Phonology [2 ed.] 9780230573147, 0230573142

Citation preview

PHONOLOGY

MODERN LINGUISTICS SERIES

Series Editor Professor Maggie Tallerman Newcastle University, UK

Each textbook in the Modern Linguistics series is designed to provide an introduction to a topic in contemporary linguistics and allied disciplines, presented in a manner that is accessible and attractive to readers with no previous experience of the topic. The texts are designed to engage the active participation of the reader, and include exercises and suggestions for further reading. As well as an understanding of the basic concepts and issues for each topic, readers will gain an up—to-date knowledge of current debates and questions in the field. Titles published in the series English Syntax and Argumentation (4th Edition) Bas Aarts

Phonology (Znd Edition) Philip Carr and jean-Pierre Montreuil Pragmatics Siobhan Chapman Linguistics and Second Language Acquisition Vivian Cook Sociolinguistics: A Reader and Coursebook Nikolas Coupland and Adam jaworski Morphology (2nd Edition) Francis Katamba and john Stonham Semantics (2nd Edition) Kate Kearns

Syntactic Theory (2nd Edition) Geoffrey Poole Contact Languages: Pidgins and Creoles Mark Sebba Further titles are in preparation

Modern Linguistics Series

Series Standing Order ISBN 9TB—O—333—7'| 701-1 hardcover

ISBN 97B—0—333-69344-5 paperback (outside North America only) You can receive future titles in this series as they are published by placing a standing order. Please contact your bookseller or, in the case of difficulty, write to us at the address below with your name and address, the title of the series and one of the ISBNs quoted above. Customer Seniices Department, Palgrave Macmillan Ltd Houndmills, Basingstoke, Hampshire RG21 6X5, UK

Phonology Second Edition PHiLlP CARR AND

JEAN-PIERRE MONTREUIL

© Philip Carr 1993 © Philip Carr and jean-Pierre Montreuil 2013

All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No portion of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, Saffron House, 6-10 Kirby Street, London ECTN 8TS. Any person who does any unauthorised act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The authors have asserted their rights to be identified as the authors of this

work in accordance with the Copyright, Designs and Patents Act 1988. First edition 1993 Second edition 2013 Published by PALGRAVE MACMILLAN

Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited, registered in England, company number 785998, of Houndmills, Basingstoke, Hampshire RG21 6XS.

Palgrave Macmillan in the US is a division of St Martin's Press LLC, ‘I75 Fifth Avenue. New York, NY 10010. Palgrave Macmillan is the global academic imprint of the above companies and has companies and representatives throughout the world. Palgrave® and Macmillan® are registered trademarks in the United States, the United Kingdom, Europe and other countries. ISBN 978-0-230-57314-7

This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental regulations of the country of origin. A catalogue record for this book is available from the British Library. A catalog record for this book is available from the Library of Congress.

Contents

Prqflrce to the Second Edition

IX

Pnjfizcr to I/Jr First Edition

x

/lckrzowledgements to the Second Edition

xii

Ac/enowledgernenrs to the First Edition

xiii

Revision of Phonetics 1. Consonants 2.

1 1

Vowels

8

1 The Phonemic Principle

13

1.1

The language of phonology

14

1.2

Phonemic rules

20

1.3

Phonological representations

27

1.4 Concluding remarks Exercises

29 29

Further reading

31

2 Altemations

32

2.1

The internal structure ofwords

32

2.2

Testing hypotheses about rules and representations

55

2.3 2.4 2.5

Morphophonological alternations Choosing between analyses Deletion and insertion

37 38 39

2.6

The ordering of rules

40

2.7

Concluding remarks

46

Exercises

46

Further reading

49

3 Features, Classes and Systems

50

5.1

Expressing generalisations

50

3.2

Features (i)

5]

3.5

General remarks

56 V

CONTENTS

3.4

Features (ii)

3.5

Features in representations

3.6 3.7

Features in rules lrnplicational relationships

Exercises

Notes Further reading 4 Problems with the Phonemic Principle

4.1 4.2

Contrast and neutralisation Contrast and the minimal pair

4.3

An alternative to the phonemic principle: generative

phonology Exercises

Further reading 5 The Organisation of the Grammar

5.] 5.2 5.3

The lexicon The location of morphology The phonological component vs the lexicon

5.4

Summing up

Exercises Notes

Further reading 6 Abstractness. Psychological Reality and the Phonetics/Phonology Relation 6.1 Ordering relations and rule application in the

6.2 6.3 6.4 6.5

SPE model Absolute neutralisation Abstractness and psychological reality

57 64 66 72 74 77 77 '79 79 84

92 93 96 97

97 101 105 112 113 ll5 ll5

117

117 122 127

Underlying representations and naturalness Abstractness, phonological change and child language

134

acquisition

138 140 143

Exercises Further reading

7 The Role of the Lexicon 7.1 Phonology and morphology revisited: lexical phonology

7.2

Lexical and postlexical application

7.3 7.4

Structure preservation, abstractness and productivity Redundancy and underspecification

145

145 151 153 155

CONTENTS

Exercises Notes

Further reading Representations Reconsidered ii]: Phonological Structure above the Level of the Segment 8.1 Lexical rules, phonotactics and the syllable 8.2

Syilabiiication and syl1ab1e~based generalisations

8.3

Extrasyllabiciry, the CV tier and abstractness

8.4 8.5 8.6

The CV tier, segment length and complex segments Stress assignment, rhythm and the Foot Symmetry, clash avoidance and the metrical grid

Prosodic domains and the syntax.’phonology relationship Exercises Notes

vii

161 163 164

166 166 171 177 182 187 197

8.7

201 206 208

Representations Reconsidered (ii): Autosegmental and Subsegmental Phonology 9.1 Nasality, segmental and suprasegmental 9.2 Vowel harmony

9.3

Dominant/recessive harmony

9.4

Feature geometry and subsegmental structure

Exercises Notes

Phonological Weight 10.1 Weight and time 10.2 The basic architecture 10.3 The weight ofcodas 10.4 The structure of geminates

10.5 Stress-to-weight and weight-to-stress 10.6 Moraic theory and compensatory lengthening 10.7 The word-filial weight asymmetry

Exercises Further reading Optimality Theory 1 1.1 The basic architecture 1 1.2 The logic ofoutput-driven models

1 1.3 Positional constraints 1 1.4 The factorial typology

1 1.5 The nature of the input

213

213 219 230 233 237 238

240 240

242 243 246 249 250 254 255 257 258

258 262 263 266 269

viii

CONTENTS

11.6 The prosody-melody interface 1 1.7 Positional markeclness vs positional faithfulness

11.8 Conclusion Exercises Further reading

12 Issues in Optimality 12.1 Opacity: problems 12.2 Output-to-output correspondence

274 277 281 283 285 286 287 291

12.3 Re-analysing cyclicity

293

12.4 Opacity: some proposed answers 12.5 Conclusion

296 307 308 310

Exercises

Further reading Feature Speczficrztiansfiar Consonant:

312

Sample Answers to Exerciser

314

References

322

Subject Index

334

Language Index

337

Preface to the Second Edition The first edition of this textbook appeared in 1993. It was intended as an introduction to generative phonology and adopted a historical perspective. lt introduced the phonemic principle, the problems with that principle and proposed solutions to those problems using the standard generative phonol-

ogy Framework of the late 19605. That model was based on Chomsky and 1-la1le’s 1968 book, The S0um1Prztrern* 0fEng[i.rh (SPE), and was known as the

SPE model. The book also covered theoretical developments in generative phonology From SPE through to the early 19905.

The book was widely used, but became outdated, and thus less useful for students. At the time of its release, a new Framework was emerging, known as Optimality Theory (OT). OT has since become widely used, so we have updated the book to include it (Chapters 11 and 12). We have also included

coverage of what is known as ‘phonological weight’, which is introduced and

exemplified in Chapter 10. In this new edition, we have cut out the final chapter of the first edition, conflated Chapters 6 and 7 and updated several

other chapters. We have increased the size of the book, but we have avoided making it too long.

We have retained the historical perspective of the first edition, so that this edition now covers an even larger chronological span than the first. The range of languages covered has been considerably expanded. We hope that this revised edition will prove accessible and useful for university students seeking to get to grips with theoretical phonology. We have retained the emphasis on

exercises: in our view, one cannot understand phonological analysis without sitting down and engaging in it.

Pun. CARR JEAN—PlERRE MONTREUIL

IX

Preface to the First Edition To the student This book is a beginners’ introduction to phonology, and concentrates mostly on what is known as generative phonology. It assumes that the reader has taken a course in elementary articulatory phonetics, but it assumes no knowl-

edge whatsoever of phonology. A revision chapter on phonetic terminology and a chart ofthe symbols which constitute the international Phonetic Alphabet are included For reference, or for those who need to brush up on the

subject. It is hoped that the present textbook will allow students to move on to more advanced textbooks, such as Kenstowicz and Kisseberth (i979),

and also to the source literature 0n the subject. The reader should also consult Hyman (1975) For Further details and issues in standard generative

phonology, Lass (l9S4a) For a broader view oiphonology (beyond generative work), and Anderson (1985) For a historical perspective on twentieth-century

phonology. Syntheses and surveys of current theoretical work, such as Goldsmith (1990) and Durand (I990), as well as the literature itself, should also be approachable after reading Chapters 8-1 1. The book begins with exercises within and at the end ofchapters, many of

which should prove easy For students who take naturally to linguistic analysis, and tractable For those who do not. After the first Few chapters, there are Fewer, and these are at the end of chapters. (Some sample answers are given

at the end of the book.) For most students, it would be rather pointless to

attempt reading the early chapters without simultaneously doing the exercises; the exercises in the later chapters are intended to help students to think

about the theories discussed by Focusing their attention on particular analyses. lt is important For students to retain all their exercise work, since earlier exercises may be referred back to, or reworked, at a later stage. The Following abbreviations are used in exercises and examples: ‘pl.’

(plural), ‘sg.’ (singular), IPS. ZPS, 3P5 (lst, 2nd and 3rd person singular); nom. (nominative), ace. (accusative), gen. (genitive), masc. (masculine), Fem.

(Feminine).

To the teacher This book is an attempt at a solution to a problem. The problem is this: how to introduce generative phonology, without trivialising the subject, within the X

PREFACE TO THE FIRST EDITION

xi

confines ofa ten-week course, to students who major in subjects other than linguistics, and thus may not take their study of the subject any further. The

problem is less acute when those other subjects require similar sorts of thinking, but when the majority of students have, for several years, spent the bulk

of their time following courses requiring quite different sorts ofthinking, difficulties arise. ln England (much less so in the US and in the continental

system, which also survives in Scotland), these difficulties are in large part a consequence of the fact that many such students have been educated in the English ‘narrow specialism’ tradition, which has tended to bolster the ‘two cultures’ arts/science divide. These students often tend to view themselves as ‘arts types’ sight of diagrams in to be introduced in and exercises which

rather than ‘science types’; for many of them, the mere a book can prove anathema. The subject therefore needs as gentle a way as possible, with much exemplification, do not throw the student in at the deep end. Thereafter,

more complex problems can be investigated, and the student can then find a way into the subject. The problem is a challenging one, and in attempting

to resolve it, one can go some way towards undermining the rather limiting ‘two cultures’ division, which fails so clearly to accommodate the discipline of linguistics. Wliiie there are excellent textbook introductions to standard generative phonology, such as Kenstowicz and Kisseberth (1979), and impressive surveys of current theory, such as Goldsmith (1990), they are simply too detailed to be used on such a course, given the time available and the sorts of

student in attendance. The book aims to introduce standard generative phonology in a relatively simple way in Chapters 1-6, and then, in Chapters 7-! l, to introduce the

student to more recent work in phonological theory. Since the text thus falls

into two parts, it is possible to teach just the first six or seven chapters if time is short; the last chapters may then be used as an introduction to a more

advanced course, for which a textbook on current work, and the literature itself, may be used. No attempt is made to present alternative, non-generative, approaches to phonology: new students have enough trouble trying to get the hang ofone framework without having to cope with several; other approaches

can be tackled once the generative approach is familiar. This is not to say that a non-critical attitude to generative work is fostered; analyses are often shown

to be tevisable, competing generative theories are discussed, and the interplay of data and theory is stressed. It is best ifstudents have seminars/tutorials in which to discuss the exercises

in each chapter before proceeding to the next, but the exercises should be tractable even ifthis is not possible; some sample answers are given at the end

ofthe book.

Acknowledgements to the Second Edition We are grateful to Daniela Muller For data From Occitan, to F. Rasooli Danesh For corrections to our Farsi data in Chapter l, and to Tajul Aripin Kassin and Zaharani Ahmad For data from Standard Malay and responses to queries.

Thanks to Nigel Love, who provided many helpful queries, and corrections

to datasets. Thanks to Maggie Tallerman, the Series Editor, and to Felicity Noble and

to Kitty van Boxel of Palgrave Macmillan, For bearing with us during the lengthy delays in getting this to press. Special thanks to Maggie Tallerman For

detailed and useful comments, suggestions and corrections on the drafts of most chapters. Many thanks also to Nick Fox and Keith Povey For all of their

work in es~> ._.?_ _

‘day’ ‘life’ ‘pastry’ ‘leaf’ ‘colourless'

You will need to establish what the distribution of the two sounds is. To do that, you will need to have an idea of what can count as an environment.

We have seen examples 0F environments in our discussion oF French and Spanish: the beginnings and ends ofwords, between vowels, after consonants

and beFore consonants. The relevant environments can be of these very broad sorts (which will suFfice for the Farsi data), or they may need to be more

particular, as we will see later. Once you have the answer, consider the status of the voiceless alveolar trill in Farsi. What is its distribution with respect to the other two sounds: parallel or complementary? Once you have this answer, you will be in a position to say whether these three sounds are realisations of

only one, or 0F more than one, phoneme in Farsi. Let us consider some environments which are narrower than these. We saw that in SSE, the voiceless alveolar tap occurs, not after (l7l_]/ consonant,

but after voiceless stops in particular. in order to manipulate phonetic data of the sort given in the problems in this book, you will need to think in terms of

classes and subclasses of sounds; the stops are a subclass of the class of consonants, For instance, and the voiceless stops are in turn a subclass of the stops. ln the Following two exercises, you will need to be a little more specific than in the Farsi problem in stating the relevant environments. B. Siamese Thai (the Thai spoken in Bangkok; data from Harris 1975, Harris and l\loss 1972) .

[l;):|

‘handsome’

2.

nl

“to play’

.

[lru:m|

‘to Forget’

4.

ak}

‘to steal’

.

[l8:n] _\iv~t»- [llin] 9. iplazul l l. J l3. kl.J:nl 15. khlail fifii

‘late meal’ ‘monkey’ ‘empty’ ‘precious stone ‘drum’ ‘dirt'

6. 8. ll). 12. 14 I6.

u:pl lizkl lntai] am] "l_~.):nl _.- _~_._U__.T,s

The distinction between voiced and voiceless alveolar trills is allophonic;

which two contexts does the voiceless allophone occur in? (The diacritic [1] in 3-6 above indicates palatalisariong you may ignore it for our purposes here.) We have now seen several cases where the distinction between a voiced

sound and its voiceless counterpart is allophonic. Looking back over the Polish data in C, can you see any cases ofphonemic voicing (cases where the distinction between a voiced sound. and its voiceless counterpart is con» trastive)? Now look at the Following data from Polish. D. Polish (data from Gussman 1980, Majewicz 1986)

+—'

.

[pompnil

‘a pump’

2.

nenodaiei]

‘on the leg‘

.

['t_reFli- ' [panil] 9. ll [£1/___ri

ALTERNATIONS

39

Loolting over the data. we see that neither hypothesis is disconl’ii'tm-tl. The lirst one says that (voiceless) pre-palatal Fricatives are realised as posl'—nlveo|a|‘

Fricatives word-finally; the second says that (voiceless) post-alveolar liricativcs are realised as pre-palatals before a high front vowel. Now, since the vowel is articulated in the palatal area, it seems perfectly plausible to suggest that a post-alveolar sound should assimilate partially to a Following palatal sound,

becoming pre-palatal. The second rule thus has a high degree of phonetic plausibility. The same cannot be said For the first rule: there is no clear phonetic sense in which a word boundary should induce a pre-palatal to move Forward to the post-alveolar place of articulation. We will therefore opt For

the second analysis, on the grounds of phonetic plausibility. Try using this criterion in the Following exercise. B. Polish (data partly From Rubach 1984)

. .

. u\-t>.wg~.a—.-

Nominative singular

Locative singular

[pus]

‘hcl t‘

[p-use-I

int/.|

‘timc'

[mic]

lbriitl

‘brother’

[bneccl

[cud] idltvopl

‘miracle’ ‘hell’

[cuje] [dir.vor'1e]

Phoizetic note

For a description of [s], [1], [6], [jl and [ti], see p. 4 on pre-palatals. The phonological form of the locative singular morpheme is /ie/; under

certain conditions, a rule applies to it which deletes the /i/. For our purposes here, you should assume an /ie/ form For this suffix. (a) State two possible phonological Forms For each ofthe noun roots.

(h) Write the rules which will have to accompany each phonological representation to give the phonetic representations. (C) Say which analysis is preferable, and why.

2.5 Deletion and insertion The rules we have written thus Far all involve changing some segment to another, but not all rules have that Function. It was just mentioned, in passing, that the I if in the suffix /ie/ in Polish gets elided (is not uttered) in certain cases. This phenomenon of elision oi: segments is a very common one, and we express it by writing rules in which the segment in question is deleted. In

our Polish case, the rule would look like this: /if-i-(Z)/_e

40

PHONOLOGY

llcrc, the symbol ‘(Z)’ means "1.ero', and we interpret the rule thus: /i/ is

deleted when immediately Followed by /e/. The converse process to elision is epenthesis, in which a segment is inserted in a specific environment. This process would be represented by a rule of the form: ®~>[i[/_c

We will see that both of these rule formats will be useful in our analyses. Try using the rule format for deletion in the following exercise. C. English (RP) in the data on p. 14, the realisations of the /r/ phoneme in SSE can be given, somewhat informally, by the following rule: _. / voiceless stops Z

/r/ —>

[ ] /voiced consonants _

li->-ili‘ The realisations of the /rl phoneme in RP, given in the adjoining column in that data, are [.1], [._t] and Q in 12-15. That is, there is deletion in 12-15

(compare them with the SSE forms in 12-15). Write a rule which gives the realisations and their governing contexts. If the data on p. 14 seems insufficient to you, the following transcriptions

of English words as uttered in an RP accent should help: .

(a)

/'\ _, N

www-.

[b(i:] [ph U9]

‘bar’ ‘poof

(b) (b)

ll‘)(l2.lll_]l [pl'(ld.'l-I]

‘barring’ ‘poorer’

[pl':):d]

‘poured’

4.

[f'3:tlvl

‘furtive’

[b3:ga]

‘burger’

6.

