation would be created.
COUNTERFEEDING ORDER: An ordering relation between two rules such that, if
the order were reversed, a feeding relation would be created.
CV-TIER: Another name for the skeleton, suggestive of its composition as a string
of Cs and Vs in early autosegmental theory.
CYCLE: A mode of rule application in which the rule applies to the smallest con-
stituent of a form first, and then to successively larger constituents, rather than
directly and exclusively to the full form.
CYCLIC DOMAIN: A domain which only allows the application of cyclic rules.
CYCLIC RULE: A rule that applies in accordance with the dictates of the cycle. Cf.
Non-Cyclic Rule.
DEFAULT RULE: A redundancy rule that supplies the missing value of a feature in
a given language, usually thought of as selected by UG.
688 Glossary
“DEFECTIVE” r: A sound produced by drawing the inside of the lower lip, onto the
edge of the upper teeth further back than for [f] or [v], and not quite close enough
to cause friction.
DEGEMINATION: The simplification of a geminate.
DEGENERATE FOOT: A foot that ‘has its growth stunted, and therefore only has
one syllable.
DENTAL: A consonant that involves an articulation on the teeth.
DEP[ENDENCY]: In OT, the (faithfulness) constraint that blocks epenthesis by
requiring every surface segment to have a lexical correspondent, on which it there-
fore “depends”. See also Fill.
DERIVATION: The mapping of a lexical form onto its correspondent surface form
in a series of steps, each defined byarule.
DERIVATIVE: A composite form which has been derived from a base, usually by
affixation.
DIACRITIC: A mark added to a phonetic symbol to implement a slight modification
in the reference of the symbol.
DIALECT: A variant of a language used by a geographical or social section of its
speakers.
DIPHTHONG: A complex vowel of non-steady quality, made up of two phases.
DIPHTHONGIZATION: The act of a vowel becoming a diphthong, a very common
tendency of many vowels in many accents of English.
DISHARMONY: The failure of some segment to harmonize in a harmony system.
DISTINCTIVE FEATURES: A standard set of parameters which together define the
segments of languages, each distinctive feature encapsulating a particular aspect
of language sound. Most distinctive features tend to be considered binary, with
the positive value indicating that the property named by the label is present, and
the negative value indicating that it is not.
DISTRIBUTED: A binary distinctive feature which is dependent on [coronal] and refers
to the length of the blade area involved in the articulation: [+distributed] indicates
a long area, and [-distributed] a short area.
DISTRIBUTION: The pattern of occurrence of sounds in the forms of any given lan-
guage, determined by the environment.
DOMAIN: An array of elements under a common scope.
DOMINANT TONE: The tone of a tonal melody that associates to the stressed syl-
lable. It is commonly known as “accented tone”.
DORSAL: A unary distinctive feature that refers to activity of the body of the tongue.
DOWNSTEP: A term that refers to a resetting of the tone register to a lower level.
Downstep is usually thought of as triggered by a preceding floating L.
DOWNSTEPPED TONE: A tone that is pronounced at a lower pitch level than it
otherwise would, as a result of downstep.
DRAG CHAIN: A type of serial sound change, particularly common in vowel
systems, which involves the shifting of sounds into positions vacated by other sounds
which have previously shifted. It is also known as a “pull chain’.
EDGEMOST: In early OT, a constraint that controlled the positition of stress on the
edge of a domain. It was superseded by ALIGN.
Glossary 689
ELSEWHERE CONDITION: A universal principle governing the interaction of

two overlapping rules, whereby the more restricted rule (with the more detailed
environment) is tried out first; if it fails to apply, the more general rule (with the
fewer qualifications) will apply instead.
ENCLITIC: A clitic that follows its host. Cf. Proclitic.
END STRESS: A process that involves the projection of an edge asterisk, particularly
frequent in languages.
ENHANCEMENT: The reinforcement of acoustic differences between segments by
means of additional articulations, which explains the frequency of some feature
co-occurrences (for instance, [+round] with [+back]).
ENVIRONMENT: The phonological frame in which a given phonological phe-
nomenon takes place, also referred to as “context”. Cf. Focus.
EPENTHESIS: A word used to refer to the insertion of a segment.
EVALUATION: In OT, the examination of the candidates to select one as the most
harmonic with respect to the ranked constraints.
EXTRAMETRICALITY: The exclusion of a designated peripheral element from the
computations in the metrical grid.
EXTRAPROSODICITY: A generalization of extrametricality to formalize the special
licensing powers exhibited by the edges of words, which allow an element on the
word edge to be excluded from some process.
EXTRASYLLABICITY: A term sometimes used to refer to syllabic underparsing,
particularly common on word edges, where it can perhaps be subsumed under
extraprosodicity.
EXTRINSIC ORDERING: The stipulative order of rules, to prevent their free applica-
tion. It is one of the devices associated with the model of phonology propounded
inSPE
FACTORING OUT: The elimination of redundancy in formal expressions by reducing

common elements to a single statement.
FAITHFULNESS: In OT, the requirement that surface forms must replicate their cor-
responding lexical forms.
FAITHFULNESS CONDITION: In metrical theory, the requirement that each grid
constituent have a head and each head have a domain.
FAITHFULNESS CONSTRAINTS: The family of constraints that enforce faithfulness
in OT.
FATAL VIOLATION: In OT, a violation that disqualifies a candidate vis-a-vis some
other candidate. It is conventionally signalled by an exclamation mark (“!”) after
the corresponding asterisk, with the boxes that follow in the tableau usually
shaded to indicate their irrelevance.
FEATURE DEPENDENCY: The fixed connection of some distinctive feature to some
other feature, determined by UG - for instance, [tround] is thought to be a
dependent of [labial].
FEATURE GEOMETRY: The full web of feature dependencies, connected to the skele-
ton by means of the root.
FEATURE VALENCY: The number of values that a distinctive feature allows for,
often one (for monovalent or unary features) or two (for binary features).
690 Glossary
FEEDING ORDER: An ordering relation between two rules such that the first rule
creates the input required by the second.
FILL: In OT, the predecessor of DEP to prevent epenthesis. The label appears to be
a misnomer, meaning as it does an injunction not to fill.
FILTER: Another name for constraint in rule-and-derivation theory (NB not used
in OT).
FLAP: A type of consonant characterized by the brevity of the contact between the
articulators: the active articulator simply taps the passive articulator on its way to
its rest position, as in the typical American t in waiting. See also Tap.
FLOATING: Said of an autosegment which is unassociated.
FLOATING TONE: A tone which is not associated to any melody or skeletal slot.
FOCUS: In a rule, the element that undergoes the change. Cf. Environment.
FOOT: A metrical constituent typically made up of two moras or syllables (binary
foot), in quantity-sensitive systems, or of several syllables (unbounded foot), in
quantity-insensitive systems. The head is placed on either the left or the right edge
of the foot.
FOOTING: The construction of feet in a given domain.
FORMALISM: A system of notation which is as concise as is compatible with full
explicitness, and which usually makes use of symbols.
FREE ELEMENT CONDITION: In metrical theory, the requirement that only met-
rically free elements may undergo metrical construction.
FRICATIVE: A consonant sound that involves friction noise made by the air escap-
ing through a narrow obstacle. Cf. Stop.
FRONTING: The articulation in the front of the mouth of a sound which is typically
pronounced further back.
FUNCTION WORD: A word which is defined by its grammatical function, such as
the in English. Cf. Lexical Word.
GA: See General American.

GEMINATE: A segment, particularly a consonant, which occurs twice in succession.
True geminates are formalized autosegmentally as one single melody associated
to two skeletal slots.
GEMINATION: The act of a single segment becoming a geminate.
GEN: In OT, a label standing for “generator”, the function that randomly generates
surface forms from each lexical representation, using only the most elementary
principles of linguistic logic.
GENERAL AMERICAN (GA): A pronunciation of English common in North
America, characterized by lacking any obvious regional traits. Also referred to as
Network English.
GENITIVE: The case used to indicate possession.
GLIDE: A label often given to non-syllabic vowels in the literature. It is a moot
point whether “glides” do constitute an independent class of sounds in some
languages.
GLOTTAL: A type of sound made by the intervention of the vocal folds.
GLOTTAL FRICATIVE: A sound made by the vocal folds narrowing the glottis, as
in the English h of hot.
Glossary 691
GLOTTAL STOP: A sound made by the vocal folds coming together to close the
glottis, thus causing a momentary interruption to the airstream; the glottal closure
is then released suddenly, exactly as happens with the remainder of the stops.
Cf. Slack Vocal Folds, Spread Glottis, Stiff Vocal Folds.
GLOTTALIZATION: The addition of a glottal gesture to another sound.
GLOTTALLING: The substitution of a glottal sound for a sound of another type, as
in the Cockney pronunciation of cutting as cu[?]ing.
GLOTTIS: The space surrounded by the vocal folds.
GRAMMATICAL WORD: See Function Word.
GREAT VOWEL SHIFT: A far-reaching chain of changes in the vowel system of English
which eventually yielded [ar], [it] and [ei], at the front, and [au], [u:] and [ou], at
the back.
GREEK LETTER VARIABLES: A set of variables available in the standard formal-
ism of generative phonology, such that the value of each variable (a, 8, y, etc.)
can be arbitrarily set as + or —, independently of the value assigned to the other
variables.
H: A shortand for a high tone. Cf. M, L.

HARD PALATE: The front and mid parts of the palate, supported by a bony struc-
ture. Cf. Soft Palate.
HARD UNIVERSALS: Universals which are inviolable, best formalized as principles.
Cf. Soft Universals.
HARMONY: (1) A situation where a designated feature has the same value
throughout a given domain, as happens with [tback] in Turkish. (2) In OT, the
property of a candidate of complying with the (ranked) constraints.
HEAD: In a constituent, the most important element of that constituent, indeed the
core element: without the head the constituent would simply not exist. See also
Faithfulness Condition.
HEAVY SYLLABLE: A syllable whose rime contains either a long nucleus or a coda.
Cf. Light Syllable.
HIATUS: The allocation of two adjacent vowels to two different syllables.
HIGH: (1) A tone characterized by a relatively high pitch. Cf. Mid, Low. (2) A binary
distinctive feature, a dependent of [dorsal], whose positive value is characterized
by a raising of the body of the tongue.
HOMORGANIC: Made at the same place.
HOST: The word that supports aclitic.
HYPER-RHOTICITY: A phenomenon involving the pronunciation of r in forms and
environments where there is no reason to pronounce it, as in the pronunciation
china[1] for china.
HYPOTHESIS: A term used to refer to an idea that explains some body of data.
IAMB: A classical word for a right-headed (binary) foot. Cf. Trochee.

IAMBIC-TROCHAIC LAW: The observation that elements of uneven intensity (and
even duration) tend to pair up as left-headed feet (= trochees), and elements of
uneven duration (and even intensity) as right-headed feet (= iambs).
IDIOLECT: The specific manner of speaking of an individual speaker.
692 Glossary
INFIX: An affix that neither precedes nor follows the stem, in the way that prefixes
or suffixes do, but is inserted in the middle of the stem. Cf. Prefix, Suffix.
INTERDENTAL: A consonant articulated (with the tip of the tongue) between the
teeth.
INTERNATIONAL PHONETIC ASSOCIATION (IPA): An association founded in 1886
to foster the phonetic transcription of languages.
INTERPOLATION: The phenomenon by which a particular phonetic gesture bridges
over a segment or stretch of segments lacking it, as when the pitch of the left tone
changes gradually into the pitch of the right tone over a tonally empty space.
INTONATION: A phenomenon that involves the modulation of pitch over phrasal
domains for functional or attitudinal purposes.
INTONATIONAL PHRASE: A phonological domain traditionally identified with the
domain of association of the intonational melodies to the segmental material.
INTRINSIC ORDERING: The ordering between two rules that obtains when the rules
are allowed to apply freely.
INTRUSIVE r: An expression referring to pronouncing an r between a non-high vowel
and a following vowel in the next word or morpheme even though the r does not
appear in the spelling. Cf. Linking r.
IPA: Acronym of International Phonetic Association.
IPA ALPHABET: The alphabet of phonetic symbols promoted by the IPA.
ITERATIVE: A word used in linguistics to indicate repetitive application within the
same domain. Cf. Cyclic Rule.
ITERATIVENESS: In metrical theory, a parameter that controls the repetition of
footing.
L: A shorthand for a low tone. Cf. H, M. -

LABIAL: A sound that involves lip action in its articulation.
LABIODENTAL: A consonant made with the lower lip and the upper teeth.
LAMINAL: An articulatory gesture involving the full blade of the tongue.
LANDING SITE: In metrical theory, the asterisk that provides the stopping point
for an asterisk that undergoes movement in the grid’s next higher line.
LANGUAGE ACQUISITION DEVICE: A device which we assume the human brain
is innately in possession of, and that allows it to acquire language spontaneously
in childhood.
LARYNGEAL: An adjective meaning ‘of the larynx’.
LARYNX: A cylindrical frame made up of cartilage which sits on the top of the
trachea, and which can visibly protrude as the Adam’s apple, particularly in males.
LATERALITY: A term that refers to the lowering of the sides of the tongue.
LATERALS: A type of consonant characterized by the lowering of one or both sides
of the tongue, but not the front, to allow the air to flow out at the sides.
LEARNABILITY: The question of how humans (specifically, children) manage to work
out and learn the grammars of languages, their phonologies included.
LEFT-HEADED: Applied to a constituent which has its head on the left edge. Cf.
Right-Headed.
LEVEL: In Lexical Phonology, each of the blocks holding together a cluster of
morphological and phonological rules, also referred to as class or stratum.
Glossary 693
LEXICAL CATEGORY: An expression referring to words which have full semantic

meaning, like cup or table, as opposed to words which chiefly perform a gram-
matical function (“grammatical” or “function” words).
LEXICAL ENTRY: A unit in the lexicon containing all the information relevant to
sound, meaning and grammar that needs to be memorized.
LEXICAL ITEM: An item which is contained as an entry in the lexicon.
LEXICAL LEVEL: In Lexical Phonology, the level or levels confined to word and
subword domains. Cf. Postlexical Level.
LEXICAL PHONOLOGY: A model of the interaction of phonology and morphology
based on the clustering of phonological rules and morphological processes into
levels or strata, hence the alternative labels of Level-Ordered Phonology or Stratal
Phonology. The strata are typically classified as lexical and postlexical, and the
lexical strata as cyclic and non-cyclic.
LEXICAL REPRESENTATION: The phonological information contained ina lexical
item, also referred to as underlying representation.
LEXICAL RULES: In Lexical Phonology, the rules which apply at a lexical level. Cf.
Postlexical Rules.
LEXICAL STRATUM: See Lexical Level.
LEXICAL WORD: A word which has full conceptual meaning, listed in the lexicon.
Cf. Function Word.
LEXICALLY SELECT: An expression used to signify that an affix cannot be attached
freely to just any base, but, rather, the information as to what bases the affix attaches
to is an integral part of its lexical entry.
LEXICON: The set of all forms making up the vocabulary of a language.
LIGHT SYLLABLE: A syllable with neither a branching nucleus nor a coda. Cf. Heavy
Syllable.
LICENSING: A relational property which allows an element to surface.
LICENSER: The element (melodic or structural) which licenses some other element.
LINE CONFLATION: In metrical theory, the deletion of line 1 from the grid to dis-
pose of all the feet but the one bearing the main stress.
LINGUISTIC SIGN: For Saussure, the basic element of language, made up of the
arbitrary conjunction of a signifier and a signified.
LINGUISTICS: The analytic study of language.
LINGUO-: A prefix signifying ‘tongue’, from the Latin lingua.
LINKING r: An expression referring to pronouncing an r between a non-high vowel
and a following vowel in the next word, in accents where that r is not pronounced
when the word is said in isolation, even though it appears in the spelling. Cf. Intrusive
fi
LIQUIDS: A type of consonant that has one part of the oral channel blocked during
its production, while another part remains unobstructed and allows the air to escape
freely.
LOOP: A device allowed in some types of Lexical Phonology, whereby a form is fed
back into the previous level.
LOW: (1) A tone characterized by a relatively low pitch. Cf. High, Mid. (2) A binary
distinctive feature, a dependent of [dorsal], whose positive value is characterized
by a lowering of the body of the tongue.
694 Glossary
M: A shorthand for a mid tone. Cf. H, L.

