The Blackwell Companion to Phonology 140518423X, 9781405184236

Table of contents :
Front Matter
1 - Underlying Representations
2 - Contrast
3 - Learnability
4 - Markedness
5 - The Atoms of Phonological Representations
6 - Self-organization in Phonology
7 - Feature Specification and Underspecification
8 - Sonorants
9 - Handshape in Sign Language Phonology
10 - The Other Hand in Sign Language Phonology
11 - The Phoneme
12 - Coronals
13 - The Stricture Features
14 - Autosegments
15 - Glides
16 - Affricates
17 - Distinctive Features
18 - The Representation of Clicks
19 - Vowel Place
20 - The Representation of Vowel Length
21 - Vowel Height
22 - Consonantal Place of Articulation
23 - Partially Nasal Segments
24 - The Phonology of Movement in Sign Language
25 - Pharyngeals
26 - Schwa
27 - The Organization of Features
28 - The Representation of Fricatives
29 - Secondary and Double Articulation
30 - The Representation of Rhotics
31 - Lateral Consonants
32 - The Representation of Intonation
33 - Syllable-internal Structure
34 - Precedence Relations in Phonology
35 - Downstep
36 - Final Consonants
37 - Geminates
38 - The Representation of sC Clusters
39 - Stress Phonotactic and Phonetic Evidence
40 - The Foot
41 - The Representation of Word Stress
42 - Pitch Accent Systems
43 - Extrametricality and Non-finality
44 - The Iambictrochaic Law
45 - The Representation of Tone
46 - Positional Effects in Consonant Clusters
47 - Initial Geminates
48 - Stress-timed vs Syllable-timed Languages
49 - Sonority
50 - Tonal Alignment
51 - The Phonological Word
52 - Ternary Rhythm
53 - Syllable Contact
54 - The Skeleton
55 - Onsets
56 - Sign Syllables
57 - Quantity-sensitivity
58 - The Emergence of the Unmarked
59 - Metathesis
60 - Dissimilation
61 - Hiatus Resolution
62 - Constraint Conjunction
63 - Markedness and Faithfulness Constraints
64 - Compensatory Lengthening
65 - Consonant Mutation
66 - Lenition
67 - Vowel Epenthesis
68 - Deletion
69 - Final Devoicing and Final Laryngeal Neutralization
70 - Conspiracies
71 - Palatalization
72 - Consonant Harmony in Child Language
73 - Chain Shifts
74 - Rule Ordering
75 - Consonant-Vowel Place Feature Interactions
76 - Structure Preservation the Resilience of Distinctive Information
77 - Long-distance Assimilation of Consonants
78 - Nasal Harmony
79 - Reduction
80 - Mergers and Neutralization
81 - Local Assimilation
82 - Featural Affixes
83 - Paradigms
84 - Clitics
85 - Cyclicity
86 - Morpheme Structure Constraints
87 - Neighborhood Effects
88 - Derived Environment Effects
89 - Gradience and Categoricality in Phonological Theory
90 - Frequency Effects
91 - Vowel Harmony Opaque and Transparent Vowels
92 - Variability
93 - Sound Change
94 - Lexical Phonology and the Lexical Syndrome
95 - Loanword Phonology
96 - Experimental Approaches in Theoretical Phonology
97 - Tonogenesis
98 - Speech Perception and Phonology
99 - Phonologically Conditioned Allomorph Selection
100 - Reduplication
101 - The Interpretation of Phonological Patterns in First Language Acquisition
102 - Category-specific Effects
103 - Phonological Sensitivity to Morphological Structure
104 - Root-Affix Asymmetries
105 - Tier Segregation
106 - Exceptionality
107 - Chinese Tone Sandhi
108 - Semitic Templates
109 - Polish Syllable Structure
110 - Metaphony in Romance
111 - Laryngeal Contrast in Korean
112 - French Liaison
113 - Flapping in American English
114 - Bantu Tone
115 - Chinese Syllable Structure
116 - Sentential Prominence in English
117 - Celtic Mutations
118 - Turkish Vowel Harmony
119 - Reduplication in Sanskrit
120 - Japanese Pitch Accent
121 - Slavic Palatalization
122 - Slavic Yers
123 - Hungarian Vowel Harmony
124 - Word Stress in Arabic

Citation preview

Front Matter

6

1 - Underlying Representations

36

2 - Contrast

62

3 - Learnability

89

4 - Markedness

114

5 - The Atoms of Phonological Representations

142

6 - Self-organization in Phonology

165

7 - Feature Specification and Underspecification

183

8 - Sonorants

206

9 - Handshape in Sign Language Phonology

230

10 - The Other Hand in Sign Language Phonology

258

11 - The Phoneme

276

12 - Coronals

302

13 - The Stricture Features

323

14 - Autosegments

346

15 - Glides

376

16 - Affricates

402

17 - Distinctive Features

426

18 - The Representation of Clicks

451

19 - Vowel Place

475

20 - The Representation of Vowel Length

500

21 - Vowel Height

526

22 - Consonantal Place of Articulation

554

23 - Partially Nasal Segments

585

24 - The Phonology of Movement in Sign Language

612

25 - Pharyngeals

639

26 - Schwa

663

27 - The Organization of Features

678

28 - The Representation of Fricatives

704

29 - Secondary and Double Articulation

729

30 - The Representation of Rhotics

746

31 - Lateral Consonants

765

32 - The Representation of Intonation

792

33 - Syllable-internal Structure

816

34 - Precedence Relations in Phonology

834

35 - Downstep

859

36 - Final Consonants

883

37 - Geminates

908

38 - The Representation of sC Clusters

933

39 - Stress Phonotactic and Phonetic Evidence

959

40 - The Foot

984

41 - The Representation of Word Stress

1015

42 - Pitch Accent Systems

1038

43 - Extrametricality and Non-finality

1062

44 - The Iambictrochaic Law

1087

45 - The Representation of Tone

1113

46 - Positional Effects in Consonant Clusters

1138

47 - Initial Geminates

1159

48 - Stress-timed vs Syllable-timed Languages

1182

49 - Sonority

1195

50 - Tonal Alignment

1220

51 - The Phonological Word

1239

52 - Ternary Rhythm

1263

53 - Syllable Contact

1280

54 - The Skeleton

1298

55 - Onsets

1320

56 - Sign Syllables

1344

57 - Quantity-sensitivity

1370

58 - The Emergence of the Unmarked

1398

59 - Metathesis

1415

60 - Dissimilation

1443

61 - Hiatus Resolution

1469

62 - Constraint Conjunction

1496

63 - Markedness and Faithfulness Constraints

1526

64 - Compensatory Lengthening

1548

65 - Consonant Mutation

1572

66 - Lenition

1594

67 - Vowel Epenthesis

1611

68 - Deletion

1632

69 - Final Devoicing and Final Laryngeal Neutralization

1657

70 - Conspiracies

1679

71 - Palatalization

1701

72 - Consonant Harmony in Child Language

1726

73 - Chain Shifts

1752

74 - Rule Ordering

1771

75 - Consonant-Vowel Place Feature Interactions

1796

76 - Structure Preservation the Resilience of Distinctive Information

1822

77 - Long-distance Assimilation of Consonants

1846

78 - Nasal Harmony

1873

79 - Reduction

1901

80 - Mergers and Neutralization

1927

81 - Local Assimilation

1954

82 - Featural Affixes

1980

83 - Paradigms

2007

84 - Clitics

2037

85 - Cyclicity

2054

86 - Morpheme Structure Constraints

2084

87 - Neighborhood Effects

2105

88 - Derived Environment Effects

2124

89 - Gradience and Categoricality in Phonological Theory

2150

90 - Frequency Effects

2172

91 - Vowel Harmony Opaque and Transparent Vowels

2199

92 - Variability

2225

93 - Sound Change

2249

94 - Lexical Phonology and the Lexical Syndrome

2271

95 - Loanword Phonology

2293

96 - Experimental Approaches in Theoretical Phonology

2318

97 - Tonogenesis

2339

98 - Speech Perception and Phonology

2369

99 - Phonologically Conditioned Allomorph Selection

2392

100 - Reduplication

2418

101 - The Interpretation of Phonological Patterns in First Language Acquisition

2449

102 - Category-specific Effects

2474

103 - Phonological Sensitivity to Morphological Structure

2499

104 - Root-Affix Asymmetries

2525

105 - Tier Segregation

2551

106 - Exceptionality

2573

107 - Chinese Tone Sandhi

2596

108 - Semitic Templates

2621

109 - Polish Syllable Structure

2644

110 - Metaphony in Romance

2666

111 - Laryngeal Contrast in Korean

2697

112 - French Liaison

2720

113 - Flapping in American English

2746

114 - Bantu Tone

2765

115 - Chinese Syllable Structure

2789

116 - Sentential Prominence in English

2813

117 - Celtic Mutations

2842

118 - Turkish Vowel Harmony

2866

119 - Reduplication in Sanskrit

2890

120 - Japanese Pitch Accent

2914

121 - Slavic Palatalization

2943

122 - Slavic Yers

2971

123 - Hungarian Vowel Harmony

2998

124 - Word Stress in Arabic

3025

This edition first published 2011 © 2011 Black\vell Publishing Ltd Blackwell Publishing \Vas acquired by John \Viley & Sons in February 2007. Blackwell's publishing progran1 has been merged with Wiley's global Sientific, Technical, and Medical business to form Wiley-Blakwell.

Registered Offie john Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, Vest Sussex, P019 8SQ, United Kingdom

Editorial Oices 30 Main Street, fV!alden, MA 02148-5020, USA

9600 Garsington Road, Oxford, OX4 2DQ, UK The Atrium, Southen Gate, Chicnester, West Sussex, P019 SSQ, UK For details of our global editorial offices, for customer services, and for information about how to apply for perms i sion to reuse the copyright material in this book please see our website at W\'fW.\'filey.com/wiley-black\vell. The right of l\1arc van Oostendorp, Colin}. Ewen, Elizabeth Hume, and Keren Rice to be identified as the authors of the editoriaJ material in th.is work nas been asse1ted in accordance with the UK Copyright, Designs and Patents Act 1988. All rights reserved. No part o f this pubication n1ay be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher. Viley also publisJ1es its books in a vadety of electronic formas. t Some content that appears in print may not be available in electronic books. Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trade1narks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book. This publication is designed to provide accurate and authoritative informaton in regard to the subject niatter covered. It is sold on the understanding that the publsher s not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought. Library of Congress Cataloging-in-Publication Data The Blackwell companion to phonology / edHed by l\1arc van Oostendorp . . . [et al.]. p. cm. - (Blackwell con1panions to linguistics series) Includes bibliographical references and index. JSBN 978-·1-4051-843·6 (hardcover : alk. paper) 1. Phonetics. 2. Grammar, Comparative and general-Phonology. 1. Oostendorp, Marc van, 1967P217.B53 2011 414-dc22 2010042206 A catalogue record for this book s available fron1 the British Library. Set in l0/'12pt .Paatino by Craphicraft Limited, Hong Kong

1

2011

Bahan dengan hak cipta

Brief Contents

VolumeI

Preface

us

General Issues and Segmental Phonology

I .

XXIX

1

VolumeII Suprasegmental and Prosodic Phonology

757

VolumeIII Phonological Processes

1363

VolumeIV Phonological Interfaces

1945

VolumeV Phonology across Languages

2561

q

Bahan dengan hak cipta

Bahan dengan hak cipta

Full e of s

VolumeI

Prfe ace

XXJX

General Issues and Semental Phonoloy 1 Underlying Representations, &fose g I nacio 2 Contrast & 3 Learnability, v 4 Markedness Eliabeth Hume 5 The Atoms of Phonological Representations, Maranne Pou.plier 6 Self-organization in Phonol, 7 Feature Specfication and Underspecfication, r lang i 8 Sonorants 9 Handshape n Sgn Language Phonoloy, Brentari i 10 The Other Hand in Sign Lae Phonoloy, Crasborn 1 1 The Phonen1e 12 Coronats 13 The Stricture Features Ellen 1 4 Autosegments, R. Leben 15 Glides V. 16 Africates 17 Distinctive Features, f{Mielke 18 The Representation of Clicks, 19 Vo-el Place, 20 The Representation of Vo\vel Length, 21 Vovel Height, 22 Consonanta. Place of Articulation i ai 23 Parta i lly 'asal Segments, Riehl & g Cohn 24 The Phonoloyof Movement Sign Language, Sandler 25 Phaeals, 26 Sln\'a

Hua/de

fennifer Cole Daniel Currie Hall Heinz fason gle

Andrew vVedel

Diana A cl el Diane Onno

Bert Botma

B. Elan Dresher T. A. Hall M. Kaisse Willia111 Susannah Levi Yen-Hwei Lin Amanda Miller Bruce vloren-Duolljti David Odden Douglas Pulleyblank Keren Rice Ab l C. Anastasia K. n Wendy Kiman1Shahin Daniel Silvennan ,

1

27 54 79

107 130 148 171

195 223

241

267

288 311 341

367 391 416 440 465 491 519

550

577

604

628 Copyrightd maerial

vrn

27 28 29 30 31

Full Table of Contents Christia. f11ann Bert Vaux & Brett Miller Jeroen van de /eijer Richard Wiese Moira Yip

The Organization of Features, The Representation of Fricatives, Secondary and Double Articulation, The Representation of Rhotics, Lateral Consonants,

643 669 694 711 730

I VolumeI

Suprasemental and Prosodic Phonoloy

32 The Representation of Intonation, Arna/in Arvaniti 33 Syllable-internal Structure, Anna R. K. Bosd1 34 Precedence Relations in Phonol, Charles Cairns Eric Rahlll( 35 Downstep, Bruce Connell 36 Final Consonants,Marie-.He/ene Cote 37 Geminates Stuart Davis 38 The Representation of sC Clusters, Heather Goad 39 Stress: Phonotactic and Phonetic Evidence,Matthew Gordon 40 The Foot Michael Hammond 41 The Representation of Word Stress, Be. Hermans 42 Pitch Accent Systems, Harn1 der Hulst 43 Extrametricality and Non-finaliyt , Brett Hvde 44 The Ian1bic-Trochaic La\v, Brett H11de 45 The Representation of Tone, Larn1M. HI/man 46 Positional Effects in Consonant Clusters,Jong/toJun 47 Initial Geminates,Astrid Kraehennrnnn 48 Stress-timed vs. Syllable-timed Languages, Marina Nespor, Mohinish Shukla &Jacques Mehler 49 Sonority, Steve Parker 50 Tonal Alignment, Pilar Prieto 51 The Phonological /ord, Anthi Revithiadou 52 Ternary Rhythm, Curt Rice 53 Syllable Contact, Misun Seo 54 The Skeleton, Peter Szigetvtfri 55 Onsets, Nina Topintzi 56 Sign Syllables, Ronnie lilbur 57 Quantiyt -sensitivity, Dra.gn Zee &

va.

757 781 799 824 848 873 898 924 949 980 1003 1027 1052 1078 1103 1124 1147 1160 1185 1204 1228 1245 1263 1285 1309 1335

Volume III Phonological Processes &

58 The En1ergence of the Unn1arked, Mich.el Becker Kathryn Finck Potts 59 Metathesis, Eugene Buckley 60 Dissimilation, Patrik Bye 61 Hiatus Resolution, Roderic F. C.sali 62 Constraint Conjunction, Megan . Crowhurst

1363 1380 1408 1434 1461 Copyrighted maerial

. M

i

6 Z 8

2

Q

l

2

. i i 6 2

8 9 Q

ll

Full T.ble of Contents Markedness and Faithfulness Constraints, Paul de Lacv Co1npensatory Lengthening, Randall Gess Consonant Mutation,Janet Gri;zenhout Lenition,Naomi Gurevich Nancv.Hall Deletion,fohn Han·is Final Devo cing and Final Lea Neutralization, GregoryK. Iverson &Joseph C. Salmons Conspiracies, Clzarles W. Kisseberth Palatalization,Alexei Kochetov Consonant Harmony n Child Language, Clara C. Levelt Chai Shifts,Anna Lubo111icz Rule Ordering,Joan Mascaro Consonant Vo el Place Feature Interactions,11e Padgett Vo,vel Epenthesis, i

l

n

-

•

v

Structure Preservation: The Resilience of Distinctive Information, Carolearadiq&Dar l e n el a C h a r t e

f Consonants, Sharon Rose Harmony, R.chel Walker ,Natasha Varner !ergers and Neutralization, Alan C. L. Yu Local Assimilation, Elizabeth C. Zsi ga Long-distance Assimilation o Nasal Red uction

IX

1

3 7

59

6

7

22

44 66 1

7 6

1

7

11 8

66 2

9

VolumeIV

Phonological Interfaces 2

Akin/Jivi Akinlabi Adam Albright s Stephen R. Anderson Ricardo Ber11 1idez-Otero Geert Booii Admn Buchwald Environment Effects, LuigiBurzio Mirjam Ernestus Frequency Effects, Ste fan A. Frisch o e Harmony: ue and Transparent Vo\vels, Adainantios I. Gafos Amanda. Dve Variability, Gregor(R. Guf Sound Change, Jose nacio Hua/de Lexical Phonoloyand the Lexical Syndrome, Ellen M. Kaisse & April McMahon Loanvord Phonoloy, YoonjungKang Exp r ntal ppr ac he in Theoretical Phonology, Sliigeto Kawahara Tonogenesis, Jolin Kingston Speech Perception and Phonoloy, Andreiu Martin & Sharon Peperkamp

Featural Affixes, . Paradi i ns, . C.tic , t . Cycliciy . Morphen1e Structure Constraints, Z Neighborhood Effects, 8 Derived � Gradience and Categor icalit y in Phonolog ical Theory, ,

0

. V w l . 3

� . 6

Z 8

e

&

ime

o

s

5

2

02

9

9 0 9

15

7 4 0 4 6

8

3 4

34

Copyrightd maerial

x

Full Table of Contents

99 Phonologically Conditioned Allomorph Selection, Reduplication, The Intepretation of Phonological Pattems in First Language Ac uisition, & 1 2 Category-specific Effects, L. Phonological Sensitivity to Morphological Structure, Root-Affix Asymmetries, Tier S tion, Exeptionality

100 101 0 103 104 105 106

Eric Rai111y Yvan Rose Sharon lnkelas q Jennifer Smith Suzanne Llrbanczyk a Adam Ussishkin Matthew Wolf

Andre1v Nevins

,

2357 2383

fochen Tro1111ner

2414 2439

2464 2490

2516 2538

VolumeV

Phonoloy across Lan.ages

Bao Zhiming 108 Semitic Templates, Outi Bat-El 109 Polish Syllable Structure, Christina Y. Bethin 110 Meta hony in Romance, Andrea Calabrese 1 - ee Yu Cho 111 La al Contrast in Korean, Yo g 112 French Liaison,Marie-Helene Cote 113 Flpgn American English, Kenneth /. de Jong 114 Bantu Tone, Laur..Downing 115 Chinese Syllable Structure, San Duanmu 116 Sentential Prominence in En li h, Carlos Gussenhoven 117 Celtic Mutations, S.[. .Hannahs 118 Turkish Vo''el Harmony, Baris Kabak 119 Reduplication n Sanskrit, l{obert Kennedv 120 anese Pitch Accent, Haruo Kubozono 121 Slavic Palatalization,[erz11Rubach 122 Slavic Yers Tobias Scheer 123 Hungar n Vo,vel Harmony, Miklos Torkencz11 124 Word Stress in Arabic,[anel C. E. lalson 107

s

2561 2 8

p

2631 2662 2685 2711 2730 2754 2778 2807 2831 2855 2879

Chine e Tone Sandhi,

e

1

g

a i

s

1

m

56 2609

2908

2936 2963 2990

Copyrighted maerial

Contributors

Akinbiyi Akinlabi is Professor of Linguistics at Rutgers University, New

Bnu1S\\'ick. He is the current President of the vVorld Congress of African Linguistics, and serves on the councils of both the West African Languages Congress and the Annual Conference on African Linguistics (USA). His expertise lies primarily in the phonologies of the Benue-Congo languages of West Africa. He has publshed articles in leading journals of theoretical and descriptive linguistics, including and His forthcon1ing book is titled:

Linguistic Inq11iry, fo11rnal of Linguistics, journal of African Languages and Linguistics. Yoruba: A phonological grammar. Adam Albright is an Associate Professor (Anshen-Chomsky Professorship in

Language, and ilind Career Development Chair) at MIT. His research interests nclude i phonology, 1norphology, and learnability, with an emphasis on using computational modeling and experimental techniques o investigate issues in phonological theory. Stephen R. Anderson is Dorothy R. Diebold Professor of Linguistics at Yale

i n 1994. Aspects of the theory of

University. After receiving s Ph.D. from MIT in he taught at Harvard, UCLA, Stanford, and Johns Hopkins Universities before coming to Yale He is the author of nun1erous articles and six books, including He has done field research on several languages roost recently the Surmiran form of Rumantsch. In addition to phonology and morphology, his research interests include animal comnn1nication systems and the evolution of human language.

1969,

clitics (2005).

,

Dian a Archangeli received her Ph.D. from MIT in

1984. She has been a faculty member at the University of Arizona since 1985 and ''as a Felio'" at the Center for Advanced Study in the Behavioral Sciences at Stanford in 2007-8. She co-authored Grounded phonology (1994) "'ith Douglas Pulleyblank, and co-edited Optima.lij Theory: An overview (1997) with D. Terence Langendoen. She is

Director of the Arizona Phonological In1aging Lab, used for ultrasound study of the articulation of language soi.mds. Amalia Arvaniti s an Associate Professor n the Department of Linguistics at the

University of California, San Diego. She received her Ph.D. fron1 the University of Ca1nbridge and has held research and teaching appointments at the Universities Copyrightd maerial

xii

Contributors

of Cambridge, Oxford, Edinburgh, and Cyprus. Her research focuses on the phonetics and phonology of prosody, with special emphasis on the experin1ental investigation and forn1al representation of rhytlun and into.nation. Bao Zhiming is a linguist working in the Department of English Language and

Literature, National University of Singapore. He has nvo main research interests: Chinese phonology and contact linguistics.

Outi Bat-El is Professor of Lingui tics at Tel-Aviv University. She is engaged in the study of Semitic phonology and morphology. In her & article, she initiated the Semitic root debate, arguing that thre is no consonantal root in Se1nitic n1orphology. Her subsequent \Vork •vithi.n the ran1evork of Optrnality Theory provided further support to this argument. She has also authored articles on Hebrevv truncation and reduplication on blends and hypocoristics as 'ell as on language acquisition

s

Linguistic Theory

1994 Natural Language

(Recherches Linguistiques de Vincennes 2003) (Language 2002) (Linguistic Inquiry 2006), (Phonology 1996) (Phonology 2005), (Langrwge Sciences 2009).

Michael Becker received his Ph.D. from the University of Massachusetts Amherst

2009,

in and is currently a lecturer at Harvard University. His \Vork focuses on the gra1nn1atical prmciples that govern lexicon organization, especially as a \vay to discover Universal biases m the phonolog ica l grammar. He a lo vvorks on the acquisition and learning of lexical patterns. Ricardo Bermudez-Otero is Senior Lecturer in Linguistics and English Language at the University of Manchester. He previously held a Postdoctoral Fellorship at

the British Academy, followed by a Lectureship in Linguistics at the University of Nevvcastle upon Tyne. His research focuses on the morphosyntax-phonology and phonology-phonetics n i terfaces, •vith particular attention to diachronic issues. language change His publications mclude rnapters m

Optilnality Theory and (2003), The handbook of English linguistics (2006), The Cambridge handbook of phonology (2007), Deponency and 111orphological n1is1natches (2007), Optimality-theoretic studies in Spanish plwnology (2007), and Morphology and its interfaces (forthcoming).