[p"ti:z1iJ]

'parsing'

2.6 The ordering of rules We have proceeded thus far by segmenting words into their component

morphs and then proposing and testing hypotheses about the phonological representations for the morphemes in question, and the rules we would

need to accompany them. We have referred to this combination of representation and rule as an analysis. What needs to be stressed now is that the notion of hypothesising and testing which we use to arrive at an analysis

is all—irnp0rtant. So much so that, in many cases, we cannot set about the segmentation of words into their component morphs without presupposirzg

an anrzlysis. This is a perfectly reasonable thing to do, as we will see shortly, but it is crucial that we notice when we have done this, and what it is we have presupposed. We must also notice whatfirllows from the claims we make

ALTERNATIONS

41

with our rules and representations. Engaging in phonological analysis is thus very much a matter of being aware of what one is claiming and what one is

committed to. We can see this in the analysis of the Following nouns in Xhosa (a Bantu language spoken in southern Africa). (8) X/Jam norms (data From Boyce 1834) Partitive

Nominative .

[udal Q I V_k

But let us assume that there is a more general rule at work here, which states that any vowel is deleted when preceded by another vowel: (13) V—>@/V____

(We will see that the order of application of this rule and our rule For the [w] and [u] allophones 0F/u/ is crucial.) In generalising the leFt—hand side of the rule to include any vowel, we

express :1 more general claim about the phonology of this language. Another way of expressing this generalisation is to say that the second of two adja—

cent vowels is elided. From this generalisation, it will Follow that class prefixes are elided when the preceding prefix ends in a vowel. Let us assume that, in checking this generalisation out against Further data From Xhosa, we find

that it is not falsified. When we succeed in getting a particular phenomenon to Follow from a generalisation in this way, we may claim to have given an

explzmatian of the phenomenon. Explanation is thus very much a matter of showing how particular states

of affairs Follow From general principles. It will be our goal throughout this book to try to show how the particular" (the dam) Follows from the general (the

ruler). Our aim, you will recall, is to build up a picture of the generalisations

in terms of which phonetic data are decoded. Notice that, in allowing for this analysis, we permit that the class prefix

morphemes, such as la! in 13 and 15, and fur’ in 1, 3 and 5. each have two allomorphs, thus:

ALTERNAUONS

( I4) la!

|a]

43

/u/

(Z)

[u]

Q?)

That is, we allow for zero morphs, where Q counts as a ‘zero realisation’ or the

morpheme. Here, we have alternations between [a] and (3, and between [ul and Q. We say that [a] alternates with Q and [ul with Q. The symbol ‘~’ is

used to express the meaning ‘alternates with’ as Follows: [a] ~ (Z) and [u] ~ (D. This analysis (let us call it Analysis A) seems perfectly Feasible. We arrived at

it in this way: the representation /ku/ was preferred over /kw/, on the grounds of the phonetic plausibility 0F the rules needed For /ku/. And the rules which went with /ku/ expressed a simple generalisation. But having decided on /ku/ as the phonological Form of the prefix, there is a question that might still be raised. We have been working with the

assumption that we analyse the words in 2, 4 and 6 as [ku + daka], [ku + buso] and [ku + ltutja], with the vowel in the class prefix deleted. We might

equally have analysed those forms as [l< + u + dalta], [k + u + buso] and [k + u + kutja], where the /u/' of the /ku/ prefix has been deleted. This would

mean allowing For three allomorphs of the partitive morpheme: [it], [kw] and

[ku]. Let us call this Analysis B. The question raised is this: is the [u] in 2, 4 and 6 the realisation ol: the /u/ in /ku/? Or is it the realisation ofthe luf in the

class prefix /uf? Under Analysis B, we would have the Following rule deleting the ful ol:/l @ / in

There is a clear advantage in adopting Analysis A. IF Analysis B is correct For 2, 4 and 6, then there is a problem with I4 and 16: there, the class prefix la], in [a —|- bantu] and [a + madoda], has undeniably been deleted. Analysis B would require us to write a rule deleting the class prefix in 14 and 16. This

would mean that Analysis B is less simple than Analysis A and less general: in A, we say simply that the /'u/ oF/l [W] applies brfiarr the rule deleting the class prefix vowel. lf 8, 10 and 12 have the phonological form /l [W] / ____i (call it Rule 2) applies before Rule 1, it will

prevent Rule I. from applying where the class prefix is /u/' (because in those

cases luf --> [w], which is not a V). This notion of rule ordering will prove to be of considerable importance to us. It is a notion which arises as a consequence of the way we have been speaking of phonology and phonetics: once we acknowledge that there is a systematic distinction to be made

between phonological representation and phonetic representation, and that phonological rules mediate between the two, the possibility of the order-

ing of rules arises. In our Xhosa case, we may view the mapping process in this way: (17) Phonological representation (//)

l Rule 2

l Rule i Phonetic representation ([

The idea is that we derive the phonetic representation from the phonological representation by applying the rules. Consequently, we refer to the process as

a derivation. This notion of derivations in phonology is of great importance; we will return to it in more depth in due course. Note the way in which we proceeded with this problem. We needed a

preliminary segmentation of phonetic representations into their component morphs in order to get started on proposing ancl testing hypotheses. We

did this, analysing forms like [kubuso] as [ku + buso], and then proceeding from there. But this preliminary segmentation prrsuppored an analysis

ALTERNATIONS

45

in which the class prefix /u/' is elided. lt is clear therefore that we do not always start absolutely from scratch, but often incorporate hypotheses into

our most fundamental starting point. This means that hypothetical notions will determine, to a certain extent, what the ‘facts’ are that we are seeking to

account for. This relationship between hypothesis and fact is of particular interest; it rather undermines the everyday notion that facts and theories are two entirely different sorts of thing. its importance therefore extends well beyond the domain of linguistics. An awareness of it will almost certainly help you to

tackle phonological problems. The following exercise requires this lcind of

awareness. D. _/apanese /r/ allophones (data from Hinds 1986)

The phoneme /r/ in japanese has at least the three allophones It], [ts] and

,_.

From the following words, say what different environments the {¥ allophones occur in: 1.

[tafsntl

‘stand’

2.

[tefsulil

‘iron’

2.

ltorurl

‘take’

4.

[Elba]

‘Chiba’

3.

[rsméil

‘dirt’

6.

[jatimici]

‘gravel toad’

(a) What is the phonological form of the word for ‘dirt’? (b) Write the phonemic rule for the realisation of the allophones of/tf. Now look at the Japarrese verb forms below and answer the questions.

E. japanese verbs (data from Hinds 1986) Stem

.

. \J"l->§.)-F -7>—' .

,—c

{matstul |fstul +cor

Ii

—|-ohs

—ant —cor

i

~—cont —ant —cor

Formulated in this way, the rule expresses the sense in which {p, la, ml and

lk, g, ti} are classes with respect to place of articulation. It does not quite

TU

PHONOLOGY

express the generalisation that the values For [ant] and [cor] in the nasal will be whatever their values are in the following stop; but we will rectify that in

a later chapter. In the following exercise, try applying the Feature-based approach to your phonemic rules which Front velar stops in Lumasaaba (see p. 25). A. Lzrmamaba The phonemic rules you wrote For the realisations of the phonemes /ld and lgf in Lumasaaba were probably as Follows:

(31) /k/—> [c] f_ Zr:

(ps-

/s/—>IJi/__[

M

l

Clearly, there is a single generalisation to be expressed here: velar stops get fronted and become palatals before non-low Front vowels. The generalisation does not make reference to voicing, but with our phonemic rules and representations we cannot express it. Your task is to express this generalisation in

the Form ofa rule written in distinctive features. Firstly, here are the phonemic vowels of Lumasaaba:

(32) high low

i +

C

3

U

U

+ +

back round

+ +

+ +

The consonant system is as Follows (with some Features omitted For ease 0F exposition):

(33) cons obs

P + +

r + +

b + +

k + + +-I-+-'1

COHI

+

high back

+

+ +

+ + +

8 + +

-l-++t/=

VOICC

cor an t

d + +

+ +

+ +-l +-I-N

+

+ + +

+

ll

J + +

+ + +

+ +

1

U1

H

+

+

+

+ + +

+

++++

+

_]1 +

+

+

FEATURES, CLASSES AND SYSTEMS

71

(11) Wt- can see From this what the Feature specifications For the velars /l [—high]

Language-specific implicational relationships depend on the system in a given language. We saw that, in Spanish, the Fricatives in the system are all voiceless. Since the Feature specification [+obs, -l-cont] picks out the Fricatives, it Follows that, in this language, [+obs, -1-cont] implies [—voice]: iF a segment is a Fricative in Spanish, it is voiceless. Thus, the grammar 0F Spanish contains an FCR oF the Form:

FEATURES, CLASSES AND SYSTEMS

73

(37) [+obs, +cont] -—> [—voice]

We shall see many more such relationships, and in due course we will con-

sider them in more derail. In adopting such FCRs, we will take it that phonological representations in a language are left unspecified for values of such features, and this will considerably reduce the amount of redun-

dant information present in the sorts of large feature matrix we have been presenting. There is another sort ofimplication which involves what are referred to as ‘unmarked’ states of affairs. For instance, sonorants are almost always voiced in human languages; we may say that they are voiced ‘in the unmarked case‘

and leave them unspecified for a value in underlying representations. We may express the implicational relationship between the two feature specifications [-obs] and [+voice] in a default rule: (33) [—obs] —> [+voice]

By “default specificatioif, we mean a specification which applies unless there is some overt statement to the contrary. While it is true that sonorants are

voiced in the unmarked case, there are other cases where the notion ‘marked’ proves problematical.“ For instance, among the Lowland Scots vowels as given on p. 59, it is clear that ifa vowel is {+ round], it is [+bacl [—ruund]

One of the difficulties with this notion offinarltedness’ is that there are often many ‘marked’ cases in languages which have been stable over long periods. The implicational relationship stated in (39) is not true of French, for instance: both front and back vowels may be [+r0und] in French (and indeed

in many languages). Furthermore, the vowel we have represented as /u/' in

Lowland Scots is very much fronted in certain accents. We will return to the idea of default rules in Chapter 8.

We have now introduced all of the features we will be using; we have assumed that there is a universal set of features which each language may

74

PHONOLOGY

draw on, although we have omitted several from the set, which characterise segments we will not be considering.

Exercises I. Pofis/a In Exercise C in Chapter l {see p. 18), you formulated a phonemic rule which

devoiced /r/, let us say, when preceded by a voiceless consonant and followed by either a word boundary or another voiceless consonant: # /r/' —> [i)"]/ voiceless consonant

{ voiceless COl’]S()fl21I1[

Given the feature specifications for Polish consonants listed below, write the rule in features (assume that all the consonants are [—syll]). r-\

/—\

v

»

pbtdkgtsdzcjcj

v

»

I

1

~

fvsasiszxmnnrllw]

C0ll'l1S sy obs

+—l—++—l--l-+-l--l-+++-F-l~++++-l~-l-+++++--~- - ~ - ~ - - ~ - - - ~ ~ ~ - - - - - ~ ~ - - - - - — -+++-l-+++++++-l-++++++-l-++ — — — — — — —-

CODI

— — — — — — — — — — ——'l'+-l"'l"-l--l"-l-+‘l-———'l'“l"+++

Cl€ll'€l — — — — ——-l--l-+-I--l-+++ — — — — — — — — — — — — — ——

Sifi

— — — — ——++-l—+++——++++ — — — — — ~ W - — —-

Voice—+—+—+—+—-+—+—"l--"l"-"l"—"l"—++*l"""l'--l"+'l'+

ant ~l-++'l———+~l-————--l—+++»-———+-i-e~+++e* cor ~—+—l--—+—l-++-———+++-l-————+—+-l--l-—~ bilfik -l--I-+++-l--l-+++——+-l~+++-l-——-l--I-+—+-l-++pigh ~————++-—--++ — — — — ——+++——+———l--l--low — — — - — — — — — - — - - - - - - - - - - - - - - - - --

lat

- - - - - - - - - - - - - - - - - - - - - - - --++--

nas

— — — — — — — — — — — — — — — — — — — ——+++———~~

Pbozrzetic flare

For reasons we will pursue later, all [+ant] obsttuents in Polish are characterised as [—l-back], i.e. with the body of the tongue retracted. This is not

relevant for this exercise. /in interlude cancerning method Your rule will need to pick out, with as few features as possible, just the segment /t/. How does one go about this task of arriving at a set of features

which pick out just the segment(s) which undergo a rule? As a first step,

FEATURES, CLASSES AND SYSTEMS

75

you should think in universal terms. Classes and subclasses of segments can be thought of as follows: is the set of segments a class, of consonants or of

vowels? If it is clearly a class of non-syllable, consonants, it will be [—syll]; if vowels, [+syll]. Let us say the class in question is [—syll]. Thar class can in turn be broken down into the two classes of obsttuents (l+obs]), on the one hand, and sonorant consonants (which are [~obsl), on the other hand; these are usually

taken to include the nasal stops and the ‘liquids’ /r/ and /l/. Let us then say the class in question is a class of obstruents. We can then

ask: is it a class of fricatives or not? If so, then [+cont]; if not, [—cont]. Ifit were [—cont], we need to ask whether it is a class ofjust stops or ofstops and

affricates. For the former, it will be designated with [~del rel]. Similarly. if the class is a class ofsonorants, we will have to ask whether it is a class of nasal stops or not; the feature [nas] would work here. and in some cases the feature

[cont], since all nasal stops are [—conr]. We can represent this set of classes and subclasses like this:

|- syll] (consonants)

|+ nl>s| (nlisrrticiits)

[+ cont] (fi'icati\'t's)

l— cont]

1+ syll] (\'u\vcls)

[—obs] (sonorants)

| + [ms] (nasals)

|_.. nus] (liquids)

2% |— tit-1 rc]_| (strips)

| + dc] rel] (affricatcs)

Let us say weive arrived at [—syll, +obs, —cont, -del rel] (stops) as the lefthand side for the rule. Since [+obs] implies [+cons], we can omit (+cons],

on a universal basis. That is the first step, and it gives you a universal feature characterisation

of the class, or single segment, you are interested in. But there are no human languages which have all possible human language segments in their phono-

logical inventory. So the next step is language-specific. Thar is where you look at the feature matrix for the language in question. lfyout rule applies to stops,

but there are no affricates in the language, the feature [del rel] can be omitted from the rule. lf there were no fricatives in the language (an unlikely event, but possible), then [wont] could also be omitted from the rule, giving you

l+obs] alone. When you are dealing with a single segment, rather than a class containing more than one member, as in this Polish case, you would expect to have to

76

PHONOLOGY

use more features than you would for larger classes. But this need not be the case (in English, the lateral approximant alone can be picked out using just

the feature [+lat], for instance). And in that kind of case, there will often be more than one means of picking out the relevant segment; this Fact simply Follows from the way phonological systems are organised. You may well be

influenced in your choice by a desire to bring out the phonetic motivation of the rule. 2. Hungarian (data from Vago 198021)

For the Hungarian exercise on p. 37, you needed six phonemic rules which showed that each of the voiceless obstruent phonemes, /p/, /t/, /l/ and /situ/. (i) Using the features [high, low, back, round], and [ATR], draw up a

feature matrix which distinguishes all nine vowels.

THE ORGANlSATlON OF THE GRAMMAR

115

Now look at the following data: 1. [o~siti-—i]

“he pierced it'

2. [3-cIrE—t]

‘he showed it’

3. le—b11—ol

‘nest’

4. l€—ho -1)}

‘stone’

(ii) Each of these forms consists 0F a prefix, a root and :1 suilix; list the

allomorphs For each prefix and suil-ix. (iii) Devise an underspecified phonological representation For the prefix

whose aiternants are [0] and [J]. (iv) Write a vowel harmony rule which will yield all of the prefix alternations,

i.e. [el ~ [C] and [0] ~ [0] (v) Because of the nature of the representations and rules of standard GP,

rules cannot take the Form: X—>Y/_____Z_,M. Nor can they take the Form X—>Y/Z. Bearing this in mind, write a single rule, using the brace notation, which will yield the suffixes and prefixes.

(vi) There is something unsatisfactory about the rule, as an attempted expression of the generalisation. Can you say what it is?

Notes 1. The ‘#' (as opposed to ‘-P) in ringer‘ and ended is explained on p. 103. 2. The notion oi’. ‘cyclicity’ has a long and Complex history in generative linguis-

tics. lt has had considerable appeal in syntactic theory and in the treatment oi stress assignment in English. as begun by Chomsky er rzl. (I956) and developed by Chomsky and Hallc (1968). The latter work (SPE) assumes that most oi the phonological rules oi‘ a language, both segmental and stress assigning,

apply cyclically, though the excmplification they give concerns stress assignment only. Anderson (I974: Ml) claimed that the putative cyclic application of segmental rules is less convincingly established than that oi the stress assignment rules, since cyclic application is most clearly motivated where syntactic structure in general and lexical category in particular are important (which is often the case with stress assignment titles). Many phonological titles, he points out, are not sensitive to such Factors. and with those the idea oi‘ cyclic application seems less useliil. However, the idea that many segmental phonological titles are indeed cyclic gained considerable ground within the theory of lexical phonology.

which we deal with later. For an overview ofcyclicity in generative phonology, sec Cole (1996).

Further reading Wliile Chomsky and Halle (I968) remains the classic text for the SPF. model. it is not easy reading For a beginning student. The reader is advised to begin with the texts on generative phonology given at the end ol'Chaptct 4, before proceeding to Chomsky

115

PHONOLOGY

and Halle (1968). Kenstowicv. (1994) is a more detailed introduction to generative phonology than was Cart (1993), but there hoe been no second edition, so it does

not cover Optimality Theory. which is dealt with in subsequent chapters of this, our second edition. Another more detailed introduction to generative phonology is Roca and _lol1nson (1999). which clues cover Optimality Theory. For a general introduction to generative linguistics, see Smith and Wilson (1979), which sets classical generative

phonology in its context and serves as a good, easy-to-read introduction to the literature on generative linguistics. Students who would like to consider an alternative to

generative phonology can consult Silverman (2006).

6 Abstractness, Psychological Reality and the Phonetics/Phonology Relation 6.1 Ordering relations and rule application in the SPE model In the SPE model of generative phonology, it was assumed that the phonological component of the grammar was an ordered block of all oi:

the phonological rules 0F the language. The grammar" was taken to be mentally represented, that is, it was assumed to constitute linguistic knowledge

stored in the mind of each speaker of the language. The rules and representations postulated by phonologists were therefore taken to be psychologically

real: phonological theory was interpreted as theory about mental states, perhaps even mental processes. Since the rules were ordered. different kinds of rule interaction were possible. Let us briefly review some of those kinds of

interaction. We begin with the Russian data presented in Chapter 5:

(1) Phonological form

Phonetic representation

l.

#gorod —|— lc —l— a#

Igorotltal

‘little town’

2.

#misensk#bi#

[mtsenzgbi]

‘if Mcensk’

3.