MANNER OF ARTICULATION: The parameter that defines the degree of constric-
tion present in the articulation of a sound.
MAPPING: The transformation of one level of representation into the next one. Cf.
Projection.
MARKED: A label given to sounds or structures which are ‘more complex’, ‘less
expected’, ‘less natural’, and the like. Cf. Unmarked.
MARKEDNESS: Nominalization of “marked”, a concept that inspires the Theory of
Markedness.
MARKEDNESS CONSTRAINTS: In OT, the family of constraints that express the
universal tendencies exhibited in the constitution of segments.
MARKEDNESS STATEMENTS: The filters or constraints that make up the theory of
markedness, prohibiting a particular feature or a combination of features.
MAXIMAL ONSET PARAMETER: See Onset Maximization.
MAX[IMALITY!]: In OT, the constraint responsible for the maximal mapping of under-
lying forms onto the surface. See also Parse.
MEDIAN LINE: The line made by the intersection of the median plane with the anatom-
ical structures.
MEDIAN PLANE: The vertical plane that cuts the human body into two symmetrical
halves, one to the left and one to the right.
MELODY: A generic label used in autosegmental phonology to refer to quality,
that is, phonetic substance, as against quantity, that is, length. A synonym for
“segment”.
METRICAL: The adjective corresponding to “meter”, which suggests measurement,
as is characteristic of rhythm.
METRICAL GRID: A formal object made out of marks (conventionally written as
asterisks) organized in rows intersecting with a series of columns, in the style of
a grid, and used to represent prominence relations.
MID: A tone pronounced at a pitch intermediate between H and L, and transcribed
as M. In a handful of languages, the mid tone is split into a high mid and a mid
proper. Cf. High, Low.
MIDDLE ENGLISH: The English that resulted from the mixture of the purely
Germanic Old English with the Norman French of the conquerors, and which became
Modern English towards the end of the fifteenth century.
MINIMAL ONSET SATISFACTION: The principle that all syllables in a given
domain need to have an onset before codas are allowed.
MINIMAL PAIR: Any pair of words that differ by just one sound in the same position.
MINIMAL SONORITY DISTANCE: A parameter that regulates the configuration of
complex syllable constituents (notably onsets) by imposing a minimal distance in
sonority between the component segments.
MODEL: A word that refers to a scientific idealization of reality.
MODERN. ENGLISH: The form of English that emerged from Middle English
towards the end of the fifteenth century.
MONOPHTHONG: A vowel which is pronounced with the same quality all through.
Cf. Diphthong.
MONOVALENT: Another word for “unary” as referred to features. Derived from
“valency”.
Glossary 695
MORA: Traditionally, a basic unit of classic versification, also used in the phono-
logical analysis of Japanese. Now also an alternative to skeletal slots, convention-
ally represented by the Greek letter “uw” (“mu” [mju] in English).
MORPHEME: A minimal unit of grammatical function.
NARROW TRANSCRIPTION: A phonetic transcription which is phonetically oriented,

rather than phonemically. Cf. Broad Transcription.
NASAL CAVITY: The cavity which is situated above the oral cavity, and opens up
to the outside environment through the nostrils.
NASAL SOUND: A sound characterized by the use of the nasal cavity as an addi-
tional resonating chamber, a situation achieved by lowering the velum, to allow
the air into the nasal cavity. Cf. Oral Sound.
NATURAL CLASS: Each of the families of segments which universally tend to be
affected by the same phonological processes. A natural class is formally defined
by a shared feature or set of features.
NATURAL PHONOLOGY: The phonology that is spontaneously present in chil-
dren in the absence of influence from marked input from the language of the
environment.
NETWORK ENGLISH: See General American.
NEUTRALIZATION: The suspension of a contrast in a certain context.
NO-CROSSING CONSTRAINT: In autosegmental phonology, the requirement that
association lines do not cross.
NON-CYCLIC AFFIXES: Affixes which do not define a domain compatible with cyclic
rules.
NON-CYCLIC RULE: A rule that only applies once, at the level of the largest con-
stituent. Cf. Cyclic Rule.
NON-INTERACTING: Said of two rules that have no influence on each other, and
therefore can apply in either order, without the results being affected.
NON-RHOTIC: An accent which does not have rhoticity.
NUCLEAR TONE: In a tune, the sequence last word tone + phrase tone + bound-
ary tone, that carries the main load of the tune.
NUCLEUS: The most sonorous element of the syllable, typically a vowel.
OBLIGATORY ARGUMENT: In syntax, a word whose meaning is integrated in the

meaning of the verb in a particularly close manner.
OBLIGATORY CONTOUR PRINCIPLE (OCP): The tendency to disallow adjacent sim-
ilar elements from all tiers but the skeleton, within a common domain or constituent.
OBSTRUENT: A type of sound characterized by the presence of a radical obstruc-
tion to the airflow in the oral cavity.
OCCAM’S RAZOR: The methodological principle in science that entities must not
be multiplied beyond necessity.
OCP: See Obligatory Contour Principle.
OLD ENGLISH: The purely Germanic form of English that eventually merged with
Norman French to give way to Middle English.
ONSET: The constituent that starts the syllable, and precedes the rime. Cf. Rime.
ONSET FIRST PRINCIPLE: The principle that onset formation takes precedence over
coda formation.
696 Glossary
ONSET MAXIMIZATION: The principle that onsets must be fully formed, in accord-
ance with the principles of the language, before coda formation is carried out with
the remnant.
OPAQUE VOWEL: In vowel harmony, a vowel prespecified for the harmonizing fea-
ture, which starts off a new harmony domain. Cf. Transparency.
OPEN: A binary distinctive feature expressing degree of tongue height in vowels in
an approach to feature geometry that aims for a uniform place specification of con-
sonants and vowels. The feature [+open] is assumed to be recursive, to reflect the
open-ended nature of vowel height.
OPEN SYLLABLE: A syllable with no coda. Cf. Closed Syllable.
OPTIMALITY THEORY (OT): A development in generative phonology that restricts
phonological grammars to a set of (violable) ranked constraints. The constraints
evaluate a potentially infinite range of candidates generated by a universal func-
tion named GEN, to find the most harmonic one.
ORAL CAVITY: The cavity inside the mouth, where most language sounds are pro-
duced. Cf. Nasal Cavity.
ORAL SOUND: A sound produced with no air coming out through the nose, as a
result of the raising of the soft palate. Cf. Nasal Sound.
ORDERING HYPOTHESIS: See Affix Ordering Hypothesis.
ORDERING PARADOXES: In Lexical Phonology, combinations of affixes which are
predicted to be illegitimate but which do occur.
ORTHOGRAPHY: The spelling system.
OT: See Optimality Theory.
PALATAL: A term referring to sounds articulated on the hard palate.

PALATE: The dome of the oral cavity, commonly referred to as “the roof of the mouth’.
PALATOALVEOLAR: A class of sounds so called because their place of articulation
straddles the palate and the tooth ridge or “alveoli”.
PANINI’S THEOREM: In OT, the Elsewhere Condition as applied to constraint inter-
action, renamed thus to honour its original discoverer.
PARADIGMATIC: Said of a relation between two elements such that they can both
(potentially) occupy the same position in the string. Cf. Syntagmatic.
PARALINGUISTIC: Interacting with language without being part of it.
PARAMETER: Acriterion for classification.
PARAMETER SETTING: The assignment of a specific value to a parameter out of
the range of values it allows.
PARENT NODE: The parent node of a node is the node that immediately dominates
it. Also called “mother node”.
PARSE: In early OT, a constraint enforcing the mapping of underlying elements onto
the surface. See also Max.
PARSING: A term which denotes allotment of elements to constituents.
PASSIVE ARTICULATOR: The inert articulator. Cf. Active Articulator.
PATH: A direct route from a subordinate feature to its superordinate feature along
a feature geometry web.
PERFORMANCE: Language as it occurs in the real world. Cf. Competence.
PERIPHERALITY CONDITION: A condition on extrametricality (or extraprosodic-
ity, in general) to the effect that it can only affect elements on the edge of the domain.
Glossary 697
PHARYNX: The backmost part of the mouth, immediately above the larynx.
PHONEME: A unit of explicit sound contrast.
PHONETIC LEVEL: The last level of a derivation, corresponding to the phonol-
ogical surface representation.
PHONETIC REPRESENTATION: The level of representation that corresponds to the
way an utterance is heard or said.
PHONETIC TRANSCRIPTION: A representation of phonetic sound by means of phon-
etic symbols, such as those of the IPA.
PHONETICS: The discipline concerned with the analysis of actual language sound:
its articulation by a speaker, its acoustic patterns in the air, and the perception of
these patterns by a hearer.
PHONOLOGICAL CONSTITUENT: A constituent in the phonology.
PHONOLOGICAL DOMAIN: The spatial bounds within which certain phonological
rules apply.
PHONOLOGICAL PHRASE (PP): A phonological domain essentially made up of a
lexical head and its specifier.
PHONOLOGICAL UTTERANCE (PU): The largest phonological domain, roughly
corresponding to a speaker’s utterance.
PHONOLOGICAL WORD (PW): A phonological domain and a phonological con-
stituent that usually corresponds to a morphosyntactic word.
PHONOLOGY: The discipline concerned with the study of linguistically significant
sound patterns, that is, with the organization of the sounds of speech.
PHONOTACTIC CONSTRAINTS: Restrictions on the patterning of sounds in the
words of any given language.
PHRASAL STRESS: The patterns of stress assigned to phrases.
PHRASE: Astring of words obeying syntactic requirements.
PHRASE ACCENT: See Phrase Tone.
PHRASE TONE: In an intonational melody, the tone that closely follows the last word
tone, just before the right boundary tone. Cf. Boundary Tone, Word Tone.
PITCH ACCENT: See Word Tone.
PITCH ACCENT LANGUAGE: A language where a fixed tonal melody is associ-
ated with each word.
PLACE OF ARTICULATION: The spot in the vocal tract where the sound is articulated.
PLURAL: A morphosemantic category denoting more than one referent.
POSTLEXICAL LEVEL: In Lexical Phonology, a level or stratum which is not word-
bound, as opposed to a lexical level, which is. Cf. Lexical Level.
POSTLEXICAL RULES: In lexical phonology, the rules which apply at a postlexical
level. Cf. Lexical Rules.
PP: See Phonological Phrase.
PREFIX: An anteposed affix. Cf. Infix, Suffix.
PRESPECIFICATION: Specified in the lexicon, contrary to expectation. Cf. Under-
specification.
PRIMARY ARTICULATION: The main articulation in segments with a complex articu-
lation. It is located along the median plane of the vocal tract. Cf. Secondary
Articulation.
PRIMARY ARTICULATOR: The articulator responsible for the sound’s primary
articulation.
698 Glossary
PRIMARY CARDINAL VOWELS: The eight basic cardinal vowels, unrounded at the
front and rounded at the back, except for [a]. Cf. Secondary Cardinal Vowels.
PRIMARY STRESS: The strongest stress in a particular domain, usually the word.
PRINCIPLE OF LATE BLOCK RULE ASSIGNMENT: In Lexical Phonology, the pos-
tulate that rules are assigned to the latest possible block in the absence of contrary
evidence.
PRINCIPLE OF STRICT CYCLICITY: The principle which restricts the application
of structure-changing cyclic rules to environments derived in that cycle, where
“derived” = resulting from a morphological process or, in some cases, from a phono-
logical change.
PRIVATIVE FEATURE: A monovalent feature, with only the positive value.
PROCLITIC: A clitic that precedes its host. Cf. Enclitic.
PRODUCTIVE: Said of a process which can be applied freely to form new material.
PROGRESSIVE ASSIMILATION: Assimilation to the preceding segment, which
thus “progresses”. Cf. Regressive Assimilation.
PROJECTION: The operation by which an element replicates itself at a higher level
of structure. Cf. Mapping.
PROPOSITION: A word used to refer to the logical content of utterances.
PROSODIC DOMAIN: An expression commonly used for “phonological domain’.
PROSODIC HIERARCHY: The hierarchy made up of the ranked prosodic domains.
PROSODIC PHONOLOGY: An ambiguous expression, used to refer both to the the-
ory of phonological domains and to the theory of phonological constituents.
PROSODIC STRUCTURE: Abstract structure over and above the structure that cor-
responds to the linear arrangement of segments and to the relations between the
features inside the segments.
PU: See Phonological Utterance.
PULL CHAIN: See Drag Chain.
PW: See Phonological Word.
QUALITY: The substance of a segment, particularly a vowel.