Christina Y. Bethin is Professor of Linguistics at Stony Brook Univesity, N' York,

vho 'Orks on prosody and syllable structure. She has 'ritten nun1erous articles on the diachronic and synchronic phonology of the Slavic languages, including Polish, Ukraiian, Russian, Belarusian, Serbian and Croatian, Czech, and Slovene, and published tvo award-vng boo ks and

, Polish syllables: The role of prosody in Slavic prosody: Language change and phonological

phonoloi; and morphology (1992) theory (1998). Geert Booij is Professor of Linguistics at the University of Leiden. He is the author of The phonology of Dutch (1995), The morphology of Dutch (2002), The grammar of words (2005), and Construction morphology (2010), and of a number

of linguistic articles in a \vide range of Dutch and international journals, with focus on the interaction of phonology and morphology, and theoretical issues in morphology. Anna R. K. Bosch has taught linguistics at the University of Kentucky since

1990,

vhere she s now Associate Dean for Undergraduate Progran1s and a faculty n1ember of the Lmguistics Program and English departments. She has written on Copyrightd maerial

Contributors

Xlll

Scottish Gaelic phonology and dialectology, and is currently ''orking on a new project on the history of phonetic transcription n European dialect studies. Bert Botma is associate professor at the Leiden University Centre of Linguistics (LUCL) and a research fello' at the Nethelands Organization of Scientfi ic Research (NWO). His main research interest is segmental phonology. He has published on such topics as nasal harmony, English syllable structure, and on the phonological contrast between obstruents and sonorants. Diane Brentari is Professor of Linguistics and Director of the ASL Program at

Purdue University. She has published widely in the area of sign language phonology and morphology. She is the author of (1998) and editor of (2010). Her current research involves the cross-linguistic analysis of sign languages.

A prosodic rnodel ofsign language Sign languages: A C.mbridge language sun1ei;

phonologt;

Adam Buchwald is an Assistant Professor at Ne" York University in Com1nun­ icative Sciences and Disorders. He holds a Ph.D. in Cognitive Science from Johns Hopkins University and had a post doctoral felJo,vship in the Speech Research Lab at Indiana University. His 'Ork is interdisciplinary and spans topics in linguistics, psycholinguistics, and communication sciences. Eugene Buckley is Associate Professor of Linguistics at the University of

Pennsylvania. His research interests include metrical and syllable structure, and the proper locus of phonetic and functional explanation. Much of his ''Ork focuses on native languages of North America. Luigi Buzio is Profesor Emeritus, Department of Cognitive Science, Jolu1s Hopkins

University. He has also taught at Harvard University, Department of Ron1ance Languages and Literatures. His research interests include syntax, phonology and (1986) and morphology. He is the author of (1994).

Italian syntax

Principles ofEnglish stress

Pa.rik Bye is currently a researcher "'ith the Center for Advanced Study in

Theoretical Linguistics at the University of Troms0. He has published scholarly articles on a variety of topics including the syllable structure, quantity, and stress systems of the Finno-Ugric Saami languages, North Germanic accentology, and phonologically conditioned allomorphy. With Martin Kramer and Sylvia Blaho (2007). he edited F ee o

r d m of analysis?

Charles Cairns is Professor En1eritus of Linguistics at the City University of Ne'" York. He received his Ph.D. in linguistics from Colun1bia University in 1968. Author of a number of articles in phonology, his most recent \VOrk is a co-edited volume with Eric Raimy,

phonology (2009).

Contemporary vie1vs on architecture and represi1tations in

Andrea Calabrese '"as born in Campi Salentina in the southeastern tip of Italy. He obtained his Ph.D. in linguistics at MT in 1988 and is currently teaching at

the University of Connecticut. His main interests are phonology, 1norphology, and historical linguistics. He has published n1ore than articles in books and journals such as Revierv, His recent book, (2005), proposes a theory integrating phonological rules and repairs triggered by 1narkedne.ss con­ straints into a derivational model of phonology.

50 Linguistic .lnquiry, Linguistic Studies in Language, Brain and Language, Journal of Neuro-Linguistics, and Rivista di Linguistica. 1arkedness and econom.y in a derivational rnodel of phonology

Copyrightd maerial

xiv

Contributors

Roderic F. Casali's nterests nclude phonetics, phonological theory, and deriptive vork on the phonology of African languages. His research has focused prmarily on vowel p hno1nena especa i lly vo\vel hiatus resolution and ATR VO\\•el harn1ony. He has done linguistic fieldwork in Ghana with SIL and currently teaclles lin­ guistics at the Canada Institute of Linguistics at Trinity Western University in Langley, British Columbia.

,

Young-mee Yu Cho is Associate Professor of East Asian Languages and Cultures at Ru tgers University and author of

Parmneters of consonantal assin1ilation (1999), Integrated Korean (2000), and Korean phonology and morphology (forthcoming). She has written on Korean .angu age and culture, theoretical lingui stics, and Korean pedagogy. Abigail C. Cohn is Professor of Lingu istics at Cornell University, Ithaca, Nev

Yo rk. Her research addresses the relationship between phonology and phoetics and is informed by laboratory phonology approaches. She also specializes in the description and analysis of a nun1ber of Austronesian languages of Indonesia. Her published Vork includes articles in and a number of edited volun1es and she is co-editor of the forthcoming

Phonology Oxford handbook of laboratory phonology.

Jenn if er Cole is Professor of Linguist ics and Computer Science at the University

of Illinois at Urbana-Champaign, and is a member in the Cognitive cience group at the Beckman Institute for Advanced Science and Technology. She is the and has served on the editorial founding General Editor of boards of the journals and

La.bora.tory Phonology, Language, Linguistic Inquiry,

Phonology.

Bruce Connell is baed at York University, Toronto. His research interests

include the phonetics of African languages, the relationships betveen phonetics and phonology, historical linguistics, and language endangerment and docu­ mentation. H.e makes regula r fieldtrips to Africa for research and is an authority on languages in the Nigeria-Cameroon borderland. He received his Ph.D. in His publications include Linguistics from the University of Edinb urgh in "The perception of lexical tone in Vlan1bila," 43, and "Tone languages and the universality of intrinsic FO: Evidence fron1 Africa," �(

1991. L.nguage and Speech

Phonetics 30.

Journal

Marie-Helene Cote is Associate Professor of Linguistics at the University of Otta va.

Mucll of her reearch n phonology relates to the role of perceptual factors n phono­ logical processes, the status of the syllable, and the treatment of variation. She also specizes in French phonology, Vith a focus on Laurentian (Quebec) French. Onno Crasbon is a Senior Researcher at the Deparhnent of Linguistis of Radboud

University Nijmegen, where he heads the sign language research theme of the Centre for Language Studies. After ompleting a dissertation on phonetic variation in Sign Language of the Netherlands at Leiden University, he has broadened his researcll interests beyond sign phonetics and phonology to sociol inguistics, discourse, and corpus linguistics. In he published the Corpus NGT, the first open access sign language corpus in the \Vorld.

(2001) 208

Megan J. Crowhurst is an Associate Professor in the Department of Linguistics

at the University of Texas at Austin. Her publications in thoretical phonology have concentrated on prosodic phenomena related to \VOrd stress and reduplication. Copyrightd maerial

Contributors

xv

In collaboration "'ith Mark Hewitt, she has also contributed to the literature on constraint conjunction. Her current research, conducted in an experimental paradign1, explores the ways \\1hich humans' perception of rhythm might con­ tribute to the form and frequency of stress patterns found in natural languages.

n

Stuart Davis is Professor of Linguistics at Indiana University, Bloomington.

Hs primary area of research is in phonology and phonological theory with a secondary area of reearch in the early history of linguistics in the USA. His work n phonology has especially focused on issues related to syllable struchrre and '"ord-level prosody. His work has appeared in a Vide variety of edited volumes and journals including and

Linguistic Inquiry, Phonology, Lingua, Linguistics, Ameri11n Speech. Kenneth de Jong is currntly Proessor of Li guistics, Cognitive Science and Second n

Language Studies at Indiana University. He has \vorked extensively on questions of ho"' prosodic organization pervades the details of speech production, and more generally how speech production and perception interact '"ith one another and ho"' this relates to the phonological system. He is author of more than articles on aspects of phonetic behavior and is currently Associate Editor of the

50

Journal of Phonetics.

Paul de Lacy is an Associate Professor n the Department of Linguistics at Rutgers

University, an Associate of the Rutgers Center for Cognitive Science, Co-director of the Rutgers Phonetics and Field\vork Laboratory, and editor of He works on phonology and its interfaces \vith syntax, morphology, and phonetics.

the Cmnbridge

handbook of phonology.

Laura J. Downing s a research fellow at the ZAS, Berlin, leading projects on the

phonology-focus-syntax interface in Bantu languages. She has published several articles on tone and depressor tone, morphologically-conditioned morphology, syllable struchrre and prosodic morphology Bantu languages, and a book,

Canonica/forms in prosodic 11101·phology (2006).

n

B. Elan Dresher is Professor of Linguistics at the University of Toronto. He has published on phonological theory, learnability, historical linguistics, and Vest Germanic and Biblical Hebre"' phonology and prosody. He is the author of and "Chomsky and Halle's revolution in phonology" n the Hs recent books include (co-ed., with Nila Friedberg, (coed., �vith Peter Avery and Keren Rice, and

Old

English and the theory ofphonology (1985) Cambridge companion to Chon1sky (2005). Formal approaches to poetry: IZecenl develop111ents in generative ·metrics 2006), Contrast in phonology: Theory, perception, acquisition 2008), The contrastive hierarchy in phonology (2009). San Duanmu is Professor of Linguistics, University of Michigan, "'here he has taught since 1991. He obtained his Ph.D. in linguistics from MIT in 1990 and from 1981 to 1986 held a teaching post at Fudan University n Shanghai.

Amanda Dye is a graduate student in Linguistics at Ne"' York University. She holds

a B.A. in Linguistics from Harvard University. Her past \Vork has focused on theoretical and experi1nental study of the n1orphophonology and syntax of n1uta­ tion in Welsh. Her research interests are in experimental phonology, segmental phonology, the study of Welsh, and the phonology-semantics interface. Copyrighted maerial

xvi

Contributors

Mirjam Ernestus is an Associate Professor at Radboud University in Nijmegen.

She obtained her Ph.D. rom the Free University Amsterdam in 200and then held several postdoctoral positions at the Max Planck Institute for Psycholinguistics in Nijmegen. Kathryn Flack Potts received her Ph.D. in linguistics fron1 the University of

Massachusetts, An1herst in University.

2007. She is a Lecturer in Linguistics at Stanford

Stefan A. Frisch is an Associate Professor in the Departn1ent of Comnnmication

Sciences and Disorders at the University of South Florida. He uses the tools of laboratory phonology to examine frequency effects in the lexicon and in meta­ linguistic judgments of well-formedness. He is also interested in speech articu­ lation and the use of experirnentally elicited speech errors r1 the study of speech production processes. Adamantios I. Gafos is an Associate Professor at Nev York University's Linguistics

Department and a senior scientist at Haskins Laboratories. His interests lie at the intersection of phonology and cognitive science. Randall Gess s Associate Professor of Linguistics, Cognitive Science, and French,

and Director of the School of Linguistics and Language Studies at Carleton University. His research interests are in historical phonology, the phonetics­ phonology interface, and French and Romance phonology. Heather Goad is an Associate Professor in Lmguistics at McGill University. She

works prmcipally on the acquisition of phonology. Her work has been published ir1 n c i and Her research is currently funded by the Social Sciences and Hun1anities Research Council of Canada and the Fonds quebecois de la recherche sur la societe et la culture. She has been an Associate Editor of smce

Li guist Revie1v, Lingua, Language Acquisition,

Second Language Research.

Language Acquisition 2004. Matthew Gordon is Professor of Lmguistics at the University of California, Santa Barbara. He is the author of Syllable weight: Phonetics, phonology, typoloi; (2006) and co-editor of Topic and focus: Cross-linguistic perspectives on 1 1ean.in.g and intonation (2008). His research focuses on prosody, includmg stress and mtonation, and the phonetic description of endangered languages.

Janet Grijzenhout is Professor of English Lmguistics at the University of Konstanz

and director o.f the Baby Speech Lab there. He.r research focuses on the phonolo­ gical representation of voicing and stricture, the phonology-morphology interface, infant speech perception, and the acquisition of prosody and morphology. Naomi Gurevich received a Ph.D. from the Linguistics Department at the Univer­

2003. 204 Lenition and conrast: F11nctional consequences ofcertain phonetically conditioned sound changes. sity of Illinois at Urbana-Champaign in The dissertation vas published in the Outstanding Dissertations in Linguistics Series ll under the title

Recently Naomi's research interests have shifted from theoretical lr1guistics to neurologically-based language disorders. She is currently vorking on a clinical certification in Speech Langtlage Pathology. Carlos Gussenhoven is Professor of General and Experimental Phonology in

the Departn1ent of Linguistics at Radboud University, Nijn1egen and Professor Copyrightd maerial

..

Contribu.tors

vu

of Linguistics in the School of Languages, Linguistics, and Film at Queen Mary University of London. One of s research topics is the prosodic structure of English, including \vord stress and sentence intonation. Other research has focused on stress and tone in a variety of languages, including Dutch, Japanese, Nubi, and a group of Franconian dialects \vith a lexical tone contra st Amon g his publi ca­ tions are and

The phonology of tone and intonation (2004)

(1998, 2005).

. Understanding phonology

Gregory R. Guy (Ph.D., Pennsylvania) is Professor of Linguistics at Ne\v York

University, and has been on the faculty at Sydney, Temple, Cornell, Stanford, and York. He has tau ght at five Lingtli stic Institutes of the Linguistic Society of America, and three Institutes of the Associa;ao Brasileira de Li ngilistica He speca i lzes n sociolinguistic s and phonological variation and change, and \vorks on Portuguese, English, and Spanish. His current research n i trests include the treat1nent of variation in linguistic theory, the relationship between individual and coonuni ty grammars, and tl1e theoretical treatment of grammatical sim il ar ity. Hs books and

.

include Towards a social science of language (1996)

quantitativa (2007).

Sociolingiifstica

Daniel Currie Hall received his Ph.D. from the University of Toronto in 2007

with a thesis entitled "The role and representation of contrast in phonological theory." He has taught on all three ca1npuses of the University of Toronto and at Queen's University and vorked as a researcher at the Meertens lnstituut of the Royal Netherlands Academy of Arts and Sciences, and is c urrently an Assistant Professor at Saint Mary's University n i Halifax, Nova Scotia. Nancy Hall is an Assistant Professor at California State Univ ersity, Long Beach. She received her Ph.D. from the University of Massachusetts, Amherst in 2003,

and has taught at Rutgers University, the University of Haifa, BenGurion University of the Negev, and Roehampton University. T. A. Hall

is Associate Professor of Germanic Studies and Adjunct Associate Professor of Li ngu isti cs at Indiana Un iversity. He has publ is hed "'idely on a number of topics in general phonological theory and Germanic phonology in journals such as &

Linguistics, Lingua, Natural Language Linguistic Theory, Journal ofGennanic Linguisics, Phonology, Linguistic Review, Journal ofComparative Germanic Linguistics, and Morphology. Michael Ha1nmond is Professor of Linguistics and Dep rtm en t Head at the University of Arzona. He received his Ph.D. in Linguistics rom UCLA in 1984. He is the author of nu1nerous books and articles. His research interests are broad, ncludi g the merical theory of stress, Optimality Theory, poetic meter, language games, computational linguistics, psycholinguistics, mathematical linguistics, s Uable structure, probabi listic p honota ctics, Englis h phonology general l and Welsh. S. J. Hannahs is Senior Lecturer in Lingu istics at Ne"•castle University. Co-author of Introducing phonetics and phonoloi; (2005), much of his research has focused on a

n

y

y

,

proodic structure, particuarly at the interface benveen phonol oy and morpholog y. His recent and ongoing \vork has concentrated on the phonology and norpho­ phonology of n1odern Welsh.

Copyrightd maerial

XVUJ

Contributors

John Harris, Professor of Linguistics at University College London, "'rites on various topics connected with phonology, including phonological theory, the interface \\1.th phonetics, language impairn1ent, and variation and mange in English. lnong his publications are the books (1985) and (1994).

English sound structure

Pl1onological variation and change

Jeffrey Heinz is Assistant Professor in the Department of Linguistics and Cognitive Science at the University of Delaware and has held a joint appoint1nent \v.th the department of Computer and Information Sciences since 2009. He received his Ph.D. in Linguistics from UCLA in 2007 and is keen to help bridge divides behveen theoretical phonology, computational linguistics, theoretical computer science, and cognitive cience.

n

Ben Hermans is a senior researcher at the Meertens Institute Amsterdan1. He was trained as a Slavist and Germanicist. In 1994 he defended his thesis "The con1posite nature of accent" at the Free University. He now focuses on the tonal accents of the Limburg dialects, for example, "The phonological structure of the Limburg tonal accents" (2009), and is particularly interested in their formal representation. He also publishes on the history of the phonology of Dutch and its dialects. Jose Inacio Hualde (Ph.D. in Linguistics, 1988, University of Southern California) is Professor in the Department of Spanish, Italian, and Portuguese and the Department of Linguistics at the University of Illinois at Urbana-Champaign. He is author of (1991) and (2005), co-author (1994) and co-editor of of

Basque phonology The sounds ofSpanish The Basque dialect of Lekeitio Generative studies in Basque linguistics (1993), Towards history oft/1e Basque language (1995), A ram111ar ofBasque (2003) and Laboratory phonologi; 9 (2007), among other books. He has also pub­ n

lished a number of articles on synchronic and diachronic issues in Basque and Ro1nance phonology.

Harry van der Hulst is Professor of Linguistics at the University of Connecticut. He has published four books, ''0 textbooks, and over 130 articles, and has edited 20 books and six journal theme issues in areas including feature systems and

segmental structure, syllable structure, "'ord accent systems, vovvel harn1ony, and sign language phonology. He has been Editor-in-Chief of the international since 1990 .nd he is co-editor of the series Jingtlistic journal Studies in Generative Grammar.

The Linguistic .Review

Elizabeth Hume is Professor and Chair of the Department of Linguistics at Ohio

State University. She holds a Ph.D. and M.A. in Linguistics fron1 Cornell Univer­ sity, an V.A. in French and Social Psychology of Language from McMaster University (Canada), and a B.A. in Frencl1 from Universite Laval (Quebec). Her research interests lie n language sound systems, cognitive science, language variation, and language change. She has published widely on topics including consonant-vowel n i teraction, feature theory, gemi.nates, n1arkedness, metathesis, ound change, the interplay of speech perception and phonology, and Maltese phonetics and phonology. Brett Hyde received his Ph.D. n Linguistics from Rutgers University in 201 and

currently has an appointment at Washington Univrsity in St. Louis. His primary researl1 interests are in n1etrical stress theory and related areas. Copyrightd maerial

Contributors

X1X

Larry M. Hyman is Professor of Lnguistics at the University of California, Berkeley He received a Ph.D. n Linguistics from UCLA in He has published several books (e.g. a nd numerous theoretical articles in such journals as & and .

1972. Phonology theory and analysis, A theory of phonological weight) anguage, Linguistic Inquiry, Natural Language Linguistic Theory, Phonology, Studies in African Linguistis, Journal ofAfrican Languages and Linguistics.

Sharon Inkelas is Professor of Linguistics at the Universiy of California, Berkeley. She specializes in the phonology-morphology interface and has branched out in recent years into child phonology. With co-author Cheryl Zoll she publshed

Reduplication: Doubling in morphology in 2005.

Gregory K. Iverson is Professor of Linguistics at the University of Wisconsin­ M i lwau k ee and Research Professor at the University of Maryland Center for Advanced Study of Language. His research interests span the ield s of historical linguisti s especialy Gern1anic, the phoneti s and phonology of laryngeal systems, especally Korean, and the acquisition of second language phonological patterns.

i

c,

c

Jongho Jun is Profe sor of Lingu istics at Seoul National University His research interests are p honeti s n phonology variation in phonology and the forn1al prop­ erties of Optimality Theory.

s c

,

.

,

Barr� Kabak is currently an Assistant Professor of English and General Linguistics at the University of Konstanz. Ellen M. Kaisse is Professor of Linguistics at the University of Washington. She has co-edited the journal since Her research concentrates on the interactions of phonology ' ith morphology and syntax, on distinctive features and on the phonology of Modern Greek, Turkish, and Spanish

Phonology "

1988.

.

,

Yoonjung Kang is an Assistant Professor in the Departn1ent of Humanities at the Univer i ty of Toronto, Scarborough and the Departn1ent of Linguistics at the University of Toronto. Her area of sp ecialization is phonology and its interface "ith phonetics and morphology, with a spe a i l focus on Korean.

s

c

Shigeto Kawahara is an Assistant Professor in Linguistics and Rutgers Center for Cognitive Science (RuCCs) at R utgers University. He was a\"arded his Ph.D. from the University of Massachusetts, Amherst His research focuses on the phonetics-phonology interface, experimental investigations of phonological judg1nents, corpus-base. studies of verbal art, and studies on intonation and accents.

in 2007.

Robert Kennedy is a lecturer the Department of Linguistics at the University of California Santa Barbara. His areas of expertise include redupliation, prosodic morphology, hypocoristics, vovel syste1ns of English varieties, and articulatory phonology.

,

n

in

John Kingston is a Professor the Linguistics Department, University of .vlassachusetts, Amherst. His publications include: "Phonetic kno\'ledge," (vith R. L. Diehl, "Lenition," "The phoneticpho ology "On the internal perceptual structure of distinctive features: The [voice] contrast," (\vith R . L. Diehl, C. J. rk, and W. A. Castleman, "Contextual effects on the perception of

Language 1994); Proceedings of the Third Conference on Laboratory Approaches to Spanish Phonology (2007); interface," Cambridge handbook of phonology (2007); journal of Phonetics 2008);

Copyrightd maerial

Contributors duration," fournal of Phonetics (vith S. Ka,vahara, D. Chambless, D. Mash, and E. Brenner-Alsop, 2009); "Auditory contrast versus compensation for coarticulation: Data fron1 Japanese and English listeners," Language and Speech (vith S. Ka\vahara, D. Mash, and D. Cha 1bless, in press). xx

n

Charles W. Kisseberth has just retired from his position as Professor of Linguistics at Tel Aviv University (vhere he taught from 1996 to 2010) and is also E1neritus Professor at the University of fllinois (\vhere he taught from 1969 to 1996). He is best known for his vork in theoretical phonology, '"here .is work on "conspiracies" laid the foundations for Optimality Theory, and his vork on Chimwiini prosody helped to lead the vay to current studies in the phonology­ syntax interface. His work over the last 30 years has focused on Bantu tonal systems, and Optimal Domains Thory has evolved out of that \Vork. He is co-author vith Michael Kenstowicz of (1979), a standard introduction to classical generative phonology.

Generative phonology

Alexei Kochetov is currently an Assistant Professor in the Linguistics Department

at the University of Toronto. He received his Ph.D. at the University of Toronto vith the thesis "Production, perception, and emergent patterns of pala talization" (published n i 2002). His research deals with various issues n i the phonetics­ phonology interface, cross-language speech production, and Slavic phonology and phonetics. Asrid Kraehenmann holds a Ph.D. n i Theoretical Linguistics rom Konstanz Univer­ sity, Germany. Her n reearch interests are phonology, phonetics, the phonology­ phonetics interface, historical linguistics, and Germanic languages. Haruo Kubozono is Professor and Director at the National Institute for Japanese Language and Linguistics in Tokyo. His main publications include (1993), "Mora and syllable" (in 1999), and "Where does loanword prosody coo�e .from?" 116, 206).

ofJanese prosody

Darlene LaCharite

The or.nization The handbook ofJapanese linguistics, (Lingua

is Professor of Phonetics and Phonology at Laval University.

Her areas of research include loanword phonology (in collaboration with Carole Paradis), the L2 acquisition of phonetics and phonology, and creole phonology and morphology (in collaboration with Silvia Kouwenberg). She has published in a variety of ln i guistics journals, including and

of Linguistics

Linguistic Inquiry, Phonology, Journal Journal ofPidgin and Creole Languages.

William R. Leben is Professor E1nerihls of Linguistics at Stanford University. He has worked on the phonology of tone in languages of West Arica and has also co-authored pedagogical '"'orks on .ausa. He continues to ''ork on phonology in Kwa languages of Cote d'Ivoire. The econd edition of a textbook he co-authored, was published n 2007.

ele111ents,

English vocabulary

Clara C. Levelt is Associate Professor at the Linguistics Departn1ent of Leiden University, and affiliated to the Leiden University Centre for Linguistics (LUCL) and the Leiden Institute for Bra in and Cognition (LIBC). She received her Ph.D. in 1994, and h.s '"orke. on child language phonology ever since. In 2007 she received a prestigious research grant from the Netherlands Organization for Scientific Research (NWO), \Vhich enabled her to set up a baby lab. Combining insights fro1n perception and production data, phonology and phonetics, she tries to uncover the source of children's deviating productions. .