#m't§€nsl [otvoice] /___(#(#)) [+obs, uvoicc]

This means that an obstruent will assimilate in voicing to a following obstruent, independently of whether there is an intervening morpheme boundary,

or a single or double word boundary. But what olthe cases, like (2) and (3) on p. 117, where more than two obsttuents are involved? Here, we must allow that when the rule ‘applies’ (in the sense of ‘gets the chance to have its con-

ditions met’), it may ‘apply’ (in the sense oF‘havc an ellect’) more than once,

like this:

(4) # mi:§€nsk## VA VA Phonetic representation

mfiensg mtscnzg lmfsenzgbill

bil# bil l7}l

In SPE, where a rule was allowed to have an ellect more than once, it was

said to apply iteratively; SPE allowed For iterative rule application. This had consequences For the idea that the grammar constructed by the linguist should be taken to have ‘psychological reality’. ll‘ speakers possess mentally

represented rules and representations, we could assume that their phonological representations are stored in memory. But what oil the rules? Are these to

be interpreted as mental processes that take place in real time as speakers pro—

duce and interpret utterances? If so, then do we assume that a Russian speaker has a stored representation ofthe form given in (4), and that two mental pro-

cesses take place, corresponding to the double application of VA, resulting in the uttering of the phonetic representation in (4). Some phonologists, such

as Bromberger and Halle (1989, 2000), have defended this position. Others, such as Hooper (i976), have argued that this is implausible. Hooper later

developed her idea (published under the name oi‘ Bybee). We will return to

those ideas below. Now let us look at the ordering of the two rules we have postulated. In SPE, it was said that two rules were aflecting if they had overlapping inputs, conditions or structural changes. Since these two rules have overlapping inputs

(both pick out the class 0F obsttuents) and overlapping structural changes (they both change values on the feature [voice]), they are clearly affecting. This means that the operation of one of them will have implications For the operation of the other. \)(/hat we need to ask, then, is how they are ordered; that is, what the direction of influence is. We can see from the morpheme

ABSTRACTNESS AND PSYCHOLOGICAL REALITY

119

/mozg/' (‘brain) that \)(/—FD precedes VA: without affixes, the phonetic form of the morpheme is [mosk], with the following derivation:

(5) # 1nov.g# W—FD

I1‘101.l
[+high] /_ l+syll, —bacl [+cor. +stril/ g [—cons, —back]

If the underlying form of the morpheme meaning ‘building’ is /gmax/, and the phonetic form of the augmentative ('big building’) is [gmasisko], what

120

PHONOLOGY

order do the rules NSP and FVP apply in? Draw up a derivation for the

augmentative form.

Derivations for both lgmaxf and fgmax + isko/' look like this:

(9) lgmaxl

/gmax -5- iskol

NSP

___

._..___.i

FVP

_.___

grnasislco

Phonetic representation

[gmax]

[gmasiskoi

What this ordering expresses is the fact that where post-alveolar [g] derives from /X/ via FVP, it does not undergo NSF. What is clear is that FVP and

NSP are affecting: FVP can create input for NSP (it can yield [g], the input segment for NSP). That is, FVP could feed NSP But in ordering the rules such that NSP precedes PVP, we prevent this from happening. Such an order-

ing was referred to in the SPE model as counter-feeding order. Note that this presupposes a restriction on the application of the two rules to the effect that, once a rule has applied, it may not subsequently reapply if its conditions are met later in the derivation. In the SPE model. it was also necessary

to allow for cases where a rule may reapply later in the derivation. Anderson (1974) addressed these sorts of problem for the standard model of generative

phonology. We have seen here only two sorts of ordering relation; others were possible in a model such as SPE, based on rules and representations. One such further ordering relation can be seen in the relationship between two rules in Lithuanian: the rules of Degemination and Metathesis. Consider

the following (data from Kenstowicz and Kisseberth 1973): (1 0) Past

Imperfect plurai

IPS future

Gloss

{lease}

[kasitel

[kasiui

‘dig’

[gore] [yteko]

[getkite] [tekite]

[gersiul [teksiu]

‘drink’ ‘How’

The form [kasiu] derives fiom /kas + siu/, and the form [tekite] derives from

/tek + kite/: the rule ofDegemination deletes the first of two identical obstruents. Now consider the following data, relating to the morpheme /dresld (‘to bind’): (51) Past

Imperfect plural

IPS future

Infinitive

[dreske]

[drekskite]

[dreksiu]

[dreksti]

In this data, the Metathesis rule reverses sequences of a fricative followed by a velar stop, where they occur before a consonant. The rules of

ABSTRACTNESS AND PSYCHOLOGICAL REALITY

121

Degimination and Metathesis affect each other, since both concern sequeiiccs

of obsttuents: Degemination has an obstruenr in its input and in its condia tions, and Metathesis has obsttuents in its input and its structural change. The derivations are as follows:

(12) Past

Imperfect

IPS future

Infinitive

plu ral

Metathesis

/drcsk + e/ _i___

/dresk + kite! drekskire

/dresk + siuzl clmicssiu

/dresk + til dreksti

Degemination

i___

i____

drel [trezlil ‘sohcr'

This creates an ordering paradox lot the standard model, since W-FD

normally precedes (and Feeds) VA, as in:

(17) W-FD VA Phonetic Representation

# mozg# mozk mosk [musk]

(‘brain’)

but W-FD appears to Follow VA in (18):

(1 3) # tre'1.v#

VA

———

W-FD

i rcrf

Phonetic Representation

ltrt-2.f']

(sob-er’)

Kiparsky took ‘/v/’ to be a sonorant underlyingly, i.e. /w/, which is therefore unmarked for voicing in lexical representations. Following Hayes (1984), he

160

PHONOLOGY

postulated a rule of W Strengthening (W5) which converts the lwl into an obstruent.6 This rule may apply only postlexically, since it violates Structure

Preservation in introducing a distinction (between a labial sonorant and a labial Fricative) which is not present underlyingly. Witli the rules Formulated and constrained as stated, let us see how they apply to the problematical cases which give rise to the ordering paradox.

( 19) IT] O Zg

trezw

(/w/ lacks a value for lvoicel)

Lexical Phonology

WFD VA WS DV

mozk

(May not apply to sonorants, lcxically)

moslc

(May not apply lexically) (May not apply lexically)

Postlexical Phonology

WFD VA W5 DV

i i Z WW

trczwy

(lwl acquires the value '-’ For Ivoiccil)

_

(/w/ already has a value)

Phonetic Representation [m0sl —>

‘water’

fur + a/

—>

‘dog’

1'caltur + a/'

-->

/pclota + ka/'

lpelotakal

[gi_|'ona] llayuiicl lure] lcalturcl

/gifon + ak/' /lagun + alt! /ur + al

[cimista]

‘lightning’

What reason can he given by the theory of Lexical Phonology For the Failure of the rule to apply in these cases?

(ii) This language also has a rule which affects voiced coronal stops in intervocalic position, such that /d/ ——> [r] /V__V. It applies in the Following sorts of case: /hide/'

lbircl

‘path’

/ari + ka + da/

l‘arilr

(i) Does the rule apply at Level 1 or Level 2 oi the English lexicon? (ii) What claim does this commit us to concerning the level ofapplication of

Nasal Assimilation in English? 3. German in Standard German, the voiceless velar Fricative [X] occurs only after back vowels (ignoring a Few loanwords From Slavic languages; data from Hall

1989);

1. lbuzx]

‘book’

2. [jprux]

‘saying’

3. llcoxj 5. [nu:x|

‘cook’ ‘after’

4. [hozxj 6. [box]

‘high' ‘brook’

The voiceless palatai Fricative [g] occurs after Front vowels: 7.

[zi:c_|

9.

[peg]

‘bad luck’

[hyzglaenl

‘booklet’

1. 1 .

_

‘sickly’

8.

[ig]

‘I’

10.

[go_['p1'c:c]

‘conversation’

12.

[g\>l'Y§ii'J:l

‘rumours’

The data in I-»I2 suggest that the two are simply in complementary distribution. Note too that the palatal Fricative occurs word-initially and afier

sonorants:

THE FIOLE OF THE LEXICON

163

1 3.

[giro rk]

‘surgeon’

14.

lcemizj

‘chemistty‘

15.

[mic]

‘such a‘

16.

[dotc]

‘through‘

17. lmuncl

‘many a’

This has led some to suggest that the underlying Form is /cl’, with a lexical

rule oF Fricative Assimilation which changes /c/' from [—baclc.o>i,u> r>l> |n,n>$l,v,r.,3>(i,fis,j‘>b,cl,g,>p.t,l
[+nas] /___ [+nas] {

Nothing in the theory of distinctive Features, as we presented it, suggests that # and C form a natural class, and there is thus no reason why they should

figure together in the rule. We therefore suspect that there is an unexpressed generalisation lurking behind the rule as formulated. An alternative, syllablebased, generalisation, which involves no appeal to Word boundary or to a

consonant Following the nasal stop, would be to suggest that a vowel nasalises if immediately followed by a nasal consonant in its rhyme. This means that

only the following structure (in which we anticipate section 8.4 somewhat) will result in nasalisation:

(10)

C

1'i:asr’

(b) (h) (h)

l l

b

O

0

21>

U

O>:11

+

O

N

Co

N

N

(Io

e

t

9

i

2

9

Schwa deletion O

C)

0

R

+

’\ N Co

b

e

t

O

21>

R

N

Co

i

2

REPRESENTATIONS HECONSEDERED (I)

175

Resyllahfzicarian 0_f'/t/

c

O

+

N

0; O

A,

21--

=1ix l)

e

t

i

2

Lowering: conditions not met.

ln giving the derivations above, we make several assumptions. We assume

that syllable structure is not given in underlying representation, but that there is a level oF initial syllabihcation and that subsequent resyllabifications

may take place. We also assume that the /ti’ in bat» is incorporated into the

rhyme at the point of initial syilabification and that Lowering Follows Schwa Deletion and Resyllabification. We do not have the space to examine all of

these assumptions, but we will consider the interesting notion that a consonant may be resyllabified into an empty onset slot. Let us approach this

idea by briefly looking at a closely related notion: the Maximal Onset prin— ciple, which is said to hold for languages in general. The principle states that,

with a given string of segments in which the consonants may in principle be syllabihed in more than one way, syllabification will take place such that

consonants which may occupy either rhyme or onset position will occur in

the onset rather than in the rhyme. Take the English word Em-ick: the /r/ must be syllabified in the onset of the second syllable, since coda sequences

/tn’ are disallowed by the phonotactics of the language. But the /ti’ may in principle occupy the coda slot in the first syllable or the onset slot, along

with the /r/, in the second, without violating the phonotactics 0F the language. Maxiinal Onset dictates that the /'t/ will occupy the onset position, gl\/lllgi

(17) 0

o

/\

/\

O

R

O

R

176

PHONOLOGY

This example concerns initial syllabiflcation, but the tendency to fill empty

onsets applies equally where resyllabification takes place. Let us return to our French example bétise: we assume that the suffix /iza/' has, at the level

of initial syllabification, an empty onset and that the /t/ in éére may ‘shift’ from the coda position to fill the empty onset position in /izo/. This process is reasonably called resyllabification: it often occurs in cases where word

formation may result in changes in the syliabic structure of morphemes; we will look at some cases in English in the exercises at the end of the chapter. Where this resyllabification occurs, the /e/ in /bet/' is not in a Closed syllable, whereas in bare, resyllabification does not take place and

the /er’ is in a closed syllable. The neutralisation rule which lowers /e/ in closed syllables therefore applies at a level after that of word formation and resyllabilication.

The resyllabification notion embodies the following claim: that speakers of French will syllabily sequences of the sort CVCVCV, as in a'es nmis

([dezami]), into a sequence of three CV syllables, rather than, say, into one CVC syllable ([dez]) followed by a V syllable ([21]) and a CV syllable ([mi]). This claim is persuasive. If you listen to speakers of French placing emphasis on each syllable of their utterances, and thus imposing pauses where they would not normally occur, the resulting utterance differs in at least one

respect from its English counterpart, as far as syllable structure is concerned. An English speaker uttering ‘Nor a ward.’ ’ with primary stress on each word

will utter it as [notlaelwfizdl (where ‘|’ indicates a pause). A French speaker uttering ‘Pas un mot!” with the same emphasis will, however, say [pa|z§lmo], with the sequence clearly syllabifiecl into a sequence of three open syllables,

i.e. with liaison of the /z/‘ of /paz/ into the empty onset of the Following syllable. Similarly, when French speakers hesitate between determiners (such as [er and mes) and nouns, and produce a hesitation vowel, it will often precede the liaison consonant, as in ‘Les..euh..Espagnols‘: [le.oe.zt:s.pa._]'iol],

with the laison consonant /2/ syllabified into the onset position of the syllable [zss]. This brings us to the role that syllable structure plays in the phenomenon

of liaison in French. There is clearly a difference between the final consonant in words like /aére (as well as the Final consonant in words like bet, met and rec, which have no underlying schwa word-Finally) and the word-final con-

sonant in words like mes. The latter participate in the liaison process: they are not realised phonetically before a pause or if the following word begins with a consonant. The former, on the other hand, are realised phonetically in all environments. This fact has consequences for the Lowering rule, since,

among that class of words with a consonant which does participate in liaison, some, like mes, have invariable [e], whereas others have invariable [8],

as in mrlis. Let us therefore examine the phenomenon of liaison from the point of view of syllable structure and then look at its consequences for the Lowering rule.

REPRESENTATIONS RECONSIDERED (I)

177

8.3 Extrasyllabicity, the CV tier and abstractness The idea oi preference For maximal onsets during the syllabification process leads naturally on, we have seen, to the notion of moving a consonant

into an empty onset slot. That notion allows us to characterise liaison in French in a rather simple way. Here are some of the data which exempliFy the phenomenon:

(13) mchu . (E1107;-lI‘flll Idesc l1:13,!

I do/,0.1 15

(mes /mus/Jrms (mes tzrnis) (c/as re’ .-'1Xes) ales ofrt" om

‘nu Y corks’ ‘m Y Friends’ ‘ 8 ta P e varieties’ ‘onionsi

u1-s’>'~.>~Jt~J —-

llCVl_]l|

([65 I/ignes)

‘the vines’

6.

Ilezoel

(Its aeigfi)

‘the eggs‘

7. 8. 9.

(nohutc'l l [nom| [nol_i

~
\V

s

w

s

W

s

A

ra

be

H9

shot

S

w

s

the hat

lot

Such a structure P resu PP oses P rinci P les whereb Y feet are assig ned relative prominence in words, phrases and clauses. For that purpose, a rule is needed

for the location of sentence stress, and the metrical versions of rules like the NSR and PSR need to assign prominence to the correct syllables in polysyllabic words, com ounds and |

1

P

I

hrases. For instance, in the Word Araoella, we P

u

Will need to assign the right foot more prominence than the left, as follows (where ‘m’ designates a phonological word, often referred to in the literature as imoti):

REPRESENTATIONS HECONSIDERED (l)

197

(45) IT! .-4-"X-5-“““-—_.

\\-'

S

S

\V

S

\\'

1\

ta

bel

la

For this purpose, a rule entitled the Lexical Category Prominence Rule was proposed by LSLP which, subject to certain conditions, assigned prominence to the second Foot, so long as it branches, which it does in this case. The rule

was augmented by a set of conditions, which we do not have the space to examine. Nor do we have the space to examine all of the other metrical rules needed to arrive at the correct representations For words, phrases and clauses

in English. Let us instead move on to another aspect of rhythm and a means of representing it. (We will return briefly in section 8.7 to the matter ofwhat each level ofthe structure in (43) might be, and how the appeal to such levels

might prove Fruitful.)

8.6 Symmetry, clash avoidance and the metrical grid In our discussion of syllable structure, we referred to the idea oF phonological rules imposing, or executing, the phonotactics of a language. As examples,

we cited Elision and Glide Formation in French, which may be said to serve to eliminate sequences of vowels disallowed by the phonotactics. IF

the phonotactics 0F a language are prefizrred structures, then the idea can be extended to metrical phenomena; that is, a language may have preferred

metrical structures, which also might be enforced by the rules which affect metrical structure or by principles or rules which restore preferred structure which has been disturbed by the application of rules. This notion of prefer-

ence for phonological symmetry is reflected in many languages. In English, For example, it is often claimed that structures with symmetrically spaced Feet

are preferred. Another, Frequently cited, example of preference For symmetry involves the symmetrical sequencing of a high tone Followed by a low tone in tone languages.” In such languages, phonological rules which delete vowels, operating on the CV tier, will often do so while leaving intact the

preferred tonal contours. For instance, if a language has a phonotactic constraint Forbidding a sequence of two high tones, and a sequence HLH (where ‘H' = ‘high’ and ‘Li = ilowi) would be transformed into a sequence NH by

a rule which deletes the first of two adjacent vowels, then the H LH sequence is often preserved (say, by Formation of a ‘contour' tone on the first vowel,

representable as HL), thus:

198

PHONOLOGY

(46) III.ll—%lll.H

l l l cv+vcv

\/ l cv+c v

This latter phenomenon, whereby a preferred sequence is maintained on what

we can call the tonal tier, independently of the operation of a rule affecting

the CV tier, has often been accounted for by appeal to the principle known as the Obligatory Contour Principle (OCP). 1" Although this principle started life as a description of a purely tonal phenomenon, the expression ‘OCP

effect’ is now used by phonologists to refer to cases where some preferred, symmetrical structure is enforced either by phonological rules or by general principles (which may come into operation when the application of other

phonological rules result in asymmetrical structures). This enforcement of

symmetry usually entails an appeal to the idea of clash avoidance (in the case we have just cited, the avoidance of a clash where two H tones end up adjacent to each other).

We may take the same view of the rules of Elision and Glide Formation in French, and indeed of the liaison phenomenon. We could regard a sequence of two vowels, in certain circumstances, as constituting a clash, and take the

Elision and Glide Formation rules to serve to repair that clash. Similarly, the liaison process could be viewed as a means of obviating clashes of the sort CC

in certain circumstances. In each of these cases. the restriction is on sequences of identical objects, whether Cs, Vs or tones of a particular sort. We might even extend the idea of clash avoidance to cases of segmental dissimilation in

sound changes, where, for instance, they function to eliminate sequences of two nasal stops. Thus, in the history of Spanish, /aomrze ('m¢m’), from Latin

/crominem, changes to /romre (and later to hombre). The matter is considerably more complex than we have suggested. (ln French, for instance, it is simply

untrue to say that any sequence of two Cs or two Vs, even word-internally, is disallowed.) But the fundamental idea does concern avoidance of sequences of identical phonological objects, and there is very probably good perceptual

motivation for such an avoidance.