QUANTITY: The length of a segment, autosegmentally formalized by its number of
skeletal associations.
QUANTUM VOWELS: The vowels [a], [i] and [u], each of which can be articulated
over a reasonably broad space with minimal effect on perception.
QUANTITY-INSENSITIVE: Said of stress systems that do not require accenting of
heavy syllables.
QUANTITY-SENSITIVE: Said of stress systems that require accenting of heavy
syllables.
R-COLOURED: An expression that refers to an [1]-type quality superimposed on

a vowel, usually achieved by curling up the tip of the tongue, in a gesture of
retroflection.
RADICAL: A unary feature that refers to the root of the tongue and dominates [+ATRI.
RADICAL UNDERSPECIFICATION: A theory of underspecification that aims for
maximal lexical simplicity, which it achieves by leaving all the feature values defin-
ing one segment (in a class) unspecified in every language. Cf. Contrast-restricted
Underspecification.
Glossary 699
RANKING METAPRINCIPLE: In OT, the requirement that a higher-ranked constraint

be complied with in preference to a lower-ranked constraint.
RECEIVED PRONUNCIATION: See RP.
RECURSIVENESS: The application of a process to its own output.
REDUNDANCY: A term applied to information which is predictable.
REDUNDANCY RULE: A structure-building rule that fills in a gap in an underspecified
input.
REDUNDANCY RULE ORDERING CONSTRAINT: In Radical Underspecification,
the requirement that a redundancy rule apply before the first (ordinary) rule that
refers to the output value provided by the redundancy rule.
REDUPLICANT: In reduplication, the affixal copy of the base.
REDUPLICATION: A process of word formation which consists in copying the whole
or part of a word in a prefixal, infixal or suffixal position.
REGRESSIVE ASSIMILATION: Assimilation to the following segment, which thus
“regresses”. Cf. Progressive Assimilation.
RESONATING CHAMBER: An enclosed space where sound gets enhanced at cer-
tain acoustic frequencies.
RETRACTED: Said of a sound when it is pronounced at a place of articulation more
towards the back of the mouth than is usually the case. Cf. Advanced Articulation.
RETROFLECTION: The action of curling back the blade of the tongue.
RETROFLEX: A sound produced with retroflection.
RHOTIC: An accent which exhibits rhoticity.
RHOTICITY: A term referring to the systematic occurrence of the sound represented
by r, whereas in non-rhotic accents r only occurs before vowels.
RHOTICS: A general term to designate a class of sounds that are phonologically
“y-like” in some way: the members of this class do not necessarily have much in
common with each other phonetically.
RHYMING: A device used in versification involving identical rimes.
RHYTHM: A pattern of repetition of certain prominent elements.
RIGHT-HEADED: Applied to a constituent which has its head on the right edge. Cf.
Left-Headed.
RIME: An immediate constituent of the syllable placed after the onset and made up
of the nucleus and the coda. Cf. Onset.
ROLLED r: An alveolar trill.
ROOT: (1) The base of the word, to which affixes can be added. (2) The topmost
superordinate node in the feature geometry, currently conceived of as made up
of the features [+consonantal, tsonorant]. (3) The section of the tongue behind the
body, mainly tucked away in the pharynx. Cf. Blade, Body, Tip.
ROUND: A binary feature dependent on [labial] whose positive value involves a round-
ing of the lips.
RP: A shorthand for “Received Pronunciation”, a term that refers to an accent of English
traditionally associated with the English upper middle classes and their “preparat-
ory” and “public” schools.
RULE: The formal statement of a mapping between an input and an output
representation.
RULE ORDERING: The strategy of applying rules sequentially, rather than simul-
taneously. The order of the rules can be intrinsic or extrinsic.
700 Glossary
SCHWA [fwal: A central mid vowel, represented as an inverted “e” ([a]) in the IPA
alphabet.
SECONDARY ARTICULATION: In segments with a complex articulation, the sub-
sidiary articulation. Cf. Primary Articulation.
SECONDARY CARDINAL VOWELS: The set of eight cardinal vowels produced by
deliberately reversing the normal action of the lips. Cf. Primary Cardinal Vowels.
SECONDARY STRESS: A peak of prominence lower than the one corresponding to
the primary stress.
SEGMENT: A cluster of distinctive feature values associated with one root in the
feature geometry.
SELF-CONJUNCTION: In OT, a special type of constraint consisting of a constraint
conjoined to itself.
SEMANTIC OPACITY: The opposite of semantic transparency.
SEMANTIC TRANSPARENCY: The situation where the meaning of a word or
expression is equal to the sum of the meanings of each of its parts. Cf. Semantic
Opacity.
SEMANTICS: The branch of linguistics that studies the meaning of words and lin-
guistic expressions.
SIBILANT: An informal label for a segment which is [coronal, +strident].
SIBLING NODES: Nodes which are immediately dominated by the same parent node.
Also known as “sister nodes”.
SIGNIFIED: In Saussurian linguistics, the characteristic meaning of each linguistic
sign.
SIGNIFIER: In Saussurian linguistics, the characteristic sound of each linguistic sign.
SKELETAL SLOTS: Each of the elements that make up the skeleton, also referred to
as timing units.
SKELETON: The central tier in the autosegmental model of phonology, from which
all autosegmental planes fan out, with the exception of those that house depend-
ent features.
SLACK VOCAL FOLDS: One of the four features concerning the state of the vocal
folds, specifically referring to their lax condition. The value [+SlackVF] is respons-
ible for both voice and high pitch. Cf. Glottal Stop, Spread Glottis, Stiff Vocal Folds.
SOFT PALATE: The soft area at the back of the palate. Cf. Hard Palate.
SOFT UNIVERSALS: Universals which are violable, best formalized as parameters.
Cf. Hard Universals.
SONORANTS: Sounds where the pressure of the air behind the constriction is the
same as the pressure of the ambient air. Sonorancy is formalized by means of the
binary distinctive feature [+sonorant]. Cf. Obstruent.
SONORITY: The amount of sound present in each segment.
SONORITY HIERARCHY: See Sonority Scale.
SONORITY SCALE: A universal ranking of segments according to the amount of
sonority. each carries.
SONORITY SEQUENCING: The property of the sonority profile of the syllable, involv-
ing rising until the sonority peaks, and then falling.
SPECIFIER: In syntax, the material that precedes the head in languages like English,
and which makes up a phrase with the head and the complement. Cf. Complement.
Glossary 701
SPELLING: The way words are conventionally written down in any one language,
in principle totally irrelevant to phonology.
SPREAD GLOTTIS: One of the four distinctive features concerning the state of the
vocal folds. The positive value [+SG] is responsible for aspiration. Cf. Glottal Stop,
Slack Vocal Folds, Stiff Vocal Folds.
STANDARD ACCENT: Said of an accent with high social prestige, favoured by the
organs of social power, such as the media.
STEM: The basic part of a word, essentially equivalent to the root.
STIFF VOCAL FOLDS: One of the four features concerning the state of the vocal folds,
specifically referring to their tense condition. The value [+StiffVF] is responsible for
both voicelessness and low pitch. Cf. Glottal Stop, Slack Vocal Folds, Spread Glottis.
STOP: A consonantal sound produced by blocking the airflow for a fraction of a sec-
ond, and then abruptly releasing the closure to allow the air to rush out.
STRATAL PHONOLOGY: See Lexical Phonology.
STRATUM: See Level.
STRATUM CONTIGUITY HYPOTHESIS: The principle by which the strata a rule is
assigned to must be contiguous.
STRATUM CONTINUITY HYPOTHESIS: Another name for the Stratum Contiguity
Hypothesis, common in the literature.
STRAY ERASURE: The rule or convention that deletes material which is not prosod-
ically licensed.
STRESS: The prominence with which certain syllables are pronounced, usually
materialized as greater loudness, greater length, greater precision, or association
with the word tone.
STRESS-AND-INTONATION LANGUAGES: Languages in which certain tone com-
binations, or tunes, provide the functional or attitudinal meaning of utterances.
STRESS CLASH: The situation that obtains between two asterisks in any grid line
when they are adjacent and there is no asterisk in between the corresponding pair
of asterisks in the line immediately below.
STRICT CYCLICITY: See Principle of Strict Cyclicity.
STRIDENT: A binary distinctive feature whose positive value is characterized
acoustically as involving high energy at the higher frequencies.
STRONG VERB: A verb which is conjugated by means of vowel alternations in the
stem, rather than through the more common affixation.
STRUCTURAL CONSTRAINTS: In OT, the family of constraints expressing the uni-
versally favoured structures for syllables, feet, and so on.
STRUCTURE-BUILDING RULES: Rules which add structure to the input. Cf.
Structure-Changing Rules.
STRUCTURE-CHANGING RULES: Rules which change some of the structure of the
input. Cf. Structure-Building Rules.
STRUCTURE PRESERVATION: The property of some phonological rules of being
strictly respectful of the basic structural properties of the language, such as the
identity of lexical segments or the core syllable structure.
SUFFIX: A morpheme concatenated to the right of some base. Cf. Prefix, Infix.
SUPRASEGMENTAL: Literally ‘above the segment’ (cf. Latin supra ‘above’), it is
referred to phenomena such as stress, tone and length, not unambiguously.
702 Glossary
SURFACE LEVEL: See Phonetic Level.

SURFACE REPRESENTATION: See Phonetic Representation.
SYLLABLE: A prosodic constituent made up of segments abstractly connected in sonor-
ity clusters.
SYLLABLE WEIGHT: The property of syllables of being light or heavy.
SYNONYMS: Words which have the same meaning but sound different.
SYNTAGMATIC: Said of a linear relation between two elements. Cf. Paradigmatic.
SYNTAX: The principles that govern the way that words are strung together.
TABLEAU: In OT, the visual representation of the evaluation of the candidates by

the ranked constraints, in the form of a grid.
TAP: A type of consonant similar, but supposedly not identical, to a flap.
TAUTO-: A prefix which signifies ‘same’.
THEORY OF MARKEDNESS: The theory made up of the set of markedness state-
ments and of the implications that logically follow from them.
THREE-SYLLABLE WINDOW: The property of word stress of not occurring bey-
ond one of the three peripheral syllables.
TIMING TIER: Another word for the skeleton.
TIP: The frontmost point of the tongue. Cf. Blade, Body, Root.
TONE: A phonological element realized by a certain relative pitch.
TONE LANGUAGES: Languages where lexical entries can be differentiated by tones.
TONE TIER: The tier that houses the tones.
TOOTH RIDGE: The edge of the palate from which the teeth stem out.
TOTAL ASSIMILATION: Assimilation in all features, resulting in two segments becom-
ing identical.
TRACHEA: The windpipe that links the lungs to the mouth.
TRANSDERIVATIONAL CORRESPONDENCE CONSTRAINTS: In OT, the con-
straints that impose correspondences between surface forms.
TRANSITIVITY: A logical property of ordered rules, such that A > B and B > C implies
PMS
TRANSPARENCY: The failure of a segment to interfere with the spread of harmony
or assimilation, as if it were not there. Cf. Opaque Vowel.
TREE: A visual representation of a network of hierarchical relations.
TRILL: A type of consonant articulated at various places in the vocal cavity making
use of a mechanism similar to voice, to take advantage of the Bernoulli effect. See
Voice.
TROCHEE: A classical word used to refer to a left-headed (binary) foot. Cf. Iamb.
TUNE: A melody made up of a string of tones.
UG: See Universal Grammar.

UMLAUT: Fronting of a vowel induced by assimilation to the front quality of a fol-
lowing vowel.
UNARY: Said of features with only one value.
UNBOUNDED FOOT: A foot with any number of elements in the baseline. Cf. Binary
Foot.
UNDERLYING LEVEL: The level that contains underlying representations.
Glossary 703
UNDERLYING REPRESENTATION: An expression usually used interchangeably with

“lexical representation”. Strictly speaking, however, “lexical representation” speci-
fically relates to the lexicon, whereas “underlying representation” refers to the
starting line of a derivation.
UNDERSPECIFICATION: The omission of feature specifications in lexical repres-
entation. See Prespecification.
UNIVERSAL GRAMMAR (UG): The set of principles for language that all humans
are supposed to be endowed with innately.
UNMARKED: Alabel given to sounds or structures which are ‘less complex’, ‘more
expected’, ‘more natural’, and the like. Cf. Marked.
UVULA: The fleshy appendage found dangling down at the end of the soft palate.
UVULAR FRICATIVE: A fricative consonant articulated at the uvula.
UVULAR TRILL: A consonantal sound produced with a vibrating uvula, in a man-
ner reminiscent of gargling.
V: A symbol that informally stands for a vowel. Cf. C.

V-PLACE: A class-node for vowels which depends on the C-Place class-node in a
theory of feature geometry favouring a uniform set of place features for conson-
ants and vowels. Cf. C-Place.
VACUOUS APPLICATION: Said of rules whose application does not change
anything.
VALENCY: The number of values that can be assigned to a particular feature —
features can mainly be monovalent or unary, binary, ternary, or multivalent or
n-ary.
VELAR: Said of sounds made at the soft palate.
VELUM: The anatomical word for the soft palate, derived from the Latin velum ‘veil’.
VOCAL CORDS: Another way of referring to the vocal folds. It is a somewhat mis-
leading expression anatomically.
VOCAL FOLDS: The two bands of muscular tissue which stretch across the larynx.
VOCAL PLAY: A stage in developmental phonology during which the child experi-
ments with a multitude of possibilities of what might be construed as consonant
articulations, at around four or six months. Cf. Babbling, Cooing.
VOCAL TRACT: The tube(s) channelling the airflow responsible for the production
of speech. The term is often restricted in use to the section above the larynx.
VOICE: The technical term used in phonetics to refer to the vibration of the vocal
folds, by the so-called “Bernoulli effect”, as a result of the rapid stream of air flowing
through.
VOICE ONSET TIME: The time span between the release of the closure for a stop
and the start of the vibration of the vocal folds for the following vowel.
VOICED: Said of a sound which is pronounced with simultaneous voicing.
VOICELESS: Said of a sound pronounced without voice.
VOWEL: A sound produced without obstructing the airflow, and formalized as
[-consonantal]. Cf. Consonant.
VOWEL TRIANGLE: The triangle made up of the three most basic, or “quantum”,
vowels ([i], [a], [u]), or, by extension, of the five most common vowels ([il, [el, [al,
[o], [u]).
704 Glossary
VOWEL REDUCTION: A phenomenon whereby the range of vowels available in a

language diminishes in certain contexts, typically involving the absence of stress.
It is not uncommon for reduced vowels to be realized as schwa.
WBP: See Weight by Position.

WEIGHT: See Syllable Weight.
WEIGHT BY POSITION (WBP): A procedure involving the structural projection of
a mora from a coda consonant, as is the case with /n/ in agenda.
WELL-FORMEDNESS CONDITION: A principle of early autosegmental phonology
formulated in connection with African tone languages. It required that at each stage
in the derivation all syllables be associated with at least one tone, and all tones be
associated with at least one syllable.
WORD TONE: In an intonational melody, the tones which are associated with
the stressed syllable of each word. Commonly referred to as “pitch accent”. Cf.
Boundary Tone, Phrase Tone.
<
X: Each of the elements that make up the “timing tier”, meant to be units of
(abstract) phonological timing.
X-TIER: Another word for the skeleton, connotative of the X elements that make it
up. Historically, the X-tier is the successor to the CV-tier.
INDEX OF LANGUAGES
Abbreviations Basque [Isolate; Basque other former British

AA = Afro-Asiatic Country areas in N Spain colonies] 17, 25, 31, 67,
C = Central and SW France] 520, 643 72, 76, 78, 80, 114, 116,
E = East Bavarian see German 126, 131, 132, 205, 294,
IE = Indo-European Bengali [JE, Indo-Iranian; 593, 631, 632, 633, 653,
N = North W Bengal (India), 657
S = South Bangladesh] 369 (stress), affricates 206
W = West 373 basic metrical structure in
Berber [AA; Northern Africa, OT 605-06
particularly Morocco and basic syllables in OT
Aklan [Austronesian, W
Algeria] 657 597-600
Malayo-Polynesian;
child utterances 236-7,
Philippines] 339-40
Catalan [IE; Romance; 238, 240
(stress), 342, 640
Catalonia, Balearic CiV-lengthening 545
Albanian [JE; Albania] 141
Islands, Valencia (Spain), complex codas 262-4
(ex.), 122 (ex.)
Roussillon (France), complex nuclei 249-50
Arabic [AA, Semitic; Middle
Andorra] 57 (ex.), 318 complex onsets 252
East, N. Africa] 17, 247,
(ex.) complex syllables in OT
597
Cayuvava [isolate; Bolivia] 600-04
Cairene (Cairo) 380 (ex.)
247 coronal palatalization
Maltese (Malta) 495 (ex.)
Chinese see Mandarin Chinese 529, 556-8, 652
Urban Hijazi (Hijaz: Saudi
Chuckchi [Paleo-Siberian; NE cyclic and non-cyclic
Arabia) 628 (ex.)
Siberia (Russia)] 141 affixes 449-51
Araucanian [Penutian; Chile,
(ex.) cyclic effects in OT
Argentina] 346 (ex.) Chuvash [Altaic, Turkic; 617-18
Axininca Campa [Arawakan;
Chuvash Rep. (Russia)] cyclic grid construction
Peru] 664
141 (ex.) 413-15
Azerbaijani [Altaic, Turkic;
Cibemba [Niger-Congo, Bantu; cyclic rules 570-6, 654
Azerbaijan and
Zambia] 644 derivations 544-79, 652,
surrounding areas] 141
Ci-Ruri [Niger-Congo, Bantu; 654
(ex.)
Tanzania] 407 (ex.) destressing 421
diphthongization 548,
Bahasa Melayu/Indonesia Dutch [IE, W Germanic; 654
[Austronesian; W Malayo- Netherlands and some effects of strict cyclicity
Polynesian; Malaysia, former colonies, Flanders 499-501
Indonesia, Brunei, (Belgium)] 68, 495 (ex.) exceptions to vowel
Singapore] 539 (ex.), 661 shortening 427-9
(ex.), 662 (ex.) See also English [IE, W Germanic; expletive infixation
Malay British Isles, USA, and 454-5, 645
Index of Languages
extrametricality in OT peripherality 455-7 syllable structure 247,