Copyrightd maerial

Contributors

.

(1

Susannah V. Levi is an Assistant Professor in the Department of Communicative Sciences and Disorders at Ne"' York University. She received her Ph.D. n i 2004 fron1 the Departn1ent of Linguistics at the University of Washington. She then con1pleted a three-year postdoctoral fello"'srup at Indiana University in the Department of Psychological and Brain Sciences. She has "'Orked on glides, Turkish stress and intonation, and most recently on speech perception and language processing in children and adults. Yen-Hwei Lin is Professor of Linguistics at M c i higan State University, and taught at the 1997 and 203 Linguistic Society of America Linguistic Institutes. Her reearch has focused on phonological representations and constraints the theoretical context of non linear phonology /morphology and Optimality Theory. She is a uthor of (2007) and editor of 5.4, 2004). -

The sounds of Chinese and PhonoloJ (Language and Linguistics

n Special Issue on Phonetics

Anna .ubowicz is Assistant Professor of Linguistics at the Universiy of Southern CaJifornia. Her research interests lie in the i nvesigation of the role of contrast in phonology and morphology, lexical phonology, and the morphology-phonology interface. She specializes in Slavic languages and holds a Courtesy Appointment in the Department of Slavic Languages at the University of Southern California.

Andrew Martin recei ved his Ph.D. in Linguistics fro m UCLA in 2007. His research is focused on understanding early phonological and lexical learning. He is also interested in hov a language's lexicon changes over time, and hoiv a "'Ord's phonological properties affect that word's ability to survive and spread n i a speech coununity. He is currently a post-doctoral researcher at the Laboratory for Lan­ gt1age Lcuning an d Development in the RIKEN Brain Science Institute nec1r Tokyo.

Joan Mascaro studied at the Universitat de Barcelona and at vIT '"'here he got his Ph.D. in lingu istic in 1976. He has taught a t Cornell University and at the Universitat Autonoma de Barcelona, where he is currently fuJI professor. His main research areas are linguistic theory, phonological theory, the phonology-n1orphology interface, and the phonological and n1orphological analysis of Romance languages.

s

.

April McMahon is Forbes Professor of English Language, and Vice Principal for Planning, Resources and Research Policy, at the University of Edinburgh. She co-edits the journal Her research focuses on the interaction of phonological theory and sound change, and methods for the con1parison and classfication of accents and languages.

English Language and Linguistics.

Jacques Mehler is the director of the Language, Cognition, and Development laboratory at the International chool for Advanced Studies, Trieste, Italy (SISSA). After obtng a Ph.D. from Harvard University in 1964, he "'orked at the National Center for Scientific Research (CNRS) in Paris, France, from 1967 until 2001. In 1972, an international journal of cognitive science, and acted as he founded Editor-in-Chief until 2007. He became Directeur de Recherche at CNRS in 1980 and was elected Directeur d'Etudes at the Ecole des Hautes Etudes en Sciences Sociales n 1982. He has published influential experimental studies of language acquisition

Cognition,

in the first year of life, and has also explored early bilngualism. He is a member of the American Academy of Arts and Sciences (2001), the American Philosophical Society (2008) and the Academic Europaea. He \Vas ararded a Doctor Honoris Causa ron1 Utrecht University (2009), and from Universite Libre de Bruxelles (1995). His publications are available at: \V"'".sissa.it/01s/Icd/publications.html.

Copyrightd maerial

xxii

Contributors

Jeff Mielke is Assistant Professor of Linguistics at the University of Otta"a and Co-director of the Sound Patterns Laboratory. He completed his Ph.D. at Ohio State University n 2004 and undertook postdoctoral research at the Arizona Phonological Imaging Laboratory at the University of Arizona before 11oving to Otta"a in 2006. His 'vork focuses on the way phonological patterns relect influ­ ences such as physiology, cognition, and social factors. He is the author of (2008).

The

efl1erge1ce of distinctivefeatures

Amanda Miller is a Visiting Assistant Professor at Ohio State University. Her research focuses on the phonetics and phonology of African languages, particu­ larly Khoesan languages. She has studied gutturals, complex segments (clicks and labial-velars), and contour segments (affricates and airstream contours). She has investigated acoustic voice quality cues and the role of acoustic similarity in a Guttural Obligatory Contour Principle constraint; as \veil as tongue root retraction in dicks and labial-velars, and its role in C-V co-occu.r.rence patterns. She has also published papers on reduplication and the tonal phonology of Khoesan languages. Brett Miller is a graduate student at the University of Cambridge. His interests include Inda-European phonology, especially stops; the phonetics-phonology interface; feature contrast, representations, and interaction in rules and constraints; and typology the comparative method.

vis-a-vis

Bruce Moren-Duollja is a Senior Researcher at the Center for Advanced Study in Theoretical Linguistics at the University of Tron1s0. He has published on synchronic and diachronic phonology, includng Slavic palatalization, Icelandic preaspiration and Thai tones. He is the author of (2001).

sonority: A uniied theory of weight

Distinctiveness, coercion and

n

Marina Nespor is Professor of General L gustics at the University of Milano­ Bicocca. She has focused her research on ho1v the phonological shape of an utterance conveys information about its syntactic structure, the so-called theory of prosodic phonology. She has also investigated how prosody is used in com­ prehension and during language acquisition. Her book (co­ authored with I. Vogel, 1986) s a citation classic and she has numerous articles in peer-revie1ved journals.

Prosodic phonology

Andrew Nevins is a Reader in Linguistics at University College London. His research has primarily focused on phonological and morphological theory, and the relations betreen different modules of the grammar. He has "orked on local­ ity, markedness, contrast, reduplication, the nature of underlying representations, and the structure of the morphological con1ponent. He is the author of (2010) and the co-editor of two books, with Bert Vaux, (2008), and with Asaf Bachrach, (2008).

in vowel harmony constraints, and phonological pheno11rena identity

LocaliIJ Rules, Inlectional

David Odden is Professor of Linguistics at Ohio State University. His areas of research specialization include phonological theory and language description, especially the structure of African languages. He served as Editor of from from 2003 to 2009 and \Vas Associate Editor of 1998 to 2008. Recent publications include (2005), "Ordering," (2008), and in Vaux and Nevins, "Tachoni verbal tonology," (2009).

Studies in Phonoloi; Arican Linguistics Introducing phonology Rules, constra.ints, and phonological phenonr.ena Language Sciences

Material com direios autorais

Contributors

XXlll

Jaye Padgett is Professor of Linguistics at the University of California, Santa Cruz.

At the broadest level his research is on the interplay of phonology and phonetics, with special attention to the role that perception plays in shaping phonological patterns. Though his research bears on larger questions it often focuses particu­ larly on Russian phonetics and phonology. Carole Paradis is Professor of Phonology at Laval University. She proposed a theory of constraints in 1987, as a visiting scholar at MIT. She co-edited a book on the special status of the coronals \Vith J.-F. Prunet in 1991. With R. Beland's

colaboration, she has drawn a parallel between aphasic errors and loan,vord adaptations. Th.is study, extended to other speech error types, has been p.lbLished and has led to the construction of more optimal in journals such as speech pathology exercises.

in

Aphasiology

Steve Parker is a Professor Vith the Graduate Institute of Applied Linguistics

n

(GIAL) Dallas. He graduated fron1 the doctoral prograin linguistics at the University of Massachusetts, Amherst in He has served as a teacher and consul tant '"ith the Summer Institute of Linguistics (SIL I nternational for 30 years In that capacity e has carried out direct fieldwork and research on a number of indigenous languages of South America and Papua Ne" Guinea, two of vhich are no" extinct.

2002.

)

.

Sharon Peperkamp is a senior researcher in the Laboratoire de Sciences ,

Cognitives et Psycholinguistique at the Ecole Normale Superieure in Paris. Her main research interests are n experimental and computational approaches to speech perception and phonological acquisition. Marianne Pouplier obtained her Ph.D. from Yale Linguistics in

2003 and is

no\v a principal investigator of an En1y-Noether Research Group funded by the Deutsche Forschungsgen1einschaft at the Institute of Phonetics and Speech Processing, University o.f Munich. She has al.so been affiliated with Haskins Laboratories for a number of years. Her research focuses on e interaction of speech planning and motor control and the phonetics-phonology interface. In particular, she has vvorked on speech errors, phonetic correlates of syllable struc­ ture, and word boundary assin1ilation. Pilar Prieto holds a Ph.D. in Romance Linguistics at the University of Illinois

at Urban.-Cha1npaign, and is currently "'orking as the coordinator of the Grup d'Estudis de Prosodia at the Universitat Pompeu Fabra in Barcelona. Her main research interests focus on ho'" melody and prosody vork in language and how they interact with other types of linguistic knorledge. She is also interested in ho\\' babies acquire sounds i1d melody together with gran1mar, and hovv these components integrate in the process of language acquisition. Douglas Pulleyblank is Professor of Linguistics at the University of British

Colun1bia. His research has focused on the phonology and n1orphology of Nigerian languages, particularly Yoruba. He has "'Orked extensively in auto­ segmental and optimality-theoretic frameworks, examining phenomena such as tone and vovel harmony. Eric Raimy is an Associate Professor at the University of Wisconsin-Madison. He received his Ph.D. fro1n the University of Delaware in 1999. He is the author

of

The phonology .nd morphology ofreduplication (2000) i1d co-edited Contemporary Material com direios autorais

Contributors views on architecture and representations in phonolo>y (2009) and Handbook of the StJllable (in press) both "'ith Charles Cairns. XXlV

Anthi Revithiadou is an Assistant Professor of Linguistics at the Aristotle Univer­

sity of Theory, "'ith emphasis on the structure of phonological representations and the issue of parallel grammars, and has published in journals and edited volumes.

(Lingua, Linguistic

Revieiu, journal of Greek Linguistics) Curt Rice

is the Vice President (Prorektor) for Research and Developn1ent at the University of Tron1s0 in Tron1s0, Non.vay. He vas the founding Director of the Center for Advanced Study in Theoretical Linguistics: A Norwegian Center of Excellence (CASTL) from 2002 to 2008, at the same university. Keren Rice is University Professor and Canada Research Chair n i Linguistics

and Aboriginal Studies at the University of Toronto. She holds an M.A. and Ph.D. in Linguistics rom the University of Toronto a.nd a B.A. in Li.ngu.istics froo� Cornell University. Her research interests lie in language sound systems, contrast and l markedness, interfaces '"ith phonology, anguage variation and language change, and Athabaskan languages as well as ethics and responsibilities of linguists in fieldwork. She has published on topics including feature theory, sonorants, markedness language change, and Athabaskan phonology and morphology.

,

Anastasia K. Rieh l is Director of the Strathy Language Unit at Queen's University in Kingston, Canada, vhere she also teaches n the Linguistics Progran1. She received

a Ph.D. n i Linguistics from Cornell University. Her research interests include the phonology-phonetics interface, endangered language docun1entation, Austronesian languages and varieties of English.

,

Jason Riggle is Assistant Professor of Linguistics at the University of Chicago,

and his main research areas are phonology, learnability, and computational linguistics. Much of his research focuses on the \vays that specific n1odels of grammar, learning, and communication interact to make predictions about lin­ guistic typology vith special emphasis on the frequencies "'ith which patterns are observed ivithin and across languages. Sharon Rose is Associate Professor of Linguistics at the University of California,

San Diego. She has published journal articles covering topics such as consonant har.mony, syllable contact, Semitic root structure, redu.p.ca.tion, phonoactics, and the interaction between tone and syllable structure. She specali es in African languages spoken in Ethiopia, Eritrea, and Sudan; her current research is an n i vestigation of the Sudanese language, Moro, funded by the National Science Foundation. t

z

Jezy Rubach holds an appointment as Professor of Linguistics at hvo universities: the University of Iowa n the United States and the University of Warsaiv n Poland. His expertise is primarily in Germanic and Slavic languages. He has published six books and 75 articles. His work has appeared in nun1erous journals, n i cluding t

Linguistic Inquiry, Language, Phonology, and Natural anguage Linguistic Theory. Joseph C. Salmons is Professor of German at the University of Wisconsin.:Madison and Executive Editor of Diachro.ica: Intenational Journal for Historical Linguistics. '

His research interests include language change, sound systems, language contact, and language shift, all often involving data from the Gern1anic languages.

Material com direios autorais

Contribu.tors

xxv

Wendy Sandler has been investigating the phonology, morphology, and prosody

of An1erican Sign Language and Israeli Sign Language for many years, beginning with her graduate \Vork at the University of Texas at Austin, \Vhere she earned her Ph.D. Sandler has authored or co-authored three books on sign language linguistics. Currently, "'ith colleagues Mark Aronoff, lrit Meir, and Carol Padden, she is investigating a new sign language that arose in an nsulated Bedouin community in the Negev desert of Israel. Tobias Scheer is currently Di.recteur de Recherche at the Centre National de la

Recherche Sci.entfique in France. He works at the laboratory Bases, Corpt.1s, Langage (UMR in Nice, of which he is the director. Being a phonologist, his main interests lie n syllable structure thematically speaking, in the (Western) Slavic family as far as languages are concerned, and n i diachronic study. In he published a book on a particular developn1ent of GP, so-called CVCV (or strict CV), and . book on the Q�jstory of the) interace of phonology and roorph.o-synta.x.

6039)

2004,

Misun Seo is an Assistant Profesor in the Department of English Language

and Literature at Hannam University in Daejeon, Korea. She received her Ph.D. Linguistics at Ohio State University; her research interests n i clude phonology, its interface vith phonetics, and L2 acquisition.

n

Kimary Shahin is a phonetician/phonologist 'vho specializes in Arabic and

Salish. She also investigates first language acquisition and contributes to the documentation and revitalization of indigenous languages n Canada. llohlnish Shukla has a background in n1olecular genetics, and holds a Ph.D. in

Cognitive Neuroscience from the Scuola Internazi.onale Superiore di Studi Ava.nzati (SISSA), Italy. He is currently doing postdoctoral 'vork at the University of Rochester, Ne' York. He is interested in human cognition, 'v.th a focus on infant cognitive development, particularly the development of linguistic abilities from phonology to syntax. He is also interested in the neural bases of cognitive beha­ vior, vhich he primarily explores using near-infra.red spectroscopy in inf.nts and adults.

1995 under the tutelage of Danca Steriade and Peter Ladefoged. He has published \videly on phonology and phonetics, including A critical introducion to phonology: Of sound, mind, and body (2006), and Neutra.zation: Rhyme and reason in phonology (forthcoming). He Daniel Silverman earned his degree at UCLA in

is currently on the faculty of San Jose State University.

Jennifer L. Smith teaches phonology, phonetics, and Japanese linguistics at

the University of North Carolina, Chapel Hill. Her research explores the 'vays in 'vhich interfaces between phonology and other doma.ins of linguistics affect phonological constraints and representations. Peter Szigetvari gives courses on phonology, linguistics, information technology,

and typography at E6tv6s Lorand University, Budapest. His n1a.n research areas include phonotactics and consonant lenition. Nina Topintzi has taught as a Teaching Fellov in England and Greece

(2006-2010) and will soon commence her appointment as Assistant Professor in Phonology at the English Department, Aristotle University of Thessaloniki She has published in journals including Natural anguage Linguistic Theory and the &

Material com direios autorais

Contributors Journal of Greek Linguistics. Beyond onsets, her research interests include stress, xxv1

the syllable, and various ''eight-based phenon1ena from a typological perspective. Mik16s Torkenczy is a Professor at the Department of English Lnguistics and

the Theoretical Linguistics Department of Eotvos Lorand University Budapest and a senior researcher at the Research Institute for Linguistics of the Hungarian Academy of Sciences. He is co-author of

The phonology ofHungarian (2000).

Jochen Tromer received his Ph.D. ro1n the University of Potsdan1, has \vorked

as lecturer at the University of Osnabruck and is currently Lecturer n Phonology at the Deparllnent of Linguistics at the University of Leipzig. His work focuses on the theoretical and typ ological aspects of phonology and morphology, especially prosody, non-concatenative morphology, hierarchy effects in agreement morpho­ logy, and affix order. His publications include "Case suffixes, postpositions, and the phonological \Vord in Hungarian," 46(1) "Hierarl1y-baed con1petition and en1ergence of tvo-argument agreen1ent in Dumi,"

Linguistics

(2008);

Linguistics 4.4(5) (26); and "The interaction of morphology and syntax in affix order," Yearbook ofll1orpholo; 2002 (2003). Suz e Urbanczyk is an Associate Professor in the Linguistics Department at the University of Victoria. She received her Ph.D. from e University of Massachusetts Amherst and has published on Salish reduplication and non-concatenative n1orpho­ logy n i Natural Language and Linguistic Theory and Linguistic Inquiry. Her current n

,

researl1 focuses on the role repetition plays in structuring language, and she is developing . model of morphology vhich is gro. nded in the mental lexicon, ''ith an empirical focus on \VOrd-formation in Salish and lakashan languages. Adam Ussishkin is an Associate Professor in the Linguistics Department at the

University of Arizona, where he also holds joint appointments n Cognitive Science

and Near Eastern Studies. His areas of research include phonology, morphology, and lexical access, vith an empirical focus on Semitic languages. Bet Vaux is University Reader n Phonology and Nlorphology at the University

of Cambridge and a Fellovv of King's College, Cambridge. He is primarily interested in phenomena that shed light on the structure and origins of the pho ologica l

component of the graar, especially in the realms of psychophonology, historical linguistics, and sociolinguistics. He also enjoys working 'vith native speakers to docun1ent endangered languages, especially dialects of Armenian, Abkhaz, and

English. Rachel Walker is Associate Professor of Linguistics at the University of Southern

California. She is the author of

Vowel patterns in language

Nasalzation, neutral seg1nents and 01incity ffects (2000)

(forthcoming). She has published 'videly on topics and involving long-distance assimilation and copy, as seen in systems of harmony and reduplication. Her current research investigates the interaction of VOvel patterns vith positions of prominence n the vord. Natasha Wamet has been a factllty n1ember in Linguistics at the University of

Arizona since

2001 and has vvorked at the Max Planck Institute for Psycholinguistics.

Her interests are in the three-way interface of phonetics, experimental phonology, and psycholinguistics, vvith language nterests in Dutch, Japanese, Korean, and English. She also \Vorks on revitalization of the Mutsun language.

Material com direios autorais

Contributors

..

vu

Janet C. E. Watson is Professor of Arabic Linguistc i s at the U niversity of Salford,

UK. She has previously \VOrked at the universities of Edinburgh and Durham, and hel d visiting posts at the universities of Heidelberg and Oslo. She has pub l ished extensively on Yemeni Arabic, and has been \vorking on the documentation of the Modern South Arabian language, Mehri. She is currently preparing a syntax of Mehri for publication. Her publ icato i ns include Phonoloy

since 2006

and 1norpltology ofArabic (2002) and A syntax of San'ani Arabic (1993).

Andrew Wedel is on the facul ty of the Departn1ent of Linguistics and the Cogn itive Science Program at the U niversity of Arizona. His primary interests lie in exploring the causes for, and the interaction of, hvo opposing tendencies evident in language change: a tendency toward pattern-coherence, and a tend­

ency to preserve emantically relevant contrasts. Relevant \Vork n i this domain includes "Feedback and regularity in and with

the lexicon," Phonology 24,

Juliette Blevins, ".1�hibited sou.nd change: A.n evolut.onary approach to lexical competition,"

Diaclironica 26 (2009).

Jeroen van de Weijer is Full Professor of English Linguistics at Shanghai

International Studies University (College of English Language and Literature) vithin the Chinese "2 -Project. He is a Distinguished University Professor of Shanghai Oriental Scholar), an award besto\ved by the Shanghai v!unicipal Education Commission. He has published widely on segmenta l structure, Optmality Theory, English phonology, and East Asian languages. His current research is focused on con1bining phonol ogical and psycholinguistic theories.

(

"

11

Richard Wiese has been a Professor of Gern1an Linguistics at the Philipps­ H is 1vork concentrates on theoretical Un iversitit, Marburg, Germany, linguistc i s, phonology, and psycholinguistc i s He has written a monograph, The

since 1996.

.

phonologi; of Ge 1an (2000) and nu1nerous articles on issues of (German) phono­ logy, morphology, psycholinguistics, and orthography, and is Co-editor of the n1

book series Linguistische Arbeiten.

Ronnie Wilbur, Professor and Director of Linguis tics and Professor of Spee�,

Language, and Hearing Sciences, received her Ph.D n Linguistics a t the University of Illinois, Urbana-Champaign. She has taught at the U niversity of Southern Cfornia Boston University, University of An1sterda1n, University dissertation "The phonology of Graz, Austria, and Uni ver sity of agreb. Her of reduplication" \Vas reissued as a "Classic in Linguistc i s" in She was Founding Editor of the journal & and

.

,

1973 1997. Sign Language Linguistics Editor fronl 1998 to 2006. Matthew Wolf received his Ph.D. in 2008 from the University of Massachusetts,

Amherst. After having been a Visiting Assistant Professor at Georgetown Univer­ sity and a Visiting Lectu rer at Yale University, e is currentl y a Postdoctoral Associate at Yale. His research has focused priarily on aspects of the phonology­ morphology interface, including process morphology allomorph selection, and paradign1 gaps, as \Veil as on opacity and serial versions of Optin1ality Theory.

,

Moira Yip is Professor of Linguistics Emerita) at University College London. Previously she 'as at the University of California, Irvine, and Brandeis Univer­

(

sity. She received her Ph.D. ro1n MIT in studying under Morris Halle. She has published extensively on tone, red uplication, distinctive feature theory, and

1980,

Copyrighted maerial

XXVll

Contributors

loanword phonology. Much of her 'Ork has been on Chinese languages. Recently she has become interestd n comparisons benveen birdsong and hu1nan phonology, and is no' publishing in this area. Alan C. L. Yu is Associate Professor of Linguistics at the University of Chicago. He is the author of and co-editor (with John Goldsmith and Jason Riggle) of the 2nd edition (forthcoming). His work on phonetics and phonology has appeared n i and

A natural history of inixation (2007) I·Iandbook of phonolo.cal theory, Language, Phonology, Journal of Phonetics, PLoS One.

Draga Zee i s Professor of Linguistics at Cornell University. She has "'Orked in several areas of phonology and its interfaces: the moraic theory of syllable struc­ ture, the representation of pitch accent, and both the phonology-morphology and the phonology-syntax interfaces.