Let us now see how the notion ‘clash’ applies to metrical phenomena. It has often been observed that there are, in English, many cases where a word alternates between two prominence patterns. Take the words antique, thirteen,

unknown, Dundee and Yiwnersee. All have main stress on the final syllable, both when uttered in isolation and when uttered such that they do not premodify a word with greater phonological prominence, as in: judy swam no/zed

in the Drrrzessee, His record collection is positively antique, Ron /rad three glasses qfwine, whereas _/rzc/tie /md thirteen and Sandrirzri w/Jereabozits in Prague are

unknown. But when they occur in a different syntactic context, such that they immediately precede (and usually modify) a word which is phonologically the

REPRESENTATIONS HECONSIDEHED ll)

199

most prominent in the syntactic structure, a shift of prominence takes place: it is striking, and intriguing, that in antique table, thirteen glasses, un/mown

artist and Tennessee Williams, the words amique, I/airreen, unknown and 72”nessee have a first syllable which is more prominent than the second. The

phenomenon is not restricted to words which normally have a final syllable which is more prominent than the first. Thus, in rrmdemir, the penultimate

syllable is the most prominent, the antepenultimate syllable is unstressed, and the first syllable has secondary stress. But the prominence patterns in amrlemir are reversed in the phrase acadenzic writer, where the stressed syllable in writer is the most prominent, the First syllable in acrzdenzic is the next most

prominent, and the penultimate syllable in academic is the least prominent

stressed syllable in the phrase. There are three stressed syllables in this phrase, but they have dilletent levels ofpromillence. The question arises: what is the

generalisation underlying this pehnomenon? To answer this question, we will consider a means of representing such aspects of metrical structure, known

as the metrical grid, originally proposed as an augmentation on metrical

trees which would clearly bring out violations of preferred metrical patterns (clashes) in a language. The grid For zmiinown would be given as:

(47) A-

unknown :§:

:i
F). informally,

IDENT-3. says: ‘do not change the features of /a/’. Constraints are universal, so both *a and IDENT-a will be found in all grammars, but a language

where /af is realized as [a] will have the same two constraints in the opposite order: F >> M.

Tableaus read from left to right. with the dominating constraints on the left. Violations are indicated by an asterisk. Exclamation marks indicate fatal

violations and the cells to the right are shaded to indicate that the information they contain does not play a role in the evaluation, i.e. the decision has been made on the basis ofthe information contained in the white cells. The hand

W indicates the optimal candidate, the winner. lt is clear to see that in an F >> M scenario, the faithful candidate is selected as the winner (compare

Tableaus 11.1 and 1L2).

260

PHONOLOGY TABLEAU 11.1

MARKEDNESS OVER FAITHFULNESS

lnput /a/'

Markedness: *a

(fiiirhfull lal '5'

Faithfulncss: 1DF.NT—;1

>k l

(optimal) [bl

*

TABLEAU 11.2 FAITHFULNESS oven MARKEDNESS lnput la! f W‘ (faithful and optimal) la]

Faithfulness: IDENT-21 i W i

lb]

Markcdness: ‘:1 _ *

=1= E

Consider a more elaborate example: (1) French

l: / film! 1:1‘ imnl

Q9

film] mn]

‘film’ ‘hymne'

English

1: / film/' l: 1' hlmn I

O: lfllmlfilm O: [him] /qymn

Let us begin with French film and /aymrze. Tableau 11.3 compares inputs

lfilml and / imn f to sets of candidates, in theory infinite, but here limited to

three of the most plausible. input forms end in a consonant cluster: outputs a attempt to resolve the cluster by simplifying it through deletion of the last consonant. By doing so, they violate a faithfulness constraint MAX-I0 which, paraphrased informally, says: ‘all segments in the input must be found in the

output’. Candidates b do not diverge from the input but they do violate a marltedness constraint against consonant clusters; ‘(COMPLEX says: ‘no corn-

plex codas’. Additionally, [imn] - but not [film] e violates a SONSEQ constraint on the sonority differential between two consonants in a cluster, i.e. sequences of two nasals, like lmn], constitute an undesirable sonority plateau ([m] and

[n] pattern together in the sonority hierarchy, as explained in Chapter 9, section Z; an [mn] sequence is undesirable because syllables should see their

sonority increase from the margins to the peak). Finally, candidates c attempt to resolve the cluster through epenthesis; thus, they violate a constraint DEPIO which says, again informally: 'all segments in the output must be found in the input’.

These four constraints are listed in a given order, again from the most

powerful to the left to the least powerful to the right. But not all rankings are crucial; for instance here there is no reason, on the basis of these data, to rank

DEI‘-IO and MAX-IO! a discontinuous line indicates that these two constraints are unranked (and could have been written in the opposite order). But the ranking between these two constraints and *coMPLF_x is crucial, otherwise all three candidates in film would have been ruled equally good and no single winner would have emerged.

OPTIMALITY THEORY

261

A continuous vertical line indicates crucial ranking. The linal order of

constraints can be summarised linearly as: DEP-I0 , MAX-IO >> SONSEQ, *co:vu'i.Ex. where the comma indicates unranl> indicates crucial ranking. Candidates a and c are soon eliminated for violating high-

ranking constraints and candidate b emerges as the winner, even though it violates a low-ranking constraint. TABLEAU 11.3

i

FRENCH FILM AND HYMNE

input a. liar!‘ b.

llilml r [fill [film]

DEP-10

c.

[filom]

-x-

input a.

/irn nil [imi]

oer»-to

c.

MAX-IO

SONSEQ

=i=c0MPLEX

*

=i=

MAX-IO

*

soussq

=l=COMPLEX

at

-it

*-

t lies) " +7 * l imnn 1

I

a_—-a _l—-oa4¢l

_ _ _ _L_ - c- _

i.__L_-_,

Now consider English film and lg/nm: TQBLEAU

| l

i11putTTTTT/T|lim/ .

‘i'w

W h. c.

1511

[film] [ftlom]

T input Tizflrilfllillf

-as i .

TTT Tb.

rims?

T [lnmn]

c.TTTTT[Vft|m£)n]

ENGLISH FILM AND HYMN

T Eoiiéiag

so ,_

stilts

giaso

cl

|

‘COMPLEX

outc

c

*

*!

Tisoifseo T >:= !

oeriiioii

1

i\iAx§i0 T T "coivri>i.r;x T

T

T E Tiiwd if

Ti

if T TT

at

E

E

ln the standard version of OT, all constraints are universal. So, as we switch to English in Tableau 1 1.4, we must find the same constraints again, but they

may apply in different orders. Indeed, English leaves untouched the final cluster in film but simplifies the final cluster in /rymn. This means that tolerance for final clusters is generally good, but sonority violations are more severely penalised. Thus, SONSEQ will be found closer to the left edge of

the tableau than it will in French. Furthermore, the English solution to bad sonority coda clusters of this type is deletion as in [him], not epenthesis as

262

PHONOLOGY

in * [hmmn]; accordingly, DEP—l0 must dominate MAX-IO (penalisation of epenthesis is more severe than penalisation of deletion). The order SONSEQ,

oer-10 >> MAX-IO >> *c0MP|.Ex is consistent with the grammar of English. It is often the case in OT that even optimal candidates violate constraints. This sets OT apart from most of the other constraint-based theories men-

tioned earlier. In this example, all winners violate at least one constraint. Although the constraint ‘COMPLEX plays no determining role here, it must be included and will become more active in other languages.

11.2 The logic of output-driven models in the short history of linear generative phonology, it soon became clear that rules sometimes seem to conspire in order to prevent a certain state of affairs (e.g. sequence of segments, co-occurrence of features) from sur-

facing. Seemingly unrelated processes often serve the same purpose: to defeat marked configurations (see Kisseberth i969 on Yawelmani, and Dixon 1977 on Yidiny). While examining any single rule gives no clue towards uncover-

ing that goal, the common effect appears clearly when the outputs of several rules are compared. Consider first the classic example discussed in Kisseberth (1969). ln

Yawelmani, rules of vowel insertion and rules of deletion ensure that every consonant is adjacent to a vowel. The first rule in (2) deletes the initial segment in /harinl, the second rule also deletes the initial segment in /hnil/'

(but the phonotactic context is different), while the third rule inserts a vowel before the final segment in /ilk/. (2)

Ce» (Z)/CC+_

hall + harin + i:n

C—> Q)/C -I-Z C

gitin + hnil + a + w gitinnilaw ‘hold under the arm‘ (locativc)

Q)—> V/C‘C {#,(l} Tillt + hin

hallatinim ‘lift up’ (future) ?ililchin

‘sing' (aorisr)

Individually, these processes seem to be unrelated; together, they conspire to

prevent sequences of three consonants. Accordingly, Kisseberth proposed the concept of output constraints, the role of which could be compared across languages. This is precisely what OT does ~ albeit differently since OT constraints are violable — when it recognizes a markedness ‘(ICC constraint,

which would rank high in the grammar ofYawelrnani. As a further illustration, consider the case of Umbundu (Pater I996). A linear generative analysis of Umbundu would recognize the following ordered

processes: l. Nasals delete before voiceless fricatives.

2. Voiccless segments delete after nasals.

OPTIMALITY THEORY

263

By rule l, /N + seval becomes [seva] ‘l cook‘ and by rule 2 /N + tumal

becomes [numa] ‘I send’. (N is a placeless nasal, adjusted homorganically prior to t-deletion.) The combined effect will be to prevent sequences of N + consonant from ever surfacing. An OT tableau will place a markedness "NC-v¢@ constraint in a dominating position in Umbundu.

While DT fails to explain why various rules consistently target similar configurations, OT exploits the so-called ‘homogeneity of target!‘ heterogeneity

of process’ concept in a way that makes typology relevant and instructive; in other words, the same goal is achieved across different languages, but the means by which it is achieved can vary considerably.

We saw in Tableau 1 1.3 that sorvssq and ‘COMPLEX are markedness constraints, while DEP-IO and MAX-IO are faithfulness constraints. A great deal of phonological activity results strictly from the tension that exists between markedness (the drive to simplify articulations and minimise effort) and faith-

fulness (the drive to maintain input distinctions, differentiate clearly between utterances and be better understood). lffaithfulness always dominated, there would be very little phonological activity or variation; if markedness dom-

inated, mergers would abound and simplification would be carried to an extreme, resulting in impaired comprehension. Understandably, this tension

separates registers of a language: in formal registers where the clarity of articulations is of paramount importance, faithfulness Constraints are placed high in the grammar. Conversely, in informal, conversational or colloquial registers, muddled articulations are frequent and marltedness constraints rank

high. If the phrase I don't know is pronounced [u|d.mnt|1.)o], constraints whose paraphrase is, in general terms, ‘do not alter segments (elements, features, etc.) of the input’ are ranked high; if the same phrase is pro-

nounced [(i5no], constraints whose paraphrase is ‘avoid difficult articulations’ have taken over. Thus, stylistic variation is expressed through a reranl-ting of

constraints. in fact, all variation is expressed in this very same way. \X/hcther diatopic (geographical), diaphasic (stylistic), diastratic (social), diachronic (historical), etc., all types ofvariation can be seen as simple cases of constraint reranlting.

Even if they result in re-analysis ofthe input — about which much more will be said later — they at least originate in constraint reranking. Because constraints are universal, grammatical differences between languages, dialects or stylistic

variants are primarily expressed as differences in the constraint hierarchy.

11.3 Positional constraints In DT, most rules are contextual, although context-free rules do exist, as discussed in Chapters 6 and 9. Similarly, in OT, a large number of constraints are positional, in the sense that they must refer to the position of a feature or a segment in a phonological constituent, such as the syllable, the

Z64

PHONOLOGY

WW

7

phonological word, the phrase, etc. A crucial understanding of inherent vs

positional markedness and faithfulness is thus required. Consider the markedness of features. lt is generally recognized, although not uncontroversially, that some places of articulation are more marked than others. ln consonants for example, much has been made of the fact that peripheral articulations, like velar or labial, are generally more marked than coronal articulations. Data from acquisition chronologies, frequency and directionality of place assimilations, speech disorders, etc., readily point

to a ranking *LAn, "vial. >> *c0R, i.e. a ranking by which constraints against n0n—coronal articulations are stronger than constraints against coronal articulations. Such constraints penalise inherent, i.e. non-contextual. markedness. However, the same expectations do not obtain everywhere. For instance, in

many languages. the occurrence of /13! at the beginning of syllables would be quite unexpected, while /n/ is normal and unmarked; on the other hand, /i;]/ may be quite acceptable and common as a coda consonant. Thus the markedness of velarity for nasal is positionally conditioned. lf position is directly

encoded in the Formulation of the constraint, then we would expect another ranking, of a positional nature; for instance: *1;/onset >> *9/coda, meaning velar nasals are more marked in onset position than they are in coda position.

just as most processes are contextual, most constraints refer to an environ-

ment, the formulation of which is encoded in the constraint itself. Take the example of the velarisation of Ill into [l‘"], as it is heard in some varieties

of northern regional French, Belgian French and Picard. Velarisation occurs word-finallyzfil is [film]. A DT rule would indicate precisely what to do: (3) H-lateral] —> [+'oack]/ V i #

To this rule may correspond a contextual OT constraint of the type ‘V[]]#. While [lm], a complex articulation involving both dorsal and laminal move-

ments, is inherently more marked than [1], a simple articulation - resulting in the constraint order *[l‘“] >> *[l] — the context requires velarisation. Thus positional markedness dominates. In addition [l‘“] is unfaithful to the /l/ input and violates a constraint loam"-coa(lar): ‘do not interfere with the Cororlality Oflateralsi. The ranking would be:

(4) *V[1]# >> ‘[l“‘l,1ot-:NT-coR(|at)>> ‘[1] Notice again that while the DT rule indicates precisely what to do. the " V[l]# constraint states what not to do (at least potentially, provided it is ranked high enough). If the candidate set was limited to two segments, [l'"] and

OPTIMALITY THEORY

Z55

ll}, the grammar would select the correct winner; but since the czmditlatc set is infinite, and includes several plausible candidates among which other

types of I (devoiced, palatal, etc.). either these must be eliminated by separate constraints, like ' [A'] or possibly the positional *\/l.-\']#, or the constraint must

be formulated as "V[+ lateral, -back]. In the above example, the context was defined as ‘post-vocalically, wordfinally’. Other types of contexts turn out to be pertinent. For example,

constraints often refer to the harmony that exists between segments. We may think of these as contour constraints, i.e. constraints that penalise

a dissimilarity of articulation between adjacent segments. For example, in the same varieties of French or Picard, back [ti] rounds into [I)_l before

[l“']: bal is [bnl‘“] and fizmilirzl is [fumlljolm]. Clearly, [(il"'] sequences need to be penalised, but the constraint is probably better formulated as

a contour constraint of roundness in VC sequences: *coNTouR-VC(rnd). Again, [ti] is not round, [l‘“] is; their adjacency thus would define a cori-

tour. This contour is undesirable and difficult to pronounce, thus it is penalised. A [nl'"] sequence. in which there is no contour in roundness, is preferred.

Note however that, by adjusting to the roundness of the environment, [D] violates faithfulness to /ti/: IDENT-RND(low). Thus we have two cases of

allophony (neither ll‘“] nor [11] exist independently), which can be recapitulated in a tableau where an input /hul/ would be matched to four candidates

(to keep it simple): a. [bul], b. [bnl], c. [btil‘“} and cl. [bol"‘]. In the following recapitulation, F denotes faithfulness, M (inherent) markedness and P positional markedness. (5)

P :

*t:oi~ri'ou|t-V(I(rn|.l)

lit-rc. no [til before ]l"‘]

l‘

1

‘VI I ]#

no coronal [l] after a final vowel

:

1171-‘.N'1'-kNl>(low)

do not inlcrfcrc with the -1-/—rnuntl quality of low vowels

:

II)!-:N‘I'-> IDENT-RND(l0W) >> *[l‘“], ‘[12], IDENT-COR(l3t) clearly shows how the P constraints are decisive. The simple example of the homorganicity of nasals in Spanish illustrates the same logic. in Spanish, syllal)ie—final nasals adjust to the point of articulation of the Following onset consonant: pan torto has a coronal [n] while

pan gringo has a velar [13]. Clearly the input is Inf and the grammar must

contain constraints reflecting the Facts that (1) [ii] is more marked than [n] as explained at the outset of this section, (2) [13] represents a distortion of /n.’ and (3) nasals are homorganic. Fact 1 is expressed by an (inherent or positional) markedness constraint, Fact 2 by a faithfulness constraint, and Fact 3

by a ‘no-contour’ contextual markedness constraint. (7) * [13]

velar nasals are marked

’“ [rflcoda

coda velar nasals are marked

IDENT-CO1‘ "CONT Nas Obs/place

input coronal Features are Found in the output no place contour in sequences or‘ nasal + ohstruent

TABLEAU 11.6 i ii

SPANISH PAN TORTO AND PAN GRINGO

/n ti [tn t]

a.

‘CONT Nas Obs/place >i< !

,_j‘“'i'F?:__ o___lP_‘L,, so c. input

[:3 t] in gl

a. l).

[m gl i n gl

"es

l

lug]

-x>

IDENT-COR >i
i= E * *con-rii§i§siOTi;i—|§isiEe K *"rsE§%i¢51£i"" is * I

I

i

*ci

W V _ _+

**

_. _ ., _ - _ ._-_,___._|

>i
> F ranking. In English, aspirated stops are not phonemic (/'p/ exists to the exclusion of /pl‘/, hence M >> F), but aspirated allophones do occur

Foot-initially: P >> M >> F. Conversely, a contrast scenario may expand to a neutralisation scenario.

Contextually, an existing contrast created by the F >> M order may be neutralised. In German, the phonemic contrast between voiced and voiceless obsttuents is neutralised in domain-final position: P >> F >> M (where P reads: ‘no final voiced codas’).

Thus we obtain a factorial typology of orders, which recognizes prototypical scenarios and ensures that the correlation between the shape of the grammar and the constitution of the lexicon is narrow and principled,

whereas it is quite loose and unpredictable in DT. This is achieved in the absence of constraints on inputs.

(3) Exclusion: Allophony: Contrast: Neutralization:

M

'TJ'Tl"T.'

>> >> >> >>

>>

F

>>

M

The general understanding is that the P version of a constraint will always dominate the M version of the same constraint. An M >> P order would

hardly make sense, as the specific case takes precedence over the general case (the ‘elsewhere’ case discussed in earlier chapters). This is only one of several

situations which shows that constraint order is itself — at least partially — constrained.

The last two scenarios, which constitute such a common stock of phono-

logical activity, deserve to be illustrated in more detail. Consider first one case of allophony discussed in Chapter 1. In Lumasaaba, [cl occurs as an allophone of /l> M ranking of IDENT—[COnt] ‘do not interfere with continuancy 0F consonants’ over *[-cont], a constraint penalising stops. However, the surfacing of stops after nasals results from the domination of a P contour constraint of the, type *CoNTOUR-NC—[contl = ‘no contour in continuancy in nasal + consonant sequences’. The resulting

tableau displays the prototypical P >> F >> M pattern:

TABLEAU 11.8 LUMASAABA [ZIMBUA] WITH I ZIMBUA I AS INPUT Wwiiwwlw/zim[luaI' if ‘W

a.

l

[zimfhia]

b.

l

[zimbua]

‘coNTouR-Nc-[cont] H

IDENT-[cont] I ‘[-cont]

*!