607-08 phonetic transcription 594-6, 637, 638
final consonants 357-9, 203 (ex.) the antics of /s/ 288-90
640, 646 phonological phrase 472, total assimilation 524-5
final stress reduction 475, 647 | transderivational
422-4 phonological utterance correspondence 614-17
Great Vowel Shift 204, 480-1, 647 truncation 619-20
214-18, 223-6, 549, phonological word 485, umlaut 143, 164-6
636 486-8, 647 (plurals), 207, 636
homophones 202 (ex.) phrasal and compound velar nasal 66
interaction between stress 321-7 velar softening 552,
morphology and place assimilation: in 554-5
phonology 451-5 nasals in stops 92-5, violations of the bimoraic
intonation 383, 387-8, 103, 105-6, 207, 647; foot 445-6
389-93, 642 34-8 violations of the three-
intonational phrase possessives 624-6, 658 syllable window
476-80, 647 postlexical rules 579, 654 446-8
[j]-deletion 559, 652, 654 prevocalic tensing 567, voice assimilation 38-41
[j]-morification 561, 652, 654 vowel alternations
654 prominence 295-8 210-14
labiodentalization 32-3, quantity-sensitivity in OT vowel lengthening 544,
90-1 609-12 566-7
lexical phonology 646 regular plurals 620-4, vowel reduction 314-15,
lexical underspecification 658 412
502-04, 512 retraction failures 310-11 vowel shortening 424-7,
long vowels in final rhythm 312-14, 319 (ex.), 565-9, 645
syllable 359-61 640 word stress 328, 412;
Middle English 199, 214, rules 544, 545, 548, 549, idiosyncratic accent
2157297 553;/958, 5597561562, 373-6; main 333-5;
no complex codas 563, 565, 567, 622 multiple 335-7;
284-7 s-voicing 544, 554-5, 654 nouns 329-31, 658;
non-cyclic accenting secondary stress 418-22, syllable weight
432-6 644 349-57, 609-12; verbs
non-cyclic processes simple syllables 242-3, and adjectives 334-5,
459-65 244 658
non-cyclic refooting small phonological words word-final syllables
418-22, 645 489 436-40
non-cyclic rules 576-9, sonority distance 256-7 American (GA) 75, 78,
654 spelling 8, 13, 21, 29 80, 119, 170, 172, 174,
non-vocalic nuclei (ex.), 30 (ex.), 83 (ex.), 175-6, 177, 179, 180,
265-8 203 (ex.) 181, 182, 183, 185, 188,
Old English 239 spirantization 562, 654 189, 190, 192, 193, 196,
onset maximization stop allophony 315-16, 197, 198, 275, 634
278-84 640 American South 172,
onset vowels 268-77 stress contrasts 301-02 178, 182, 184, 186, 188,
ordering of affixes 468, stress distribution in 192, 196, 197, 198, 200,
470 personal names 201
ordering paradoxes 471, 302-04 Australian 173, 174, 175,
647 stress retraction under 177, 183, 194, 634
past tense 57 (ex.) clash 304-09, 643 Birmingham 188
Index of Languages 707
British (RP) 75, 119, 137, Welsh 186 Japanese [Isolate; Japan]
ZONA 745 75 07 6, Yorkshire 123, 128, 171, 237-8, 239, 363, 395-7
177, 179, 180, 181, 182, 175, 176, 177, 179, 180, (Tokyo; pitch accent), 398,
183, 185, 186, 188, 198, 182, 190 515, 541 (ex.), 637, 642,
192, 196, 197, 199, 200, 643 (Tokyo), 663
633 Finnish [Uralic, Finno-Ugric;
Canadian 173, 196, 197, Finland] 168 (ex.), 247, Klamath [Penutian; Oregon
634 373, 441 (ex.), 582 (ex.) (USA)] 247
Caribbean 173 French [IE, Romance; France
Central US 175, 182 and former colonies, Lardil [Australian; Mornington
E New England (Boston) Wallonia (Belgium), West Is. (Australia)] 583 (ex.)
123) 128 2A 734175, Switzerland, Quebec Latin [IE; Romance; formerly
177, 182, 196, 200 (Canada)] 25, 67, 68, 80, in the Roman Empire]
Estuary 171 81, 124, 125, 126, 127, 130, 68, 379 (ex.), 628 (ex.)
Irish 73, 78, 198 PSS? 8183 41350205; Lithuanian [IE, Baltic;
Northern 138, 171, 247, 284, 292 (ex.), 313-14 Lithuania] 398, 411, 412,
727,193 (rhythm), 328 (stress), 497 643
Southern 171, 175, 178, (ex.), 638
182 Macedonian [IE, S Slavic; F.Y.
Liverpool 188 Gaelic (Scottish) [IE, Celtic; Republic of Macedonia]
London (Cockney) 74, Scotland, particularly W 338-9 (stress), 342, 343,
170, 172, 174, 176, 178, Islands] 13 376 (loan phonology), 640
179, 180, 192, 193, 194, German [IE, W Germanic; Malay 67; see also Bahasa
196, 231 (ex.) Germany, Austria, Melayu/Indonesia
Middle Atlantic 192 Switzerland] 13, 80, 124, Malayalam [S Dravidian;
New York City 172, 176, 125, 1315, 132) 13631357 Kerala (India)] 76
177, 178, 180, 182, 188, 143, 160-03 (umlaut), 164, Mandarin Chinese [Sino-
196, 198, 200, 614, 619 16572077210, 21ieee Tibetan; China and
New Zealand 138, 173, (stress retraction), 636, outposts] 394 (tones),
175, 177, 183, 188, 230 638, 640, 646 395, 398, 642, 643
(ex.), 634 Gooniyandi [Australian; Maori [Polynesian; New
Northern England 135, Australia] 516-17, 521 Zealand] 247
170, 171, 177, 186, 192 Greek [IE, Hellenic; Greece] Maranungku [Australian; Dali
Philadelphia 619 17, 77, 486, 497 (ex.) Is (Australia)] 340
Scottish 73, 74, 76, 78, Guajiro [Arawakan; (stress), 342, 343, 344, 640
IDA 25) 128) 138) 170) Venezuela, Colombia] Margi [AA, Chadic; NE
WA A274 77, 178; 380 (ex.) Nigeria] 406 (ex.)
179, 180, 181, 185, 186, Mende [Niger-Congo; N
188, 190, 192, 193, 197 Hua [Papuan] 247 Sudanic; SC Sierra Leone]
South African 173, 177, Hungarian [Uralic, Finno- 399 (tones), 400, 643
184, 192, 197, 634 Ugric; Hungary] 141 Mid-Waghi [Indo-Pacific,
South Carolina 196, 197 (ex.), 347 (ex.), 636 Trans-New Guinea;
South West England 170, Papua-New Guinea] 73
182 Icelandic [IE, N Germanic; Mohawk [Troquian; E USA]
Southern British 171, 192, Iceland] 629 (ex.) 520
193, 194, 196, 197, 198 Italian [[E, Romance; Italy] Mongolian (Khalkha) [Altaic,
Southern Hemisphere 67, 68, 72, 141 (ex.), 496 E Mongolian; Mongolia]
176, 192 (ex.) 369-71 (stress), 372-3,
Virginia 196, 197 Itonama [Paezan; NE Bolivia] 534-7 (rounding
W Pennsylvania 192, 614 141 (ex.) harmony), 641, 650
708 Index of Languages
Norwegian [IE, N Germanic; Serbo-Croatian [IE, S Slavic; disharmony), 161, 167

Norway] 138, 398, 636 Bosnia-Herzegovina, (ex.), 207, 229 (ex.), 498,
Croatia, Yugoslavia] 398, 502, 504, 534
Papago [Uto-Aztecan; SC 580 (ex.), 643
Arizona (USA)] 141 (ex.) Shona [Niger-Congo, Bantu; Vietnamese [Austro-Asiatic;
Persian [IE, Indo-Iranian; Iran] Zimbabwe] 390-402 Vietnam and
141 (ex.) (tones), 403, 498, 502, 504, neighbouring area] 136
Polish [IE, W Slavic; Poland] 643
338 (stress), 342, 343, 496 Slovak [IE, W Slavic; Slovakia] Warao [Paezan; Venezuela]
(ex.), 640, 645, 646, 648, 441 (ex.), 641 346 (ex.)
657 Spanish [IE, Romance; Spain Welsh [IE, Celtic; Wales] 72,
Portuguese [IE, Romance; and former colonies, 138 (Northern), 292
Portugal, Brazil, and other particularly in the (Southern) (ex.)
former colonies] 68, 82, Americas] 24, 67, 68, 72, Weri [Indo-Pacific, Trans-New
137 77, 78, 80, 181, 247, 259 Guinea; Papua-New
Putonghua see Mandarin (ex.), 276 Guinea] 347 (ex.)
Chinese Sundanese [Austronesian, Western Aranda [Australian,
W Malayo-Polynesian; Pama-Nyungan; N
Russian [IE, E Slavic; Russia W Java (Indonesia)] 141 Territory (Australia)]
and outposts] 77, 505-07 (ex.) 347 (ex.)
(assimilation), 521, 649 Swedish [IE, N Germanic; Winnebago [Siouan; C
Sweden, E Finland] 76, Wisconsin, E Nebraska
Sanskrit (Vedic) [IE, Indo- 138, 398, 643 (USA)] 340-1 (stress),
Iranian; formerly in India] 342, 640
376-8, (stress), 641 Tangale [AA, Chadic; Nigeria]
Scots [IE, W Germanic; 581 (ex.) Yawelmani [Penutian, Yokuts;
Scottish Lowlands] 17 Tiv [Niger-Congo, Bantoid; California (USA)] 260
Selayarese [Austronesian; Nigeria] 395, 402-05 (ex.)
Indonesia] 113 (ex.) (floating tones), 415-17 Yidin” [Australian, Pama-
Selkup [Ostyak-Samoyed; W (cyclic tone association), Nyungan; Queensland
Siberia (Russia)] 371-2 643, 644 (Australia)] 340 (stress),
(stress), 641 Turkish [Altaic, S Turkic; 341, 342, 488-9, 640
Senufo [Niger-Congo, W Turkey, N Cyprus] 136, Yoruba [Niger-Congo, Defoid;
Sudanic; Ivory Coast, 143, 149-54 (vowel SW Nigeria, Togo, Benin]
Burkina Faso, Mali] 247 harmony), 154-60 (vowel 98, 540 (ex.)
INDEX OF NAMES
Abaglo, P. 650 Chafe, W. 651 Golston, C. 656

Abercrombie, D. 640 Chambers, J. 634 Gussenhoven, C. 639, 640, 647
Al-Ageli, H. 657 Chomsky, N. 99, 205, 208,
Algeo, J. 636 249, 329 Hale, K. 640
Allen, M. 645 Christdas, P. 650 Hale, M. 655
Anderson, J. 635 Clark, J. 631 Hall, T.A. 659
Anderson, S. 634, 635, 642, Clements, G.N. 635, 636, 637, Halle, M. 99, 205, 221, 222,
651, 652, 653 638, 641, 643, 650, 657 223, 329, 632, 634, 635, 636,
Aoki, H. 636 Cohn, A. 658 639, 640, 641, 644, 645, 648,
Archangeli, D. 640, 649, 650, Cole, J. 644, 645, 654 649, 650, 653, 654, 655, 658
651, 655 Coleman, J. 642, 644, 653 Hammond, M. 642, 644, 657,
Attridge, D. 639 Cook, V. 638 658
Crothers, J. 633 Haraguchi, S. 642, 643
Bagemihl, B. 642 Cruttenden, A. 642 Hardcastle, W. 632
Bao, Z. 643 Hargus, S. 644, 645
Bauer, L.. 616 Daniloff, R.G. 632 Harris, J. 635, 638
Beckman, J. 640 Denes, P. 631 Harris, R. 634
Beckman, M. 642, 643, 649 Diehl, R. 632 Hawkins, P. 634, 638
Bell, A. 638 Dikken, M. den 635 Hayes, B. 636, 639, 640, 641,
Benua, L. 640 . Dinnsen, D.A. 659 644, 646, 647, 649, 650
Bird, S. 636, 653 Dogil, G. 640 Hayward, R. 637
Blevins, J. 637, 640, 643, 658 Donegan, P. 649 Hendriks, B. 639
Bloomfield, L. 651 Downing, P. 639 Hobbs, J. 642
Bolinger, D. 641, 642 Dresher, E. 640 Hockett, C. 637
Booij, G. 646, 647, 648, 651, Duanmu, S. 643 Hooper, J. 637, 638, 651, 652,
654, 658 Durand, J. 659 657
Borden, G. 631 Hualde, J.1. 637, 643, 645,
Borowsky, T. 638, 645, 646, Ewen, C. 635 650
647, 648, 653, 654 Hulst, H. van der 635, 636,
Brame, M. 644 Fabb, N. 647 639, 643
Branford, W. 634 Fischer-Jorgensen, E. 635 Hume, E. 636, 650
Broe, M. 650 Franks, S. 640 Hyman, L. 639, 640, 641, 643,
Bromberger, S. 653, 655 Fromkin, V. 642 644, 653
Bronstein, A.J. 633 Fry, D.B. 631
Broselow, E. 641 Idsardi, W. 640, 641, 658
Bruce, G. 643 Giegerich, H. 634, 638 Ingria, R. 637
Burzio, L. 653, 658 Gimson, A.C. 633, 634, 638, Inkelas, S. 643, 644, 646, 647,
640 648
Calabrese, A. 649 Gnanadesikan, A. 655 ItOyys %637 4657,, 608
Catford) J= /631,1632 Goldsmith, J. 636, 642, 643 Iverson, G. 640, 647, 651
Index of Names
Jakobson, R. 226, 241, 634, McCawley, J. 642, 643 Roach, P. 633