Elizabeth C. Zsiga is an Associate Professor .in the Department of Ling uistics at GeorgetO\Vn University. Her research primarily addresses the phonetic and phonological patterns that occur at and across 'Ord boundaries in conncted speech. Her publialions have examined a range of processes in different languages, includ­ ing final consonant deletion and palatalization n English, VOvel assJnilation in Igbo, palatalization i n Russia n tone simplification in Thai, voicing in Setsvana, pitch accent in Serbian, and most recently nasalization and voicing in Korean and Korean-accented English. ,

Copyrightd maerial

Preface

Progress n phonology, like other disciplnes grows out of debate. Every journal article, every conference paper, every book chapter, every book on phonology, can be seen as a contribution to one or more discussions on son1e theoretical topic. New data are sought to sharpen theoretical claims and new theories are proposed to accommodate previously undescribed data. Already familiar data paradigms are requently appealed to n arguing that some new theory fares better than its co petitors the description of various phenoena. Arguably, the synchronic study of phonology is currently celebrating its first centena ry: Ferdinand de Saussure \Vas teaching his Cours de linguistique generate (it 'vas published posthumously in One hundred up to his death in years of debates have yielded many insights into the sound structure of human language. ve nov know much more about a \vhole range of specific phono­ logical phenomena, including topics such as vo\vel harn1ony, the typology of word stress systems, and the structure of affricate sounds. We kno\v more about how sound systems interact with morphological and syntactic systems, and about the importance of taking actors like variation and frequency into account in the study of phonology . Phonologists have not developed a theory \Vhich completely captures each of those phenon1ena (let alone one which captures every aspect of aU of them. in a uniform vay). Bt.1t a.t le.st "'e have a much better picture of some of the properties which a successful theory of phonology should have. Alongside external and universal challenges, such as university adn1inistrators who do not see the in1portance of somethng as esoteric as the study of sound systen1s, we can identify at least l\vo internal dangers which challenge our field. The irst is that n1any debates are abandoned at some point, and then orgotten, \Vith the issues involved sometimes being rediscovered much later, ''ithout the earlier research necessarily being knovn to the ne1 generation of researchers. The reasons for this are often perfectly understandable, and there is probably no 'ay to avoid this state of affairs con1pletely. After lengthy discussions n the litera­ tLlre on, say, the relationship bel\veen continuancy and place of articu.lat.on, appar­ ently involv i ng arguents for and against almost every logically poss i ble vie'" on the issue, the topic may seem to be intractable, and scholars may tire of it and move on to ne' topics of debate. Thus Ve n1ay avoid unruitful atten1pts to solve problems for '''hich \Ve just do not have the right tools at that particular mon1ent. ,

1

in

1913

1916).

i

Copyrightd maerial

xxx

Preface

Ho·ever, these discussions 'vill frequently have led to at least partial agreement on in1portant properties of the phenomenon at hand, even if only that the pheno­ n1enon s extren1ely con1plicated: just positing, for instance, son1e feature-geon1eric structure n '"hich continuancy and place of articulation are in an unambiguous universal relation to each other 'vill encounter many problems that help us better understand the phenomenon, even if they do not lead to a generally accepted solution. However, by abandoning the debate on the topic, \Ve run the risk of losing that knov]edge. The danger is that ne\v generations will have to rediscover the subtleties of the phenomenon \vhen the topic is taken up again. The second danger is over-specialization. Saussure \vas an aJl-round li ngu ist, vith a deep understanding of much of (Western) linguistics as it 'vas kno''n at the time. Today there are very fev, if any, people vho can make such a claim. Scholars '''ith a thorough w1derstanding of phonology n1ay have some t10\'­ ledge of neighboring disciplines sum as morphology or phonetics, but usually not even of both of those. There are many more phonology talks given every year at specialized phonology conferences than there are phonology talks at genera! linguistics conferences. And even within the field itself there are fairly \Vell­ established dividing lines: an expert on intonation is unlikely to be co1npletely up to date on the literature on coronals; somebody \vho studies the Ian1bic-Trochaic La\v may skip the talks on sign language phonology in the local phonology vork­ shop; somebody \vho 'vorks on stress may have little knowledge of segmental phonology. his means that crucial insights vithin one sub-discipline of phono­ logy are becoming less and less accessible to phonologists '"orking V.thin other sub-disciplines. does not have the ambition to offer solutions to these problems; indeed, they are probably unavoidable. Yet a tool such as this allo,vs us to document at least some of the n1any insights into hun1an language that phonologists have gathered in the past decades, and also to give an overview of what at present sen1 to be the n1ajor issues that those interested in sound sructure a re think ing and a rgu in g a bout. The is in essence an encyclopedia of case studies. Each chapter addresses some topic '"hich has been debated in one \vay or another in the his­ tory of our field. Authors were invited to concentrate pri.n1arily on the empirical arguments that have been put forvard by the various sides in sum debates. Because of this concentration on case studies, there are many topics we have ignored . For instance, there is a chapter on coronals, but not one on labials, simply because there has been much more discussion in the phonological literature on the inter­ nal structure and the behavior of coronal sounds than on their labial counterparts. Si.nUlarly, there is a cllapter on palatalization, but not one on labialization, again because the former has been discusse. broadly in the phonology literature while ilie latter has not. Some chapters have turned out more like the case studies 'e originally had in mind than others. Inevitably, some chapters had to be organize. differently, for instance those concentrating n1ore on a specific theoretical device (such as constraint conjunctio n or rule ordering) than on some empirical phenoo.eno n. However, even ilie authors of these chapters were asked to provide some discussion of the data vhich Jed scholars to develop such theoretical concepts in the first place. that many We are, of course, concious of the fact that the reader vill possible topics are missing fron1 the including son1e hotly debated ones

The Blackwell Companion to Plionology Companion

Co11 p.nion,

d

Copyrightd maerial

Preface 124

.

XX e A Á D/ appear in pre-r position (e.g. here, care, car, sure, more), while all other vowels are neutralized to /// (which may reflect a merger with historical /r/), as in bird, word, heard. The general tendency of rhotics not to undergo palatalization, as discussed by Walsh Dickey (1997) and Hall (2000), provides another example of (negative) rhotic–vowel interaction. (b) As discussed by Walsh Dickey (1997: 91–92), rhotics in Australian languages are often prohibited from occurring in word-initial position (see chapter 86: morpheme structure constraints). One such language is Mullukmulluk, in which the two rhotic phonemes /7/ and /r/ can appear in any position except wordinitially (Birk 1975: 61). Given that Australian languages typically allow for two or three rhotic phonemes, this constraint focuses on rhotics as a class, and not on a single phoneme which happens to be a rhotic. There are other more detailed constraints on the placement of rhotics: for Warlpiri, Nash (1980: 76) notes that in a CVC sequence (with heterosyllabic Cs), the two consonants cannot be identical rhotic phonemes (i.e. two tokens from one of the phonemes /7 È P/). This constraint seems to have exceptions, but it still captures a pattern significant in its exclusive reference to the class of rhotics. (c) For languages that allow clusters of more than one consonant to appear in onset and/or coda position, rhotics are typically assigned to the position immediately adjacent to the vowel of the respective syllable. That is, a template of the type CrVrC describes the phonotactic placement of rhotics rather well, with C standing for one or more consonants other than /r/. (8) below exemplifies such a patterning from German, but many languages with complex syllable constituents behave analogously (chapter 49: sonority). (d) In many languages, there is a great deal of allophonic variation for the usual single r-phoneme, with the allophones standing either in free variation or in complementary distribution. But the large number of these allophones are all drawn from the inventory of rhotic sounds. The Persian language (Farsi) provides an instructive example: the phoneme /r/ has trilled [r] as its main allophone according to Majidi (1986: 63–64, 2000: 41–43), but has three to four additional rhotic allophones in complementary distribution, as shown in (5). The phoneme /r/ in Persian does not have non-rhotic allophones. (5)

Distribution of rhotics in Persian (Majidi 1986: 63–64) flap [7] intervocalically /tare/ ta’[7]e ‘chive’ b. voiced fricative [p] in word-initial position /ruz/ ’[p]uz ‘day’ c. partially or completely devoiced trills [ã] adjacent to voiceless consonants and word-finally7 /babr/ ’bab[ã] ‘tiger’ /xCrkan/ xC[ã]’kan ‘collector of blackthorn’ d. voiced trill [r] elsewhere /arzCn/ a[r]’zCn ‘cheap’ a.

7

Majidi (1986: 64) sees a tendency to distinguish the devoiced trills in terms of either partial or complete lack of voicing.

The Representation of Rhotics

7

Walsh Dickey (1997: 94) concludes from her survey of alternations involving rhotics that “rhotics were overwhelmingly associated with other rhotics in alternations.” In addition, rhotics of all types can be involved. Looking beyond single variants of a language, languages with a sufficiently large number of variants/dialects may display almost all types of rhotics in the various dialects. This is true for English, for example: Ladefoged and Maddieson (1996: 235) point out that for English “we can exemplify nearly all the different forms of rhotics we have been discussing simply by reference to this one language.” The same is also true for German, as the discussion in §2.3 will demonstrate. Not surprisingly, the same massive variation is found in language change. Historical change from one type of /r/ to a different one is widespread. But in striking contrast to the variability noted here and further exemplified in the next section, rhotics are quite invariable in terms of their phonotactic patterning: they are, as discussed above, the consonants occupying the position closest to vowels (where vowels include glides). (e) Examples of restricted patterning among rhotic phonemes have been given above for the rhotic phonemes of Spanish and Catalan, for which there is partial neutralization between the two rhotics. A further example is provided by Irish Gaelic, which has /r/ and palatalized /r j/ as rhotic phonemes, but with the restriction of the palatalized version to word-internal contexts (Ní Chiosáin 1999: 553). In simple word-initial onsets, the distinction between the two rhotic phonemes is neutralized to /r/.

2.3

Variability of rhotics

In the fields of dialectology and sociolinguistics, many studies of sound change and variation consider the behavior of rhotics to be important linguistic variables. Most prominently in the history of sociolinguistics, Labov’s studies of sound change in American English demonstrate the absence or presence of postvocalic /r/ as an important sociolinguistic variable in the East Coast varieties of American English (Labov 1966). In general, the enormous variability of rhotics makes this set of sounds highly suitable as a linguistic marker of regional, social, or age distinctions; see Scobbie (2006) for comment and more examples. The present section is meant to demonstrate the extent of the variability of rhotic sounds and the possible speed of change for these sounds. Examples here are taken from German, but similar points could be made for Dutch, French, English, and other well-documented languages (see the contributions on various European languages in Van de Velde and van Hout 2001). In the field of German dialectology, Göschel (1971) provides a survey on variants of prevocalic /r/ across local dialects in Germany of the 1930s. As shown in the map in Figure 30.1, /r/ across German dialects could be alveolar, retroflex, or uvular. It could also be either tap/flap, trill, approximant, or fricative. Thus, nearly all variants of /r/ listed in (1) above can be found in the dialects of a single language at one particular point in time. Focusing on Austrian German and Swiss German variants, Ulbrich and Ulbrich (2007) found a total of 12 different phonetic variants of the phoneme /r/, plus a zero realization. The unity of rhotics in this case consists in the fact that all of these varieties have exactly one rhotic phoneme, usually in corresponding positions in cognate words, but its phonetic realization varies widely across the varieties of German,

Richard Wiese

8

Königsberg

Hamburg Stettin

Bremen

[r]

Münster

[r]

Magdeburg

Kassel

[È]

[ú]

Köln

Breslau

Dresden

Erfurt

[;]

Koblenz

[È]

Frankfurt Trier

Prag Würzburg

[r] [r] [r]

Regensburg

The articulation of r in German

alveolar [r] [7] uvular [r]

Ulm

retroflex [È], also [P]

München

voiced uvular or velar fricative sounds [ú] [:] From tape-recordings of 1200 speakers in 300 locations from “Sound memorial of German dialects in 1936.”

Jnnsbruck 0

50 km

100

Transcribers: G. Heike, F. Schindler, J. Göschel

Figure 30.1 Variants of prevocalic /r/ across local dialects in Germany in the 1930s (from Göschel 1971: 74)

with no obvious limitations. In the same study, Göschel (1971: 98) also demonstrates that the languages of Europe display the complete range of r-sounds distinguishable by articulatory parameters. Nearly all of these languages have exactly one rhotic phoneme, confirming the point that there is a large difference between phonetic variability on the one hand and phonological uniformity on the other (see chapter 89: gradience and categoricality in phonological theory).8 Wiese (2003) analyzes the changes of rhotics found for the speakers of the dialect in southwest Germany, west of the river Rhine, as described in the dialect atlas of Bellmann et al. (1999). This dialect atlas compares dialectal features across two generations of (male) dialect speakers, those from an older generation, around 70 years old, and those from a younger generation, around 40 years old, and presents separate maps for the two generations. The comparisons presented in (6) give the results of counting all the r-related changes documented in three different lexical contexts in Bellmann et al. (1999).9 8

I am not aware of any European language without a single rhotic phoneme. Some Romance languages have two rhotic phonemes, although the distinction between them is partially neutralized (see discussions of Spanish and Irish Gaelic above). 9 A small number of points on the maps are rendered as “hard to segment.” These were not counted.

The Representation of Rhotics

9

These counts demonstrate that, from one generation of dialect speakers to the next, up to 60 percent of rhotics are changed in their phonetic realization. What remains is a single rhotic phoneme in the respective word. (6)

Intergenerational changes a.

b.

c.

Berg ‘mountain’ change in type of /r/ no change Rose ‘rose’ change in type of /r/ no change fahren ‘go, drive’ change in type of /r/ no change

194/327 ≈ 60% 133/327 ≈ 40% 93/296 ≈ 31% 203/296 ≈ 69% 106/313 ≈ 34% 207/313 ≈ 66%

Word-initial or syllable-initial rhotics change in about a third of the tokens, while rhotics in the coda, as in the lemma Berg, showed a large number of r-realizations for both generations of speakers. Furthermore, the most r-realizations changed across the generations. There is no reason to think that this phenomenon is restricted to this particular dialect. An even more striking case of such a change in rhotic realization is reported by Enderlin (1911: 168). In his study of the Alemannic (Swiss German) dialect of Kesswil, he notes that at the local school all 1st grade students had uvular [ö], while all 9th grade students had alveolar [r]. He also reports a 50 percent realization of both forms for 4th and 5th grade students. In general, postvocalic, rhymal rhotics seem to be subject to more variation than rhotics in onset positions. In Table 30.2, these changes are classified according to the type of change found in the same dialect map, that for Berg ‘mountain’ in Bellmann et al. (1999: 463). The r-sounds listed next to “older generation” are those found for the older speakers; while the first row lists rhotics found for the younger generation. The check marks denote intergenerational changes for particular pairs of rhotics (i.e. in some particular location as selected in the dialect atlas) through the comparison of the two maps.

Older generation

Table 30.2 r-conversions found for one lemma (Berg ‘mountain’)

r

ö

Younger generation > > H

7

✓

✓

✓

–

✓

–

✓

ö

✓

✓

✓

–

✓

–

✓

>

–

–

–

–

–

–

–

>

–

–

–

✓

✓

–

–

H

–

✓

✓

–

✓

–

✓

C

–

–

–

–

✓

–

–

Ø

✓

–

–

–

✓

–

✓

C

Ø

10

Richard Wiese

Although r-pronunciation may vary considerably as a result of dialectal change, the two related varieties studied display a large range of rhotics. In other words, dialectal change does not lead to a neutralization between different realizations of /r/ or to a proliferation of rhotic phonemes. The continuous and omnipresent change from one type of rhotic to another is also demonstrated on a different level in the acoustic measurements presented by Ladefoged and Maddieson (1996: 219f.) for individual tokens: their spectrograms of r-sounds from a speaker of Standard Italian show that these sounds, usually classified as alveolar trills, may display an approximant, possibly vowel-like, phase, preceding and/or following the opening/closing phase characteristic of trills. Many other languages display a range of phonetic variants of /r/; in §2.2 above, Persian was shown to have a trill /r/ phoneme, with a whole range of allophonic variants, all of which are rhotics. For more examples of this type see Lindau (1985: 158–159). Walsh Dickey (1997: 96–97) reports that different dialects of Portuguese show a similar and very wide range of variants of the (two) rhotic phonemes. Here and in the case of the southwest German dialect discussed above, even individual realizations of the r-phoneme are not confined to particular manners or places of articulation.

3

The featural basis of rhotics

(1) demonstrated the articulatory variability of rhotics. There is simply no articulatory feature there which is shared by all rhotics, and it is hard to see what other, possibly more general, articulatory feature might do the job. Major class features as used in theories of phonology since the proposals advanced by Chomsky and Halle (1968) are of no help either (see chapter 13: the stricture features; chapter 17: distinctive features). In order to see this, consider the values for the classificatory features [continuant], [sonorant], and [consonantal], taken here to be binary features. Rhotics are not uniform with respect to any of these features: taps and flaps are usually classified as [−continuant], while approximants and trills are [+continuant]; fricatives are [−sonorant], while all other rhotics are [+sonorant]. As for [consonantal], rhotics are [+consonantal], but this class is far too comprehensive in order to be of help in the classificatory definition. Furthermore, rhotic vowels might well be included in the class of rhotics; see discussion in §4. (For further discussion of these features for rhotics, see Hall 1997: 124–128.) Phonetic studies have therefore explored the possibility that a description of rhotics might be given in acoustic rather than articulatory terms. Lindau (1978), in particular, explores the acoustics of rhotics, finding that many r-sounds across languages are characterized by a lowered third formant. However, for other rhotics, again cross-linguistically, the third formant was actually found to be very high (for summaries, see Lindau 1985 and Ladefoged and Maddieson 1996: 244–245). The conclusion drawn by Lindau (1985) and Ladefoged and Maddieson (1996) is that there is no possibility of a description establishing the unity of all rhotics as a class by means of phonetically based features, either articulatory or acoustic. Lindau further concludes that there are family resemblances between neighboring subclasses of rhotics, but that there are no overall characteristics common to

The Representation of Rhotics

11

all the subclasses. Similarly, Kohler concludes that a positive characterization of the phoneme /r/ in German, encompassing all its allophonic variations, is not possible, even for a single speaker: “only a negative characterization is possible” (Kohler 1995: 156). Ladefoged and Maddieson (1996: 245) summarize on an even more pessimistic note: Although there are several well-defined subsets of sounds (trills, flaps, etc.) that are included in the rhotic class, the overall unity of the group seems to rest mostly on the historical connections between these subgroups, and on the choice of the letter “r” to represent them all.

This conclusion would deny the existence of rhotics as a phonetically defined class, and is rather pessimistic on the possibility of providing any coherent descriptions in phonetic terms. It falls back on the conventions of alphabetic writing, while these are themselves obviously in need of explanation. Conceivably, the spelling of some (class of) sounds by means of the letter exerts some influence on the paths of historical change of these sounds. But to assume that this spelling has a pervasive cross-linguistic influence and thereby constitutes the sole basis of the development of a class of rhotic sounds worldwide does not seem to be well founded. Other well-documented sound changes such as spirantizations, consonant losses, or vowel shifts do not seem to be restricted by the spelling systems. For example, the Second Germanic Consonant Shift, changing /p/, /t/, and /k/ to fricatives or affricates (Iverson and Salmons 2006), was not prevented by the fact that the spelling of affected words was changed. Both Lindau (1985) and Ladefoged and Maddieson (1996), while noting the lack of a convincing segmental definition, emphasize the role of rhotics as a phonologically relevant class, along the lines discussed in §2 above. In order to express this unity, Hall (1997) proposes to use a feature [±rhotic] as a classificatory feature for the rhotic/non-rhotic distinction. However, a substantive definition of this abstract feature does not seem to be available. In yet another attempt, Walsh Dickey (1997) proposes to define rhotics by means of the feature Laminal, the use of the tongue blade as opposed to the tongue tip, as an articulator subordinated in the feature hierarchy to the articulator Coronal, as presented in (7). (7)

Feature structure of rhotics (Walsh Dickey 1997: 106) [liquid] Coronal Apical Laminal

However, it is questionable whether all rhotics make use of this feature structure. It is certainly not the case for uvular rhotics. Furthermore, it is unclear why such a marked segment class (expressed here by a deep hierarchical stacking of several

12

Richard Wiese

place features, as opposed to an underspecified representation indicating an unmarked class; see chapter 27: the organization of features; chapter 7: feature specification and underspecification) should be present in the majority of languages. The question of manner features for rhotics is even more pressing: if the degree of opening for rhotics may range from vocalic to fricative, as the survey given above seems to suggest, it is unclear how the class of rhotics may be characterized as a whole. Furthermore, this raises the puzzle how segments at the extreme ends of this dimension are to be classified: for example, when is a voiced uvular fricative [ú] a rhotic, and when is it not? The following section will propose an approach from a different angle. The failure to find a common denominator for rhotics in terms of acoustic or articulatory features does not preclude the possibility that it can be found eventually, but chances seem to be slight.

4

Alternative proposals

Featural descriptions proposed for rhotics, as discussed in the preceding section, have in common that they attempt to characterize rhotics in purely segmental terms. But we have already seen above that rhotics are tightly connected to their positions within larger phonotactic patterns, at least with respect to the conditions for their allophonic variants or for patterns of complementary distributions. This observation raises the question whether rhotics should in fact be described in terms of purely segmental categories, or in segmental terms at all. An alternative view would be to capitalize on the observation that rhotics appear in a particular well-defined syllabic position, namely the one immediately adjacent to the vowel. This view, proposed by Wiese (2001), and previously by Selkirk (1984), relies primarily on the apparently uniform behavior of all types of rhotics in terms of their syllabic constraints, and suggests that this is in fact the defining and constant property of the class of rhotics. On this view, the search for constant segmental properties is futile, because it starts from an incorrect presupposition, i.e. that classificatory features are by necessity segmental features. In contrast, it seems that rhotics are very stable with respect to their phonotactic behavior. In particular, their slot in the structure of a syllable does not seem to change with a change in their segmental make-up. For example, Hall (1993) discusses a (lower Rhine) variety of German in which the rhotic phoneme, in coda position, varies between a vowel and a voiceless fricative (Tor [tos] ‘gate’ vs. Sport [œpDøt] ‘sport’). In other words, a fricative rhotic obeys the same constraints on syllabic placement as a trill or approximant rhotic, or even a rhotacized vowel. The proposal then is that rhotics are defined as a particular relative point on the sonority scale, the point between vowels and laterals. Another pattern discussed in Wiese (2003) casts doubts on the segmental approach to rhotics as a class. (8) presents those onset clusters of present-day Standard German which consist of any stop followed by a sonorant (the velar nasal /I/ is excluded from such clusters on principled grounds). Examples given in the cells of the table include rather marginal clusters (such as /pn/ or /tm/ occurring in a few rare words only), but in all of these rare cases, there is no tendency to replace the clusters in question by a more natural cluster or to break it up by a process of epenthesis.

The Representation of Rhotics (8)

13

Initial clusters of stop + sonorant in Standard German /p/ /b/ /t/ /d/ /k/ /g/

/r/ Preis braun Traum drei Kreis grau

/l/ Platz blau — — klug Glaube

/n/ Pneu — — — Knie Gnade

/m/ — — Tmesis — Khmer Gmünd

Not all possible stop–sonorant clusters are attested. As shown in (8), /r/ can be combined with any of the existing stops, while all other sonorant consonants are restricted in some way or other with respect to the preceding stops. Basically, all homorganic clusters, such as /tl/ or /pm/, are ill-formed in (8), but there might be additional restrictions ruling out /bn/ and /dm/. For discussion of these, see Hall (1992: 65–80) and Wiese (2000: 261–269). But crucially, in contrast to the nonrhotic sonorants, no restrictions at all hold for stop–sonorant clusters involving /r/. As the examples in (8) demonstrate, this phoneme productively combines with any preceding stop, and none of the clusters involving /r/ is marginal. While it would be possible to simply make /r/ exempt from the ban on homorganic clusters, such a move would simply beg the question why this is the case. It seems more plausible to say that place features of /r/ simply do not count, or, in an underspecification approach to features, are not present (see chapter 22: consonantal place of articulation; chapter 7: feature specification and underspecification). The homorganicity ban in place for other sonorants cannot apply, then, leading to the complete set of stop–/r/ clusters shown in (8). Similar patterns can also be demonstrated for other clusters in German (such as fricative–stop clusters), and for similar clusters in other languages (e.g. English, Italian, Basque, Lithuanian). This line of argumentation, arguing that rhotics are defined as those sounds which bear a sonority value between that of vowels (including glides) and the next lower sonority class, is supported by the fact that the freedom of rhotics to combine with a preceding stop is independent of the particular type of the r-phoneme present in the respective variety of German – place features as well as manner features of the rhotic phoneme are always irrelevant. Finally, the proposal gives an answer to a puzzle noted above, namely that one and the same segment may sometimes be classified as a rhotic and sometimes not. This is particularly obvious in the case of voiced uvular fricatives: for Standard French, [ú] is generally seen as a rhotic; for Classical Arabic, what is apparently the same segment is not (Watson 2002: 13). But if rhotics are defined in terms of their phonotactic behavior, this is less mysterious: in French, the segment in question appears between obstruents and vowels, as in the initial cluster in frais [fúe] ‘fresh’ or in the final cluster in carte [kaút] ‘card’, while in Classical Arabic the respective segment patterns with other fricatives in terms of phonotactics, even after the rhotic [r] as in farı [farú] ‘width’. In addition, Arabic has a tap or trill /r/ which is analyzed as a rhotic, because it behaves accordingly in terms of phonotactic patterning: e.g. San’ani Arabic sirt ‘I/you (masc sg) went’; hribt ‘I fled’ (Watson 2002: 67, 73). Dutch is another language which shows both a phoneme /r/ (with a great deal of variation) and a voiced velar or uvular fricative /:~ú/ (see Booij 1995: 7–8).