V

Wif

>l=

|

V

H=i
> MAX, DEP evaluates the candidates [CV], [V]

and [CVC]. Since M-constraints dominate F-constraints, the [CV] output will always be selected as the winner. regardless ofwherher the input is CV, V or CVC.

TABLEAU 11.10 [CV] WINS WITH INPUTICVI l

J

/I CV /

is‘ Riv; r r [Vi 1 lC\/Cl

ONSET

NOCODA

-1!-

MAX

DEP

-X- _.-_ . -.i ~31»

*

TABLEAU 11.11 [CV] wms WITH |NPUT/V/ 1/

1'

ONSET

NOCODA

|r;~ii'

_-

MAX

DEP -)9

I‘/to ICVCI

-X*

‘X-*

OPTIMALITY THEORY

Z71

TABLEAU 11.12 [CV] WINS WITH INPUT/CVC/'

” Tflrisiii TO

[CV]

_,_ [V] WW” L WTTCVTCTWO ii

W

*!

Z _,_ O” a.-__.l_-.1l,’-

-I-*

iiiiiiiiiiiiii *

_ ._. -.L‘_._-_-1

Faithfulness cannot play a role in determining the winner since this has already been done by M-constraints, and F-constraints are in the shaded area.

Lexicon optimisation compares optimal mappings through the Tableau des Tableaus procedure, recognizes that the mapping /CV/: [CV] provides the cleaner line For the winner, and therefore Tableau ll.lD reflects the correct grammar of language L. The effect of this is to prohibit positing abstract inputs gratuitously. In the absence ofan aiternation, the input of an M >> F grammar will be identical

to the output. If [n] is less marked than [13] and a tableau winner is [nat], it is true that the input could be /nat/ or /nan’ (or an infinity of other Forms) and the same winner would be predicted; try it. But lexicon optimisation imposes an lnatl 2 [nat] matching, which violates neither markedness not faithfulness

over a /nan‘ 2 [nat] which would violate faithfulness. The consequence is similar to what was obtained in linear generative phonology by postulates such as Hooper’s No-Alternation Condition: In non-altemarirzg items, the input is

identical to the output. This means that OT must be a concrete model of phonology (prima Facie, any model containing an unchecked ROB should be

interpreted as abstract; but OT is not such a model, and lexicon optimisation eFFectively ensures a high degree oi-concreteness). Note however that the result in OT was achieved by mere stipulation. ln OT, it is directly observable from the grammar. ln the presence ofan alternation, the analysis cannot be so simple. Lexicon

optimisation is then tempered by lexicon minimisation, i.e. the need to simplify the lexicon by positing 21 single input For several alternants. Like most

models of generative linguistics, OT ascribes to the principle - discussed in Chapter 5 — that what is explainable belongs to (the phonological component of) the grammar and what is arbitrary is encoded lexically (stored in

the lexicon). Predictable information then must be abstracted away from the output.

Consider a simple example. in some varieties of northwestern French, the

length of vowels in open syllables is predictable on the basis of morphological information: vowels are short in the singular, long in the plural. The word For ‘cat’ is U(1l, while ‘cats’ is []'(1:]. Marltedness constraints state that long vowels

are more marked than short vowels, but, in spite of that, [_]'r1:] wins in the

plural because ofa dominating constraint against short vowels in the plural: *SuoR"rV(PLuR). The simple tableau with no indication otinput would reflect only the domination oil’ over M:

272

PHONOLOGY

TABLEAU 11.13 CHATS 'CATS' IN REGIONAL FRENCH 7

7

"W (input: plural

iii

[Jo]

as

[jtu]

’S1—t0RiV(P?tiR)

* LongV

* !

*

Once inputs are added, faithfulness constraints come into play: add some

faithtulliess to length constraint after *Si-l0RTV(P1.UR). if lexicon optimisation applies blindly, the input to the plural Form must be /fa:/. This is because if length is not lexical and the input is /_|'(1/, the winner violates faithfuliiess, whereas with a /I011’ input, there is no violation oflaithfulness and the winner

line is cleaner. But if lexicon optimisation does not apply blindly but is tempered by lexicon minimisation, then length will be abstracted away from the input. This

effectively means that both alternants are evaluated in pairs and the analytic question comes down to: is the grammatical alternation [lo] / Um], resulting

from a P > M, F grammar, best explained From an input /I01’ or an input /f(1:/ . The answer is clearly the former, since the matching /_ft1:/ : UH] maltes little sense (no shortening, long vowels occur in the specific context, short vowels in the general context). This impacts on the ranking of faithfulness constraints. Consider the

pronunciation of similar words in a related language, a dialectal variety

of Québécois French (data From Picarcl 1978). Some dialectal variants of Québécois French display the Following [A] vs [a] distribution:

(12) l.|i.’\l

‘cat’ (masc.)

[lat]

‘cat’ (Fem.)

[calm]

‘tobacco’

[tal)a3i]

‘smoky place’

I plal Idrnl

'Hat' (masc.) 'shcct’

[plat] [clrapri]

‘llat' (fem) ‘drapery’

lrnil

‘rat’ (masc.)

[rat]

‘rat’ (fcm.)

The vowel [A] exists independently in some positions as a type ‘o’ vowel, as in sorta [sntl ‘silly’ (Fem) or propre [prnp] ‘clean’. But it does not occur in unstressed positions; rather, it is encountered only in domain-final syl-

lables (which carry a degree of stress); For example, the article la and the possessive ma, always unstressed, are always [la] and [ma] (never *[lA] and

*[mAl), while the pronoun gr: has a variable pronunciation since it can occur under stress: Q1 lsa] commence comm: ya [sax] ‘it (unstressed) begins like this

(stressed)’. How would OT account For this situation? What is the grammar and do inputs have the Front vowel /a/ or the central vowel /A/? A positional

markedness analysis would require a constraint order where two P constraints

‘a and "AC dominate two ranked F constraints: lolznr-lnl > > IDENT-/a/.

OPTIMALITY THEORY vi

a. ‘AC. [DENT-/M

= = =

no la] in open final syllables no A in closed final syiiables do not alter the features of /.-xi’

ro1~:N'r—/a/

=

do not alter the Features of /a/

273

Now consider the words in pairs, so we can account For alternations such as

masculine vs Feminine in nouns and adjectives. While IDENT-/A/ and IDENT/a/' could be unranked if we evaluated items individually, there is a clear

advantage in claiming that faithfulness to marked items dominates Faithfulness to unmarked items roam"-/.\l > > roam"-/a/. The paired evaluation gives the Following tableau of matchings (Tableau l L14). where the pair b and e

constitutes a better matching than the pair d and g. Thus, the input with the Front vowel is retained:

TABLEAU 11.14 CHAT 'CAT' AND CHATTE 'CAT, F’ IN DIALECTAL OuEBEco|s FRENCH ‘a. a. /]'a/ :

[Ia]

W_fl*7.}Q.V*iV "lf]VZ)glV3NT:/.-\I' .

[fa]

=l_: s I candidate F in Tableau l 1.15

Other options are conceivable, some of which are shown in (14) (rhymes only), but must be discarded by the grammar. They correspond to candidates a—e in the tableau.

(14) I-L Q

'4"

a

Gm

b

ll

Pun

m_

,-,_

v__ c

o—*: _-—-1»: d

95

"9

-»—1: _>": e

OPTIMAUTY THEORY

Z75

In order to select the winner, various prosodic constraints are used. Some, like

“unit, refer to the size of prosodic units, others like ALIGN refer to the way information is aligned at various levels (here the word and the mora), others reFer to the way moras associate to segmental information. The tableau selects the correct winner, dubbed candidate F in the tableau. *uup. = bimoraicity (upper limit oF2 us in a s)

uottot. = offglides are moraic *|LDlP = no monomotaic diphthongs

"1|.si-[AR]-L = no double associations (segments don’t share moras)

ALlGN—FlNAL—|L = align the final pt to the right edge of the prosodic do main

TABLEAU 11.15 /MEJS / ‘MONTH’ ||\| RAETO-ROMANCE I mejsl

l

‘pun

ALIGN-FINAL-ill p

a.

*1

b.

l~LGLIl2§iW ittoir

__’usi~ntRE

>t= *1

1 T éfw T11}? i TTTWW cl.

Z

1

it

*1

0. Wt‘.

*!

>t< =r

Many suprasegmental constraints relate to stress, Foot structure and the well‘ Formedness ofptosodic units. Consider what needs to be evaluated to account

for the Following data (Fitzgerald 2002). ln Tohono O’odham, a Uto-Aztecan language which has borrowed heavily from Spanish, the Following Forms are Found:

(15) ['musigol

‘niusiciauf

but

['mu-msi,go]

‘musicians

l'?asuga.ll

‘sugaf

but

I 'Yasu_ga.l-tl

‘make suga I‘-IMPERFECT,

The words in the left-hand column are monomorphemic, while those in the right-hand column are polymotphemic (morphologically complex): the plu-

ral in [‘mu-msi,g0] is made by reduplication and the linal [t] in ["1’asu,ga.l-t] is the imperfect marker. On the basis ol-numerous similar examples, one could observe that the primary stress Falls on the initial syllable and that Following

stresses ~ if any ~ all fall on odd syllables. Note however that monomerphemic words do not have a secondary stress on the third syllable, while polymorphemic words do. The constraints must select the appropriate feet and the stressed syllable within those feet. This is (a Fragment oi‘) the grammar

proposed in Fitzgerald (2002):

276

PHONOLOGY

FTBIN = foot binarity: feet must be binary *cLAsi-1 = no adjacent stresses *L.APs1=. = no adjacent weak beats PARSE = no unfboted syllable

MSP = morpheme to stress principle: every morpheme must be stressed ln the Following tableau For monomorphemic ['i’asuga.l], the winner violates PARSE, since its last syllable is not Footed. But a is worse, since it has a mono-

syllabic Foot, and c is the wo rst For having several unstressed syllables in a row and no stress associated to the morpheme.

TABLEAU 11.16 /‘?AsuoA.1/’suoAR' IN TOHONO 0'ODHAM /"r’asuga.l/

‘CLASH

"LAPSE

MSP

Frliml

a. ('1’asu) (lgall) FF b. ("Pasu) gal

mass I

W

c. "Basuga.I

*!

=|
PAR$E

|

a. ('mu-msi) go

Ila‘ b. ('mu-msi) (go) c. (l mu-m) (jsigo)

Finally, in the tableau For bimorphemic ["r’asu,gal-t], b violates MSP since -t is not in a footed environment, associated with stress.

TABLEAU 11.18 /'?Asu,oA.I-T/‘suoARs' IN TOHONO O'ODHAM /"fasujga.I-t/ "F a. ('?asu) (‘gal-t)

b. ("Pasu) gal-t

_;&;.‘lgI;TT

*LATPSE

TMSP T7 VFTBIN

fi

mast

=|=

=i—c]w,,,,| ‘do not insert a consonant at the end of a word’ or CONTOUR-vce,,|,,

‘do not vary the voice quality in obstruent clusters’. For instance, it has long been observed that speakers do not use word-internal epenthesis or deletion

Z78

PHONOLOGY

as they do epenthesis or deletion across words. Consider just such a Case in French, in the broader context of stylistic variation.

ln con temporary standard French, the last word of the sentence je ne /hipar flzit exp!-Es! ‘l didn’t do it on purpose’ would be expected to surface as [tlkspre] in careful style, and [espre] in a more relaxed, conversational style. Let us recognize a faithfulness constraint MAX-C militating against consonant deletion and a markedness constraint *cccc militating against heavy (four—consonant) clusters (ofcourse a more careful constraint would ultimately be needed, and

the grammar would need to reflect the fact that we delete [k] rather than [s],

[pl or [r]). If we consider only these two candidates, the stylistic difference will easily be captured by the reranlcing of these two constraints. Expect— edly, the more careful articulation will favour faithfulness, while the easier articulation will favour the dominance of rnarkedness. Careful style: MAX-C >> *CCCC generating lekspre] Relaxed style: *cccC >> MAX—C generating [espre] in the sentence On nestplus all un exprés ‘We’re not counting his/her former

spouses', we still expect [8l> M grammar in careful style and an M

>> F grammar in relaxed style. Careful style: DE!‘-V >> *CcCC generating [eksprfi] Relaxed style: *CCCC >> D1211-V generating lilksoepre]

A combined tableau in the grammar of the careful style is given below. But although this grammar selects the correct optimal forms, it is not adequate. The non-crucial order of DEP-V and MAX-C would suggest that epenthesis and deletion are equally bad in both cases. This is false: leltsoepre] is never

a possible realization for je ne liar pas fizir exprés! in any style, and would be considered quite inappropriate. A tableau must do more than determine

the winner; it must indicate the second best, third best, etc., and effectively attempt to determine the relative desirability of all candidates. Here. the relative desirability for careful exprés is a >> c >> b (suggesting DEF >> MAX), but that for careful exprés is a >> b >> c (suggesting MAX >> DEP). There cannot be two different orders for the same speaker in the same style.

In relaxed style, markedness dominates, but the same problems are revealed, only in reverse. [EIsprt?} is never a possible realization for ex prér in

any style, and has to be ranked as the worst of the three candidates. It soon appears that we have to break MAX and DEF into two positional variants of themselves and take the phonological domain into account. The grammar

OPTIMALITY TH EOFIY TABLEAU 11.20

279

PROVISIONAL TABLEAUZ EXPRES AND EXPRES IN CAREFUL STYLE

'"“:‘:*”"“"7aasaJ"e“‘:’T’1i5If**‘ “sacs _ a._

"F

[cl M >> F ranking will

be roam"-vct/oas onset >> ‘cont Obs-Obs /vce >> roam"-vcs/oss. if both markedness and faithfulness can be positional, it follows that OT will conceivably present at least two scenarios for many alternations, which

are amenable to a positional markedness analysis and a positional faithfulness analysis. Many papers have been written which recast one analysis into the other, and it is important to realise that these analyses express different,

sometimes opposing, generalisations.

Consider again our just mentioned example of final obstruent devoicing, whereby filial neutralisation takes place: both /pad/ and /pat/ are [pat], not

*[pacl], but of course contrast remains in onsets: /bat/' is [bat] and fpatl is [pat]. TABLEAU 11.22 A POSITIONAL MAHKEDNESS (PM) ANALYSIS OF DEVOICING

W

/ pad/

‘vcs],,,,,,|

[p a cl ]

*1

[ pat I

IbatI

4

l p a tl

I r 1 b J I

losnrm.

‘vet

* *

‘vcs],,,,,,,

in ENTM.

*!

‘vcs

I * WW7

OPTIMALITY THEORY

Z31

'li»vo analyses compete: one argues for positional inarkcdtiuss. as we have sketched it above, in which the PM >> F >> M order applies. l|)E.N'|'-vcc

dominates *vcE, hence the general phonemic contrast, but it is itself dominated by the PM constraint *VCE.]“.m5, hence the specific neutralisation. A Positional Faithfulness analysis turns the argument around. Voice is marked for obsttuents and normally would not surface, but it is protected in initial

(strong) position, favouring (barf : [bat] and /ban’ : [bat] matchings (and of course / b a d/ : [bat] and / p a t 1' : [pat]). Here it is the PF >> M order that

establishes the contrast.

TABLEAU 11.23 A POSITIONAL FAITHFULNESS (PF) ANALYSIS or DEVOICING 7 W 7

7 [P 3 dl __ _ 7

iP_F'_iIiI\=£lQ§5

_ __,,;vCE

[ pa d l

=i=!

lpa t]

*7’

/batI

lDEN'1‘,u.-ODS

iflflw 7

l pa t l

>ii> DEP-V: it is more important to avoid awkward clusters than to avoid inserting vowels. Deletion corresponds to *[l1] >> MAX—C: it is more important to avoid {la} than to avoid deleting consonants.

TABLEAU 12.2 ANAPTYXIS IN GASCON HLU (NON-H-DELETING DIALECTS) a. b.

gm Ff

/him’

ctusren

hlu hulu

*1

DEP—V ____W

iiiiiiiii

"rim

TABLEAU 12.3 [Hl—DELETl0N IN GASCON HILU ‘THREAD’

_

__.......,a_e.fW£:,..tetW1 a. hilu *1

-sWseih>t

_Hm.Wes.

mlsaestzjee- __LW c_* LL,

The real problem is that in forms without [h], non-epenthetic outputs should be optimal. ln DT, a form to which the second rule applies, but not the

first one, will always be less marked. in OT, transparent *[iu] will always be preferred over opaque [ulu]. Again, there is no constraint order that can make [ulu] the winner, since [lu] violates a subset of the constraints violated

by [ulu]. TABLEAU 12.4

WW OTTO)”

‘SAD’ TABLEAU FDR [ULU]

/hlu I T 7

CLUSTER

hulu

>+=[h]

MAX—C

>-kl

>i=

l

l

=i=*

it ® P—nP‘r=

mt

7

>i IDENT-V[hi] u)E.N"r~G1ide: a glide in the input surfaces as a glide in the output *\/[hi] + V: no sequences of high vowel + vowel [DE.NT—\/[hi]: a high vowel in the input surfaces as a high vowel in the

output TABLEAU 12.11

FFOUA WITH GLIDING AND LEXICAL ROI IN FRENCH

frual ‘roua’ F

V5‘

/rwa/ ‘roi’

I

Ioeisrri-glide

>i> IDENT-\/[hi] >>"‘V[hi] + V, and a [rua] vs [rwa] contrast would result. The challenging part was mentioned in our note l in Chapter 8. When two consonants precede the glide, Glide Formation is blocked. Thus, from a root

rrou [tru] ‘hole’, we might expect rroua *[rrwa] ‘(she) nailed’, if Gliding had applied as in roua. Yet, troua can only be pronounced in two syllables: [tru.a].

If roua can be pronounced in one syllable, but troua cannot, the difference

must be in the onset which precedes the glide, and indeed a phonotactic constraint (a filter, suggested as early as the 19705 and mentioned in our

first edition on p. 208), was proposed which read: *OLGV ‘no sequences of obstruent -1- liquid + giide + vowel’. However, there exists a perfectly good lexical form trait [trwa] ‘three’, which, like mi, can only be monosyllabic. Clearly, various analyses could be, and have been, offered. A structural one

was suggested in Chapter 8, note 1, which would exploit the 2 + 2 onset + nucleus grouping in rrois vs (a blocked attempt at) a 3 + 1 grouping in trawl.