635, 637, 649 Maddieson, I. 631, 632, 633 Roca, I. 638, 657, 658, 659
Jensen, J. 653, 654 Marantz, A. 647, 648 Ross, J.R. 640
Jones, C. 636 Mascaro, J. 645, 654 Rubach, J. 638, 641, 645, 646,
Jones, D. 118, 169, 329, 633 Mester, A. 637, 657, 658 647, 648, 653, 654, 657, 658
Mohanan, K.P. 636, 638, 641, Russell, B. 634
Kager, R. 639, 640, 641, 644, 645, 646, 648, 650, 653, 654 Russell, K. 658
655 Moulton, W. 634
Kahn, D. 638, 640 Myers, S. 638, 643, 645, 646, Sagey, E. 635, 636, 637, 642,
Kaisse, E. 645, 646, 647, 650 650, 651 650
Kawasaki, H. 637 Sainz, S. 640
Kaye, J. 635, 639, 640, 653 Nespor, M. 634, 638, 640, Saussure, F.de 42, 43, 53,
Kean, M.-L. 645, 649, 654 646, 647 225, 634, 638
Keating, P. 636, 649 Newson, M. 638 Schane, S. 635
Kenesei, I. 646, 647 Newton, I. 631 Schuh, R.G. 642
Kenstowicz, M. 631, 644, 656, Ngunga, A. 643 Scobbie, J.M. 653
657, 659 Nolan, F. 632 Selkirk, E. 638, 639, 640, 644,
Kent, R.D. 631 Norton, R. 655 646, 647, 657
Kenyon, J.S. 633 Noyer, R. 658 Sezer, E. 636, 637
Keyser, S. 637, 638, 650 Shaw, P. 636, 645
Kimura, A. 637 Occam, William of 44 Sherrard, N. 655
Kiparsky, P. 639, 640, 641, Odden, D. 636, 638, 641, 643, Shibatani, M. 637, 643
645, 649, 651, 652, 654 645, 646, 650 Siegel, D. 645
Klausenberger, J. 653 Ohala, J. 632, 637, 642 Smith, N. 636, 643
Kornai, A. 638 Ohman, S. 637 Smolensky, P. 640, 653, 655,
Koutsoudas, A. 640, 651, 652 657, 658
Krapp, G.P. 633 Padgett, J. 636, 650, 657 Snider, K. 643
Kreidler, C.W. 633, 634, 638 Panini 327 Spencer, A. 638, 647
Kubozono, H. 643 Paradis, C. 658 Sproat, R. 647, 648
Passy, P. 8 Stampe, D. 649, 654
Labov, W. 633 Pierrehumbert, J. 632, 642, Steriade, D. 635, 638, 650
Ladd, R. 639, 642 643, 649 Stevens, K. 633, 650
Ladefoged, P. 631, 632, 633, Pike, E. 637 Strang, B. 636
634, 638, 641, 642, 650 Pike, K. 637, 642 Sweet, H. 8
Ladusaw, W. 632 Pinson, E. 631 Szpyra, J. 646, 658
Lass, R. 633, 634, 636 Poser, W. 645
Laver, J. 631, 632, 633, 642 Prince, A. 637, 639, 640, 644, Tesar, B. 655, 658
Leben, W. 642, 643 648, 653, 655, 656, 657, 658 Thomas, C.K. 633
Levin, J. 637, 638 Pulleyblank, D. 635, 636, Tranel, B. 638
Liberman, P. 639, 640, 642 642, 643, 644, 646, 650, 655, Trubetzkoy, N. 635
Lieber, R. 648 658 Tsujimura, N. 637, 643
Lindau, M. 636 Pullum, G. 632, 638, 640 Turkel, W. 658
Lindsey, G. 635 Pyles? le 636 Turner, G. 634
Local, J. 642
Lodge, K. 636 Ramsaran, S. 633 Vago, R. 636
Lombardi, L. 635, 637 Read, C. 631 Vennemann, T. 637
Lovins, J. 637 Reiss, C. 655 Vergnaud, J.-R. 636, 639,
Rice, K. 638, 646, 647 640, 641, 644, 645, 648,
McCarthy, J. 635, 636, 637, 638, Rietveld, A. 647 649
645, 648, 650, 656, 657, 658 Ringen, C. 636 Vihman, M. 636, 637
Index of Names 711
Vogel, I. 634, 638, 640, 646, Wetzels, L. 637 Yallop, C. 631

647 Wheeler, D. 640 Yip, M. 635, 637, 638, 641, 643
White, Eagle E. 640 Youmans, G. 639
Weijer, J. van der 636, 639 Wiese, R. 636, 640
Wells, J. 634, 636, 638 Woodrow, H. 641 Zec, D. 643, 646
INDEX OF SUBJECTS
Note: starred pages contain Northern England 170, 171 allophony, English 281-3
definitions (vowels) (stop), 315-16 (¢-
Northern Ireland 171, 172 weakening), 215-16 (stop
accent, foreign 126-7 Received Pronunciation) (see aspiration), 638, 639, 640
accent, metrical 352-7 (and alsoRP) *119 (consonants)
syllable weight), *354, RP (Received Pronunciation) alternant *53
373-8, 427 (violations), "LTO, 170, 1717-633 alternation *53
641 Scotland 170, 171, 193 (no alveolar *10
accent, pitch see pitch accent intrinsic vowel length) alveolus 10
accent, tonal 643 South Africa 173, 634 An Outline of English Phonetics
accenting 430 & 644 (non- South West England 170, ooo
cyclic), 432 172 apical *106
accents of English (see also Southern Hemisphere 173 approximants 269
variation in English) Southern Ireland 171, 172 articulation *6, 315 (schwa),
169-73, 169 (“standard”) Yorkshire 171 (vowels) 633 (vowel)
American 634 (vowels) acoustics 54, 64 (resonating and phonetic symbols 30
Australia 173, 634, chamber), 631
(ex.), 84 (ex.)
BBC (see also RP) 119 act of speaking 3-5 coarticulation . *54
Canada 173 affix *92, 457 (productive vs
lip rounding *122, 125
Caribbean islands 173 non-productive), 458-9
manner ‘*7 (fricative), 69
Cockney (see also London) (exceptionality), 470 (non-
(continuant), *71
70), Ve occurrence of legitimate
(laterality)
Eastern New England 172, strings), 645 (division into
173 two sets), 647
multiple 633
Estuary 171 (combinations), 648 (as physiology 632
GA (see also General phonological words) place *7 (labiodental), 62 &
American, North America) cyclic and non-cyclic 71 & 72 (primary), 71
S149¥,170 *449-51, 457, 468, (order (alveolar), 76
General American (GA) (see of concatenation), 648 (retroflection)
also GA, North America) (interspersing) r-colouring *187
AOR 72 ordering 465-6, 470-1 sound source coupling 69
Ireland 172 (problematic), 645 articulator *6, *13 (active and
London (see also Cockney) Affix Ordering Hypothesis see Passive), 97 (active,
172 Lexical Phonology, Affix relevance for distinctive
Network English (see also Ordering Hypothesis features), 109 (passive,
General American) 119 allomorph *206 relevance for distinctive
New York City 172, 196 allomorphy 206 (plural and features), 632 (anatomy)
New Zealand 173, 634 past tense in English), 621 articulators 632
North America (see also GA) (plural (regular) in alveolar ridge 10
172, 172 & 173 (US South), English) palate *11, 60-1 (soft)
633 (American) allophone *53 pharynx *11
Index of Subjects 713
tongue *9 (blade), 10, 11, [u:]_ English diphthongal 522 (see also Contrast-
*12 (body), *9 (tip) realizations 194 restricted
uvula *80 [u] 123, 178 (English Underspecification)
“ash” [ze] 128, 176 correspondents) cooing 236
aspiration 281-*2, 316-17 secondary 130-3 (front), corner vowels 122-3
(voiceless stop), 528 *131ff., 134-7 (back) correspondence 616
autosegmental phonology see [@] 132 Correspondence Theory see
phonology, autosegmental [ce}se133 OT, Correspondence
[], »133 Theory
babbling 236 [Al 1385 crying 236
base 616 [pb] 134, 182 (English CV-tier (see also skeleton,
bilabial *20 correspondents) timing tier) 637
bracketing paradoxes *471 {#135 cycle (see also Strict Cyclicity)
(unhappier), 489 (prefixes), [w] 136 412ff., 415-17 (tone
490-2, 647, 648 ly] 131 association), *415, 432-6,
(rebracketing) case *150 617-18 (OT), 644
breaking (see also child phonology, early 236-7, (criticism), 657 (and level
diphthongization) 194, 588 (naturalness), 637, 654 ordering)
215 (simplifications) cyclic 415
classification 90 cyclic and non-cyclic affixes
Canadian raising 195, 634 clitics *393 468 (order of
cardinal vowels 115-26, coarticulation *54 concatenation)
130-7, 169, 633 compensatory lengthening cyclic construction 643
ah (see also primary [a]), 229, 637 (Turkish: ex.) cyclic prefixes 468
116 competence 208 cyclic rules 570-6
corner vowels 119-23, complement 474
173-8 (English complementary distribution dental 7
correspondents) 259 (in Spanish: ex.) derivation 52-*4, 543 634,
ee (see also [i], primary) 117 complex segments 205, 637 651, 653
numbering 125 compounds 301, 321-2 derivative 616
primary 121ff. (stress), 485-6 (Modern diacritic(s) *28, 207
perceptually intermediate Greek), 639 diphthongization (see also
123-5, 178-80 (English consonant + vowel sequence breaking) *171, 189
correspondents) 237 (English tendency), 215
[a] 116, *118, 121, 178 consonants 115 (articulatory diphthong(s) *190, 195-6
(English correspondents) difference with vowels), (raising), 252
[e] 123, 179 (English 275 (tautosyllabic), 530 Canadian raising 195, 634
correspondents) (unified features with centring 199-200 (historical
[9] 124-5, 180 (English vowels), 602 (word-final evolution), 200 (in non-RP
correspondents) licensing) non-rhotic accents), 200
[2] 128, 175-8 (English constituent *414 (monophthongization)
correspondents) constraints (see also OT heterogeneous 195-9
[a] see [z] constraints) 209 homogeneous 190-5
[e] 123, 178 (English (phonotactic), 509, 655 [a1] 195 (historical source),
correspondents) (universal) 196 (English
[iz] English diphthongal Contrast-restricted correspondents)
realizations 193 Underspecification 521 [au] 197 (historical source),
fi], 117, *118, 121, 174 (contrastive lexical 197 (English
(English correspondents) features)-*522, 534, 536-7, correspondents)
[o] 124, 180 (English 649, 650 [er] 191, 192 (English
correspondents) Contrastive Underspecification correspondents)
714 Index of Subjects
[1u] 563, (idiosyncrasy) feature valency 635 [+ front] (pseudo-feature)

[ou] 192 (English binarity 89, 519, 635 149
correspondents) monovalency 537, 635 [+ high] 144, 147
[o1] 197-8 (variation), 199 privative 534-7, 635 [labial] 146
(historical origin) single-value (see also [+low] 144, 147
distinctive features see features monovalency, unary) [+ open] 530, 650
distribution *53 97-9 [radical] *147
dorsum 98 third value ({[OF]) 519 [tround] 145, 146, 534
drag chain (or pull chain) 217 unary 98 (place of (evidence against
drastic constriction 144 articulation co- [—round])
Duke of York Gambit 344, occurrence), 534, 635, 650 [+slack vocal folds] 528
640 features (see also Feature [+spread glottis] 528
Geometry) 85ff., 87-8, *89 [+stiff vocal folds] 528
economy, typographical 129 (complementary labels), 89 [+voice] 528, 529
enclitic *396 (+, — operators), 96 filters (see also constraints)
epenthesis *597, 622 (schwa:-z (omission from rules), 97 509
— -9Z) (and active articulator), foot *332, 365-8 (foot
eth 23 102-5 & 636 (functional structure and universal
extrametricality *323-4, unity), 109 (and passive rhythm), 366 (constraints
359-60 (blocking by long articulator), 110
on quantity-sensitive feet),
vowel), 375, 424 (first grid (tabulation), 144-6 & 636
424 (binarity), 483 (as
element), 426 (English phonological domain), 657
(for vowels), 147 & 513
adjectival suffixes; -ic (peripheral: enhancement)
(implicational
exception), 438, 607-9, 640 degenerate *356, 366
relationships), 162 & 165
extraprosodicity 438 & 241 (floating), 521
moraic 425, 426, 444-6
extrasyllabicity 438 (trimoraic?), 445
(contrastive, lexical
(bimoraicity violation), 645
specification), 525
feature dependencies 105-9, (bimoraicity)
(autosegmental
146-7, 522-3 et passim, unbounded 371-8, *372,
behaviour), 529-31
524 (web), 636, 650 641
(relations between vowel
Feature Geometry *524-31, foot construction 337 & 657
and consonants), 530
533 (implications in (direction), 372 (not in
(place unification for
Feature Geometry and in Mongolian), 612-13
vowels and consonants),
markedness), 545 (in (secondary stress)
587 (physical
rules), 650 (vowel and non-cyclic (see also refooting)
consonant unification) contradictions between 418-22, 429, 644
class nodes 526-8, 533, 650 values), 635, 636 (common foot inventory
Aperture 530, 650 subscript), 643 (of tone), asymmetric 366, 368
C-place 530 650 (enhancement), 650
(performance factors), 641
Laryngeal 527 (organization) (empirical evidence)
Places. 527 [+anterior] *108 symmetric 365
V-place 530 [+ ATR] (Advanced Tongue foot typology 332, 605
tree root (R) 524, *525, 650 Root) *147, 183, 566, 567 formalism 511 (implicational
feature redundancies 531-3, ([— ATR]) statements in logic)
650 (following from [+back] *145, 147
implications) [+ constricted glottis] 528 gamma MW “y” 17
feature transparency 504- [+continuant] *89 genitive *150
(as underspecification), [coronal] *96-7 glide(s) 269, 559
507 (Russian), 649 [+ delayed release] *205-6 glottalization 528
feature underspecification [+distributed] *106 glottis *16, 528 (spread)
504-7 (transparency), 635 [dorsal] *98, 147 Great Vowel Shift 214-20,
*215, 218 & 223 Japanese pronunciation 237 (unhappier), 489 (prefixes),
(naturalness), 218-23 490-2, 647, 648
(synchronic), 220 labial 7 (rebracketing)
(undecomposable basic laminal *106 levels *469, 495 (lexical in
vowels analysis), 221 (SPE language 41-4 (organization), Maltese Arabic: ex.), 614
analysis), 223ff. (in English 49 (acquisition device), (suffixes, levels 1 and 2),
plurals), 636, 645 301-2, (infiniteness), 533 617 (level 2 processes), 657
(alternations) (universals) (level ordering and the
laryngeal 526 cycle)
harmony (see also vowel larynx *3-4, 14,15 loop 491, 648
harmony) *149, 636 learnability 49 (language morphological constituency
acquisition device), 325, and phonological domain
(consonants)
head *249 341, 352, 658 491-3, 647
high tone (H) *386 length 207 (diacritic; principles
segmental), 249 Bracket Erasure
homophones 202 (ex.)
(contrastive in vowels),
host (to clitic) *396 Convention *459, 646
260 (alternation in Late Block Rule
hyper-rhoticity *171
Yawelmani vowels: ex.), Assignment 576, 654
hypothesis *33, 36, 154
564-9 (alternations in Stratum Contiguity
vowels)
Hypothesis 578, 654
iamb 368 in vowels *171
rule blocks 578 (multiple
idiolect *119 (distinctive), 175
membership), 578
infix(es) *454, 454-5 (-bloody-) (variation), 193 (intrinsic),
(postlexical)
interpolation 391 210 (length-based
rules
(intonational), 649 alternation), 260
lexical vs postlexical 483
(phonetic) (alternation in Yawelmani:
(properties), 654
intonation 383-4 (basic data), ex.), 564-9 (alternations),
postlexical 482-3, 579
384-5 (mechanics), *385, 566 (lengthening), 641
(properties), 647
385-6 (primitives), 391 (word-final lengthening)
strata see levels
(interpolation), 392-3 lengthening, vowel 229
lexical selection *457
(sentence), 398 (in tone (compensatory in Turkish:
languages), 398 (Japanese), ex.), 442 (Slovak: ex.), 566
lexical word, dual identity of
476 & 493 (association 487
(English), 637
domain), 641, 642 (English (compensatory), 641 lexicon *45
and Japanese), 642 (pitch), licensing 364, 387, 438, 445
(English word-final)
642 (typology) lexical entry *45, 212-14 (word-final consonant
sequence), 456 (by
and stress 389-90, 642 (selection), 504 (full
autosegmental 387-93, 642 specification), 649 extraprosodicity), 601 &
intonational tones 391 (simplification) 648 (by PW node), 602 (by
boundary (T”) 391, 392 lexical item see lexical entry word edge)
nuclear *393 Lexical Phonology *469-70, by syllable 243, 250-1, 602,
phrase accent (see also word 470 (non-occurrence of 646
tone) 391 legitimate affix strings, lingua 10
phrase tone (T~) 391, 392 645-6, 646 (English linguistic sign *42, 43
pitch accent (see also word phonology; evaluation), (arbitrariness), 44 (official)
tone) 391 648 (integration with linguistics *42, 634
pre-nuclear 393 prosodic phonology) (Saussurian)
word tone (T*) 391, 392 Affix Ordering Hypothesis linguo- 10
intrusiver *171 492-3, 645, 648 linking r *171
intuitions 281 (abandoned) loans, English: adaptation to
IPA alphabet 8, 632 bracketing paradoxes *471 Japanese 637
Index of Subjects
loop 491, 648 Macedonian (literary) nasal cavity 64