Richard Wiese

14

In summary, there is some evidence for a description of the rhotic class in terms of sonority (chapter 49: sonority). First, this hypothesis explains the constant phonotactic position of rhotics in the syllable, and second, it explains the similarity of rhotics to vowels across languages, namely the phenomena of r-vocalization and of rhotic–vowel interaction. This view also abstracts away from all types of segmental realizations for rhotics, a move justified by their non-uniform behavior with respect to segmental variability, and is able to accommodate the fact that a segment sometimes acts as a rhotic and sometimes does not. The proposal does not mean that specific rhotics are necessarily unspecified for place and/or manner; obviously such a specification is needed at least for languages with more than one rhotic. Rhotacized vowels, which have largely been ignored so far, can also be brought under the sonority-based definition of rhotics. Consider the final segments in the words in (9), from American English and German. These are, arguably, vocalized variants of /r/ in the two languages. In other words, rhotacized vowels are rhotics functioning as syllabic nuclei or as glides in postvocalic position. (9)

Rhotacized vowels a.

American English ladder [lædÌ] beer [b>r]

b. German Leiter [la>tô] Bier [bis]

‘ladder’ ‘beer’

In terms of segmental features, it would be difficult, if not impossible, to give a definition of rhotics which includes these or similar vowels. In terms of a sonoritybased definition, these vowels naturally fit into the class to which consonantal rhotics belong as well. This is because all sonorant consonants (/r l n m/) in English and German can be realized as syllabic nuclei. The only difference between /r/ and the remaining sonorants is that the latter are not vocalized.

5 5.1

Related phenomena Rhotacism

Changes of some sounds into an /r/ are repeatedly found in languages, and have been given a name of their own, rhotacisms (Barry 1997; Catford 2001).10 Rhotacisms may lead to systematic alternations, such as those in (10) for Latin and Germanic. (10)

Rhotacisms a.

10

Latin nom sg ius os opus

gen sg iuris oris opera

‘law’ ‘mouth’ ‘work’

Speech disturbances involving r-sounds, especially in language acquisition, are also called rhotacisms. These will not be discussed here. It seems that the alveolar trill is particularly difficult to acquire, again raising the question of why its overall frequency is so dominant (see §2.1).

The Representation of Rhotics b.

Germanic maiza huzd dius (gen diuzis) marzjan

West Germanic mero hord, hort deor, tior meriian, merren

15

‘more’ ‘treasure’ ‘animal’ ‘disturb’

Latin rhotacism (10a) is analyzed by Walsh Dickey (1997: 82) as intervocalic voicing, on the basis of the assumption that there are no alveolar continuant consonants in the Latin phonemic system other than /s/ and /r/. The transfer of [+voice] from a neighboring vowel to /s/ would then lead to /r/ if this rule is constrained by structure preservation (because there is no /z/). Germanic rhotacism, exemplified in (10b), seems to be a similar phenomenon, but is not restricted to intervocalic contexts. The sources (e.g. Braune and Reiffenstein 2004: 80) argue that the /z/ undergoing rhotacism was a voiced sound already, itself derived from Germanic voiceless /s/ through Verner’s Law. This /z/ would spontaneously undergo rhotacism in West Germanic and North Germanic (but not in Gothic). The sound change from /s/ (perhaps via /z/) to /r/ seems to be the most common type of rhotacism, as in the examples of (10), but other changes, such as /n/ to /r/, are found as well (see Catford 2001).

5.2

r-epenthesis and metathesis

Rhotics are also found quite often as epenthetic consonants. Given that epenthetic consonants are usually those regarded as unmarked (glides, /?/, and /t/ being the most obvious examples), this is somewhat surprising. On the other hand, the phonetic variability of rhotics, ranging from a fricative to a glide, may make an r-sound particularly suitable as an epenthetic segment. Varieties of English can be sorted into rhotic dialects (roughly those with a rhotic in the syllable coda) and non-rhotic dialects (those without a rhotic in the coda). The latter dialects, in turn, are often classified into those displaying so-called “linking r” and those displaying “intrusive r.” In the “linking r” varieties, /r/ is absent in, e.g. near ([n>H]) and car ([kA(]), but surfaces in near him ([n>P>m]) and car is ([kAP>z]). The “intrusive r” type of varieties goes beyond linking r, and provides a well-known example of r-epenthesis in additional contexts, as in India and [>nd>HrHnd]. Uffmann (2007) discusses the phenomenon of r-epenthesis in these English dialects, and argues that the choice of /r/ instead of some other unmarked consonant is by no means arbitrary. Rather, /r/ provides the optimal consonant in a non-margin position because of its value on the sonority scale (see also Ortmann (1998: 58): “r is the default approximant”). A margin position is one at a domain edge, such as a foot or word boundary, while all other positions are non-margin positions. Thus /r/ is the consonant to be chosen in the internal position of the English example above, while glottal stop or some other unmarked stop is typically inserted in a margin position. This analysis again relates rhotics to a particular place on the sonority hierarchy, without reference to their segmental features. Rhotics may also be involved in metathesis between vowels and consonants (chapter 59: metathesis); see Blevins and Garrett (1998: 516ff.) for a discussion of rhotic metathesis in the Le Havre dialect of French. Given that rhotics occur adjacent to vowels, as noted above, and that metathesis reverses the order of adjacent segments, this is not surprising. Blevins and Garrett’s analysis involves

16

Richard Wiese

the assumption that rhotics involve a lowering of the third formant in r-sounds, an assumption, however, which does not seem to be valid for rhotics in general (see discussion in §3 above).

5.3

Liquids

Consonants of the class comprising both rhotics and laterals, the latter most commonly represented by the alveolar approximant /l/ (chapter 31: lateral consonants), are often called “liquids” (from Latin liquida ‘fluid (adj; neut pl)’, but the concept is utilized in Greek metrics to refer to the larger class of rhotics, laterals, and nasals). Greek and Latin grammarians noted at an early stage that such liquids share some properties, and form a unified class. In particular, Latin clusters consisting of a stop followed by either /l/ or /r/, the so-called muta cum liquida, behave like single consonants with respect to Latin syllabification. Furthermore, the muta cum liquida are the only true consonantal clusters allowed as complex onsets, along with clusters with initial fricatives, as in frater ‘brother’ and flumen ‘river’ (Marotta 1999: 298). In other languages as well, rhotics and laterals form a single phonological class, thus reinforcing the unity of liquids as a class. In a number of languages, e.g. Korean, Maori, and Japanese, there is widespread variation between a rhotic (flap) [7] and lateral [l] as realizations of a single phoneme. This phoneme may therefore most adequately be interpreted as an otherwise unspecified liquid consonant. The variation may be one either of free variation or of complementary distribution (see Rice 2005 for more examples). Another property shared by liquids is their similarity in terms of the sonority hierarchy: if rhotic and lateral liquids are different at all with respect to their sonority value taking the same position within the syllable, they occupy adjacent positions on this hierarchy, with laterals occupying the less sonorous position. For Germanic languages, there is evidence that the latter is true: rhotics are more sonorous than laterals and thus occupy the position closer to the vocalic nucleus: e.g. curl in rhotic dialects of English or German Kerl ‘guy’.

6

Final discussion

The present chapter has provided rich and varied evidence for the phonetic variability of rhotics. This variability, between and within languages, dialects, and even idiolects, is in stark contrast to the phonological unity and consistency of rhotics. It is this unity which justifies a unified representation of rhotics. Therefore, the present discussion of rhotics turns out to be a significant topic for the study of the relation between phonetics and phonology. On the one hand, no analysis has been successful in demonstrating that rhotics form a class in phonetic terms. Furthermore, the variability of rhotics within individual languages and across languages is impressive. On the other hand, there are various strands of evidence showing that rhotics do form a single class in terms of their phonological behavior. This uniform behavior may best be described as a syllable-related patterning. One conclusion therefore is that the phonological level is not necessarily one of segmental abstraction from the phonetic one. If this picture comes close to the truth, it also provides evidence for a view on the relation between phonetics and phonology according to which there can

The Representation of Rhotics

17

be a considerable difference between these two levels. In some recent theories of phonology, emphasis has been placed on the seamless integration of phonetic and phonological representations. In the model of Articulatory Phonology, for example, all representations consist of articulatory gestures, i.e. movements in the vocal tract, which constitute the domain for phonological as well as phonetic representations (for a survey see Browman and Goldstein 1992; chapter 5: the atoms of phonological representation). In a treatment of English /r/ and its alternations with “zero,” McMahon et al. (1994: 303) argue that /r/ consists of two articulatory gestures, a palatal constriction and a pharyngeal constriction. It remains to be seen to what extent this approach is adequate for other rhotics, in particular the prototypical trill and the uvular varieties. The discussion of rhotics in this chapter raises the question of whether a gestural – or any other – representation in terms of uniform and rather concrete units is possible for the characterization of rhotics.

REFERENCES Bammesberger, Alfred. 1982. Essentials of Modern Irish. Heidelberg: Carl Winter Verlag. Barry, William J. 1997. Another R-tickle. Journal of the International Phonetic Association 27. 34–45. Bellmann, Günter, Joachim Herrgen & Jürgen Erich Schmidt. 1999. Mittelrheinischer Sprachatlas, vol. 4: Konsonantismus. Tübingen: Niemeyer. Birk, David B. W. 1975. The phonology of MalakMalak. In Sharpe et al. (1975), 59–78. Blevins, Juliette & Andrew Garrett. 1998. The origins of consonant–vowel metathesis. Language 74. 508–556. Booij, Geert. 1995. The phonology of Dutch. Oxford: Clarendon Press. Braune, Wilhelm & Ingo Reiffenstein. 2004. Althochdeutsche Grammatik, vol. 1: Laut- und Formenlehre. Tübingen: Niemeyer. Browman, Catherine P. & Louis Goldstein. 1992. Articulatory phonology: An overview. Phonetica 49. 155–180. Catford, J. C. 2001. On Rs, rhotacism and paleophony. Journal of the International Phonetic Association 31. 171–185. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Enderlin, Fritz. 1911. Die Mundart von Kesswil im Oberthurgau: Mit einem Beitrage zur Frage des Sprachlebens. Frauenfeld: Huber & Co. Foulkes, Paul & Gerard J. Docherty. 2000. Another chapter in the story of /r/: “Labiodental” variants in British English. Journal of Sociolinguistics 4. 213–222. Göschel, Joachim. 1971. Artikulation und Distribution der sogenannten Liquida r in den europäischen Sprachen. Indogermanische Forschungen 76. 84–126. Hall, T. A. 1992. Syllable structure and syllable-related processes in German. Tübingen: Niemeyer. Hall, T. A. 1993. The phonology of German /r/. Phonology 10. 83–105. Hall, T. A. 1997. The phonology of coronals. Amsterdam & Philadelphia: John Benjamins. Hall, T. A. 2000. Typological generalizations concerning secondary palatalization. Lingua 110. 1–25. Harris, James W. 1969. Spanish phonology. Cambridge, MA: MIT Press. Haspelmath, Martin, Matthew S. Dryer, David Gil, Bernard Comrie & Hans-Jörg Bibiko (eds.) 2005. The world atlas of language structures. Oxford: Oxford University Press. Hulst, Harry van der & Nancy Ritter (eds.) 1999. The syllable: Views and facts. Berlin & New York: Mouton de Gruyter. IPA. 2007. Handbook of the International Phonetic Association: A guide to the use of the International Phonetic Alphabet. Cambridge: Cambridge University Press.

18

Richard Wiese

Iverson, Gregory K. & Joseph C. Salmons. 2006. Fundamental regularities in the Second Consonant Shift. Journal of Germanic Linguistics 18. 45–70. Jagst, Lothar. 1975. Ngardilpa (Warlpiri) phonology (language of the Warnayaka tribe, a subtribe of the Warlpiri tribe). In Sharpe et al. (1975), 21–57. Kohler, Klaus J. 1995. Einführung in die Phonetik des Deutschen. 2nd edn. Berlin: Schmidt. Labov, William. 1966. The social stratification of English in New York City. Washington, DC: Center for Applied Linguistics. Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’s languages. Oxford & Malden, MA: Blackwell. Lindau, Mona. 1978. Vowel features. Language 54. 541–563. Lindau, Mona. 1985. The story of /r/. In Victoria A. Fromkin (ed.) Phonetic linguistics: Essays in honor of Peter Ladefoged, 157–168. Orlando: Academic Press. Lipski, John M. 1990. Spanish taps and trills: Phonological structure of an isolated opposition. Folia Linguistica 24. 153–174. Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press. Majidi, Mohammad-Reza. 1986. Strukturelle Grammatik des Neupersischen (Farsi), vol. 1: Phonologie: Eine paradigmatisch-syntagmatische Darstellung. Hamburg: Buske. Majidi, Mohammad-Reza. 2000. Laut- und Schriftsystem des Neupersischen. Hamburg: Buske. Marotta, Giovanna. 1999. The Latin syllable. In van der Hulst & Ritter (1999), 285–310. McMahon, April, Paul Foulkes & Laura Tollfree. 1994. Gestural representation and Lexical Phonology. Phonology 11. 277–316. Nash, David. 1980. Topics in Warlpiri grammar. Ph.D. dissertation, MIT. Ní Chiosáin, Máire. 1994. Irish palatalisation and the representation of place features. Phonology 11. 89–106. Ní Chiosáin, Máire. 1999. Syllables and phonotactics in Irish. In van der Hulst & Ritter (1999), 551–575. Odden, David. 2006. Kimatuumbi: Phonology. In Keith Brown (ed.) Encyclopedia of language and linguistics. 2nd edn, vol. 6, 198–206. Oxford: Elsevier. Ortmann, Albert. 1998. Consonant epenthesis: Its distribution and phonological specification. In Wolfgang Kehrein & Richard Wiese (eds.) Phonology and morphology of the Germanic languages, 51–76. Tübingen: Niemeyer. Padgett, Jaye. 2003. Systemic contrast and Catalan rhotics. Unpublished ms., University of California, Santa Cruz. Rice, Keren. 2005. Liquid relationships. Toronto Working Papers in Linguistics 24. 31–44. Schiller, Niels O. 1999. The phonetic variation of German /r/. In Matthias Butt & Nanna Fuhrhop (eds.) Variation und Stabilität in der Wortstruktur: Untersuchungen zu Entwicklung, Erwerb und Varietäten des Deutschen und anderer Sprachen, 261–287. Hildesheim: Olms Verlag. Scobbie, James M. 2006. (R) as a variable. In Keith Brown (ed.) Encyclopedia of language and linguistics. 2nd edn, vol. 10, 337–344. Oxford: Elsevier. Selkirk, Elisabeth. 1984. On the major class features and syllable theory. In Mark Aronoff & Richard T. Oehrle (eds.) Language sound structure, 107–136. Cambridge, MA: MIT Press. Sharpe, Margaret C., Lothar Jagst & David B. W. Birk (eds.) 1975. Papers in Australian linguistics 8. Canberra: Australian National University. Uffmann, Christian. 2007. Intrusive [r] and optimal epenthetic consonants. Language Sciences 29. 451–476. Ulbrich, Christiane & Horst Ulbrich. 2007. The realisation of /r/ in Swiss German and Austrian German. In Jürgen Trouvain & William J. Barry (eds.) Proceedings of the 16th International Congress of Phonetic Sciences, 1761–1764. Saarbrücken: Saarland University. Van de Velde, Hans & Roeland van Hout (eds.) 2001. ’r-atics: Sociolinguistic, phonetic and phonological characteristics of /r/. Brussels: Université Libre de Bruxelles.

The Representation of Rhotics

19

Walsh Dickey, Laura. 1997. The phonology of liquids. Ph.D. dissertation, University of Massachusetts, Amherst. Watson, Janet C. E. 2002. The phonology and morphology of Arabic. Oxford: Oxford University Press. Wheeler, Max W. 2005. The phonology of Catalan. Oxford: Oxford University Press. Whitley, M. Stanley. 2003. Rhotic representation: Problems and proposals. Journal of the International Phonetic Association 33. 81–86. Wiese, Richard. 2000. The phonology of German. Oxford: Oxford University Press. Wiese, Richard. 2001. The phonology of /r/. In T. A. Hall (ed.) Distinctive feature theory, 335–368. Berlin & New York: Mouton de Gruyter. Wiese, Richard. 2003. The unity and variation of (German) r. Zeitschrift für Dialektologie und Linguistik 70. 25–43.

31

Lateral Consonants Moira Yip

Laterals are extremely common, and yet they are something of a phonological puzzle. For most consonants, there is fairly general agreement on which natural classes they belong to, and clear expectations of how they pattern in phonological processes. So for example [m] is Labial and [+nasal], and as such it patterns consistently with the other labials [p b], and the other nasals [n I]. But for laterals this is not the case, as we shall see. Their behavior is highly variable across languages. This chapter gives an overview of their somewhat perplexing behavior, summarizes extant proposals for their phonological representation, and ends by advocating a proposal using violable constraints that allows for this variability. I begin in §1 with some background on their articulation and acoustics, and how they are acquired by the child. §2 summarizes the types of laterals found in natural language. §3 surveys their roles in syllable structure. §4 discusses their sonorancy, voicing, and continuancy characteristics. §5 looks at alternations between laterals and other sounds. §6 asks whether there is a feature [lateral]. §7 looks at the positioning of [lateral] in a theory of feature geometry. I show that there are serious problems associated with any single choice of superordinate node, and instead in §8 I propose an approach based on violable feature co-occurrence constraints. §9 concludes.

1

Phonetics, perception, and acquisition

1.1 Types of laterals and their frequency in the world’s languages Over 80 percent of languages have one or more lateral consonants (Maddieson 1984). Laterals are defined by Ladefoged and Maddieson (1996: 183) as “sounds in which the tongue is constricted in such a way so as to narrow its profile from side to side so that a greater volume of air flows around one or both sides than over the center of the tongue.” The palatograms in Figure 31.1 show clearly that the tip of the tongue makes contact with the roof of the mouth, but the sides do not. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0031

Lateral Consonants Dark/l/

2

Light/l/

AK

PK

MI

SC

Figure 31.1 Linguo-palatal contact profiles of dark and light laterals for four speakers of American English, showing mid-sagittal contact in the alveolar and pre-palatal regions but no or limited contact at the sides. From Narayanan et al. (1997)

The approximant versions have some turbulence at the (incomplete) stricture; resonant versions like Standard German coda /l/ do not. Since the vocal tract is not fully obstructed in the typical approximant /l/, they are among the most sonorous of consonants (see chapter 49: sonority). They frequently contrast with other sonorants, especially rhotics, and the class of laterals and rhotics is referred to as the liquids (see chapter 8: sonorants). In other languages (such as Japanese) a language has only a single liquid, and it may vary between a lateral [l] and a more rhotic tap or flap [7]. The variation may depend on context (such as syllable position) or they may be in free variation. Speakers of such languages famously have trouble perceiving and producing the difference between [l] and [r] when learning languages like English (see Iverson et al. 2003). There are also lateral obstruents, which will be discussed below.

1.2 Articulation, acoustics, and acquisition of laterals Most laterals are dental or alveolar in articulation, but the tongue body is also frequently implicated, as shown for English by Sproat and Fujimura (1993). Gick et al. (2006) studied the articulation of laterals in six languages: Western Canadian English, Quebec French, Serbo-Croatian, Korean, Beijing Mandarin, and Squamish Salish. They found that all their laterals had an anterior tongue gesture in all syllable positions. More interestingly, in coda position all also have a dorsal gesture, which starts slightly earlier than the anterior gesture. Some, but not all, have a dorsal gesture in onset position too, in which case it is roughly simultaneous with the anterior gesture. The dorsal gesture in coda position may result from biomechanical causes, such as “active lateral compression of the

3

Moira Yip 4000

3000

2000

1000

Hz Figure 31.2 (2006)

[ l

e

d ]

[ j

e

l ]

Spectrograms of the English words led and yell, adapted from Ladefoged

tongue [. . .], which could result in a consistent but small non-phonological tongue dorsum backing, simply as the result of volume displacement” (Gick et al. 2006: 69). In onset position, however, the dorsal gesture may be phonological. The amount of dorso-palatal contact may vary from language to language, and may also depend on the vocalic context and the syllable position. In some languages there may be a continuum from the more anterior “clear” l to the more posterior “dark” l; in others there may be a categorical distinction. See Recasens and Espinosa (2005) for a useful overview. Taking these articulatory facts seriously, Walsh Dickey (1994, 1997) opts for a representation of laterals that explicitly labels them as featurally both Coronal and Dorsal. See §6 for further discussion. Acoustically, laterals have strong formant structure with a low F1 and a high F3. F2 is variable depending on place of articulation, and whether /l/ is clear or dark. The spectrogram in Figure 31.2 shows English /l/ in initial and final position. The raised F3 is detectable not only in the immediate vicinity of laterals, but up to five syllables preceding the lateral itself (Heid and Hawkins 2000). The low F1 groups them with the high vowels, and when they vocalize (synchronically, historically, or in acquisition), it is usually to [w/u], possibly because the acoustic spectra for [l] and for [w] are quite similar. See Ohala (1974). There are also possible articulatory reasons for this outcome: see §5.4 for an account along these lines. Laterals are often acquired late, and frequently children go through a phase in which they replace laterals by glides, as in Inkelas and Rose’s (2008) study of E, who replaces onset /l/ with [j] and coda /l/ with [w] (see chapter 101: the interpretation of phonological patterns in first language acquisition). It is likely that this is the result of both the motoric difficulty of phasing the complex tongue tip and body gestures required for /l/, and the variable nature of /l/ in English adult speech. The choice of high glides matches their acoustic and articulatory characteristics. See Johnson and Britain (2007) on the difference between the acquisition of clear l (often early) and dark l (often late, which is unexpected if its articulation is biomechanically unavoidable).

Lateral Consonants

2

4

Types of laterals

A suitable theory must account for the full range of lateral sounds, as summarized in this section, which is based heavily on Ladefoged and Maddieson (1996). Most laterals are coronal, and they may have any of eight coronal places of articulation: dental, alveolar, postalveolar, and palatal, each either apical or laminal (see chapter 12: coronals). All of these can contrast, though the largest set of contrasts within a single language is four, as in some Australian languages, such as Kaititj. Several others have a three-way contrast. Two-way contrasts are quite common. (1) shows the most common voiced lateral approximants by place of articulation: further distinctions are shown by the use of additional diacritics when necessary. (1)

Most common lateral places of articulation dental alveolar postalveolar (retroflex) palatal velar

* l Î O l

There are no known labial laterals, not even linguo-labial ones formed by the tongue and upper lip, even though they occur in child language, and even though in adult language linguo-labial stops and fricatives (but not laterals) are found in languages of Vanuatu (Maddieson 1987). For some English speakers, [f] and [v] have central closure only, and could thus be called labio-dental laterals, but they do not pattern with the true laterals phonologically. Phonetically, velar laterals are found, and their articulation is clearly velar, not coronal. The back of the tongue makes contact with the velum, and the sides of the tongue are lowered so air can escape past the molars. Their velarity is further attested to in some cases by their ability to spread their velarity to adjacent segments. For example, in Zulu, nasals before the velar lateral ejective affricate are [I], not [n]. The examples below come from the Papua New Guinea language Mid-Waghi, which contrasts the two coronals, dental [*] and alveolar [l], with the velar [l]: (2)

a*a a*a alala alale

‘again and again’ ‘speak incorrectly’ ‘dizzy’

Nonetheless, Blevins (1994) argues that velar laterals always have the phonological feature Coronal. The Papua New Guinean language Yagaria shows a coalescence of a lateral and a glottal stop. The result is a voiceless coronal stop. The lateral in question is phonetically a velar lateral, which Blevins argues to be phonologically Coronal, because the output of the coalescence is [t]. In (3a) the process changes the initial /s/ of the suffix to [p] after /?/, and in (3b) the initial velar lateral /l/ of the suffix to changes to [t] after /?/:

Moira Yip

5 (3)

a. b.

/igopa-si?/ /jo?-si?/ /igopa-lo?/ /gipa?-lo?/

igopasi? jopi? igopalo? gipato?