A purely morphological analysis could also be developed, whether in DT or in OT, which would exploit the difference in morphemic composition, yet remain within the spirit of our discussion of the ability to distinguish between

derived segments and lexical segments. *OLGV cannot be a surface constraint in DT, since trois is fine as a monosyllabic. *OLGV does not prevent a state of affairs, it blocks a process. Again, classical OT, which has no insight into

ISSUES IN OPTIMALITY

301

intermediate forms or processes, will be challenged by such a case (even ii‘ nu

counter-Feeding or counter-bleeding has taken place). What we need to be able to say is: ‘[trwa] is Fine as a faithful output ~

i.e. as in trais f but it is blocked by *OLGV when it is an unfaithful candi— date — i.e. as in rroua’. This is precisely what an M86}: conjunction says. in Tableau 12.12, *OLGV has been added to the grammar but its ranking is

low. Many surface forms violate it, like trois (B/air ‘a city’, pluie 'rain’). However, its high-ranked conjunction with 1DENT~\/[hi] has exactly the desired effect: to penalise any OLGV sequence that matches a /OLVMVI input; in DT terms: any derived OLGV sequence. The ltrual : [trwa] potential matching

is penalised, while trois is untouched.

TABLEAU 12.12 TROUA WITH BLOCKED GLIDING AND LEXICAL TROIS IN FRENCH Irma! ‘troua’ tosnt-glide ____ T T£,lwa1 W

W[it1't1a]

V H WW

=ii=OLGV IDENT-V[l1l] p H "7 W l * *

V V

=l=

V WW’ if

Wfliiw

ltrwal ‘tr0is‘ IDENT-glide *0LGV 54 IDENT-Vh, >i=V|,, +V| W

[trwa] ltrua]

[DENT-Viliil

l *5

*

*

|

Thus we see that F8513 conjunctions can handle some cases of unclerapplication and M&F constraints help in sorting away derived elements From

lexical elements. This is not 0F course the extent orwhat can be accomplished through conjunction, merely a quick illustration. However, notice what cannot be accomplished: local conjunction cannot defeat the (wrong) transparent

winners in overapplication scenarios. Recall Tableau 12.4, repeated here For convenience as Tableau 12.13, where opaque b Fails to win and transparent d is wrongly predicted. No conjunction can help, since d violates only one

constraint.

TABLEAU 12.13 ‘SAD’ TABLEAU FOR lutul 1N GASCON lhlul @

rm-

a.

hulu

b. c.

ulu hlu

d_

T1uTTTTl

ctustsa

DEP-V

MAX-C

=14-4_

>:= ._.

|

* TTT

=l=!

T

VT’

T TH” V

K

T

Q

Some of the problems that we have addressed have been the targets of another approach, which, like LC, expands the phonological constraints (but without

302

PHONOLOGY

expanding their inventory, as we shall explain). It is known as Comparative Markedness. The fact that grammars will generate forms in which a process fails to occur, even though it is known to be very regular and pro-

ductive in the grammar, could be related to distinctions between original infractions of markeclness and derived infractions of markedness. The ability to make such distinctions constitutes a characteristic of derivational models

but is foreign to the classical model ofOT. In recent work, McCarthy (2003 and subsequent work) proposes that each markedness constraint be split into two versions which he calls old markedness (UM) and new markedness (KM). OM refers to violations that are shared by the faithful candidate, while NM refers to marked configurations which are absent from the faithful candidate.

By ranking faithfulness constraints between the OM and NM, it is possible to

account for counter-feeding opacity. Consider again our European Portuguese example (Tableaus 12.8 and 12.9) and consider the consequences of splitting the "E + PAL markedness

constraint into two versions of itself: ,,"s + PAL and ,,*s + PAL. (As before, T, O and F are used in Tableau 12.14 to help identify the transparent. the opaque and the faithful candidate, respectively).

TABLEAU 12.14 A COMPARATIVE MAHKEDNESS ANALYSIS OF UNDERAPPLICATION

T if K/isi:'i'i§;‘ifi‘ciherry’ T -av [M35] 7 F

*5],

lsiregfil

=i=!

lvcspfil ‘wasp’ O

F

‘FW

W‘

.,*@+ of iDE~eiv,,,,;,. T if 1* on + mt

7 7 7

tosur-V,,|;,,.,.

[vsfpa]

lvcspii]

ilvajipilfi

§§v§;L1~n. T "*2 +1»/u. =1=

*1

1

*!

In /sir£'35/, the faithful candidate is defeated because it violates ,,*rI + PAL

(obviously, F can only violate Old Markedness). However, in /vtspfi/, where O violates ‘E -1- PAL, it is _.,*€ + PAL that is violated, since F does not violate it. With respect to F, O introduces a new violation of markcdness and

because ioEN'r-V,,1,Cc is sandwiched between ,,*£ + mt. and ,,*s + PAL, O wins out over T. Comparative Markeclness has a number of other applications, which can-

not be covered here, but it offers an interesting solution to some types of opacity by developing the notion of new markedness. in this fashion, it retains the strength and parallel integrity ofQT while respecting the insight of

DT. The effects of a derivation are achieved, even though there is no derivaw tion. However, it too has its limitations and specifically is not designed to handle opacity by overapplication.

ISSUES IN OPTIMALITY

303

Enriched outputs There are a number of cases where opacity emerges From the interaction of segmental and suprasegmental processes. These cases may well constitute a

special category of scenarios, For which specific solutions can be introduced. One of the prevalent proposals has been to make use of enriched outputs, i.e. outputs which encode more information than just the mere phonetic Form. For instance, Germanic languages have been argued to display ambisyllabic consonants, i.e. consonants that are associated with two syllables (see

a Fuller definition of ambisyllabicity in Carr Z008: 12). Thus, we find that, in English, the consonants that occur between a Fully stressed syllable and

a Fully unstressed syllable are ambisyllabic, especially when they Follow a lax vowel, as in copy, petrol, college. l-low much of‘ this information is included in

the outputs themselves? If outputs are enriched to include inFormation about the segmental/suprasegmental information, then some cases of opacity may evaporate. As an illustration, recall the distribution of [X] and [cl in Standard Ger-

man, exemplified in Chapter 7, exercise 3. The back F1-icative occurs after back vowels as in [bux] ‘book’ while the non—back Fricative occurs after Front vowels, as in [peg] ‘bad luck’. Additional data show an interesting interaction

with morphology. ln muchen ‘to smoke’, the fricative is in the stem and surfaces as back [X], while in Fmuc/Jen ‘little woman‘, the liricative is in the suffix

and surfaces as non-back [c]. The derivation is the Following:

(13) tau /ct’ + en

Frau + fcf en

tautosyllabic assimilation

rau [x]

-------------

lcvel 2 affixation

rau [X] + en

Frau + [glen

rC$yll£lblFiCation

rau.[x|cn

ii'a1u.[ IDENT-[arr], and

resyllabification by the ranking onset >> ALiGN-Rlstem, tr], we obtain the Following ‘sad’ tableau:

ISSUES IN OPTIMALITY

305

* [atr]—C](,: no tense vowel in closed syllables ONSET: syllables must have onsets |DENT—[atr]: the tense/lax value in the input surfaces as unchanged in the

output ALIGN-R[stem, 0}: the right edge ofa stem is aligned to the right edge oFa

syllable TABLEAU 12.16

if

7 11

p7

'SAD' TABLEAU IN QUEBECOIS PETITE YVONNE

_fl-atr]-CL,

7 Mint . i..7 7 W____p_p O

ONSET

ALXGN-RlS[L21T1, 0]

IDENT-[arr]

>l=

=|= *'

p__ ____=l=1

..t1.ti..

..tit.i..

T

>z!l=l

-_s ,_-s

*-ie

Stratum 2 For petite Yvonne

opacity, and these cases are so numerous that Stratal QT has indeed enjoyed considerable success in the literature.

However, we have seen that opacity can occur in the absence of morphological distinctions. For instance, in our very first example in this chapter,

epenthesis and deletion interacted in simple Gascon roots. To eliminate the opacity in such a case, Stratal OT would have to posit that epenthesis occurs

at Stratum 1, but deletion at Stratum 2. Both processes, incidentally, are

word-level processes. So stratal distinctions could not be made on the basis of word vs phrase, as in Québécois (since both epenthesis and deletion are lex-

ical, we would end up with within-stratum opacity and the problem would persist). ln such cases, the basis For stratal distinctions is unclear and the

question would arise about whether this procedure is ad hoc. Even when the morphological bases are well-understood, the number of strata is con-

troversial (are there two, three, Four lexical strata in English? Recall our discussion at the end of Chapter 7). But iF multiple strata can be used in

the absence of morphological activity, then there is no principled limit to strata proliferation.

12.5 Conclusion The advent of OT has been viewed by many researchers as a paradigm shift of sorts which challenges previous analytical principles and procedures. The

model undeniably has much to offer, but with new answers come new ques-

tions. At the time of writing, OT has been discussed For more than 15 years, and it is a constantly evolving model.

In this chapter, we have only been able to point to some of the problematic issues — but also, hopefully, some of the promising areas — associated with the OT model. We have briefly discussed a number of new approaches, keeping in mind that, as we pointed out earlier, every one of these new

308

PHONOLOGY

proposals challenges one or more of the tenets of the original 1993 proposal. For instance, in Local Conjunction analyses, constraints can be much more

complex than in the original model; O-O correspondence suggests a new theory of faithfulness; for Comparative Markedness, constraints are not necessarily unitary; enriched outputs raise the question of what it is exactly that we evaluate; Stratal QT in many ways challenges parallelism, a cornerstone of

the original proposal. And there are, not surprisingly, other proposals, which could not be cov-

ered in this chapter, which have challenged Richness of the Base, or the untargeted nature of constraints (‘do anything but . . .’), the infiniteness of the

candidate set, the universality of constraints, and more. Even hybrid models, mixing rules and constraints, have been argued For.

lt would take much more space to even begin to give an idea of the range of applications that have been proposed For the OT model. OT has been used to evaluate the desirability of phonetic gestures, the pressure effects of linguistic

systems, the movement of intonation curves, the emergence of exceptions, and much more. Before any of this is attempted, the logic of the enterprise must be Well-understood; this and the previous chapter have sought to make it clear and approachable.

Exercises 1. Camzdirzn raising

Probably one of the most-cited cases of opacity by overapplication is that of Canadian Raising or Shortening, where [ai] shortens to

before voiceless

consonants, as in [roit] ‘write’; compare with [raid] ‘ride’. Opacity is created by the Fact that word-internal I tI frequently flaps to [r], which is a voiced sound: I[aitorf ‘writer’ is pronounced [pair-or], just like ‘rider’, In other words,

flapping removes the motivation For shortening. (i) Write the rules and the derivation For the /r_airorI : [roiror] matching. (ii) Using two markedness constraints, simply called *ait and FLAP, and two

faithfulness constraints (determine what they should be), draw a ‘sad’ tableau that illustrates the opacity problem.

(iii) Which conjunction of constraints could help to select the correct Winner? 2. Hezlg/at 5/nfi in Malmfi Swedish

in Malmo Swedish, a two-step height shift occurs in long vowels — written here as doubled vowels —- generating the following surface forms (Lindau 1978, Schane 1996):

ISSUES IN OPTIMALITY

liil : feel :

lei] [cc]

/t:r:/ :

[see]

/yy/: /no/1

[oy] [mo]

/uni:

lo!-ll

/nu/: /oo/:

309

lcul [fol

Notice that the first member of the diphthong must be a Front vowel and exactly one degree lower than the second member. In addition, first vowels that emanate from back rounded vowels must become unrounded. (i) Formulate a constraint that will protect the identity of the second vowel

(the head)? Which Family of constraints would it belong to? (ii) Which Faithrulness constraints are violated? '\Y/hat type of conjunction might be welcome?

(iii) Using a simple CONTOURDEGREE constraint, paraphrased as ‘diphthongs have a rising height contour of exactly 1’, draw the OT tableau from an input /uu/. Consider the following candidates.

a. [uu]

b. [ou]

c. [oe]

d. [eu]

e. [cu]

F. [ou]

5’. Dz];/it/10n_gis¢zti0n in Northern Iris/.1 English Reread the section on O-O correspondence and the discussion of dentalisa-

tion in Belfast English. A similar case olundetapplication by counter—feeding, also in Northern lrish English, is cited in Kramer (Z008). Diphthongisation in closed syllables precedes the atiixation of the plural marker -2, leading to the Following DT analysis:

1. Diphthongisarion (closed syllables} 2. Affixarion oF plural ~z

/def ‘day’ /d£1+ 1/ ‘days’ -------------------------» dfiz

/dcz/' ‘dam’ dloz ------'-

Had morphology occurred first, the two words 'days’ and ‘daze’ would have

been undifferentiated and both pronounced [dial]. How does OT account For this situation?

4. Devoiczézgin Tb/nz Batu/e In Toba Batak (Hayes 1986), voiced obsttuents can occur Freely in initial or intervocalic position within roots, but never root-finally: [marisap] [manihutl lpedclekj

‘sn1oll] 'ravine', [bogo] ‘bow’,

b. Before coronals: [toss] ‘horse’, [bodo] ‘floor’, [trotta] ‘sidewalk’, [horn] ‘horn’ ln addition, umlaut fronts /o/' in the plural, giving alternations like: c. [boga] ‘bow’ lbogo] ‘bows’, [bodo] ‘Hoot’ [bode] ‘floors’

ln the neighbouring dialect of Kesswil, the same grammar applies with one difference: lowering also applies to derived fol and the word For ‘floors’ is [boed9]. Words with non-derived /o/ do not undergo lowering: ‘Frog’ is [Fro_l'_['], not * [frceff].

(i) in what ways is the grammar of Kesswill less transparent than that of Schaffhausen? (ii) Do a tableau For ‘Floor/floors’ in Kesswil Swiss German. 6. Cyclic stress in English

English has a stress principle whereby monomorphemic words which have five syllables, the first three of which are light, must be stressed 6 o o ti cr rather than *cr Er 0' 6' 0, as in lollrzpalooza or rz/mzcrzdabra, but derived. words

like imagiizarion are correctly stressed. cr Er o ti 0 because they are cyclically derived from imagine. ln both cases, the grammar must recognise intermediate Forms. How does OT, with its strict requirement of parallel evaluation,

approach this problem? Can O-O correspondence be used effectively in this instance?

Further reading john McCarthy’s (2()07b) Hidden Generalizations, in addition to developing the Optimality with Candidate Chains (OT-CC) Framework, presents a clear and thorough assessment of opacity-related issues. lt is part of a new series called Advances in Optimality Theory, published by Equinox Books. and although some of the contributions will be too technical For some readers, the series is worth a very close examination sincc it includes studies with a broad typological Focus and thus discusses

at length the universal properties that underlie variation.

ISSUES IN OPTIMALITY

311

The Rutgers Optimaliry Arrbiries (RDA) at http://roa.rutgcrs.cdu now oili-rs more than L200 papers written in the Optimality framework. Most olithem are conct-|'nt-d with phonology, although readers might enjoy studying OT papers in syntax. language acquisition, metrics, morphological exponcnce, pragmatics and the like. Again. some contributions arc too advanced For many students. but it will he enlightening to

read about issues that this short chapter could not explore or sometimes even mention: gradient Constraints, Sl'()Cl1H.S[iCl|Iy= optimal paradigms, perceptual constraints, numerous issues developing the computational perspective, and much more. This

huge database grows on a weekly basis.

Feature Specifications for Consonants Obstruents P

I

‘L

C

k q

b

d

Ci i‘

g (3

ES

dz

v

C

J F U

S

f

é

¢* B

V

6

Z

3

» Z

J

Y

H

C0115

+- -F 4- —F i- -F i- -F %- -F %— %* —+ %- 4* %* A? i- i- %- %- i- -F %- i- -F i- i- %— -F i- -F if 4—

voice obs

—F %~ -+ i- -F %— i- i--? -i-F -+~F -P~+%—%~ 4~~% AF %--+ —F ~F

2

%— %—%* i- -i—F-F -F-F-F -FAF-F 4~4-+i%—-P %—%—iie %-4- 4-4- i%- %~-F-%%4%— %~~% %--P %--F -i-%i-

con:

-+ +- -i —+

d.re| lat

r1as a|1r cor high

lovv back round stri

i4-%~ ~F

%-i‘ %-i- -F -k

-F~F —% -F + —F %-

-F i-

+- i-

i-

—F —F

+-i- +i-

—%

%—i- %-i- -F

i%- %— +- |

4-‘F %~%— %—

-++

—F -F

i- -i —F i-

-F %- -F —F i- ~P - %—

%— -P %*

-+ i- i-

-F

Sonorants I11

1']

IL

_]!

U

N

COHS

+

+

+

+

+

+

voice obs

-i"

+

+

+

+

+

Syn

Cflnt

I5

W

F

Cf»

.1

i

Y

if

l

l

A’

T

R

I‘

II

+

-§-

+

+

+

+

+ +

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

d.rcl

lat nas ant COY

high

+ +

+ + +

+

+

+

+ +

+

+ +

+

+

+

+ +

+ +

+ +

+ +

+ +

+

IO“!

back round

+

+

+ +

+ +

+

+

Sample Answers to Exercises Chapter 1 C. Polis/1

The devoiced allophone occurs between voiceless sounds. and between :1 voiceless sound and a word boundary. There is phonemic voicing For velar stops: lgurul (‘a mountain’) vs [kum] (‘:1 hen’)

0. Ykmil (3) #_= m_; v_v

’P'-> lliii 4 $5] (C) we

i ELEV]

/1 Q

(iii) lt must he ordered after those rttles. (iv) IF French-ttcquiring children never heatt [I1]. it is hard to see how they could have phonological representations containing /11/. The analysis appears overly

tthstt':tct. (v) The presence oi‘ a written in the spelling does not induce speakers to assume that these words begin with an /h/' (vi) It might he suggested that these loanword data support the SPE-type analysis. httt. given the response to question (v), we cannot reasonahly say that the presence ofa written in a word indttces the speaker to assume an underlying /h/. Given that speakers oi French are treating -initial English loanwords as ‘h-aspire‘ words. supporters ofthe SPE analysis might :1 rgue that native speakers have nativised the -initittl English loanwords hy assimilating them to

the rules and representations postulated in SPE. But pronunciation of such words with Elision and tt"uncation would equally well constitute nativisation. Supporters of’ the exception ttnztlysis cottld argue that speakers 0F French are treating these English loanwords as exceptions. (vii) This ttudermines the SPE analysis. An alternative would he to violate the prin-

ciples ol'SPE and assume, Following Byhee, that sequences like le /taro are stored in the lexicon. But if‘ retention oliitregular forms is said to he correlated with high token Frequency, we would expect /e hero to he high in tolten Frequency,

httt !'/)e'ror'srn¢* to he low in token lirequency.

Chapter 7 1. Om/rlrrorrrt lirtsqw (i) The rule applies in derived environment only. Because ofthis, these cases cannot undergo the rule, even where they have an /a: at a word Fortnatitxn hracket, preceded hy a high vowel, e.g. [pinta] and [Eirrtista] where ']' = word formation

hrttckct. (ii) The rule is a lexical rule and thus dues not apply in cases like ‘seven Sundays’. (iii) lt applies postlexiczllly.