low tone (L) *386 338-9 natural class 96, 112 (ex.), 635
Maranungku 340 natural processes 99
marked *226, 508 Polish 338-9 naturalness 90-1 (formal
markedness 226, 507-11 (and Winnebago 340-1 economy), 588 (coronal
underspecification), *508, Yidiny 340 palatalization), 585-8
533 (based on physical metrical principles (phonological inventory),
necessity), 540 (ex.), 585-8, Continuous Column 588 (child phonology)
587 (economy of lexical Constraint *324, 428, 654 notation 8* (square brackets),
features), 588 (and Faithfulness Condition 357, *26 & 191 (tie bar), 51
inventory size), 589 (least *358, 369, 386, 654 (horizontal line in rules),
marked vowel system) Free Element Condition 51 (slash), 53 (slashes), 54
constraints 589, 654, 656 *356, 421, 640 (NA), 89 (+, —), 127
implications 533, 649 Iambic-Trochaic Law 368, (diacritics), 129
statements 509 (vowels), 641 (typographical economy),
510 & 585 (consonants and Peripherality Condition 175 (length diacritic), 207
vowels), 512 (universal), 324, 438, 456-7, 455-7 (X), 221-2 (Greek letter
586 (prohibited segments), (scope), 640 variable), 222 (braces), 239
649 metrical structure 344 (line 1 (C, V), 363 (mu, u), 438 &
meaning and stress 300 construction in 552 (angled brackets), 523
melody *209, 425 Maranungku), 345 (foot (autosegmental association
(preservation), 226 (tier) lines), 638 (X-bar in
and word layers), 352
metrical accent and syllable syntax)
(uniformity), 605-6 (basic)
weight 352-7 CV see syllable, core
mid tone (M) 390
metrical constituents 343 R see tree root
minimal pair *53
(decapitated) subscript 104, 526, 527 (vs.
model (see also phonology,
metrical constraints (see also class nodes), 533, 636, 650
models) *49
OT, constraints) 657 tableaux see OT, tableaux
monophthongization of.
metrical foot 331—*2, 640
metrical grid 299-300, 299 centring diphthongs 200
object of sympathy *658
(stress line), 303 (number
moraic syllable structure 644
obligatory argument 478
of lines), 376-84, 412-15 moras 361-5 (and syllable
Obligatory Contour Principle
(staged construction), 422 weight), *363, 563 (non-
see OCP
(asterisk retraction), 640, moraic /i/), 640, 641 (vs.
OCP (Obligatory Contour
640 (asterisk insertion), skeletal slots)
Principle) *273, 283, 284,
640 (line conflation) morpheme *92, 403 (tonal)
289, 401-2, 459 [grey box],
metrical parameters 338-42 morphological constituency
623 (OCP constraint), 638
extrametricality *323-4, and phonological domains
odd one out 83 (ex.)
359-60 (blocking by long 647
opaque surface forms 658
vowel), 375, 424 (first grid morphology *150, 637
opaque vowel 158
element), 426 (English (template-based), 658 (OT) oppositions, binary and
adjectival suffixes; —ic interaction with phonology multiple 635
exception), 438, 607-9, 640 451-5, 465-7 (model), 495 Optimality Theory see OT
iterativeness *343, 640 (Dutch: ex.), 643ff., 645 OT 584-5, 589ff., 590 (faithful
line conflation *343 (interleaving with forms), 594 & 596 & 655
quantity-sensitivity SOOTe phonology), 646 (GEN), 594 & *595 & 656
609-12 (relationship with (evaluation), 596 (basic
metrical parameters: language phonology) machinery), 617-18 (cyclic
settings mother node see parent node effects), 653, 654-61, 654
Aklan 339-40 mouth 116 (faithfulness), 655 (level
English 338 (nouns) mu (u) 363 mapping), 656 (EVAL),
656 (Emergence of the *STRUCTURE *610 (multiple settings in a

Unmarked), 656 (Fallacy *TENSE, ow *615, 619-20 single language), 533
of Perfection), 658 V/u = *605 for vowels 123 (backness-
(derivational) VorcE HARMONY *623 frontness and height), 125
OT constraints 590 (conflict), WEP [Weight-by-Position] (roundness), 144 & 530
604 (self-conjunction), 609 *610 (height)
(English noun stress), 613 WTS [Weight-to-Stress] phonetic 31 (differentiation
(parallelism), 655ff., 656 *610 of obstruents), 86, 87
‘(obedience to higher- OT constraint families 592 (advantages and
ranked) alignment 605, 606, 620, disadvantages)
*[e] 619-20 621, 657 parent node *257
*[eC] *615 correspondence 613-16, parsing 278
ALIGN LEFT-RIGHT p;uRaLstem *614, 657, 661 (identity) performance 202, 368 (in foot
*621 faithfulness 592, 654, 661 inventory)
ALIGN LEFT-RIGHTpogsgssive- identity 656, 661 personal names 302-4
stem “625 (correspondence) (stress)
ALIGN-LEFT *612 markedness 592 phoneme *53, 635
ALIGN-RIGHT 606 sonority 593-4 phonetic symbols 8, 30 (and
ALIGN-RIGHTHEADFT *612 structural 592-3 articulation: ex.), 78
ANCHOR *657 syllable 656 (liberal use), 84 (and
ANCHOR-LEFT;, ~*623 OT constraint ranking 591, articulation: ex.), 90-1
BD-IDENTeszpZ<g 618 599 (mutually unranked), (disadvantages), 129 (lax
BD-IDENTygnsz 0*616 608 & 611 (reversal), 616 use), 137 (front round
BR-IDENT 661 (strata differences), 655, vowels), 207 (diacritics),
*COMPLEX 592, 603 658 (reranking) 632
*COMPLEX©OP4 *603 OT constraint violation 598 “ash” [e] (and [a]) 128, 176
*COMPLEX&COMPLEX *604 (number), 599 (fatal), 656 phonetic transcription 7-9, 8
CONTIGUITY *657 OT morphology 620-4 (systems), *129 (narrow
DEP[ENDENCY] *597, 656 (English plurals), 624-5 vs. broad), 141-2 (phonetic
Frent*597 (English possessives), 658 symbols and vowel
F[oo]r Brn[Arity] *605 OT principles descriptions: ex.), 202-3
F[oo]T TYPE,,yn *605 Panini’s Theorem (OT) 603, (ex.), 203 (sound-to-
F[Oo]T TYPE;RocpHer) *605 612, 657 spelling: 632 IPA alphabet
I[Nput]O[utTruT]- Ranking Metaprinciple 8, ex.), 632
IDENT[ITY]reatporE] “091 591, 598 phonetics *54
IO-IDENTygnsp 615 OT rules 655 (in GEN), 658 acoustic *54, *631
*1 610 (outside GEN) and phonology 54-5, 531,
*M/V 593 OT subtheories 632
MARKEDNESS 591 Correspondence Theory articulatory *31, *54, 631-2
Max[IMALity] *598, 656 *616, 624-5 (English auditory (or perceptual)
SNY/ GIP 93 Possessives), 653, 657, 658,
*54, *631
No-Copa 592 661 phonological constituents vs.
No-Larse 641 Sympathy Theory *658 phonological domains
Non-Fin[ALiTy] *607 OT tableaux 595-*6, 596
*484, 648
Non-IniTIAL *613 (dotted line), 597 (solid
phonological domains 481-2
Nucieus 592 (properties), 483 (word-
line), 598 & 599 (box
OCP on SIBILANTS *623 internal inconsistencies),
shading), 601, 656
ONSET 592, 614 483-6 & *484 (sub-
PARSE *598 phrasal), *484 & 648 (vs.
SLH *601 paralinguistic *386 phonological constituents),
Son[oriTy] *601 parameter(s) 65, 323 643, 646-8, 647 (and
morphological Correspondence Theory) plane 227

constituency), 647 (direct- *616 plural *150
mapping from syntax) Declarative Phonology prefix(es) *92, 468 (cyclic),
Intonational Phrase 647 644, 653 524 (in-)
(domain of intonational Dependency Phonology prespecified vowel 158
association) 635 principles
Phonological Utterance Derivational Theory Association Convention
480-1, 481 (formation and 651-4 *400, 415, 416, 417
merger), 647 Government Phonology Bracket Erasure Convention
Phonological Word 485, 635 *459, 646
487-8 (affiliation of Natural Phonology ED09, Continuous Column
extrasyllabic consonants), 649 Constraint *310, 414, 640
648, 486-8 (as segmental Optimality Theory (see also Elsewhere Condition
licenser), 487 (dual status), OT) 584-5, 653, 654-61, 326-7, 582-3, 582
488-91 (small), 490 658 (derivational) (Western Finnish: ex.), 583
(implications of PW Particle Phonology 635 (Lardil: ex.), 603, 640, 647,
status), 648 (as segmental “rule-and-derivation” 651, 657
licenser), 648 (dual status), 584-8 Faithfulness Condition 355,
648 (affixal) phonology, autosegmental 372, 640
clitic group 647, foot, 99-*102 et passim, 107 Free Element Condition
syllable, mora 483 (multiplanar *356, 421, 640
(criteria), 479 (as non- representation), 207 Iambic-Trochaic Law 368,
intonational domain), 480 (formalism), 226 (melody 641
(extension) tier), 227 (plane), 619 Late Block Rule Assignment
Phonological Phrase 472-5, (word formation), 636, 642 576, 654
473 (vs. syntactic phrase), (intonation), 661 Minimal Onset Satisfaction
*474, *475 (formation), 647 (reduplication) *279, 283-4, 614
phonologies, differences association principles No Crossing of Lines
between 655 399-401, 400 & 643 157-60, *159, *400, 417,
phonology *3, 87, *55, 226-8 (original), 642 (basic) 506, 642
(multidimensional), 253 Association Convention Occam’s razor *44, 55, 87,
(goal), 382-405 (tonal), 634 *400, 415, 416, 417 96, 344, 401, 402
(history), 658 (English No Crossing of Lines (violation), 634
truncates) 157-60, *159, *400, 417, OCP *273, 283, 284, 289,
and phonetics 54-5, 531 506, 642 401-2, 459 [grey box], 623
(interface), 632 Well-Formedness (OCP constraint), 638
(relationship) Condition *400, 401 Onset First 284
interaction with morphology phonotactic constraints 209 Onset Maximization
(see also Lexical phonotactics 205 (word- 279-84, *283
Phonology) 451-5, 465-7 initial clusters), 206 (word- Panini’s Theorem (OT) 603,
(model), 643ff., 645 initial sC- clusters), 612, 657
(interleaving with 249-57, 262-90, 275 ({i] Peripherality Condition
morphology), 646 plus vowel other than [u]), 324, 438, 456-7, 455-7
(relationship with 292 (S. Welsh rimes: ex.) (scope), 640
morphology), 648 phrase, syntactic vs. Ranking Metaprinciple (OT)
(integration of prosodic . phonological 473 591, 598
phonology with Lexical pitch 642 (in intonation) Redundancy Rule Ordering
Phonology) pitch accent (see also tone) Constraint *518-19, 649,
models *394, 395-8 (Tokyo 651
Correspondence Theory Japanese), 398 (European Sonority Sequencing 437,
(see also OT, languages), 641, 642, 643 438, 439, 460, 464, 638
Stratum Contiguity representation (see also overapplication *651