‘into the land’ ‘into the house’ ‘on the ground’ ‘at the door’

In Blevins’s feature-geometric analysis, phonetically velar /l/ is phonologically Coronal. A default rule adds secondary Dorsality. Most laterals are voiced approximants, but 9.5 percent of languages (Maddieson 2008) have lateral obstruents, such as the voiced fricative [\], and some languages such as Zulu contrast the two. Not all laterals are voiced: there are truly voiceless lateral approximants such as Toda [l] or the laterals following voiceless stops in English words like please, and there are many voiceless lateral fricatives like the Welsh and Zulu [x]. (4)

Zulu Toda Welsh

lálà ‘lie down’ kal ‘bead’ lo(n ‘road’

\álà kal xon

‘play’ ‘study’ ‘glad’

xàxá

‘cut off’

Ladefoged and Maddieson (1996: 199) point out that it is not always easy to tell voiceless lateral fricatives from voiceless lateral approximants. Acoustically, the approximants have “a lower amplitude of noise, a greater tendency to anticipate the voicing of a following vowel,” and lower frequency energy than the fricatives. Lateral affricates [tx] and [d\] are quite common, as are clicks with a lateral release, shown by || (chapter 18: the representation of clicks): (5)

Tlingit Zulu

xaa k||ók||à

‘melt’ ‘narrate’

txaa g||álà

‘be big’ ‘stride’

d\aa

‘settle (of sediment)’

Like other clicks, lateral clicks behave phonologically like obstruents, and there do not seem to be any cases where they pattern with lateral approximants. Finally, pre-stopped laterals and pre-lateralized stops are reported in Icelandic and West Norwegian, and also among Australian languages: see Evans (1995) for details. In sum, at least the following possibilities must be accounted for, with sample languages included: (6) approximant fricative affricate click

voiced

voiceless

l \ d\ g||

l x tx k||

Zulu Zulu Tlingit Zulu

Toda Zulu Tlingit Zulu

I shall assume that the approximants are sonorants (even if voiceless), and all the others are obstruents. This distinction is based on their acoustics – lateral approximants frequently interact with adjacent vowels, taking on the coloring of both following and preceding vowels – and their phonotactic behavior (see below), particularly their ability to become syllabic and occupy nuclear position. The affricates and clicks I take to be lateral stops, and I shall not have room to discuss how they are distinguished from each other featurally.

Lateral Consonants

3

6

Phonotactics

Lateral approximants can occur in almost any syllable position. In English, for example, they can be onsets, as in lick, the second part of onset clusters, as in click, codas, as in kill, and even nuclei, as in the last syllable of tickle. They may show allophonic variants depending on syllable position, as in US English light [l] in leaf and dark [l] in feel. It is not entirely clear whether this is categorical, as Scobbie and Wrench (2003) argue, or whether it involves two ends of a continuum, as Sproat and Fujimura (1993) propose. In some languages, flapped versions of laterals are found intervocalically, and in these very short closures it is not always clear whether the release is or is not still lateral. For example, some Japanese speakers use approximant [l] wordinitially but flapped [7] intervocalically (Ladefoged and Maddieson 1996). The phonotactic behavior of laterals shows clearly that along with the rhotics they are the most sonorous type of consonant, closest to the vowels. Not only may they act as nuclei in many languages, but they are typically the innermost member of a consonant cluster, so that we find onsets like [gl], but not *[lg]. Many languages require members of a cluster to maintain a minimum sonority distance from each other. Based on this fact, laterals seem to be more sonorous than nasals, since a language may allow Cl onsets but not Cn (e.g. Spanish), but the reverse is not true. They appear to be less sonorous than rhotics, although the evidence here is not quite so striking. Rhotic dialects of English, for example, allow [rl.] in codas in words like Carl, but not *[lr.]. See Botma et al. (2008) for an interesting discussion of coda liquids within Government Phonology. One particular aspect of their clustering behavior has received much attention. In many languages, [bl] and [gl] are found, but [dl] is not permitted. Flemming (2002) and Bradley (2006) point out that the transitions for the [d] release before [l] are acoustically shifted closer to those for the [g] release, explaining why the contrast is not very robust. Indeed, children find [dl] and [gl] hard to distinguish (Macken 1980; Hale and Reiss 2008; Smith 2010), and as a result may produce puddle as [pZgHl] but pickle as [p/tHl]. This line of argument explains why [dl] and [gl] are not in contrast, but not why [gl] is preferred cross-linguistically to [dl]. A common suggestion is that [dl] is banned by the OCP (Obligatory Contour Principle), since both /d/ and /l/ are coronal. However Bradley (2006) claims that there are languages, such as the Phúhòa dialect of Katu, which allow [dl] but not [gl].

4

Phonological behavior

We now come to the most controversial aspect of laterals: how are they represented phonologically, or more precisely what distinctive features do they carry? Almost every aspect of their featural representation is under dispute, and here I shall try to summarize some of the major areas of disagreement. For reasons of space I shall limit myself to feature theories that are, in some broad sense, descendants of the Chomsky and Halle SPE system. On the representation of laterals in element/dependency theory, see Anderson and Ewen (1987) and Botma (2004: ch. 2). See also chapter 17: distinctive features and chapter 27: the organization of features for more general discussion of feature theory.

7

Moira Yip

4.1 Sonorancy and voicing There is general agreement that the vast majority of laterals are sonorants. But how sonorant are they? Based on widely accepted phonological evidence, Parker (2008) posits the following sonority hierarchy for sonorant consonants (see also chapter 49: sonority): (7) Relative sonority of sonorant consonants glides > flaps > laterals > trills > nasals Using data from Quechua (Peruvian) Spanish and US English, he then sets out to show that this ranking has a measurable phonetic correlate, namely acoustic intensity. Although he finds a very good statistical match for his overall notion, [l] in onset position frequently comes out as louder (and thus more sonorous) than the glides, contra expectations. Parker suggests that this is because the glides may become more “obstruent-like” in onset position, especially in Spanish. According to Harris and Kaisse (1999), this is rather pervasive. In Argentinian Spanish it is carried to an extreme, as shown by alternations such as le[j] ‘law’ vs. le.[Ú]es ‘laws’. But even in Castilian, the onset glide [j] frequently hardens to the non-strident palatal fricative [ ù] or even to its non-continuant counterpart. Once we look outside the prototypical approximant laterals, we quickly encounter laterals that are not sonorant at all. Examples of obstruent laterals include not only the obvious fricatives, affricates, and clicks, but also languages in which [l] patterns with the voiced obstruents, such as Southern Min, which has [l] instead of [d]. Modern Southern Min completely lacks [d]. Not only is [l] the modern reflex of historical *d, but underlying /p t k/ voice to [b l g] foot-internally (Hsu 1996), and /b l g/ nasalize to [m n I] before nasal vowels. In some Bantu languages, like Ikalanga, historical *d has become /l/, and synchronically under velarization /l/ becomes the stop [gw], suggesting that it may still be an obstruent. (A reviewer correctly notes that in both these cases there could be intermediate steps involved, and /l/ may not be underlyingly obstruent.) There are even a few languages in which all the laterals appear to be obstruents: Tlingit for example has two lateral fricatives and three lateral affricates, but no lateral approximants. Finally, turning to voicing, like other sonorants laterals are typically voiced, but voiceless versions are found, for example in Toda. The theoretical implications of the fact that most (but not all) laterals are voiced sonorants will be discussed in §7.

4.2 Continuants or not? There has been little agreement as to whether ordinary approximant laterals behave like stops or like continuants, and thus whether they are [±continuant]. Alternations with stops are common, but alternations with fricatives do not seem to be found. Nonetheless, Holt (2002) argues that they are both [+continuant] and [−continuant] and Mielke’s (2005) survey, as his title “Ambivalence and ambiguity in laterals and nasals” makes clear, finds an almost even 50/50 patterning as [±continuant]. He attributes this to their ambiguous phonetic cues, and suggests that it is precisely phonetically ambiguous segments that are likely to behave ambivalently in the phonology. I shall return to this issue in §9.

Lateral Consonants

8

Mielke gives these examples of each pattern. In Basque, laterals (but not rhotics) pattern with nasals in assimilating in place to a following consonant. The class of segments that undergo the rule is [+sonorant, −continuant]. Mielke points out that there is no way to define this class other than by using [−continuant] to include /l/ and the nasals, but not /r/. (8)

/l/ patterning with non-continuants in Basque (Hualde 1991: 96) egu[m] berri egu[|] denak egu[J] ttiki gu[I] gorri ata[*] denak ata[O] ttiki

‘new day’ ‘every day’ ‘small day’ ‘red day’ ‘every section’ ‘small section’

(I will suggest an alternative explanation for the Basque case in §8.3.) Conversely, in Finnish stems may end in the five coronals /t s n r l/. Before /n/-initial suffixes, [−continuant] /t/ undergoes total assimilation, but [+continuant] /s r l/ trigger total assimilation. (9)

/l/ patterning with continuants in Finnish (Sulkala and Karjalainen 1992: 387– 388) a. b.

/avat/ /pur/ /nous/ /tul/

active potential [avannut] [purrut] [noussut] [tullut]

2nd part active [avannee] [purree] [noussee] [tullee]

‘open’ ‘bite’ ‘rise’ ‘come’

Mielke quotes Kaisse (2000) on the issue. She points out that the status of laterals hinges on whether [−continuant] is defined in terms of complete occlusion in the oral tract (“vowel tract” in SPE; Chomsky and Halle 1968: 318) or complete occlusion in the mid-sagittal region of the oral tract (see also chapter 13: the stricture features). Laterals have complete occlusion only in the mid-sagittal region, not elsewhere, so they qualify as [−continuant] only under the latter definition. Mielke suggests that this makes laterals phonetically ambiguous, and that the boundary for the natural classes of [+continuant] and [−continuant] may vary cross-linguistically, placing laterals in different classes in different languages. The implication is that feature values may not be universally fixed for such segments, but may “emerge” on the basis of observable phonetic properties. One might also note that other rationales for cross-linguistic differences in feature specifications have been advanced, particularly the system of contrasts in the language in question. See Morén (2006) for an account on these lines for Serbian laterals.

5

Alternations involving laterals

When laterals alternate with other segments, it is usually with ones that are minimally different either articulatorily, acoustically, or both. Since most laterals are coronals, a small change in the type of closure so that it is complete will

9

Moira Yip

produce a coronal stop. If the lateral finds itself subject to nasal spreading, so that the velum is lowered, a nasal sonorant is the likely result. If the tongue shape is inverted, so that the closure is made with the sides but not the midline of the tongue, then a rhotic is formed. And if the closure is eliminated, but the tongue body remains high, a high vocoid results. Some of these changes also produce a sound that is acoustically still quite similar to the lateral, meaning that the change may have originally been driven in whole or in part by a misperception of the signal. All of these are common changes, and are discussed below.

5.1 Stops In some languages, /l/ alternates with stops; more specifically, it appears to replace the voiced coronal stop /d/ in some contexts. In Palenquero Spanish (Piñeros 2003), /d/ is in free variation with [l] in some words: (10)

/{e{o/ → [’le.lo] ~ [’{e.Ïo] ‘finger’

In some Bantu languages, like Ikalanga, historical *d has become /l/, but under velarization /l/ becomes the stop [gw], suggesting that it may still be an obstruent. Conversely, in Southern Min (which has /l/ instead of /d/ in its phoneme inventory), if syllable-final /t/ ends up intervocalically (especially before an unstressed vowel), it voices and becomes not [d] but [l] (although descriptions vary, and as it is a brief flap or tap, it is not entirely clear how lateral its articulation is; see Hsu 1996).

5.2 Nasals Historically, Cantonese had two distinct phonemes /l/ and /n/. Both could occur syllable-initially, as in the contrast between [lei] ‘reason’ and [nei] ‘you’. Only /n/ could occur syllable-finally. However, in the last 50 years a gradual sound change has been taking place, and is now nearly complete for younger speakers (Bauer and Benedict 1997). Initial /n/ is being replaced by /l/, so that ‘you’ and ‘reason’ are now both pronounced as [lei]. As a result, [l] and [n] can be treated as allophones of a single phoneme, with [l] as the syllable-initial variant and [n] as the syllable-final one. However, actual alternations do not exist, because the language has essentially no resyllabification. In the Min dialects of Chinese (including Southern Min, mentioned earlier), voiced stops and nasals are in complementary distribution, with [b g] occurring only before oral vowels and [m I] only before nasal vowels. These data are from the Chaoyang dialect: (11)

bi(?11 ‘hide’

mg(53 ‘fast’

The language has no alveolar voiced [d], but the reflex of historical *d is [l] before oral vowels and [n] before nasal vowels. As a result [l] and [n] are in complementary distribution, and this is productive. In Chaoyang onomatopoeia, words are reduplicated and one onset is replaced by [l]. However, if the vowel is nasal it is replaced instead by [n]. See Yip (2001) for details.

Lateral Consonants (12)

tsi tsiau liau jãh njãh

10

‘sound of a crowd talking’ ‘sound of door opening’

5.3 Rhotics Laterals and rhotics have a very close affinity to each other and, as we mentioned earlier, some languages have a single liquid that varies between the two. (On which feature distinguishes the two, see §4.2 on [continuant] and §6 on [lateral].) In Korean, [l] occurs only in codas, and [r] only in onsets. Most authors assume they are reflexes of a single phoneme, here given as /L/. These examples are taken from Iverson and Sohn (1994). (13)

a. b.

/paL-cHn/ /mi-taL/ /saLam/ /Ladio/

[palc’Hn] [midal] [saram] [radio]

‘development’ ‘shortage’ ‘person’ ‘radio’

A more unusual alternation is found in Yanggu Chinese (Yip 1992; Yu 2007). Yanggu has a diminutive suffix which usually surfaces as the rhotic [P], as in (14a), matching its historical source as a rhotic suffix. However, if the word begins with a dental or alveolar, it also adds an infix to form an onset cluster, and this infix surfaces as a retroflex lateral [Î], as in (14b). The intervening vowels are also rhotaicized (not shown). (14) a. b.

root

diminutive

pu ke tao

puP keP tÎaoP

‘cloth’ ‘cover’ ‘knife’

Onset clusters are not found elsewhere in Yanggu: the lateral is found only as part of the diminutive formation process. Yu argues that it arises from a misperception of the drastic transition from an anterior consonant to a retroflex vowel: see Yu (2007: 146) for details.

5.4 Glides or vowels One of the most widespread of alternations is l-vocalization. It is found in many dialects of English, as when milk becomes [miwk] or table [te(bu], and alternations are present in words like feel [fiw] vs. feeling [fi(l/I]. Although commonly transcribed as [w] in coda position and [u] in nucleus position, lip rounding is often absent. Scobbie and Wrench (2003) show clearly that l-vocalization is categorical for many English speakers. Johnson and Britain (2007) draw attention to the articulatory basis of this change, and propose a markedness account in OT. Vocalization happens in coda position, which means its source is always dark l (not clear l). They extensively document the historical and geographical distribution of dark l in English, and show that l-vocalization correlates with the presence of dark l. Recall that dark l has two gestures, with Dorsal coming first in coda position. They suggest that vocalization is driven by markedness reduction,

11

Moira Yip

leaving only the more vocalic Dorsal gesture (see chapter 75: consonant– vowel place feature interactions). A rather different but equally well-known case comes from the Cibaeño dialect of Spanish, in which both /l/and /r/ become the palatal glide [j] in coda position (Guitart 1985; Harris 1985; Alba 1988). (15)

celda cerda

[sejda] [sejda]

‘cell’ ‘bristle’

Since the gesture that is preserved here is the Coronal one, Johnson and Britain’s approach cannot deal with these facts unmodified. However, there is evidence that, unlike English coda /l/, Spanish coda [l] is clear, not dark, with a smaller Dorsal gesture. This might then explain why /l/ vocalizes as [j], not [w], although the picture is less clear if one studies a range of Spanish and Portuguese dialects. See Quilis et al. (1979) and Recasens and Espinosa (2005) for details. Many dialects of Spanish, especially in Latin America, have replaced the palatal lateral [O] by the glide [j]. This sound change is known as yeismo. (In Buenos Aires Porteño Spanish, this change has gone one step further in a change called zheismo, with [j] spirantizing to [Ú]. See Harris and Kaisse (1999) for details.) (16)

llorar ella

[O]orar e[O]a

[j]orar e[j]a

‘to cry’ ‘she’

In Serbian, rather unusually, /l/ vocalizes to [D]. See Morén (2006) for an interesting account within a Parallel Structures Model. Historically, vocalization of laterals is also common. In Germanic, compare English old to Dutch oud. In Polish, dark [û] changed to [w] everywhere, even in onsets. In Romance, Latin *l has developed variously to [w/u], e.g. in French (caldus to chaud), or to [j/i] (in both onset and coda), as found in Italian flos to fiore and Portuguese multus to muito. The inverse of vocalization of laterals, the lateralization of glides, seems to be much rarer (chapter 15: glides). Li (1974) documents a case in which *j > [l] in some Formosan languages, and a voiced coronal fricative [z Ï] in others. Interestingly, in no case does *w > [l], even though in l-vocalization /l/ becomes [w] more often than [j]. This suggests that there must be a coronal gesture already present (as there is for [j]) for the creation of a novel lateral.

6

The feature [lateral]: Is it necessary?

Phonologists routinely use a feature [lateral] to distinguish /l/ from /r/, but some linguists (Spencer 1984, Brown 1995, and most recently Walsh Dickey 1997) have argued that it can be dispensed with. If a language has [l] but no [r], one might define [l] by the features [+consonant, +sonorant, −nasal], and [lateral] would be redundant. However, if [l] contrasts with [r], as it does in many languages, this will not suffice. They could perhaps be distinguished by the feature [continuant], with /r/ as [+continuant] and /l/ as [−continuant], but this is not without problems (see §4.2, and also van der Weijer 1995). Walsh Dickey (1997: 55) distinguishes them by means of Place features, with /l/ having complex Corono-Dorsal place and /r/ non-primary Laminal. Dissimilation of /l/ to [r] is loss of a secondary

Lateral Consonants

12

Dorsal articulation. The lateral articulation of [l] is, for her, “a necessary phonetic consequence of a phonological Corono-Dorsal complex place structure.” It is worth having a slightly closer look at her arguments and her proposal. Walsh Dickey points out, correctly, that the strongest arguments for a feature come from its role in defining natural classes. The many languages with only one lateral can of course never offer this type of evidence. She lists three types of potentially significant evidence from languages with more than one lateral: (i) co-occurrence restrictions on different types of laterals; (ii) positional restrictions which cover all types of laterals in a language; and (iii) phonological processes which need to refer to all laterals, both sonorant and obstruent. She concludes that no such evidence exists, and I have also not encountered any convincing cases. However, she pays a high price for the absence of [lateral]. In particular, she has to greatly complicate the internal structure of the coronal node by including Laminal (which in turn may or may not be [dental]), Apical (which in turn may or may not be [back]), and secondary Dorsal (which may have yet another Dorsal specification below it, to account for velarized “dark” laterals). Finally, any other sound which might have been thought to involve secondary Dorsal articulations, such as velarized coronal consonants, would require some extra specification if they did not have a lateral release. I therefore tentatively conclude that the feature [lateral] is still useful, and probably necessary. Positive evidence for the feature [lateral] comes from its active role in the phonology of many languages, despite Walsh Dickey’s claims to the contrary. In Eastern Catalan (and Sanskrit), for example, [lateral] spreads onto nasals to create a lateral nasal: /nl/ → [ml] in /son les tres/ → [somles tres] (Mascaró 1976). There are wellknown phonological processes that involve only [l] and [r], and in which they either dissimilate, as in Latin, where the suffix /-alis/ surfaces as [-aris] after a lateral root: nav-alis vs. milit-aris (Steriade 1987), or assimilate, as in Sundanese, where the infix /-ar-/ surfaces as [-al] after a preceding /l/: [k-ar-usut] vs. [l-al-Hga] (see Cohn 1992 for details). Several of these processes are long distance, and can cross over other Coronals, making a Place feature account tricky. I conclude that the feature [lateral] cannot be dispensed with. I should note that for the remainder of this chapter I shall treat [lateral] as a privative feature, but the results would not be materially affected if it were to turn out to be binary, as Steriade (1987) argues.

7

Feature geometry and the feature [lateral]: Two competing models

It has been proposed that distinctive features are related to each other by a hierarchical feature geometry (Sagey 1986, Clements and Hume 1995, and many others; see also chapter 27: the organization of features). For example, the features Labial, Coronal, and Dorsal are dominated by a Place node. In such a model, we must then ask where the feature [lateral] is located. Early proposals spent little time worrying about the placement of [lateral], and tended to put it directly under the root node. Clements and Hume (1995: 293) opt for this, but admit that its position is open to dispute. Subsequently, two competing detailed proposals for the placement of lateral have been put forward, and are shown below: it might

13

Moira Yip

be under the Place feature Coronal, or under a node related to its sonorant nature, called Sonorant Voicing (SV), which is argued to be responsible for voicing in sonorants but not (most) obstruents. (17) a.

Under Coronal (Blevins 1994)

b. SV model (Rice and Avery 1991)

Place Labial

Coronal

Sonorant Voicing Dorsal

[lateral]

[nasal]

[lateral] These proposals were motivated by two observations about laterals: they are normally Coronal (hence the Coronal proposal), and they are normally voiced sonorants (hence the SV proposal). But problems arise because, as we have seen, these are tendencies rather than universals. The problems become even more vexing when we look at other predictions of these two proposals. In particular, [lateral] requires the presence of its superordinate node, and anything which affects that node (such as spreading it, delinking it, or deleting it) will also affect [lateral]. The trouble is, the evidence is contradictory.

7.1 Coronal node model Let us look at these predictions in more detail, starting with the predictions of the Coronal node theory (see Table 31.1).

7.1.1 Supporting evidence for the Coronal model The evidence in support of this theory is well known. Additional data for this and subsequent sections can be found in Yip (2004, 2005). First, as we have already Table 31.1 Predictions of the Coronal node theory Prediction

Under Coronal? FOR

AGAINST

All laterals should be Coronal

Laterals are usually Coronal: many languages

Placeless laterals: Cambodia Velar laterals: Yagaria

Place spreading should spread [lateral]

Yes: Selayarese

No: Chukchee

When [lateral] spreads, it should seek Coronal targets

Yes: Teralfene Flemish, Yanggu Chinese

Laterals skipped by harmony that targets Coronals: Tahltan

Place spreading onto laterals should remove [lateral]

Yes: Moroccan Arabic, Cuban Spanish

No: English, Basque

Loss of Place contrasts should eliminate [lateral]

No cases known

No: Caribbean Spanish

Lateral Consonants

14

seen, most laterals are coronal. Secondly, place spreading in languages like Selayarese also spreads laterality, as can be seen below: (18)

Selayarese place assimilation (Mithun and Basri 1986) anna[I] anna[mp]oke anna[Jj]arang anna[nt]au anna[nr]upa anna[ll]oka

‘six’ ‘six spears’ ‘six horses’ ‘six persons’ ‘six kinds’ ‘six bananas’

Thirdly, in some languages when lateral spreads it seeks out specifically Coronal targets. Yanggu Chinese has a very unusual affix (Yip 1992). It may be rhotic or lateral, or both, surfacing variously as [r l È Î], where the last two are retroflexed. All that concerns us here is that the lateral variants only surface in words with surface coronals, and they are attracted to the rightmost coronal. This mobility suggests that [lateral] is a floating feature that docks onto coronals: (19)

a.

No coronals

b.

Initial coronal

c.