Chapter 8 I. London Englir/1 (i) bored/hoard. (ii) /3:1’ is rc.1lised as loo] in open sylluhles. and as [On] in closed syllables.

(iii) /h.);rt' (bore) is realised as [boo]. (The fr! is not realised. since there is no empty onset position for it to occupy.) Sttllixation then occurs to yield the past tense Form; thus the realisation lh.).1d] (bored). The word btmrd, unlike bored, is

313

SAMPLE ANSWERS TO EXERCISES morphologically simple. /bo:d/ is realised as [bond] because the /d/ occurs in coda position, so the /ozf is in closed syllable; thus the Iouil realisation.

Chapter 9 2. japanesc The Supralaryngeal tier will have to be delinked to yield the forms with /h/. To rep-

resent the transition From /h/ to /g/, the supralaryngeal tier will have to be relinked, and specifications will have to be added for the feature lcontinuantl, on the manner tier, and the Feature [high], under the dorsal node, which in turn is in the place tier.

Chapter 10 1. l/owe! qmmtity in lmni Persirm

All vowels are short except low [a], whose greater sonority allows For minimal

contrasts. 2. Wéighr-sensitivity in C/zpanrz/atia The Capanahua system is quantity-sensitive: closed. syllables attract: stress. Main stress occurs on the second syllable ifclosecl, otherwise on the first one. 5. Lengthening in Carib (i) Stressed syllables are lengthened. VVhen the Words have an odd number of

syllables, the final syllable is stressed (and its vowel lengthened). (ii) Coclas are weight bearing. (iii) Unstressed vowels are deleted. 4. Seczmdmjr stress in Bas Varmemis Breton‘ Secondary stress Falls on the heavy syllable closest to final stress. If there are no heavy

syllables, initial stress is the default. Codas are weight-bearing.

5. Weight and time in Italian geminare: These facts are not directly accounted for by weight theory, since onsets are involved; not all length phenomena are related to weight.

Chapter 11 I. Scotti:/1 Engfis/9 spent/Jeri: A plausible grammar For SE film would be SONSEQ >> MAX—IO >> ‘COMPLEX >> DEP'IO.

3. Hiatus resolution in Bengali The fact that glides keep their height and roundness means that 1DENT[vhigh] and tDENT[tnd] are undominated. Hiatus resolution means that hiatus is penalised by a

constraint ONSET (or *‘\/V) dominating IDENTVOC. However, the Fact that /tfala-if!’ : [t]'ala~]] is the winning match shows that MAX-V is dominated by some contraint on final *GV']W sequences (it could also be a prosodic

SAMPLE ANSWERS TO EXERCISES

319

constraint on the weight ofrimes, but that cannot be determined on the basis oitltese limited data). The grammar fragment could be ordered as Follows:

rnsurlhighl, IDENTlrncll >> *[GV]w, ONSET >> MAX-V, IDENTHK 4. Normrm /uzfzztrzlisrzrion A positional markedness (PM) analysis would use a domain-restricted version oi

the contour: constraint, restricted to apply only morpheme-internally. \X/hen this

cowroua constraint dominates faithfulness (IDENT-baclt), candidate b emerges as the winner (both ll and d satisly the matkedness constraint, but d violates identity

twice). By contrast, a positional faithfulness (PF) analysis would use an unrest1'ictecl version OFCONTOUR, but would protect morpheme-final segments From palatalisation with a PF version oi‘ IDENT-back. The PF >> M >> F grammar would then be:

IDENT—b;lCl> couroua-CV[-back] >> lnsnr-back Wliile both analyses select the same winner, the order of desirability For candidates dililiers. PM predicts a.3, b.1, c.4, d.2. PF predicts a.2, b.l, -2.4. (L3. Thus, iii lack of palatalisation in the first syllable is possible. but presence oi‘ palatalisation in the second syllable is nor, the PF analysis imposes itself as more accurate.

5. Vowel qoent/Jeri: in Am/mric

Constraints: GEM-INTEGRITY : no epenthesis inside a geminatc structure ‘CC DEP-V

Ranking:

: no sequences oiconsonants (oi the wrong sonority) : no vocalic epenthesis

{DENT-GEMINATE >> *cc >> n1=.r>~v Tableau:

/drill’ E“

E&

b. l [dadai] C. [dodla] lea. iaddalj

deli-rnrscstpj 7 . 00

oar-v

M]

. *2 p

.3

* l e * fl *~

Chapter 12 I. Crzmzdirm raising Raisin g oi‘

I to [oi] occurs before a voiceless consonant.

Flapping oi: ld. tl to it] (which is voiced) occurs intervocalically.

lraidl

raising

—

morpheme-adjunction flapping

[raider] [rairor]

‘ride’

/raid

‘writc'

[roit] ‘rider’

[roitor] [roiror]

‘writer’

320

SAMPLE ANSWERS TO EXERCISES

fraitarl l

V

lT9iT9l'l

O

®

FLAP 1 *ait l

s|

i-

IDENT—[lOwlJ roam-lvce] *

WWW’

________V____V_" ___"____"W_V_T_V_V:_::

[roirorl l

F

lraitarl

T

[miter]

*! *!

*

*

No conjunction can penalise the transparent candidate. 2. Hezlg/at s/Jrfi in Malmfi Swedish (i) An IDENT-Head constraint, paraphrased ‘do not alter the features of the head oi‘ diphth0ngs', would ensure that the second part of the diphthong remains unchanged. This is positional Faithflllness.

(ii) [DENT-[belt] and IDENT-[rndl are both needed. In addition, a contour constraint on rounclness is required that will penalise sequences where both members are [+rouncll. The M BC F conjunction otthis constraint with IDENT-['l>cl IDENT-[vce] >> [vce]. '

I

iii) An O-O constraint would Force identity between the devoiced [pl in the root [marisap] and the [pl in the derived Form [marisapanlz O-Olvcel >> IDENT-[vce] >>* lvcel.

References

Allwood, _|., Anderson. L.-G. and Dahl, O. (1977) Logic in Linguistics (Cambridge: Cambridge University Press).

Anderson, M. (1969) ‘Syllabic or non-syllabic phonolog)/?', jrmrmzl 0fLr'ngt/isrics, 5: 136-42. Anderson. ].M. (1986) ‘Structural analogy in case grammaf, Lingurt, 70: 79-129. Anderson, ].M. (1991) ‘Contrastivity and non-specification in dependency phonol-

ogy,. Studio Anglim Poznanienm, XXIV. Anderson, J. M. and Durand, J. (eds) (1987) E.xplomti0m in Dependency Pbr.-rmlngy (Dotdtechtz Foris).

Anderson. ].M. and Ewen, C]. (1987) P:'inriplcs 0fDepena'enr_y P/nonorfiogy (Carnbridge: Cambridge University Press). Anderson, ].M. and Jones, C. (1974) ‘Three theses concerning phonological representations’. journal 0fLinguim't'.t, 10: 1-26. Anderson. S. R. (1974) T/Je Orgrznisrttian offlronoiogy (New York: Academic Press). Anderson, S. R. (1976) ‘Nasal consonants and the internal structure of segments’. Lrmgrmge, 52: 326-44. Anderson, S. R. (1980) ‘Problems and perspectives in the description ol‘ vowel

harmony‘, in R. Vago (cd.), Issmzr in Vowel Harmony (Amsterdam: Benjamins):

1-48. Anderson, S. R. (1981) 'VV11y phonology isn't natural’, [.ir:gm'stic Inqtrir)/, 12:

493-553. Anderson, S. R. (1982) ‘The analysis of French schwa: or how to get something, lot nothing', Langrmge. 58: 534-73. Anderson, S. R. (1985) P/90110/ogy in t/zre Yiumttet/J Cmmry (Chicago: Chicago

University Press). Archangeli. D. (1988) Uzzderspectfiratian in K111/elrnarai P/Jamafogy and /1/[tn];/rralngy

(New York: Garland). Archangeli. D. and Langendoen, D. T. (1997) Optimality T/wary: An Ozwttiew (Oxford: Blackwell).

Archangeli, D. and Pnllcyhlank, D. (1989) ‘Yoruba vowel hatmony', 1,ingr-tistic Inquiry, 20: 173-2l7. Asher, R. E. (1982) Wzmil (London: Croom Helm). Baetens-Beatdsmote, H. (1971) Le Fmngais Régiunai dc Bruxrlles (Brussels: Presses

Universitaircs dc Btuxellcs).

Beckman, J. (1998) ‘Positional FairhFulness', Doctoral dissertation, University 0F Massachusetts, Amherst, MA. Baker, B. (2008 [19‘)9]) 1%rdSt:'|1cmre in 1Vgnla£*gan (Stantordz CSLI). Bakovic. E. (2007) ‘A revised typology oFopaque gcneralisations’, Phonology, 24(2):

217-59. 322

REFERENCES

323

Bendot-Samuel. T. (1960) ‘Some problems of segmentation in the pl1u|lulugical analysis olTerena‘, Word, 16(3): 348-55. Reprinted in F. K. Palmer (1970),

Prosodic /lnnzj/ris (Oxford: Oxford University Press): 214-21. Benua, L. (1997) ‘Transderivational ldentity: phonological relations between words". PhD dissertation, University of Massachusetts, Amherst, MA (Also ROA-259).

Bcnua, L. (2000) Plronologiml Relations between 11’/(ml: (New York: Garland Press). Bermudez-Otero, R. (2011) ‘Cyclicity’, in M. van Oostendrop, C. Ewen, E. Hume

and K. Rice (eds). The Blackwell Comprmiort to Pborzology. Volume 4 (Maiden, MA: \)Viley-Blackwell): 2019-48. Blumenfeld, L. (2006) ‘Constraints on phonological interactions‘, PhD dissertation. Stanford University. Bird, S. (1990) ‘Collstraint-based phonology‘, PhD dissertation, University of

Edinburgh. Bolognesi, R. (1998) ‘The phonology oFCampir‘1anian Sardinian: a unitary account oFa sell.-organizing structure‘, PhD dissertation, University oFAmsterdam. Booij, C. (1983/84) ‘French CIQJ alternations, extrasyllabicity and lexical phonol-

ogy’, Linguistic Rezwieui, 3: 181-207. Booii, G. and Rubach, ). (1987) ‘Postcyclic vs postlexical rules in lexical phonology’, Linguistic Inquiry, 18: 1-44. Boyce, W. B. (1834) /1 Grammar of the Kofir Language ((_iraharn's Town: Wesleyan Mission Press). Broadhent, (1991) ‘Linking and intrusive r in English‘, University College 313,

London Working Papers in Linguistics, 3: 281-302. Bromherger, S. and Halle. M. (1939) ‘Why phonology is cliflerent‘, Lirzglrirtic lnqruirj/, 20: 51-70.

Bromberger, S. and Halle. M. (2000) ‘The ontology oli phonology (rcvised)', in N. Burton-Roberts. P. Carr and G. Docherty (eds), Phonological Knowledge (Oxford: Oxford University Press): 19-38. Broselow, E. (1995) ‘Skeletal positions and Moms‘, The Handbook offlmrtologiml T/reorjy (London: Blackwell): 175-206.

Btoselow, E., Chen, Su-1. and Ht1Fli11an, M. (1997) ‘Syllahle weight: convergence 01‘ phonology and pltonc-tics‘, Phonology, 14(1): 47-82 (Cambridge University Press). Brown. G. (1972) Plwrmlrigicrtl Ruler rmrl Dialect 1/rzrirttioru A Study ofthe Plvonology t.fL:rmo.rrzrzoo (Cambridge: Cambridge University Press). Brock, A., L1Galy. B. and Fox, G. (eds) (1974) Paperrfiom the Parrzsersion on Natural Phonology (Chicago: Chicago Linguistics Society). Bullock, B. (2000) ‘Consonant geminarion in Neapolitan‘, in L. Repetti (ed.), Phonological Tlreozjt and t/.1: Dialects of Italy, (Amsterdam: john Benjamins):

45-59. Bybee.

(2001) P/Jonology rind L/mgmzge Use (Cambridge: Cambridge University

Press). Carr, P (1990) L1'rzgu1'stiz' Re/rlities (Cambridge: Cambridge University Press). Carr, P. (1991) ‘Lexical properties of postlexical rules: postlexical derived environment and the Elsewhere Condition’, Lingo/1, 85: 255-68.

Carr, P. (1992) ‘Strict cyclicity, structure preservation and the Scottish Vowel Length Ru1e’,]ournrrl ofLing:n'rricr, 28: 91-1 14.

324

REFERENCES

Carr, P. (I993). Phonology (London: Macmillan). Carr, P (2000) ‘Scientific realism, sociophonetic variation and innate endowments

in phonology’, in N. Burton-Roberts, P. Carr and G. Docherty (eds), Phonological Knowledge (Oxlord: Oxford University Press): 67-104. Carr, P. (2008) A Glossary ofPhonol0gy (Edinburgh: Edinburgh University Press). Casagrandc,

(1984) The Sonncl System of French (Georgetown: Georgetown

University Press).

Chatctte, M. (1991) Condition: on Phonological Government (Cambridge: Cambridge University Press). Chomsky, N. (1965) Aspect; ofthe Theory ofS)/ntax (Cambridge, MA: MIT Press).

Chomsky, N. and Halle, M. (1968) The Sound Pattern ofEnglish (New York: Harper 86 Row).

Chomsky. N.. Halle, M. and Lukoli, F. (1956) ‘On accent and juncture in English’, in M. Hallc (cd.), For Romm/z>/nhohson.' Essrtys on the occasion ofhir sixtieth birthday,

Volume 1 (The Hague: Mot1ton): 65-80. Christdas, P. (1988) ‘The phonology and morphology 0F Tamil’, Dissertation, Cornell University. Clements, G. N. (1976) ‘Neutral vowels in Hungarian vowel harmony: an autosegmental interpretation’, in Kagl, D. Nash and A. Zaenen (eds), Proceeding: of the 7th /lnnnnl Meeting of NELS (Cambridge, MA: MIT, Linguistics Department). Clements G. N. (1981) ‘Akan vowel harmony: a non—lineat account’, Hzzrurtm'.S’t1tdir: in Phonology, 2: 108-77.

Clements, G. N. (l935a) ‘The geometry oi’ phonological Features’, Phonology lénrhooh, 2: 225-53. Clements, G.N. (198511) Compensatory Lengthening rind Consonant Gerninotion in Lug/znrln. Studio: in Conipenrntrny Lengthening (Dordreclit: Foris): 37-77. Clements, C. N. and Keyser, ]. (1983) C V Phonology: A Generative Theory ofthe

Syll/zhle (Cambridge, MA: MIT Press). Clements, G. N. and Se"/.er, E. (1982) ‘Vowel and consonant disharmony in Turl P/aonemie and Phonetic System: (Poznan: Wydawnictwo Naukowe Uniwersyteru Im. Adama

Mickiewicza). Mascaro. (1976) ‘Catalan phonology and the phonological cycle’. PhD dissertation, MIT, clistrihuted by Indiana University Linguistics Club. Mester, R. A. (1989) Studies in Tier Structure (New York: Garland). Milner, C. (1.967) ‘The French truncation rule’, Q1.ozrter{'y Pr'og1'c‘ss Report, 86: 273-83 (Ca1'nh|.'iclge, MA: MIT Press).

Mohanan, K. P. (1986) The Theory ofLexz'col Phonology (Dordrccht: Reidel). Moron, B. (2001 11999)) Distinerioeness, Coercion and §onor'i{y.' /1 Unified Theory of 1/%:'g/at (New York: Routledge). Morin, Y. C. (1987) ‘French data and phonological theory', Linguistics, 25: 81 S-43.

Nauton, P. (1957-63) Atlas linguisrique er eahnogrop/nque do Massif Centml (Paris: Publications de l'1nstitut de Linguistiquc Romano dc Lyon). Nespor, M. (1990) ‘Vowel deletion in Italian: the organisation of the phonological component’. Linguistic Review, 7: 375-98. Nespor, M. and Vogel, I. (1986) Prosodic Phonology (Dordrecht: Foris).

O'Connor, j. D. (I973) Phonetics (Hatmondsworth: Penguin). Odden, David (2011) ‘The representation of vowel length’, in M. van Oostendorp, C. Ewen, E. V. Hume and K. Rice (eds), The Buzrkweli Companion to Phonology (London: john Wiley 86 Sons): 465-90. Onn, F. M. (1980) /1.gveets of Malay Phonology and M0172/vology (Kuala Lumpur: Universiti Kehangsaan Malaysia). Ohala. (1974) ‘Phonetic explanation in phonology’. in A. Bruck, B. LaGaly and G. Fox (eds), Pop-errflom the Porrzsession on Natural Phonology (Chicago: Chicago

Linguistics Society). Palmer, F. K. (ed.) (1970) Prosoo'ieAnolysir (Oxford: Oxford University Press). Paradis, C. (1988) ‘On constraints and repair strategies’, .Li'ngm'stz'c Review,

6: 71-97. Paradis, C. and Prunet,

F. (1989) ‘On coronal transparency‘, Phonology. 6: 317-48.

Parker, S. (2002) ‘Qt1antiFying the sonority hierarchy’, PhD dissertation. University

of Massachusetts. Parker, S. (ed.) (2012) ‘The sonority controversy’, Phonetics and Phonology, 18 (Berlin: De Gruyter). Pater. ]. (1996) “NO. Proceeding: oft/1e Non‘/1 East Linguistic Society, 26: 227-39 (Amherst. MA: GLSA). Persson, A. and Persson._1. (I980) Sudanese Colloquial/Irooiefor Beginners (Hots-':ly's Green, High Wycombe, Bucks: Summer Institute 0F Linguistics). Picard, M. (1973) ‘Les voyellcs ouvertes du québécois ct Fordonnancc cxtrinseque', Co/nets do linguistique, 7: 71-6. Pierrehumhert, ]., Beckman, M.E. and Ladd, D.R. (2000) ‘Conceptual Foundations of phonology as laboratory science’, in N. Burton-Roberts. P. Carr and

REFERENCES

331

C. Docherty (eds), Phonological Knowledge (Oxford: Oxford University Press):

273-304. Piggott, G. (1987) ‘On the autonomy of the feature [Nasall’, in A. Bosch, B. Need and E. Schiller (eds), CLS 23: Proceedings ofthe Porosession on /lutosegrnentol and 1)/letricol Phonology (Chicago: Chicago Linguistics Society): 223-37. Piggott, G. (1988) ‘A parametric approach to nasal harmony’, in 1-1. van derHu1st and N. Smith (eds), Features, Segmental Structure emel Harmony Processes (Dordrect: ForiS): 131-67.

Pike, K. (1947) Phonemics (Ann Arbor: University of Michigan Press). Pisoni, D.B. (1997) ‘Some thoughts on “normalization” in speech perception’, in K. Iohnson and Mullincx (eds), iitlker Voriohility in Speech Processing (San Diego: Academic Press). Poppe, N. (1965) lntrocluction to /‘Iltoic Linguistics (Wiesbade11: Otto Hartasowitz).