Hypothesis 578, 654 derivation) 45-8 (basic persistent application 651
Stray Erasure *459, 428-9 and derived), 107 underapplication 553 (and
Strict Cyclicity 427-9, *428, (multiplanar), 511-12 overapplication of Velar
439, 499-501, 501, 502 (economy), 514 Softening), *661
(contravention), 503, 571, (redundancy-free) rule ordering 326 (conflict),
572, 577, 645, 648, 654 representation, levels of 651 518 (language-specific),
Strict Layer Hypothesis (mapping) 552-6, 647, 651
479, *482, 483, 488 lexical or underlying *46, (transitivity), 647
(violation), 601, 647, 657 *53, 93 (selection of (universal), 651 (universal
Structure Preservation *470 consonant), 651 (and principles)
Well-Formedness Condition surface) English 556, 560, 570, 576
*400, 401, 403, 404, 405 phonetic or surface *53, (cyclic)
(counterevidence) 634, 651 (and lexical) rule ordering principles
projection *249 resonating chamber 64 Elsewhere Condition 327,
prominence 295-6 (syllable), rhoticity *171, 172 651
296-8 (word) rhyming 244, 271 Redundancy Rule Ordering
proposition *383 rhythm 312-14 (English vs. Constraint *518-19, 651
prosodic 482 French), 313 (binary), 319 rule ordering types
prosodic domains *481, 647 (English: ex.), 319 (poetic: bleeding 550, 652
(see also phonological ex.), 365-8 (universal and
complement *514-15, 517,
domains) foot structure), 639 (stress 649
prosodic licensing (see also and verse), 641
counterbleeding 550, 553,
licensing) 445 (word- ROA (see Rutger’s Optimality
652, 653
final consonant sequence) Archive)
counterfeeding 553, 555,
Prosodic Phonology (see also root *150
647, 652
phonological domains) rule(s) *48, 103ff.
disjunctive 651
*481, 648 (integration with (autosegmental
extrinsic 326, 647, 651, 652
Lexical Phonology) formalization), 105
feeding 223, 546, 559, 560,
prosodic structure 294, 656 (vacuous effect), 411ff.
568, 651, 652
prosody 294 (modes of application),
free 651, 652
pull chain (or drag chain) 217 444ff. (domains), 470
intrinsic 647, 651
(properties, 545 (feature
geometry formalization), non-interacting 560, 653
quantum vowels 129-*30
570-6 (cyclic), 634 (in
simultaneous application
r 74-5 ([1]), 77 (“defective”; classical generative 218
rolled: see also speech phonology), 643 (domains rule orderings, language-
sounds, alveolar trill) and grammatical specific
r colouring *187 constituents), 651 (level e-Tensing 617
rinsertion 480 mapping), 654 (segmental transitive s-Voicing > Vowel
r, intrusive and linking *171 in English Shift 554, 556
Radical Underspecification exceptions 213, 501, 504, counterbleeding
511-16, *512, 516-21 545, 554 Velar Softening > Vowel
(problems), 537, 649, formalism 49-52, *51 Shift 553
650-1 (focus and environment) Vowel Shift >
Redundancy Rule Ordering rule interaction 325 (End Diphthongization 549-50
Constraint *518-19, 649, Stress [right] and counterfeeding
651 Extrametricality [right], s-Voicing > Velar
recursive *474 543 Softening 554-5
refooting see foot construction, rule application effects Velar Softening > Vowel
non-cyclic normal *661 Shift 553
Vowel Shortening > CiV- Axininca Campa 664 Labial Assimilation in

Lengthening 566 (ex.) Nasals 95
English 564-5, 578 (non- Aspiration 579 Labiodentalization 54
cyclic) assibilation in Finnish 441 liaison in French 497 (ex.)
feeding (ex.) Meeussen’s Law 401-*2,
Palatalization > [j]- assimilation 32-41, *34, 403, 643
Deletion 559 35-8 (place in stops, total), Nasal Simplification *464
Velar Softening > *37 (basic alternative; nasal spread in Bahasa
Palatalization 560 progressive vs. Melayu-Indonesia 539
Prevocalic Tensing > regressive), 38—41 (voice), (ex.)
Vowel Shift 568 92-5 (place in nasals), 479 Palatalization 557-60, *558,
CiV-Lengthening > s- (optional nasal place), 518 562, 564, 577
Voicing 646 & 649 (voice in Russian), past tense in English 55
non-interactive 622 (-z — -s), 524 (total), (ex.)
s-Voicing, Vowel Shift 529, 540 (Yoruba vowels: Place Assimilation in Nasals
642 ex.), 634 (English), 647 104, 264, 647
Vowel Shift, (nasal place) plural in English 164, 223
Diphthongization 560 CiV-Lengthening *545, 565, (Old), 224 (-i drop in Old
Palatalization > [j]- 573, 646 English), 620, 658
Morification > [j]-Deletion coronal palatalization 558 (interaction with
561 Coronalization of Nasals possessive ‘s)
Serbo-Croatian 580 (ex.)
105 Plural Devoicing 622
deletion *462 (word-final Plural Epenthesis 622
Spirantization > [iu]-
Formation > Palatalization
g), 463 (word-final b), 464 possessive in English 624
(word-final n), 464-5 (pre- (plural), 658
564
word-final nasal g) Postnasal Voiced Stop
Tangale 581 (ex.)
deverbal nouns in Icelandic Deletion *463-4
rule types
629 (ex.) Prenasal G Drop *465
cyclic 654
Diphthongization 549-50, Prevocalic Tensing 567, 577
default 507, 513 (for
568, 569, 577, 648 r-insertion 480
vowels), 518 (no language-
dissimilation of voice in Raddoppiamento Sintattico
specific ordering)
Japanese 541 (ex.) in Italian 496 (ex.)
lexical 654
Distributed Assimilation in reduplication 113
non-cyclic 576-8
Nasals 108 (Selayarese: ex.), 661 (and
post-lexical 482-3, 579
Dorsal Assimilation in consonant coalescence in
(properties), 654 Nasals 103 Bahasa Melayu/ Indonesia:
redundancy 508, *587
final r in Bahasa Melayu/ ex.)
structure-building 503, 508
Indonesia 662 (ex.) Rendaku (Japanese) 663
via 657 Final Syllable Stress (ex.)
rules and processes (see also Retraction 465 s-Voicing 499
stress rules) 99 (natural), Flapping 480,579 (prerequisites), *500, 501
459-65 (non-cyclic) front vowel unrounding (cyclicity), 497 (Greek: ex.),
z- Tensing 617 224 543-4, 545-6, 571, 572
[ee] tensing to [AE] 614 glottalization 528 (cyclicity problem), 648
allophonic in English Glottalization 579 (morpheme-internal
281-3 (stop), 315-17 (E- [1]-Deletion 578 application)
weakening), 638, 639, 640 [iu]-Formation 564, 574 schwa epenthesis (-z — -9z)
(consonants) [j]-Deletion *559, 561 622
Anterior Assimilation in (failure to apply) shortening 646 (failure
Nasals 109 [j]-Morification *561 before some suffixes)
anti-hiatus epenthesis in I-Velarization 579 Sonorant Devoicing 579
Sonorant Voicing Default signified *42 (resonants: see also

508 (final formulation), signifier *42, 44 (official sonorants), 647 (vowels
511 selection) and consonants of British
Spirantization 563, 562, sister node see sibling node English)
564, 574-5, 576 skeleton (see also timing tier) glottal 14-16 (fricative [h]),
Stressed Tense Vowel *228, 636 17-18 (fricative (voiced)
Lengthening *568, 577 skeletal slots 228, 641 (and [fi]), *21 & 315 (stop [?])
t-weakening 315-16 moras) laryngeal fricative see
tensing of word-final non- smoothing up of centring glottal (fricative)
low vowels 569 diphthongs 200-1 speech sounds: obstruents 6ff.
Velar Softening 552, *553, sonority *241-2, 259 (production), *23, 31
574 (non-cyclicity), 577 (differences: ex.) (phonetic differentiation;
Voice Assimilation in distance 256-*7, 264 (in parameters), 55 (Catalan:
Russian *506, 519, 649 codas), 274 (in onsets) ex.), *59 (differences), 268,
Voiced Velar Nasalization in hierarchy (see also sonority 633 (stops)
Japanese 663 (ex.) scale) 253-6, 638 affricates *25-6, 204,
voiceless stop aspiration in syllable 241-2, 248 205-6, 209, 227, 633, 637
316-17 (peak; profile), 265 alveolar fricative 10
vowel fronting in Old (codas), 266 (profile), 268 (voiceless [s]), 17 (voiced
English 224 (minimum in nuclei), 637
[z])
vowel lengthening 442 (alternation) alveolar fricative lateral [+]
(Slovak: ex.), 566 (English) scale 255 (1st version), 274 72
Vowel Lengthening 544 (2nd version), 288 alveolar stop 20 (voiceless
Vowel Shift 548, 549-50, (potential) [t]), 21 (voiced [d])
576, 577 sequencing 255, 266 bilabial stop 19 (voiceless
Vowel Shortening 424-7, sound and spelling in English [p]), 20 (voiced [b])
*425, 427 (exceptions), 436, ZAP
dental fricative 23 (voiced)
439, 565, 573, 645 Sound Pattern of English (see also
[d]d voiceless [9])
Word-Final Vowel SPE) 98, 100, 205, 220
dental stop 24 (voiced [d]
Lengthening 361 (Great Vowel Shift
and voiceless [t])
Rutger’s Optimality Archive analysis), 329
“dropped t” 21
(ROA) 659 sound to spelling (see also
fricatives 6-7, 9-18, 10
spelling to sound) 29
(hissing), 23-5
schwa [a] *138 (French), 187 (ex.), 83 (ex.)
interdental fricative 28
(r-coloured), 315 source coupling 69
(voiced [9] and voiceless
SPE (see also Sound Pattern of
(reduction to), 328
(French), 622 (epenthesis -z English) 100, 218, 473, (81)
480, 634, 635, 639, 640, 641, labiodental fricative 6
— -2Z)
647, 651, 653 (voiceless [f]), 17 (voiced
Second World War 173, 186
segmental inventory 586 specifier 474 [v])
speech distortion by children labiodental stop 32 (voiced
selector *658
237: [RB] and voiceless [¥])
semantic compositionality
458 speech sounds 3-5, 10 labiodorstal stop 98
semantics *42 (linguo-alveolar), *19-21 (voiced [gb] and. voiceless
semiconsonants 269 (stops), *23 & 268 [kp])
semivowels 269 (obstruents), *24 palato-alveolar fricative 24
sentence intonation 392-3 (palatoalveolars), *59-60 (voiceless [f]), 25
sentence stress see phrasal & 268 & 633 (sonorants), (voiced [3])
stress *61 (oral vs. nasal), 68 uvular fricative 80 & 81
sibilant cluster avoidance 622 (palatal), 76 (retroflexes), (voiced [s] and voiceless
sibling node *257 *77 (approximants), 647 [x])
velar fricative 13 (voiceless (“defective”; rolled: see (quantity-insensitive), 367

[x]), 17 (voiced [y]) also alveolar trill) (quantity-insensitive
velar stop 21 (voiced [g] tap [rc] *79-80 systems), 368ff. (non-
and voiceless [k]) uvular trill [R] 81 rhythmic), 373 & 443-5
[] 28 velar lateral [L] _ 73 (three syllable window),
[J] 28 velar nasal [n] 64-7 374-7 & 641 (irregular),
382, 389-90 & 642 (and
[6]¢#:23: [p] 65-7
[3] 25 [x] 67-8 intonation), 390 (and tone
[¢]/ 72. [4] 73 association), 393 (nature
[ws] 80, 81 [rR] 81 and mechanics), 605
ly] 17 [Ae 72 (quantity-sensitive
[6] 23 [Ll ee76 systems), 639 (and
[f] 24 [a] 74-5 rhythm; differences in),
[x] 80, 81 [vu]aa 639 (focus and), 640
[b] 20, 32 {l] 70 (effects of final
[RB], 32 [m] 63 consonants), 644 (well)
[d] 24 [m] 61-2 retraction 304-6 (under
[dq] y21 [n] 63-4 clash), 307-9 (word-
if] 6 fr]> 277278: internally), 307
(permanent shift), 309
[gb] 98 spelling 7 (system), 8
[g] 21 (English system), 21 (landing site), 310-12
{h] 14-16 (letter x in English), 68 (failures), 422-4 & 465
[kp] 98 (diversity), 103, 297 (word-final), 472ff. (under
[khazh (conventions), 212 (and clash), 640, (failures), 643
[p] 19, 32 sound in English), 555, (English phrases) 644 &
LP loe 558, 621 647 (word-final)
[s] 10 spelling to sound (see also secondary *337, 418 & 644
[t] 24 sound to spelling) 30 (English)
[ele:20 (ex.) stress cycle, word-internal
[viz spread glottis 528 432-6, 433 (formal
bits stop allophony 315-17 mechanics), 435 (primary
[alee Stratal Phonology see Lexical stress carry-over), 644
speech sounds: sonorant Phonology stress, English 302-4
consonants stress (see also accent, metrical, (personal names), 321-2
alveolar lateral [1] 70 and under metrical) 294ff., (compound & phrasal),
alveolar lateral, velarized [t] *296, 298 (contrastive), 299 328-9 (word), 329-31
73 (as projection; bearers), (nouns: basic), 333-5 &
alveolar trill [r] SLO OL 300 (contrastive and 644 (word main), 350
bilabial nasal [m] G1=27 63 meaning), 300-2 (parameter settings), 351
(voiceless [m]) (motivation for contrasts), & 419 (line conflation), 420
bilabial trill 78 *304 & 639 (clash), 311-12 (cyclic), 421 (non-cyclic),
dental nasal [n] 63-4 (direction of movement in 640 (final consonants
flap *79-80, 315 shift), 314-17 (segmental effects; iteration)
labiodental approximant [v] evidence), 327ff. (word), algorithm 333 (no. 1), 334
ve 327-8 & 640 (French), (no. 2), 350 (no. 3), 357
laterals *70—4 335-7, (multiple), *336 (no. 4), 360 (no, 5), 432
liquids 69-73, *70 (iterative), 338-42 (update), 448 (failure to
nasals 60-8, *61 (typology), 341 (principles reapply)
palatal nasal [x] 67-8 and parameters model), stress patterns
patalal lateral [A] 72 342-5 (word-level), 364 Arabic (Cairene) 380 (ex.)
rhotics *74-81,77 (Weight by Position), 366 Araucanian 346 (ex.)
Bengali 369 clusters: ex.), 292 (high complex 262-4, 285