Initial and final coronals

xou pe tu na tsh uHn

xour per tlur nlar tsh uHÎ

‘monkey’ ‘card, board’ ‘rabbit’ ‘to press’ ‘village’

Fourthly, if laterals are the targets of rules that spread the Place node, their laterality should be lost, as is the case in educated Havana Spanish (Guitart 1976; Harris 1985; Padgett 1991: 228). Liquids assimilate in Place, Manner, and nasality to the following consonant. Before stops, they always remain voiced, but before voiceless fricatives they devoice. In all cases they lose their laterality: (20)

albañil tal droga pulga el pobre el tres tal mata el fino

a[bb]añil ta[dd]droga pu[gg]a e[bp]obre e[dt]res ta[mm]ata e[ff]ino

‘mason’ ‘such a drug’ ‘flea’ ‘the poor man’ ‘the three’ ‘such a shrub’ ‘the refined one’

7.1.2 Counterevidence to the Coronal model It is also clear that there is considerable counterevidence to this proposal from cases like those in the right-hand column of Table 31.1. Firstly, not all laterals are Coronal. We have already seen that velar laterals exist. An example of phonologically placeless laterals comes from Cambodian (Nacaskul 1978). There are co-occurrence restrictions on identical Place features, but they do not treat /l r/ as Coronal, even though Place restrictions cross-cut obstruents and sonorants, stops and fricatives, nasals and glides. Instead, /l r/ behave like [h ?] in co-occurring freely with all other sounds. Somewhat similar facts hold in Javanese, where /l r/ also fail to trigger nasal assimilation or coalescence, unlike all other obstruents, nasals,

15

Moira Yip

and glides. See Yip (1989) for details. As a final note, these laterals are of course articulated with the tip or blade of the tongue, so this would have to be viewed as the phonetic implementation of a phonologically placeless lateral. In a model committed to full output specification, like classic OT, this poses a challenge. Secondly, there are languages like Chukchee where laterals as triggers of Place assimilation spread only Coronal, not [lateral], as can be seen by the last example below: (21)

Chukchee (Clements and Hume 1995: 270) tHI-Hû?-Hn (/tEI-/) tan-tsai tam-pera-k tan-ran tam-vairgin ten-jHûqHt-Hk tam-wa:Hr:-Hn but ten-leut *tel-leut

‘good’ ‘good tea’ ‘to look good’ ‘good house’ ‘good being’ ‘to sleep well’ ‘good life’ ‘good head’

Thirdly, there are languages where harmonies that target all other Coronals skip the lateral series, as if they did not have Coronal nodes at all. Tahltan (Shaw 1991) has five series of coronals, shown here by their voiced affricate members: /d dl dÏ dz –/. At each place of articulation, there are voiced/voiceless and glottalized, as in [dz ts ts’]. At the last four places, there also voiced and voiceless fricatives, as in [z s]. Within a word, the last three series harmonize with the rightmost participant. In the following data, the first dual subject prefix /Ïi(d)/ stays [.] before non-coronals or the /d/ and /./ series, but becomes [s] before the /dz/ series, and /œ/ before the /–/ series: (22)

.i(t.ædi nisi(t’a(ts uœi–e

‘we ate it’ ‘we got up’ ‘we are called’

The lateral /dl/ series are not triggers or targets, and are transparent: /na.iba(tx/ ‘we hung it’. Fourthly, in Basque, as we have already seen in §4.2, there is general Place spreading onto sonorants, but it does not remove laterality, as shown in (23) (some of these data appeared earlier in (8)). Nasals assimilate to all places of articulation, as can been seen in the left-hand column below. Laterals also assimilate before coronals, but are unchanged before other places (Hualde 1991). (23)

egu[n]a egu[m] berri egu[,] fresku egu[|{] enak egu[J] ttiki egu[I] gorri

‘the day’ ‘new day’ ‘cool day’ ‘every day’ ‘small day’ ‘red day’

ata[l]a ata[l] berri ata[l] fresku ata[*{] enak ata[O] ttiki ata[l] gorri

‘the section’ ‘new section’ ‘cool section’ ‘every section’ ‘small section’ ‘red section’

Lateral Consonants

16

Similar facts hold in Tamil (Beckman 1998), and in English: we[*.] wealth, but whe[lk] whelk, although the English case may be purely phonetic, since it is not structurepreserving. The interesting fact is that in Basque and all these other languages laterals do not lose their laterality as the targets of Place assimilation, contra the predictions of the Coronal model. Finally, there are languages in which Place contrasts are lost – a phenomenon described by Trigo (1988) and others as the delinking of the Place node – but laterality survives. In Caribbean Spanish (Trigo 1988: 71) place features are neutralized in codas: /d/ deletes, /s/ becomes [h], and all nasals become velar; /r/ and /l/ are unchanged. (24)

a. b. c.

/ßerdad/ /ines/ /album/ /tren/ /desdeJ/ d. /tonel/ /par/

→ → → → → → →

[ßerÏa] [ineh] [albuI] [treI] [desdeI] [tonel] [par]

‘truth’ ‘Ines’ ‘album’ ‘train’ ‘disdain’ ‘barrel’ ‘pair’

(optional)

The data in this section are problematic for the Coronal model.

7.2 The sonorant voicing (SV) model What about the sonorant voicing (SV) theory? Rice and Avery (1991) base their claims on three main arguments. Firstly, the SV node is the target node when lateral spreads to other sonorants. Secondly, a process of de-sonorantization can be seen as delinking of the SV node. Thirdly, rules which spread both nasality and laterality can be viewed as spreading the SV node (see Table 31.2).

Table 31.2 Predictions of the SV model Prediction

Under sonorant voicing? FOR

AGAINST

All laterals should be voiced sonorants

Laterals are usually voiced sonorants: many languages

Voiceless laterals: Tahltan Obstruent laterals: Min, Bantu Affricate laterals Tahltan, Zulu

SV spreading should spread [lateral]

Yes: Sanskrit

No: Polish

When [lateral] spreads, it should seek SV targets

Yes: Toba Batak

Laterals skipped by harmony that targets sonorants: no cases known

SV spreading onto laterals should remove [lateral]

Yes: Itsekiri

No: English

Loss of SV removes [lateral]

Yes: Yagaria

No: Koyukon, Angas

17

Moira Yip

7.2.1 Supporting evidence for the SV model Apart from the observation that most laterals are voiced sonorants, there are four other types of evidence in support of the proposal. Firstly, SV spreading sometimes spreads [lateral], as in Sanskrit. Before laterals, all obstruents voice, and if they are coronal they also lateralize (Whitney 1889: 54; no examples of non-coronals are given): (25)

tat labhate trin lokan

→ →

tal labhate trim lokan

Secondly, [lateral] spreading sometimes seeks SV targets to attach to, as in Toba Batak (Hayes 1986). Coronal sonorants assimilate to a following liquid: (26)

Spreading

No change

nr → rr nl → ll rl → ll

ln rn lr mr ml Il Ir rr ll nn

If what spreads is the SV node, and if [lateral] is its dependent, then laterality will be carried along too. Thirdly, when SV spreads onto a lateral target, replacing the original SV specification, the original lateral specification may be lost. The following facts from Itsekiri (Nigeria; Piggott 1991, cited in Brown 1995: 64) are often cited, and very similar facts hold in Southern Min Chinese and in Yoruba. (27) lã → nã

‘ask the price of’

This type of nasal harmony can be seen as spreading the superordinate SV node, carrying nasality with it, and removing the original SV node, laterality and all. Fourthly, if an SV node delinks, lateral may be removed too, as in Yagaria. Here a coalescence process removes voicing and converts a sonorant [l] to an obstruent [t], and in the process laterality also goes. (28)

gipa?-lo? → gipato? ‘at the door’

For further data on Yagaria, see §2, where these data are discussed in a different context.

7.2.2 Counterevidence to the SV model There are various types of counterevidence to the SV model, as listed in the right-hand column of Table 31.2. The first problem is the existence of laterals that are not voiced sonorants, being either voiceless, obstruent or both. But there are other problems. Firstly, SV spreading does not always spread [lateral]. The following data show a post-lexical process of voicing assimilation in Krakow and Posnan Polish (Dorota Glowacka, personal communication; MadeÓska and Witaszek-Samborska 1998):

Lateral Consonants (29)

brat syn

→ →

bra[d] Doroty/Natalji/Iwony/Luizy sy[n] Luizy

18

‘brother of X’ ‘son of Luiza’

Similar facts hold in all dialects between verbal prefixes and roots: (30)

s-kDJŒqtu ‘to end’

z-bitu ‘to break’

z-liŒqtu ‘to count’

In the SV model, voicing in sonorants is represented by the presence of an SV node, and so this must be the active node that spreads. If [lateral] is under this node, it too should spread, but it does not, contra the predictions of the SV model. Secondly, SV spreading into lateral targets sometimes leaves the laterality intact. This is the case in English, where liquids after voiceless aspirated stops become voiceless: (31)

[bl]eak [br]eam

[pl]ease [pã]een

[gl]eam [gr]een

[kl]ean [kã]eam

If sonorant voicing is denoted by the presence of an SV node, the devoicing would presumably mean that the SV node had been delinked, and one would then expect loss of [lateral] as well, but no such thing happens. Thirdly, if the SV node of a lateral delinks for other reasons, rendering it voiceless, laterality should also be lost, but this is not the case in the Athapaskan language Koyukon (Rice 1994), which devoices syllable-final sonorants and continuants, including /l/. For the lateral, the result is a voiceless lateral fricative [x]. Final stops are plain voiceless unaspirated. (32)

[nH:ælH] ‘your (sg) trap’ [sH?D:H’] ‘my snowshoes’ [nizuni] ‘that which is good’

[xæx] [?Dx] [nizun]

‘trap’ ‘snowshoes’ ‘it is good’

Under the SV hypothesis, where [lateral] is under SV and devoicing of sonorants means removal of the SV node, laterality should also disappear, but it does not, contra the predictions of the SV model. In the case of [lateral], then, the need created by theories of universal feature geometry to commit to a single location for the feature in all languages creates problems. Luckily, there are alternatives to these two proposals. Hegarty (1989) and Bao (1992) argue that [lateral] is simply a dependent of the Root node. Yip (2004: 5) goes further, and agrees with Padgett (1995, 2002) that (at least with respect to the behavior of [lateral]) features can be treated as an unstructured set of which [lateral] is a member, and that feature geometry as such is redundant. The next section lays out this proposal.

8

A feature co-occurrence constraint approach

8.1 Inventories Suppose we capture the fact that the least marked lateral is a coronal sonorant by means of two universally fixed hierarchies of feature co-occurrence markedness constraints.

19 (33)

Moira Yip a. b.

*LateralObstruent >> *LateralSonorant *LateralLabial >> *LateralDorsal >> *LateralCoronal

The idea is that these constraints are violated if a segment bears both features, at least if both are primary articulations. For example, the intention is that *LateralLabial bans bilabial or labio-dental segments with a lateral release. If the labiality is secondary, as in [lw], it may be acceptable. Both the fixed rankings are phonetically grounded. Lateral release means the airflow is not easily obstructed, so lateral obstruents are more marked than lateral sonorants (33a), and lateral release is easiest if only the tip of the tongue is used to make closure, so coronals are the least marked laterals (33b). Placing the faithfulness constraints at different points in these hierarchies gives us a typology of segmental inventories like those below: (34)

Typology of lateral place of articulation a.

Either no laterals (Maori), or placeless ones (Cambodian) b. *LatLab >> *LatDors >> Faith >> *LatCor Common type, with Coronal laterals only (English) c. *LatLab >> Faith >> *LatDors >> *LatCor New Guinea type, with velar and coronal laterals (Mid-Waghi), or perhaps palatal laterals d. Faith >> *LatLab >> *LatDors >> *LatCor Laterals at all POAs (unattested)

(35)

*LatLab >> *LatDors >> *LatCor >> Faith

Typology of lateral sonorants and obstruents a. b.

*LatObs >> *LatSon >> Faith *LatObs >> Faith >> *LatSon

c.

Faith >> *LatObs >> *LatSon

Languages with no laterals Common language type, with sonorant laterals Languages with both obstruent and sonorant laterals

In this way, we can correctly describe the inventories of laterals found in the world’s languages, with the possible exception of Tlingit, which is reported to have lateral obstruents but no lateral sonorants. The complete absence of labial laterals also remains unexplained.

8.2 The spreading of [lateral] How can these mini-grammars characterize the behavior of laterals shown in §7? First, let us consider the behavior of [lateral] in spreading processes, such as in Place assimilation or perhaps the spreading of SV under pressure from the Syllable Contact Law. Let us assume that assimilation involves a violation of the Ident family of faithfulness constraints, such as Ident-Place and Ident-Son,

Lateral Consonants

20

under pressure from higher-ranked constraints such as Share-F and Syllable Contact. I start with cases in which [lateral] is (potentially) spread by the process in question. Any assimilation process that creates the ordinary sonorant Coronal lateral [l] from an underlying non-coronal or non-sonorant will violate at least one of Ident-Place and Ident-Son (and of course also Ident-Lat). The ranking of these constraints with respect to the constraints causing assimilation, here abbreviated as Share-F, will determine which segments may undergo the process. If IdentSon >> Share-F, targets must be already sonorant. If IdentPlace >> Share-F, targets must be already Coronal. If the output is always Coronal and sonorant, *LatObs and *LatDors are always high-ranked, and *LatCor and *LatSon are always low-ranked. The following typology results: (36)

a.

Target must be sonorant: *LatObs, IDENT-SON >> SHARE-F >> *LatSon a’. Target need not be sonorant, but output will be: *LatObs >> SHARE-F >> IDENT-SON, *LatSon b. Target must be Coronal: *LatDors, IDENT-PLACE >> SHARE-F >> *LatCor b’. Target need not be Coronal, but output will be: *LatDors >> SHARE-F >> IDENT-PLACE, *LatCor

By combining one of the sonorancy rankings with one of the Place rankings, we get the following mini-grammars: (37) a & b.

Target must be sonorant and Coronal: Toba Batak *LatObs, *LatDors, IDENT-PLACE, IDENT-SON >> SHARE-F a & b’. Target must be sonorant, but need not be Coronal: Selayarese *LatObs, *LatDors, IDENT-SON >> SHARE-F >> IDENT-PLACE a’ & b. Target must be Coronal, but need not be sonorant: Sanskrit, Yanggu *LatObs, *LatDors, IDENT-PLACE >> SHARE-F >> IDENT-SON a’ & b’. Target need not be Coronal or sonorant, but output will be both: ? *LatObs, *LatDors, SHARE-F >> IDENT-SON, IDENT-PLACE

In all these cases, *LatCor and *LatSon are low-ranked, since these languages have underlying laterals, and also allow the creation of new ones from some subset of targets. The more interesting cases are those where underlying laterals survive, but new ones are not created, as in Chukchee and Polish. The central idea is that this will arise in any ranking where *LatSon and *LatCor (and of course also *LatObs and *LatDors) are ranked above Share-F, blocking laterality from surfacing at all on the target. To be more precise, the difference between the grammars of Selayarese (in which new laterals are created by Place assimilation) and Chukchee (in which they are not) lies in the relative ranking of *LatSon and Share-F, where Share-F requires adjacent consonants to have identical values for as many features as possible. It will be held in check by markedness constraints such as *LatSon, and faithfulness constraints such as Ident-Voice (in a language without voicing assimilation). In Selayarese, Share-F is ranked above *LatSon, and so it can create new lateral sonorants. In Chukchee, Share-F is ranked below *LatSon, so all the other Place

21

Moira Yip

features are shared, but not laterality, and the segment remains a nasal. For full details, see Yip (2003, 2005). We see then that variability in the spreading behavior of laterals in assimilation is handled without great difficulty by these minimally different OT grammars.

8.3 Laterals as the targets of spreading I now turn to cases where lateral is (potentially) lost on the target of assimilation, or in neutralization. Whether or not it is lost depends on the relative ranking of Share-F and Ident-Lat, as shown below: (38)

a. b.

Loss of [lateral] under (Place) assimilation: Educated Havana Spanish Share-F >> Ident-Lat, Ident-Place Retention of [lateral] under (Place) assimilation: Basque IdentLat >> Share-F >> Ident-Place

A parallel analysis can be constructed for the cases in which SV spreading does or does not obliterate laterality. The Basque case deserves a little more attention, because the Basque facts were used by Mielke to argue that laterals behave as [−continuant], since like nasals they undergo place assimilation, whereas [+continuant] [r] does not. Under the account offered in this section, however, there is another explanation available, as suggested by a reviewer. Nasals assimilate to all places, but laterals assimilate only to coronals, since *LatLab and *LatDors are high-ranked. Suppose there is also a set of constraints regulating the co-occurrence of a feature [rhotic], such that only a single rhotic exists. If these constraints and Ident[rhotic] all outrank Share-F, the rhotic will fail to assimilate, and continuancy need not be invoked. Finally, laterality may or may not be lost when Place or SV is neutralized, for example in codas. Here the grammar is modeled on the ones in (38), but with *Coda-F instead of Share-F. I have now outlined analyses that explain all the variation in the behavior of laterality, without reference to feature geometry, and using only rather simple markedness constraints on the co-occurrence of [lateral] with the various Place features, and with the feature [sonorant]. These interact with familiar faithfulness constraints and with markedness constraints that create pressure for change, such as Share-F, SyllableContact or *Coda-F. The clear preferences for coronal and sonorant laterals are captured by positing universal rankings of the relevant feature co-occurrence constraints. As always in OT, these constraints are violable, and this correctly predicts the observed cross-linguistic variation in lateral behavior that is such a problem for a fixed universal feature geometry.

9

Conclusion

The topic of this chapter has been laterals. As I hope is clear, they do not fit tidily into current phonological theories, especially when it comes to their distinctive feature make-up. Considering how common they are in language, this is something of an embarrassment. After all, they are neither unusual, prone to disappear over time, nor hard to acquire (that is, even if the child acquires them late, they get

Lateral Consonants

22

them sooner or later, which is more than can be said for [s] and [P] for some speakers of British English). It is therefore incumbent on a good theory to accommodate them. We have seen that their behavior is more variable across languages than that of most other sounds. So, for example, [t] always behaves like a voiceless coronal stop, and its distinctive features and their placement in the feature geometry are rarely if ever in dispute. So why should laterals be different, and are there other classes of sounds that exhibit comparable variability? Variable behavior might be seen whenever the features are most readily produced and perceived on a certain type of segment, but can with some effort also be produced and perceived on sounds of other types too. For example, [strident] sounds, in which the turbulence produced at the point of constriction is sufficiently strong, and/or where the ensuing airstream then hits a sharp obstacle like the teeth, is easy to produce with the tip or blade of the tongue, but hard to produce elsewhere. We derive from this a constraint hierarchy *[Labial, strident] >> *[Coronal, strident]. Languages which contrast [f] and [z], like Ewe, arguably violate the former as well as the latter. Turbulent airflow also requires a period of incomplete closure, or continuancy, so we also derive *[−cont, strident] >> *[+cont, strident]. Languages that violate the former have strident affricates, which have often been argued to be strident stops. In principle, then, the interactions of these constraints might also produce comparable variation to that we have seen with laterals. For other features, no such variation is to be expected. [anterior] and [distributed] refine the type of contact the tip or blade of the tongue makes with the roof of the mouth. As such they can only be present in Coronals, and a sound that is [Dorsal, +anterior] is phonetically uninterpretable. Mielke, by contrast, takes the variability in behavior of “ambivalent segments” like laterals to be an argument against universally defined distinctive features. Instead, he argues for “emergent distinctive features” based on phonetic similarity. Laterals, for example, may pattern with either continuants (16 languages) or noncontinuants (61 languages) because, like continuants, they do not have totally blocked airflow, but like non-continuants they do have “a blockage of airflow past the primary structure.” It is not clear how his proposals bear on those cases in §7.1 where the variability of laterals concerns their coronality rather than their continuancy, and thus where no appeal to variation in continuancy would seem to explain their behavior. It might be possible, however, to develop an extension of his approach from the starting observation that laterals are produced with both coronal and dorsal tongue gestures. Languages might therefore differ in which gesture they choose to interpret as a distinctive place feature for laterals. I leave the fleshing out of this idea for future research. A substantial part of this chapter has focused on where the feature [lateral] sits in the feature geometry, but the proposal outlined in §7 and §8 makes no use of feature geometry at all. In this respect it is entirely compatible with the proposals of Padgett (1995, 2002), in which features form classes, but in which these classes are not embodied as superordinate nodes (with the exception of a root node). Instead of asking which node dominates [lateral], one would ask which class(es) of features [lateral] belongs to. There is then no reason why it might not belong to more than one class, for example Place and SV (or Manner), or to none. The feature co-occurrence proposal goes further, however, in that as far as [lateral]

23

Moira Yip

is concerned it does not even make use of feature classes. For die-hard proponents of feature geometry, I should note that it would be possible to combine the feature co-occurrence proposal with feature-geometric representations if other phenomena make this desirable, but only if [lateral] were directly under the root node.

REFERENCES Alba, Orlando. 1988. Estudio sociolingüístico de la variación de las líquidas finales de palabra en el español cibaeño. In Robert M. Hammond & Melvyn C. Resnick (eds.) Studies in Caribbean Spanish dialectology, 1–12. Washington, DC: Georgetown University. Anderson, John M. & Colin J. Ewen. 1987. Principles of dependency phonology. Cambridge: Cambridge University Press. Bao, Zhiming. 1992. A note on [Lateral]. Unpublished ms., Ohio State University. Bauer, Robert S. & Paul K. Benedict. 1997. Modern Cantonese phonology. Berlin & New York: Mouton de Gruyter. Beckman, Jill N. 1998. Positional faithfulness. Ph.D. dissertation, University of Massachusetts, Amherst. Published 1999, New York: Garland. Blevins, Juliette. 1994. A place for lateral in the feature geometry. Journal of Linguistics 30. 301–348. Botma, Bert. 2004. Phonological aspects of nasality: An element-based dependency approach. Ph.D. dissertation, University of Amsterdam. Botma, Bert, Colin J. Ewen & Erik Jan van der Torre. 2008. The syllabic affiliation of postvocalic liquids: An onset-specifier approach. Lingua 118. 1250–1270. Bradley, Travis G. 2006. Contrast and markedness in complex onset phonotactics. Southwestern Journal of Linguistics 25. 29 –58. Brown, Cindy. 1995. The feature geometry of lateral approximants and lateral fricatives. In Harry van der Hulst & Jeroen van der Weijer (eds.) Leiden in last, 41–88. The Hague: Holland Academic Graphics. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clements, G. N. & Elizabeth Hume. 1995. The internal organization of speech sounds. In Goldsmith (1995), 245 –306. Cohn, Abigail C. 1992. The consequences of dissimilation in Sundanese. Phonology 9. 199–220. Evans, Nicholas D. 1995. Current issues in the phonology of Australian languages. In Goldsmith (1995), 723–761. Flemming, Edward. 2002. Auditory representations in phonology. London & New York: Routledge. Gick, Bryan, Fiona Campbell, Sunyoung Oh & Linda Tamburri-Watt. 2006. Toward universals in the gestural organization of syllables: A cross-linguistic study of liquids. Journal of Phonetics 34. 49–72. Goldsmith, John A. (ed.) 1995. The handbook of phonological theory. Cambridge, MA & Oxford: Blackwell. Guitart, Jorge M. 1976. Markedness and a Cuban dialect of Spanish. Washington, DC: Georgetown University Press. Guitart, Jorge M. 1985. Variable rules in Caribbean Spanish and the organization of phonology. In Frank H. Nuessel (ed.) Current issues in Hispanic phonology and morphology, 28–33. Bloomington: Indiana University Linguistics Club. Hale, Mark & Charles Reiss. 2008. The phonological enterprise. Oxford: Oxford University Press. Harris, James W. 1985. Autosegmental phonology and liquid assimilation in Havana Spanish. In Larry D. King & Catherine A. Maley (eds.) Selected papers from the 13th Linguistic Symposium on Romance Languages, 127–148. Amsterdam: John Benjamins.