Port, R. (2010) ‘Rich memory and distributed phonology’, Language Sciences, 32(1): 43-55. Prince, A. (1985) ‘Relating to the grid’, Linguistic inquiry, 14: 19-100.

Prince, A. (1990) ‘Quantitative consequences of rhythmic organiratiou’, CLS Parmession on the Syllohle (Chicago: Chicago Linguistic Society): 355-98. Prince, A.and Smolenslty. P. (2004 (1993)) Optimality Theory: Constraint interaction

in Generative Grammar (Maiden, MA and Oxford: Blackwell Publishing) (Also

ROA-537). Pullurn, G. and Zwicky, A. (1988) ‘The syntax-phonology inteifacc’. in F. Newmcyer (ed.), Linguistics: The Cambridge Survey, Volume 1 (Cambridge: Cambridge University Press): 255--80.

Pyun, K. S. (1987) Korean-Swedish lnterlrmguoge Phonology, Koreanological Studies 2 (Stocl-tholm: University of Stockholm institute of Oriental Languages). Roca, 1. and W. Johnson (1999) A Course in Phonology (Oxford: Blackwell). Rubach, (1984) Cyclic and Lexical Phonology: The Structure of Polish (Dordrecht: Foris). Ryan, K. M. (201 1) ‘Gradient weight in phonology’, PhD dissertation, University of California, Los Angelcs. Sagey, E. (1986) ‘The representation ol’ features and relations in non-linear phonology’, PhD dissertation, M1T.

Sagey, E. (1988) ‘Degree ofclosute in complex segments’, in H. van der Hulsr and N. Smith (eds), Features, Segmental Structure oncl Harmony Processes, Part 1 (Dordrect: Foris): 169-208.

Saltarelli, M. (2005) ‘Durational asymmetries and the theory of quantity’, Romance Languages oncl Linguistic Theory (Amsterdam: john Benjamins): 219-55. Sanders, G. (1990) ‘On the analysis and implications of Maori verb alternations’, Lingoo, 80: 149-96. Sampson, G. (1989) ’Language acquisition: growth or learning?’, Philosophical Papers,

18(3); 203-40.

Saporta, S. (1965) ‘Ordered rules, dialect differences and historical change’, Longuage, 41 : 218-24. Schane, S. (1968) French Phonology rmcl Mrirphology (Cambridge, MA: MIT Press).

Schane, S. (1973) Genenztive Phonology (Englewood Cliffs, N]: Prentice Hall). Schane, S. (19'/4a) ‘There is no French Truncation rule’, in R. ]. Campbell, M. G. Goldin and M. C. Wang (eds), Linguistic Studies in Romance Languages (Washington: Georgetown University Press): 89-99.

332

REFERENCES

Schanc, S. (19741)) ‘How abstract is ahst1'act?', in A. Bruck. B. Lz1Galy and G. Fox (eds) (1974) Papemfi-om the Pamsession on Natural Phonology (Chicago: Chicago

Linguistics Society): 297-518. Scobhie, James M. (1991) ‘Attribute-value phonology’, PhD dissertation, University of‘ Edinburgh. Scobhie, James M. (1993) ‘Constraint violation and conflict From the perspective of

Declarative Phonology’, Canadian journal 0fLingui.tric.t, 38: 155-69. Reprinted in Charles Kteidlet (ecl.) Phonology: Critical Concepts. Wilume IV From Rules to

Constraints (New York: Routledge, 2000). Selkirk, E. O. (1980) ‘Tho role of prosodic categories in English word stress’, Linguistic Inquiry, 1 1: 563-605. Selkirk, E. O. (1982) ‘The syllahle’, in H. van der Hulst and N. Smith (eds), The Structure ofPhonological Reprcsenmtions. Part 11 (Dotclrecht: Fotis): 335-83. Selkirk, E. O. (1984) Phonology and Syntax: The Relation between Round and Srrucmr? (Cambridge, MA: MIT Press). Selkirk, E. O. (1986) ‘On derived domains in sentence phonology’, Phonology

lfrrzrhook, 3: 371-405. Sever, E. (1985) ‘An autosegmental analysis ofcompensatory lengthening in Turkish’,

Stmlies in Compensatory Lengthening (Dordrcchtz Fotis): 227-50. Shokri, N. (2001) ‘Swedish lexical word stress’, Proceeding: ofthe Eighteenth Scandinrwizm Conference on Linguistics, 1: l03#13, Lund University. Sibillc, ]. (2004) ‘L’év0lution dcs parlers occitans du btianqonnais, ou comment

la diachronic se teptoduit clans l’espace’. Cahiers dc Gmrnnuzire, 29: 121-41. Toulouse le Mirail.

Silvetman, D. (2006) A Critical lnzroelocrion to Phonology (New York: Continuum). Smith, N. V. and Wilson, D. (1979) Modern Linguistics." The Results of Chomsky}

Revolution (Harmondswotth: Penguin). Smolensky, P. (1995) ‘On the structure ofthe constraint component CON’, R0/1-86. Stetnbcrger, (1935) ‘CV phonology and French consonants: a concrete approach’, journal offingnistics, 21: 453-7.

Swadcsh, M. (1944) ‘South Greenlandic (Eskimo)’, in H. Hoijct et al. (eds) Linguistic Structures ofNotive America, Viking Fund Publications in Anthropology and

Linguistics, (6): 30-54. (Reprinted 1971, Johnson Reprint Co.. New York). Szpyra, J. (1989) The Phonology—Morjo/Jology Interfirce (London: Routlcdge). Ternes. E. (1970) Gmmmoire structtmtle do lvreton ole l’lle ole Groix (Heidelberg. Germany: Carl Winter).

Tcsar, B. and Smolensky, P. (1998) 'Lcarnability in optimality thcoty', Linguistic: Inquiry, 29: 229-68. Topintzi, N. (2010) ‘Onsets: suprasegmcntal and prosodic behaviour’, Cambridge Studies in Linguistics. 125. Tranel. B. (1981) Concrercness in Gcrzernrivc Phonology (Berkeley: University of

California Press).

Trancl, B. (1991) ‘CVC light syllables. gcminatcs and moraic theory’, Phonology.

s(2)= 291-502.

Trancl. B. (1995) ‘On phonetic evidence For the phonological mom‘, Phonology on/1 Phonetic Evidence — Paper: in Lolromcoiji Phonology, 4: 188-201 (Cambridge, UK:

Cambridge University Press).

REFERENCES

333

Truhctzkoy, N. (1939) Grunc/ziigc der P/Jonologie. Travaux du Ccrcle Linguistique dc Prague 7. Republished in 1958: N. Truhetzlcoy (1958). Grunclziige der Phonolo-

gie. Gottingen: Vandenllocck 81 Ruprecht. Translation: C. A. M. Baltaxe (1969): Principles 0fPhonoi'ogy (l.osAnge1es: University of‘ California Press). Virgo, R. (198021) The So:-indPz1ttcrn ofHungarian (\)%1shington, DC: Georgetown University Press). \/ago R. (ed.) (198013) lssncr in Vowel Harmony (Amsterdam: Bcnjzunins). Walker, D. (1984) T/ac Pronnnciririon offlznadiitn French (University of Ottawa). Wate1'son, N. (1956) ‘Some aspects of the phonology of the nominal forms of the Turkish word’, Bulletin oft/Jc School ofO1"icnt/11' and Afiicrzn Studies, 18: 578-91

(Reprinted in F. K. Palmer (1970), Prorociic Analysis (Oxford: Oxford University Press): 174437). Weidert, A. and Subba, B. (1985) Concise Limbo Grammar and Dictionary (Amster-

dam; Lobster). Wier, T. (2004) ‘Compmative markcdness and opacity in Mcskwaki palatalizat'ion', Berkeley Linguistic Sociciy. 30 (Paper presented). \)(/indiuhr, G. L. (1979) Persian Gmnnnrzr (The Hague: Mouton). \X/oocl, S. (1975) ‘Tense and lax vowels: degree oF constriction or pharyngeal vo1ume?’, Worildng Papers in Linguistic: (Lund University): 1()9—34.

Zaharani. A. and Nor Hashima,]. (2010) ‘lsu imhuhzin /di-i/ Bahasa Melayuz apitan atau gabungarf, jurnal Mclayn, 5: 105—26. Zec, D. (1988) Xonoricy Constraints on Prosodic Structure, PhD dissertation, Stanford University.

Zcc, D. (1995) ‘sonority constraints on syllable structure’. P/ionoiogy, 12: 85~129. Zec, D. (1996) ‘The role of moraic structuri-: in the clisrrihution of segments within syllables’, in]. Durand and F. Kzttamba (eds), New Frontiers in Phonology (London: Longman):149—79.

Zec, D. (2003) ‘Prosodic weight’. in C. Péry and R. Van dc Vijver (eds), The Syllrzbic in Optimality T/Jeoijy, (Cambridge: Cambridge University Press). Zcc, D. (2007) ‘The sy|lah|e'. T/Jr Cambridge Book of Phonology (Cambridge: Cambridge University Press): 61-195. Zimmer, K. E. (1967) ‘A note on vowel harmony’, lntcrnntionoijonrnctl 0fAmcrimn Linguistics. 33(2): 166-71. Zimtner, K. E. (1969) ‘Psychological correlates ofsome Turkish morpheme structure conditions’, L/mgicrzge, 45: 309-21.

Subject Index Absolute neutrztlizrttion

Abstractness

124—7

Derived environment

127, 140

Derived environment constraint

Allomorph 33, 240, 285 Allophone 16, 267 Alternations 37. Z49, 253, 268-9, 273,

Anaptyxis

Derived environment effect Dizictitics 134

Diphthongisation

134, 271, 273

287, 289

Assimilation 23, 259, Z64, 280, 295, 303 Autosegmenuil phonology 215

164

286

309

Direct syntax hypothesis

280, 294, 299, 310 Alternntion condition

147, 294

203

Disharmonic roots rec vowel lmrmony Distinctive Features 52-74 Distribution 15, 243, 245-7, 257, 269, 272, 292, .303

Autosegments: ii-cc vs lexicztlly bound 218

Bleeding rec’ Rule ordering

Elision 39 Elsewhere Condition Encliticisation 196

Boundaries

Environments

32, 249. 277

Bounded Vs unbounded feet Bmce notation 21 Brackets 35

see Feet

round brackets and optionality brackets in iexicai phonology

132

15, Z52, 294

Epcnthcsis 40, 259, 261, 277, 287, 289 Eurhythrny 200 Extrametricaliry 138, 242, 254

35

Extrasyllabicity

reeSy1|:tbificati0n

148

Falling diphthongs

184

Chain shifts 286, 295-6 Clash avoidance 197-8

Feature co-occurrence restrictions Feature geometry 233-7

Ciities

Feeding

104

Competence/performance distinction Complement rules 156

Complementary distribution Complex segments

204

sec Rule ordering

Feet

bimotaic Feet 243 binary-branching Feet

15. 269

182

degenerate Feet

16, 79-80

188, 248, 25.5

Filters

258

Fortition

23, 263

Free variation

27

258 Generative grantrnar

1)efi1tilttti1es 73. 156 Degemination 120-1

Glides

Degenerate Feet

Greek letter variables

Derivations

190

quantity sensitivity (weight-sensitivity)

Counter-bleeding 126, 288, 290 Counter-Fcecling 120, 287, 291

CV tier 234 Cyclicity 103, 192, 293-4

191

195, 256

Foot Formation 241 leFt- vs right-lieadedncss

258

Declarative phonology

194

bounded vs unbounded feet

Constraints and repair strategies theory Contrastivity

rec Feet

93, 258

266-7, 274-5. 283, 300

106

44, 250-3, 259, 261-3, 287,

289 derivational morphology derivational theory (D1)

iambic reversal

101 258, 299

72, 262

200

lmplicational] relationships

72

Indirect syntax hypothesis

204

33 4

SUBJECT INDEX Inflectionnl morphology 101 Insertion 262 Iterative appiication 1 18

335

Optimality theory (OT)

*AMBlSYL1,

3044)

‘COMP1,F,X

259-61

CON Left— vs right-headecl Feet tee Feet Length Chapter 10 pimim, 271-2 Lenition Z3. 295 Level ordering 146

lexical category

197

lexical lneaning

34

Lexical phonology Lexical rules

Lexicon

Z58

contour constraints

DEP-IO 259-63 enriched outputs 286, 303, 305. 308 EVAL 258, 282

fititlifhlitess constraints

146-55. 293

259-60, 263,

Z66, Z70, Z72, Z73-9, Z82 GEN 253

145-53. 306

{DENT 250, Chapter 11 p/.4.rsim,

(1l1£l|')[L‘l' 12 prminz Maximal onset principle

265, 267-8, 280

exclusion 266-7 Factorial typology Z66

97, 267, 270-1, Z93, Z99

Markcdiiess

258»-9, 310-1 1

see Syllahification

Chapter ll ;m.m'm, Chapter 12

pasrim Input Chapter ll passim, Z86, Z88, 291-2, 294, 299-301, 305-(1, 309

Metziphony 295 Metricul grid 199

lexicon optimization 299 local conjunction Z97-8, 301, 308

Metriczll trees

markedness constraints

139-97

259-60. 263,

Metricnl ponology 189, 241 Minimal pairs 16, 84-92 Modularity 97

266-7, Z71. 273, Z78, Z32, Z36, Z88, 291, 298-300, 302-3, 308 MAX-IO 260-3

Mora

M86}; conjunction

304

output

Moraic theory

amotaic 242 245-6 bimoraic Chapter 10 przssirn, monomoraic

299, 301

Chapter 11 pasrim, Chapter I2

p1l.~'.tim

275

Z41-2, 245-7, 254, 275

output-to-output correspondence 291-3 positional constraints

263, 268

Morpheme 3'2, 27(1-7,283, 293, 303 Morpheme bount.l:1ry see Bountlttries Moi-plteme structure rules 99

positional markcdness

Z64—7

$()NSPiQ

259-61

Morphological complexity

Strata] (")1

306

trirnoraic

247

32

Morphology 240, 291, 293, 303, 309 Morphoplionemic representations 80

Morpliophonemic rules 80, 136 Morphophonologicztlalternations

richness of the base

turbidity Optionnliry

Ordering see

269, 308

304 35

see Rule ordering

Overappliczttion

see Rule ordering

Alternations Multilateral oppositions

221

Parallel distribution

Parallelism

15

308

Natural classes 75 Natural generative phonology 135 Neutralisation 80, Z6?-8, 270, 279-81

Parameters 190, 241, 231 Pattern congruiry 2.3

Neutral vowels

Phonemic principle

Nucleus

see Vowel harmony

see Syllnble

Pliortemt:

13-31, 79-92

Z7, 79-92

Phonetic content 128 Phonetic representation Phonetic similarity I6

92

Obligatory contour principle 198 Obstruents 53, 245. 252, 267, Z79-80,

Phonological representation

295, 309 Onset rec Syllablc

Plionotacties 99, 250 Postcyclic rules 103

Opacity

Postlexical rules

Cltapterl Z pzmint

Oppositions

221

92

151-3, 306

Prague Sclttml of Lingllistics

221

336

SUBJECT INDEX

Pri\r.ttive oppositions 221 Procliticisarion 196 Prosodic domains

Syllabification syllable

201-6, 243, 275

Prosodic hierarchy prosodic worcl 241, 269

Foot

closed syllable heavy syllable

284; see alto Feet Chapter 10 parsim, Chapter ll

138, 249, 254

light syllable

138, 242, 249

open syllable

169, 247-8, 271

onset

parrirrl, Chapter 12 passim

mora

169, 247-8, 289-90,

304-5, 309

241, 243, 249, 256, 26.7, 275-6,

syllable

171-6

168, 241-2, 246-7, 250-2, 264,

266, 270-1, 279-81, 283,

304

Chapter 12 prmim rhyme (I time)

Quantity sensitivity

Realisation Recu|'$iun

tee Feet

nucleus

coda

16, 258

Resyllahification see Syllabificarion Rhyme see Syllable Rhythm rule see lambic reversal

Rising tliphthongs 185 Root 32, 273, 276-7, 279-80, 293. 300, 307, 30‘) Rule ordering bleeding 121, 287 counter-bleeding 126 counter-Feeding 120

I69, Chapter 10 pzrsrim, 259, 261

298-9, 304 syllable weight 248, 257 zero syllable Tonal tier

Tones 197 True (ieneralisation Condition

135

see Rule ordering rev

Phonological representation Underspecification 100, 155. 235 theory 155 Utterance accent 205

over-application 287,289 underapplication 290-1

Via rules Sentence stress 205 Selitence/utterance distinction Seriality 258 Sonorants 52, 290 303-8

204

Sonoriry hierarchy see Syllabification Stern 32 Straral OT see Optimality theory

Strengrlrening we For-titinn Stress Feet see Feet Stress suhortlination convention

136

Vowel harmony

Sonority

Structural change

198

Underlying representation

128

directionality 223 disharmonic roots 222 clominanrlrecessive harmony neutral vowels 122 opaque morphemes 232 opaque vowels

192

164, 294

105

Structural description 105, 269 Structure preservation 153, 286 Surface representations see Phonetic representation

Weigl1t—to—stress

230- 3

222

\1(/eight~by-position

24-9

Strict cyclicity condition

195

Undcrapplication

feeding 1 19, 287, 290 ordering paradoxes 131

Stress-to-weight

51, 241, 245, 300

264, 266-7, 270-1, 279-81, 284.

292-3

extrinsic ordering

168, 240-3, 274

242-/1,274

249

Well-Formedtress conditions Word-Formation braclosts

258. 276 56¢‘ l’>1'ilClt1:ts

Word stress assignment 242, 250 primary stress 242, 256, 275 Zero morphs 43, 277 Zero syllables tee Syllable

Languagelndex

Akan (Kwa) 60. 1141-5, 133. 220-3 Arabic 5, 6. 244, 246, 249, 254

Eiision 140 /c/ Lowering

Hasquc

Glide Fnrmarion 166-7, 29‘)Liaison 130, 177

161-2

11¢-.113 C0013

Bengali

Middic Frc11c11 251 nasal vuwc-1 lowering 87 Nasalisarion 84-6, 150-I Normandy French 252, 293

245

242, 283

licrbcr

245

Breton (Celtic)

172

256

Norrliwestcrn French (lapanaiiua (iarib

Québécois 1-‘ranch 245, 272-5 305-7 Schwa Dclcliuii 85-6

255

256

(I.1ra1a.n (Ro|n;mce) (Ihaha

'1'runcari¢m

273, 310

Friulian

245

American 1".ng1is11 Canadian Engiisli

151-2 308

compound stress ru1c

flapping

2/15, 253

Swiss (.]crm;m 310 (lokana (Niger Kurdnfanian)

192

147

1'7.ng1is11 Stl'