English 308 (verb and vowels in French: ex.), 437 (differences with complex
noun merger), 334-5 (word-final consonant), onset in English), 284-8
(verbs and unsuffixed 438 (extraprosodicity & (English), 285ff. (skewed
adjectives), 640 extrasyllabicity), 510, 638 English distribution), 637
Guajiro 380 (ex.) (CV rule) (motivation), 638
Hungarian 347 (ex.) syllabification parameters (English), 641 (moraic
Latin 379 (ex.), 628 (ex.) Minimal Sonority Distance formalization)
Macedonian 376 256-*7, 285, 638 nucleus *239, 638 (two
Mongolian 369-70, 372 Onset Maximization timing unit maximum)
Selkup 371-2 279-84, *283, 614, 638 complex 249-52, 271
Vedic Sanskrit 376-7 Maximal Onset Satisfaction (with three members), 600
Warao 346 (ex.) 638 (and complex onsets)
Weri 347 (ex.) Minimal Onset Satisfaction consonants in 265-8,
Western Aranda 347 (ex.) *279, 283-4, 614, 638 460-2, 638
stress rules Onset First 284 onset *239, 271-3 & 277
Destressing 421, 430, 434, Sonority Sequencing 437, (high vowels), 278-9
644 438, 439, 460, 464, 638 (fulfillment), 289 (three-
End Stress [right] 922,1325 syllabification rules *461-2 membered?), 638 (high
(interaction with (Sonorant Nucleus vowels; maximal)
Extrametricality [right]), Formation)
complex 252-3, 253
369, 337 (at level 1) syllable 236ff., *238, 247
(gaps in English), 254
Rhythm Rule *305-6, 309, (implicational syllable (obstruent + liquid
312 (German), 423, 475, structures), 248 (as
clusters in English), 285
644 sonority peak), 248-9
(differences with complex
Strict Cyclicity 427-9, *428, (nature), 261 (usefulness of
coda in English), 600 (and
439, 499-501 English syllable), 275
complex nuclei)
strong verbs *224-5 (tautosyllabic consonants),
rime *244, 353-4
suffix(es) *153, 395 483 (as phonological
(correspondences with
(morphological function), domain), 501 (appendix),
stress)
488-9 (as Phonological 600-4 (complexities), 637
syllable weight 352-7
Words), 614 (level 1 and as licenser 243, 250-1 (of
(metrical accent), 353-4
level 2) segments), 646 (of a word-
(correspondences stress-
‘s possessive 624 & 625 final consonant)
(mechanics), 625 core (CV) 238-41, 637, 638 rime structure), 353 (heavy
(compatibility with (CV rule) vs. light syllable), *354
role of sonority (accent), *356, 357-8
plural), 658
-(e)s plural 610, 625 241-8,(peak; profile), 255 (English word-final
(compatibility with & 266 (profile), 268 consonant), 359-60 (final
possessive) (codas), 280 (profile), 282 long vowels), 448 (heavy
-i 224 (Old English) (minimum in nuclei), 637 penult), 640
/-i/ 428 (non-metricality), (alternation) moras 361-5, *363, 563
563 (non-moraicity) typology 245-7 (basic), 637 (non-moraic /i/), 640, 641
-ic extrametricality syllable structure 244-5, (vs. skeletal slots)
exception 426 436ff. (word-final), 596-9 quantity-insensitive systems
-ion 545 (English basic), 628 (Urban 366, 367
-ity 210 Hijazi Arabic: ex.), 638 quantity-sensitive xOD/,
syllabification 275-7 ([iu]), (anchoring on the 609-12
287 (word-final coronal skeleton) Weight-by-Position 364-5,
obstruents), 288-90 & 639 coda 242-*3, 285 (English 641
(/s/), (Polish consonant minimal sonority distance) synonym 33
Index of Subjects
Syntactic Structures 301 stress-and-intonation voice *16-18, 77, 385, 528,

syntax *41 languages 394 529, 633 (phonation types),
tone languages 393-5, 642 (production)
tauto- 275 *394, 641-2 voice box (see also larynx) 14
tautosyllabic consonants 275 tone tier 387 voiced and voiceless stops in
The Sound Pattern of English see tongue root 150 English 281
Sound Pattern of English trachea 3 vowel(s) (see also cardinal
theta 23 transcription see phonetic vowels) 115-16
tier 204, etc. (timing), 227 transcription (differences in sound
(melody), 387 (tonal) tree 239 quality), 115 (articulatory
timing tier (see also skeleton, triphthongs, centring 200 difference with
CV tier) 204, 207-9, *208, trochee 368 consonants), 117 (number
226, 388 (as autosegmental truncation 619-20, 658 in the world’s languages),
baseline), 636 (phonology) 127-9 (basic: [i], [ul], [a]),
timing unit 204, 207 tunes 383-*4, 641 139 (chart), 141-2
Timmy 236, 637 (descriptions and phonetic
tonal morpheme 403 umlaut 160-3 (German), symbols: ex.), 158 (opaque
tonal patterns in languages 160-4 (English plurals), vs. prespecified), 169-70
African languages 643 636 (German and English) (variation in English),
Chinese 643 underspecification (see also 225-6 (alternations in
Ci-Ruri 407 (ex.) Radical English), 225 (natural
English 642 Underspecification, alternations), 227-8 (long-
Margi 406 (ex.) Contrast-Restricted short alternations), 250
Mende melodies 399 underspecification) 150, (long), 268-9 (in disguise),
tone 387, 390-1 (non-lexical), 502-22, 504-7 (feature 530 (unified place features
392 (complex), 529 (and transparency), 505-7 (in with consonants), 569
voice), 637 & 657 Russian sonorants), 507-11 (non-low word-final
(contour), 641-3, 641 (and markedness), 515 tensing), 633 (articulation)
(tunes), 643 (distinctive (maximal), 649, 650 back high lax [u] 184, 185
features) (Combinatorial back mid-low [a] 185-6
downstep *404, 643 Specification) central *137—8, 186-90
floating 401-5, *404, 643 in Turkish suffixes 152-4, low raised: [ve] 186
functions of 394 153, 502, 507 mid [3] 187-8
(intonational), 394 underspecified segment 515 [a] (see also schwa) 138,
(lexical), 395 & 649 (asymmetric 186, 187 (r-coloured: [a*])
(morphological), 641 behaviour), 519 (and fe] AMis7,
(lexical), 641 epenthesis), 520 (identity [4] 138
(morphological) with lexically specified [u] 138
rules (Meeussen’s law) segment) English (see also cardinal
401-2, 403, 643 universal grammar *89, 513 vowels, English
types of *386-7 (highH & universal tendencies 508 correspondents) 119
low L), *390 (mid M) universals, hard vs. soft 533 (system complexity), 126
tone association 387ff., 390 unmarked 508 & 127 (vs. French), 169ff.,
(to main stressed syllable), 176 (raised [z]: [z]),
394, 399-405, 400 & 643 variation in English 169-73 210-11 (alternations), 663
(original principles) velar 11 French 126 & 127 (vs.
415-17 (cyclic), 641, 644, velum 11 English)
646 vocal chords see vocal folds front high lax [1] 183-4
tone, types of function 641 vocal folds *15 lax non-primary non-central
pitch accent languages vibration 77, 641 181-6 (English
*394, 395-8, 641-2, 643 vocal play 236 correspondents)
schwa see central [9] (distinctive), 175 vowel typology 127-9

[a] 130 (variation), 193 (intrinsic), (triangle), 633
[ce] 179 (English) 210 (length-based vowels, quantum 129-*30,
[A] 185-6, 186 (absence alternation), 260 531,633
from N. England), 186 (alternation in Yawelmani: vowels, undecomposable basic
(English correspondents) ex.), 564-9 (alternations) 219-20, 220 (Great Vowel
[3] 187, 188 (English vowel lengthening in English Shift analysis)
correspondents) 566, 641 (word-final)
[1] 183-4 (English vowel parameters wheezing 236
correspondents) backness-frontness 123 word formation: truncation
[vu] 184-5 (English
height 123 (monosyllabic in English),
correspondents) 619-20, 658 (phonology)
roundness 125
vowel disharmony (see also word(s) 150 (Turkish
vowel reduction (see also
vowel harmony) 154-7 structure), 237 (Japanese
schwa) 187, 314-15, 318
(Turkish) (see also Great pronunciation of English),
(Catalan: ex), 335, 394, 640
Vowel Shift) 262 (possible), 296-8
vowel sets 141 (ex.)
vowel features 144-6, 636 (prominence), 457-9
vowel shift (see also Great
(phonetic foundations) (internal cohesion), 487
vowel harmony (see also vowel Vowel Shift) 204 (dual identity of lexical
disharmony) *149, 535 (English), 230 (shifts: ex.) words), 602 (word-final
(blockers), 636 vowel space 120 consonant licensing)
Finnish 168 (ex.) vowel systems 127-9
Khalkha Mongolian 534 (triangle), 141 (ex.), 589 X-bar 249 (template), 638
([+ round]) (least marked) (notation in syntax)
Turkish 149-54 ([+ back]), American English 634 X-tier see timing tier, skeleton
167 (ex.), 636 ([+ back]) Northern England 171
vowel length *171 Yorkshire 171 yod 269
= todd Vouk
Syriac ee ennisit frwresy
pee “e! Efe (okgrinhd)
OE-2ST— eutanup e@laweyv
tawio~ - 275 Oa LE
ene
Tanai sdatapntogiteete =
Sorrel F qdberets Sigiee
imete Zo
lee 204. ete. (ir wegdtiedw dite
noite Sibir bow ea opKaadis

| Wnlezignie
ndseinite adetrany”
—oygebbiewiy 34, Beeig’ 0%
etl ifads ;
=aae,
GH leper .
220ASSEEY OSI" wlio Uwrcataen

SONIA
ai iby
Sh JS ni
leveov
— sedrraty
¥Sfrurournte Biger _ (everibe
ee AGogeatanical
(iti
nélthiiiiasg ees
| - Fimne-adifadiifzeog)
Sa6 Uo JF, eae neat) (Jeuceeptig
> Sori PARES nartiftho rg) cursobetat (a?) ie ig Aperoy ponte ysis
loop titolo (invent? ate) i tootBade tego
i ixarts) cu +S, i like twp ; thy woweioadeten
debra ube fda
LCS erlelipal BEARER
ediwal OS). -queqe iswoy Teele ne rieiste
Ly: vik: tretage leveo

Petes), Mitel aresht)
renee
Lhe*
Nike thfooty (ev), Jaf
Tay evraA (bbs Myae
- i | ea saeale i,
ex, UE & ‘ear boy
PS piel mearlal ((osaehsq}e2 |
Rate (epee aaa! i Oy Py aye faye
(toss), Af? 1, 44) (are aught), 515 ens), 632 (articulati
thas.) OT: (diatowsive (maswaah, At? 6501 echbigirhasfidh “184,48
- beaerryts) ifarmmbineticsal back mktelowe fe] 8-6 :
oettiep “4, of” bigest central “13764, 186690. —
Herptary 5 “WM, 64? in Paderch safes 1904 levew aleed: fe) G6
fantHos at a “3 502, SP: mid in}. 1878
Oritanciienal’. 39 wlerspeciinn| anginedtt 515 ie) Ge Rye soda) TSR)
= evra), Sat deBt) (n0Vinoetric Le. LF? (p-coknumeets fr])
drtkrphoingical). O¢1 bet viewed, AIS fond fe) 17
-fheatoe’), 44) ‘penthed=) S20 (identity i) 7 18
Canter ie 4) em ledceoile ane fied jab 28
ales Chawisaiy’s law) wgitee nt) i rghish(are leocastinad
Di, WS ea “WE grenemqar Sr, 635 vivre, English. : 5 fe
Geet “Sab-7 (pe H & “itera Gervlerties WO conespondenis) 719|
= ie te mid tf). eras. hordore anit O55 \aytiem canplexity), 26.
tone atencastion Se 80 evrrp is Ke Ate
& 127tea. French); 26096.
Go wien steerenk ay Mab’
24, FMS HO ieed - dapletion te Piaget Ve-7
(resp ral prince }(ieal) velar +I
: a5-1)
aaa. (ela):
™ b4), 404, col) 24
vootchon ne eotal (atis
tune, lppes of finches G4) vucal folds . +14
piet overd Gny eos viwhon 3?) sey
704, ©. 0 Al, fA) wel piny 284
+ we ie© 6 ee - eee ae eee; ee ee =
a eas ee vr
m~ es
:
:
>
\\
‘Roca has done an impressive job of covering the broad
field of phonology in a comprehensive yet accessible
manner. Beginning with the basic principles of
phonological investigation, the text leads the reader
through all major theoretical frameworks, up to and
including Optimality Theory. Professor Roca’s
experience as an educator is evident in both content
and quality; constant self-checks ensure the active
participation of the reader.’
KEVIN VARDEN, Meiji Gakuin University
A Course in Phonology presupposes no prior knowledge of phonology or phonetics,
and takes the learner step by step through the various stages and areas of the
discipline without sacrificing rigour or breadth of coverage. The book’s extraordinary
clarity makes it readily understandable by anyone with a keen interest in phonology.
The substance of phonology is more constant and more general than the
limitations inherent in any particular theory or formalism might suggest. The book
therefore provides a coherent account of phonology for beginner-students, rooted
primarily — but by no means exclusively — in the phonology of English. No particular
theory or formal apparatus is preferred over another: instead, the essence of
phonology is provided in the most neutral way possible, for the maximum benefit of
the reader. The book avoids entanglement in doctrinal disputes and formal minutiae,
and aims instead for the general and the permanent.
The book reviews all the major advances that have taken place in generative
phonology over the past thirty years, including Optimality Theory. Its many
pedagogical features encourage interaction with the reader, and include a wealth of
check-points, chapter previews and summaries, lists of key points, and exercises for
further practice.
Iggy Roca and Wyn Johnson

SEEEEEEEEEEE ee
Iggy Roca is Professor in the Department of Language and Linguistics
at the
University of Essex. He specializes in phonology and is the author of Generat
ive
Phonology (1994). Among the volumes he has edited is Derivations and Constra
ints in
Phonology (1997).
Wyn Johnson wrote her doctorate on the lexical phonology of French,

and has been
teaching phonology at the University of Essex since 1983.
Cover design by Design Deluxe, Bath
Printed in Great Britain .
Visit our website at

http://www.blackwellpublishers.co.uk
Visit our abstracting service, Linguistics Abstracts Online, at
http://www.blackwellpublishers.co.uk.labs
BLACKWELL
as eee

A Course in Phonology

Uploaded by

Copyright:

Available Formats

A Course in Phonology

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Course in Phonology

Uploaded by

Copyright:

Available Formats

IGGY ROCA AND WYN JOHNSON

Stuart Davis, Indiana University

Consonants (Non-Pulmonic) Vowels

Voiced implosives Front Central Back

© Bilabial 6 Bilabial > Examples: Close 1ay eu weu

Palatoalveolar g Velar k° Velar Open-mid [SUG ) \o ——T AWS o)

Alveolar lateral ec Uvular S’ Alveolar fricative Gs 4

Saivoiced epiglottal fricative tions can be represented by two uprasegmentals

Diacritics ' founa' tifon

h Aspirated th db _. Linguolabial Laminal t

Less rounded 2 Palatalized di Nasal release

Advanced U ¥ Velarized t® d¥ Lateral release Tones and Word Accents

Syllabic n Lowered e (8 = voiced bilabial approximant) e { oa € A OW EES

Iggy Roca and Wyn Johnson

First published 1999

Blackwell Publishers Ltd

Blackwell Publishers Inc.

British Library Cataloguing in Publication Data

Library of Congress Cataloging-in-Publication Data

This book is printed on acid-free paper

Part |: Phonetics and Phonology 1

1 How Are Sounds Made? The Production of Obstruents 8

2 Introducing Phonology: Assimilation 31

General Properties of Sonorants 59

4 Natural Classes of Sounds: Distinctive Features 85

Descriptive Phonetic Parameters 86

5 Vowel Sounds: Cardinal Vowels 114

1 On What Vowels Are and How They Are Made 143

Key Questions 140

6 Phonological Processes Involving Vowel Features 143

7 The Vowels of English 169

8 The Timing Tier and the Great Vowel Shift 204

A Puzzle with Affricates 205

Part Il: Suprasegmental Structure 239

9 The Syllable 235

The Shape of Children’s Early Utterances 236

10 Syllable Complexity: English Phonotactics 261

my Complex Codas 262

11 The Phenomenon of Stress: Rhythm 294

M Stop Allophony 315

12 Metrical Principles and Parameters 321

English Phrasal and Compound Stress 321

13 Syllable Weight. Further Metrical Machinery 349

— An Additional Pattern of Stress in English 349

14 Tonal Phonology 382

The Phenomenon of Intonation 383

10 Pitch Accent Languages 395

Part Ill: Advanced Theory 409

15 Modes of Application: The Cycle 411

Staged Grid Construction 412

16 Domains of Application: Lexical and Prosodic Phonology 444

17 Aspects of Lexical Representation: Underspecification,

Effects of Strict Cyclicity 499

18 Rules and Derivations 543

Rule Ordering: Feeding and Counterfeeding 543

19 Constraints: Optimality Theory 584

Naturalness of Phonological Inventory: Markedness 585

11 Correspondence Constraints 613

20 Looking Back and Moving On 630

A book of this kind inevitably owes an incommensurable debt to many more