Lateral Consonants

24

Harris, James W. & Ellen M. Kaisse. 1999. Palatal vowels, glides and obstruents in Argentinian Spanish. Phonology 16. 117–190. Hayes, Bruce. 1986. Assimilation as spreading in Toba Batak. Linguistic Inquiry 17. 467–499. Hegarty, Michael. 1989. An investigation of laterals and continuancy. Unpublished ms., MIT. Heid, Sebastian & Sarah Hawkins. 2000. An acoustical study of long-domain /r/ and /l/ coarticulation. Proceedings of the 5th Seminar on Speech Production: Models and Data, 77–80. Munich: Institut für Phonetik und Sprachliche Kommunikation, LudwigMaximilians-Universität. Holt, D. Eric. 2002. The articulator group and liquid geometry: Implications for Spanish phonology present and past. In Caroline R. Wiltshire & Joaquim Camps (eds.) Romance phonology and variation, 85 –99. Amsterdam & Philadelphia: John Benjamins. Hsu, Chai-Shune. 1996. Voicing underspecification in Taiwanese word-final consonants. UCLA Working Papers in Linguistics 2: Papers in Phonology 3. 1–24. Hualde, José Ignacio. 1991. Basque phonology. London & New York: Routledge. Inkelas, Sharon & Yvan Rose. 2008. Positional neutralization: A case study from child language. Language 83. 707–736. Iverson, Gregory K. & Hyang-Sook Sohn. 1994. Liquid representation in Korean. In Young-Key Kim-Renaud (ed.) Theoretical issues in Korean linguistics, 1–19. Stanford: CSLI. Iverson, Paul, Patricia K. Kuhl, Reiko Akahane-Yamada, Eugen Diesch, Yohich Tohkura, Andreas Kettermann & Claudia Siebert. 2003. A perceptual interference account of acquisition difficulties for non-native phonemes. Cognition 87. B47–B57. Johnson, Wyn & David Britain. 2007. L-vocalisation as a natural phenomenon: Explorations in sociophonology. Language Sciences 29. 294 –315. Kaisse, Ellen M. 2000. Laterals are [–continuant]. Unpublished ms., University of Washington. Ladefoged, Peter. 2006. A course in phonetics. 5th edn. Boston: Thomas Wadsworth. Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’s languages. Oxford & Malden, MA: Blackwell. Li, Paul Jen-Kuei. 1974. Alternations between semi-consonants and fricatives or liquids. Oceanic Linguistics 13. 163–186. Macken, Marlys A. 1980. The child’s lexical representation: The ‘puzzle-puddle-pickle’ evidence. Journal of Linguistics 16. 1–17. Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press. Maddieson, Ian. 1987. Linguolabials. Journal of the Acoustical Society of America 81. S65. Maddieson, Ian. 2008. Lateral consonants. In Martin Haspelmath, Matthew S. Dryer, David Gil & Bernard Comrie (eds.) World atlas of language structures online, ch. 8. Munich: Max Planck Digital Library 2008. http://wals.info/feature/description/8. MadeÓska, L. & M. Witaszek-Samborska. 1998. Zapis fonetyczny. PoznaU: Wydawnictwo Naukowe UAM. Mascaró, Joan. 1976. Catalan phonology and the phonological cycle. Ph.D. dissertation, MIT. Mielke, Jeff. 2005. Ambivalence and ambiguity in laterals and nasals. Phonology 22. 169–203. Mithun, Marianne & Hasan Basri. 1986. The phonology of Selayarese. Oceanic Linguistics 25. 210 –254. Morén, Bruce. 2006. Consonant–vowel interactions in Serbian: Features, representations and constraint interactions. Lingua 116. 1198–1244. Nacaskul, Karnchana. 1978. The syllabic and morphological structure of Cambodian words. In Philip N. Jenner (ed.) Mon-Khmer studies VII, 183–200. Honolulu: University Press of Hawaii. Narayanan, Shrikanth S., Abeer A. Alwan & Katherine Haker. 1997. Toward articulatoryacoustic models for liquid approximants based on MRI and EPG data, part I: The laterals. Journal of the Acoustical Society of America 101. 1064–1077. Ohala, John J. 1974. Phonetic explanation in phonology. Papers from the Annual Regional Meeting, Chicago Linguistic Society: Parasession on natural phonology. 251–274.

25

Moira Yip

Padgett, Jaye. 1991. Stricture in feature geometry. Ph.D. dissertation, University of Massachusetts, Amherst. Padgett, Jaye. 1995. Feature classes. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 385 –420. Amherst, MA: GLSA. Padgett, Jaye. 2002. Feature classes in phonology. Language 78. 81–110. Paradis, Carole & Jean-François Prunet (eds.) 1991. The special status of coronals: Internal and external evidence. San Diego: Academic Press. Parker, Steve. 2008. Sound level protrusions as physical correlates of sonority. Journal of Phonetics 36. 55 –90. Piggott, Glyne L. 1991. The geometry of sonorant features. Unpublished ms., McGill University. Piñeros, Carlos. 2003. Accounting for the instability of Palenquero voiced stops. Lingua 113. 1185 –1222. Quilis, A., M. Esgueva, M. L. Gutiérrez Araus & M. Canterero. 1979. Características acústicas de las consonantes laterales españolas. Lingüística Española Actual 1. 233–343. Recasens, Daniel & Aina Espinosa. 2005. Articulatory, positional and coarticulatory characteristics for clear /l/ and dark /l/: Evidence from two Catalan dialects. Journal of the International Phonetic Association 35. 1–25. Rice, Keren. 1994. Laryngeal features in Athapaskan languages. Phonology 11. 107–147. Rice, Keren & Peter Avery. 1991. On the relationship between laterality and coronality. In Paradis & Prunet (1991), 101–124. Sagey, Elizabeth. 1986. The representation of features and relations in nonlinear phonology. Ph.D. dissertation, MIT. Scobbie, James M. & Alan Wrench. 2003. An articulatory investigation of word final /l/ and /l/ sandhi in three dialects of English. In M. J. Solé, D. Recasens & J. Romero (eds.) Proceedings of the 15th International Congress of Phonetic Sciences, 1871–1874. Barcelona: Causal Productions. Shaw, Patricia A. 1991. Consonant harmony systems: The special status of coronal harmony. In Paradis & Prunet (1991), 125–157. Smith, Neil V. 2010. Acquiring phonology: A cross-generational case-study. Cambridge: Cambridge University Press. Spencer, Andrew. 1984. Eliminating the feature [lateral]. Journal of Linguistics 20. 23–43. Sproat, Richard & Osamu Fujimura. 1993. Allophonic variation in English /l/ and its implications for phonetic implementation. Journal of Phonetics 21. 291–311. Steriade, Donca. 1987. Redundant values. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23(2). 339–362. Sulkala, Helena & Merja Karjalainen. 1992. Finnish. London & New York: Routledge. Trigo, Loren. 1988. On the phonological behavior and derivation of nasal glides. Ph.D. dissertation, MIT. Walsh Dickey, Laura. 1994. Representing laterals. Papers from the Annual Meeting of the North East Linguistic Society 25. 535 –550. Walsh Dickey, Laura. 1997. The phonology of liquids. Ph.D. dissertation, University of Massachusetts, Amherst. Weijer, Jeroen van de. 1995. Continuancy in liquids and in obstruents. Lingua 96. 45 –61. Whitney, William Dwight. 1889. Sanskrit grammar, including both the classical language, and the older dialects of Veda and Brahmana. 2nd edn. Cambridge, MA: Harvard University Press. Yip, Moira. 1989. Feature geometry and cooccurrence restrictions. Phonology 6. 349 –374. Yip, Moira. 1992. Prosodic morphology in four Chinese dialects. Journal of East Asian Linguistics 1. 1–35. Yip, Moira. 2001. Segmental unmarkedness versus input preservation in reduplication. In Linda Lombardi (ed.) Segmental phonology in Optimality Theory, 206 –228. Cambridge: Cambridge University Press.

Lateral Consonants

26

Yip, Moira. 2003. Some real and not-so-real consequences of comparative markedness. Theoretical Linguistics 29. 53–64 Yip, Moira. 2004. Lateral survival: An OT account. International Journal of English Studies 4(2). 25–51. Yip, Moira. 2005. Variability in feature affiliations through violable constraints: The case of [lateral]. In Marc van Oostendorp & Jeroen van de Weijer (eds.) The internal organization of phonological segments, 63–91. Berlin & New York: Mouton de Gruyter. Yu, Alan C. L. 2007. A natural history of infixation. Oxford: Oxford University Press.

32

The Representation of Intonation Amalia Arvaniti

1

Introduction

It is a well-known truism that no utterance is ever produced in a strict monotone; all utterances, in all languages, show some pitch modulation. Such changes in pitch – impressionistically described as rises and falls – are due to changes in fundamental frequency or F0, the physical property of the speech signal that is determined by the basic rate of vibration of the vocal folds and gives rise to the percept of pitch. Although pitch modulations exist in all languages, their origin and function differ, in that pitch patterns may be specified either at both the lexical and phrasal levels or only at the phrasal level, resulting in more or less dense tonal specifications, respectively (Gooden et al. 2009). The term intonation is used to refer to phrasal tonal patterns, while the terms pitch accent and tone are traditionally used to refer to lexical tonal specifications (chapter 45: the representation of tone). Simplifying somewhat, in languages like English, Italian, Greek, and many other European languages the entire F0 contour is specified at the phrasal level by means of a complex interplay between metrical structure, prosodic phrasing, syntax, and pragmatics; these factors determine where pitch movements will occur and of what type they will be. In languages referred to as tone languages – such as Mandarin, Thai, and Igbo – most syllables are lexically specified for tone and tonal changes affect lexical meaning; in languages often referred to as pitch accent languages – such as Japanese, Swedish, and Serbian – tone operates in a similar fashion, except that at most one syllable in each word is lexically specified for tone. In both tone and pitch accent languages additional tonal patterns are specified at the phrasal level. Here the focus is on languages without lexical tonal specifications, since it is the intonation of these languages that has been mostly examined. Determining the structure of pitch modulation and the primitives that make up pitch contours in languages without lexical tone is challenging, since F0 changes are not as discrete and easily identifiable as in “tonal” languages, their connections to segmental material are less easy to determine, and associated meanings are harder to pinpoint since they deal with information structure and pragmatic interpretation rather than lexical semantics. The following examples illustrate these points. In Figure 32.1, two utterances are shown, Me?! and A The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0032

2

Amalia Arvaniti

Pitch (Hz)

375

Me?! 100

A

ballgown

designer?!

0

2.575 Time (s)

Figure 32.1 Waveforms and F0 contours of two English utterances illustrating the incredulity contour (Hirschberg and Ward 1992); on the left, Me?!; on the right, A ballgown designer?!. Vertical lines indicate word boundaries

ballgown designer?!, both using the rise-fall-rise melody that implies incredulity (Ward and Hirschberg 1985; Hirschberg and Ward 1992). They are plausible responses to a career advisor’s pronouncement that, according to test results, designing ballgowns is the recommended career choice for the speaker, who has all along dreamed of becoming an aerospace engineer (for similar examples, see Ladd 2008: 45–46). Although the short contour can be informally described as rise–fall–rise, the longer contour cannot be described in a similar fashion, as it shows a long low-level stretch between a rise–fall and a final rise. In Figure 32.2, Greek contours very similar to the English ones in Figure 32.1 are shown, though in the case of Greek these contours are used for wh-questions (Arvaniti and Baltazani 2005; Arvaniti and Ladd 2009). As can be seen, the same issue with overall shape arises here as well. Further, as Arvaniti and Baltazani (2005) note, the Greek melody in Figure 32.2 can also be used for polite requests employing an imperative; e.g. [’Ïose sti ma’ria ’li:o ne’raci] ‘give Maria some water’ (lit. give to Maria a-little water-dim). Finally, Figure 32.3 illustrates two instances of another English melody: unlike the contours in Figures 32.1 and 32.2, which look different from each other but convey the same meaning in each case, the contours of Figure 32.3 are realizations of the same melody but convey different meaning, depending on the utterance: the melody used is

Pitch (Hz)

325

pça 125

pça mama tilefonise sti

0

nosokoma 3.32

Time (s) Figure 32.2 Waveforms, transcriptions, and F0 contours of two Greek wh-questions, on the left, [’pça] ‘which (fem)’, on the right, [’pça ma’ma tile’fonise sti noso’koma] ‘which mom called the nurse?’ Vertical lines indicate word boundaries

The Representation of Intonation

3

Pitch (Hz)

300

That’s really 150

awesome

That’s

0

twenty

dollars 2.323

Time (s) Figure 32.3 Waveforms and F0 contours of two English utterances; on the left, That’s really awesome; on the right, That’s twenty dollars. Vertical lines indicate word boundaries

the default for That’s twenty dollars, but sounds blasé or sarcastic when used with That’s really awesome. Note that this is not because the melody is wrong for That’s really awesome: the use is legitimate and meaningful (if the speaker wishes to be sarcastic or convey her indifference). These examples illustrate three main points about intonation. First, they show that the shape of intonational contours with a given pragmatic interpretation can vary substantially, depending on the segmental material with which they are uttered. Such differences are not random, but related to the overall prosodic structure of the utterance with which the contours are associated, including the number of syllables and the position of stresses (where applicable). Second, the examples show that contours do not have a constant meaning, either within or across languages; within a language, their interpretation may well depend on lexical and other choices that accompany the use of the melody; across languages differences can be arbitrary. A successful theory of intonation should be able to capture these properties: it should be able to explain the connection between intonation and meaning and make generalizations from surface F0 data with sufficient predictive power to generate new contours of the same basic melody to “fit” new utterances of varying lengths and structures. Although the above observations are by and large shared by most intonational models, the ways in which they treat these basic properties show substantial differences. As discussed in more detail in §2.1, many researchers have treated F0 contours as gestalts or configurations, that is, as holistic pitch movements that encompass entire utterances and have a uniform meaning. In other models, melodies are seen as being composed of primitives of some sort. These primitives are considered to be either local configurations (or dynamic tones), such as local rises and falls, or level tones, such as high, mid, and low. Here I review both the controversy between advocates of gestalt approaches to intonation and those who proposed analyses based on the decomposition of melodies into smaller elements, and the disagreement between researchers who use dynamic tones (that is, local configurations) as the primitives of intonational structure and those who advocate the use of level tones instead. As I show, however, focusing only on the form of intonational primitives avoids an even more fundamental question, namely which parts of a contour should be

4

Amalia Arvaniti

represented at all. This question is addressed in more detail in §3.2, where the main argument is advanced that an inordinate attention to phonetic form and a reluctance to accept that intonation is part of a language’s phonology have hampered research and have led to analyses that by and large aim at reproducing entire melodies, but have little predictive power and cannot successfully generalize beyond the F0 contours they reproduce. As argued in §3.2 and §3.3, sparse representations that aim at capturing only the linguistically significant aspects of each contour can better handle both intonational form and intonational meaning.

2 2.1

Configurational models Melodies as gestalts

As mentioned above, many researchers have treated intonation contours as gestalts, such as Bolinger (1951), Jones (1972), Cooper and Sorensen (1981), Hirst and Di Cristo (1998), Grabe et al. (2003), and Xu (2005). In these works, it is most often the case that intonational contours are seen as holistically and directly reflecting certain functional or structural aspects of speech, such as the distinction between questions and statements or that between levels of phrasing. According to Jones (1972: 279) – who in the last edition of An outline of English phonetics followed several earlier intonational analyses, notably those of Armstrong and Ward (1926) and Kingdon (1958) – English has two basic tunes, Tune 1 and Tune 2. These are a fall and rise respectively, which “may be spread over a large number of syllables, or [. . .] be compressed into smaller spaces.” Bolinger (1951: 208) also concludes his critique of level-based analyses (see §3.1) by saying that “intonation could not be a more appropriate illustration of the Gestalt.” More recently, Cooper and Sorensen (1981) presented a series of experiments in which contours are treated as undivided wholes and peak height is taken to directly reflect levels of phrasing. Modern versions of the gestalt approach include INTSINT (International Transcription System for Intonation; e.g. Hirst and Di Cristo 1998; Hirst et al. 2000), OXIGEN (Oxford Intonation Generator; Grabe et al. 2003), and PENTA (Parallel Encoding and Pitch Target Approximation; e.g. Xu 2005). In INTSINT, entire intonation contours are transcribed, using symbols that represent pitch movements. The movements, however, are not seen as primitives but rather as a means to transcribing the course of F0 over an entire utterance (hence their descriptive labels Higher, Lower, Upstepping, Downstepping, Same, Top, and Bottom, which express the course of F0 relative to preceding points and the overall range of the speaker). In OXIGEN, polynomials are used to model overall contour shape differences between statements and questions in British English. Finally, in PENTA, each syllable in a contour has its own pitch specification, while global aspects of the overall melody are directly linked to functional effects (a feature shared with OXIGEN); e.g. the use of an utterance as statement or question is said to lead to changes in overall pitch shape from fall to rise (Xu 2005). Configurational approaches have been quite popular for several reasons. First, their appeal is intuitive: F0 contours are more or less continuous, so, as Bolinger (1951: 206) put it, intonation can be seen as “a pattern [. . .] in the fundamental, down-to-earth sense of a continuous line that can be traced on a piece of paper”

The Representation of Intonation

5

(though, as noted in Arvaniti 2007, the fact that F0 looks continuous on paper or a computer monitor does not necessarily mean that it is perceived in this fashion). Second, the relationship between shape and function seems sufficiently natural in many cases that it has been argued to derive from the biological underpinnings of pitch production (e.g. Ohala 1983; for a thorough discussion of the biological code, see Gussenhoven 2004: ch. 5, who, however, does not adopt a configurationalist approach to intonation). This view is supported by certain general trends, such as the use of rising F0 for questions and falling F0 for statements for which it is possible to find empirical evidence in several languages (e.g. Grabe et al. 2003). Despite their popularity, configurational approaches face a major problem when it comes to accounting for intonational form. Specifically, if melodies were undivided wholes, they should simply shrink and stretch to “fit” the segmental duration of the utterance they co-occur with. There is plenty of evidence, however, that when tunes are matched with utterances of varying lengths and different metrical structures they are not realized in this simple manner. This was observed by ’t Hart et al. (1990: ch. 4), among the first researchers to use instrumental rather than impressionistic data for intonation research (e.g. Cohen and ’t Hart 1968; ’t Hart and Cohen 1973; ’t Hart and Collier 1975). They noticed that in their Dutch corpora sequences of pitch movements would appear on a single syllable in some instances, but would be separated by several syllables in others (cf. Figures 32.1 and 32.2). Importantly, ’t Hart et al. found that this elasticity, as they called it, did not affect the melodic identity of the contour (determined by means of perceptual experiments; see §2.3), even though it radically altered the overall contour shape (thus, the concept of elasticity can be juxtaposed with the compression envisaged by Jones 1972, which implies greater uniformity in the squeezing and stretching of contours). Results from several later studies support the original observations of ’t Hart et al. that certain aspects of the contour are important for listeners, while overall contour shape is not. Pierrehumbert and Steele (1989) varied in steps the position of the pitch peak in English melodies that can be holistically described as rise–fall–rise, and found that listeners imitating these stimuli produced not a continuum but two different contours, one with an early and one with a late peak. Similar results have also been presented by Redi (2003), following Pierrehumbert and Steele’s imitation protocol (argued by Gussenhoven 1999 to be the best way to examine categoricality in intonation). Similarly categorical responses to intonational continua that would be holistically seen as instances of the same contour have also been obtained by Rietveld and Gussenhoven (1995) for Dutch, D’Imperio and House (1997) for Neapolitan Italian, and Botinis (1989) for Greek, inter alia. The contours in Figures 32.1 and 32.2 also illustrate this general point. In the monosyllabic utterances, the rise–fall–rise stretches over the entire syllable. In the longer utterances, however, the rise co-occurs with the first stressed syllable (with some peak delay) and the final rise is realized on the last syllable, while the fall and subsequent low-level stretch vary depending on the language and length of the utterance. As a result, the contours of the longer utterances are not stretched-out copies of the shorter contours, nor are the shorter contours compressed versions of the longer contours; these differences, extensively discussed in Arvaniti and Ladd (2009), are illustrated in Figure 32.4, using the Greek contours of Figure 32.2.

6

Amalia Arvaniti

Pitch (Hz)

325

125

0

3.32 Time (s)

Figure 32.4 F0 contours of the Greek wh-questions shown in Figure 32.2; the solid gray lines are compressed and stretched-out copies of the long and short contour respectively

Differences in contour shape may also relate to the number and position of stressed syllables in an utterance and the location of the word in focus. This is experimentally shown in Arvaniti et al. (2006a), who studied the default melody of Greek polar questions in which the position of focus can vary. They show that the shape of the contour is strongly affected by the position of the stressed syllables and of the word in focus (see also Arvaniti and Baltazani 2005; Arvaniti 2007). The focus effect in particular is illustrated in Figure 32.5, which shows the two contours that can be used with the sentence [’pinun lemo’naÏa] ‘they drink lemonade’ when it is uttered as a question with focus on the verb (dotted line) or the noun (solid line). As can be seen, no description in terms of overall shape can possibly cover both contours; at best, the late focus question would be characterized as rise–fall–rise–fall and the early focus question as rise–fall, but this ignores the location of the rise–fall part that the two contours share and the significance of this location both for understanding the pragmatics of the two questions and for their naturalness (for details, see Arvaniti et al. 2006a). Holistic approaches, then, suffer from two problems. First, they cannot represent in the same manner contours that superficially look different, like the contours in Figure 32.5. At the same time, holistic approaches cannot account for systematic differences between realizations: e.g. they cannot account in a principled manner for the fact that the syllable [nun] in Figure 32.5 is low in the dotted, early focus contour, but rising in the solid, late focus contour. In short, configurational approaches cannot represent different instantiations of the same melody in a way that can either capture their similarity or predict their differences, thereby failing one of the main criteria for an adequate theory of intonational phonology mentioned in §1.

2.2

Intonational gestalts and meaning

In addition to issues with intonational form, gestalt approaches encounter problems with intonational meaning. In gestalt models, overall contour shape is said to associate with differences in meaning. Yet, as illustrated in §1, the relationship between melody and meaning is not one-to-one: the same melody can lend different pragmatic nuances to different utterances, while the same meaning can

The Representation of Intonation

7

Pitch (Hz)

300

pi 100

nun

le

mo

0

na 1.001

Time (s) Figure 32.5 Waveform, transcription, and F0 contours for the phrase [’pinun lemo’naÏa] ‘they drink lemonade’, uttered as a question with focus on [’pinun] ‘they drink’ (dotted contour) or on [lemo’naÏa] ‘lemonade’ (solid contour); the former could be glossed as ‘is lemonade something they would drink?’ and the latter as ‘is it lemonade that they are drinking?’ Vertical lines indicate syllable boundaries

be expressed by superficially different-looking contours. This lack of one-to-one correspondence has been repeatedly noted over the years (among many, Bolinger 1964; Lehiste 1970: 95ff. and references therein; Ladd 1980; Pierrehumbert 1980; ’t Hart et al. 1990: ch. 4), and prompted Pike (1945: 23ff.) to strongly caution against the practice of investigating contour meaning on the basis of grammatical structure (such as pitting statements against questions). Overall, then, melodies do not appear to have specific functions, and indeed attempts to describe the melodies of specific pragmatic nuances, such as irony, have proved unsuccessful (e.g. Bryant and Fox Tree 2005). In addition, cross-linguistic research has shown that functional effects of the sort favored by gestalt approaches are expressed in language-specific ways. Such findings abound and strongly argue against a natural or direct relationship between intonational form and function. For instance, it has been argued that focus is universally expressed as local pitch expansion with a concomitant reduction in pitch range post-focally (e.g. Xu 2005). Yet a review of the literature clearly shows that this is far from a universal mechanism for the prosodic marking of focus. For example, in Greek polar questions the word in focus has the lowest F0 in the entire utterance and pitch expansion is associated with the post-focal region (Arvaniti et al. 2006a; see Figure 32.5 for an illustration). Taiwanese relies on duration to mark focus rather than changes in pitch (Pan 2008), while in many other languages pitch expansion is just one, optional, manifestation of an overall prosodic reorganization under focus (e.g. Chen 2006, 2010 for Mandarin; de Jong 1995 and Harrington et al. 2000 for English; Baltazani 2006 and Arvaniti et al. 2006b for Greek; Jun 2005a for Korean; Venditti et al. 2008 for Japanese). Perhaps the strongest counter-argument against the view that the relationship between focus and pitch range is natural and direct is the fact that not all languages can use intonation to mark focus (e.g. Swerts et al. 2002 on Italian; see Ladd 2008: ch. 6 for a discussion). If the relationship between focus and intonation is natural and direct, there is no good explanation of why some languages do not avail themselves of this option. The presence of an arbitrary relationship between intonation and meaning (as in all aspects of linguistic structure) is also evident in cross-linguistic data from

8

Amalia Arvaniti

questions, which, as mentioned, are often assumed to end in rises while statements are said to end in falls (e.g. Ohala 1983; Gussenhoven 2004: ch. 4; Xu 2005). But this idea is not supported cross-linguistically. Low or falling intonation is used with questions in typologically diverse languages, such as Bengali (Hayes and Lahiri 1991), Chickasaw (Gordon 2005), Bininj Gun-wok (Bishop and Fletcher 2005), many Niger-Congo languages (Rialland 2007), Greek (Arvaniti et al. 2006a, 2006b; Arvaniti and Ladd 2009), and Romani (Arvaniti and Adamou, forthcoming). Conversely, it is also the case that statements end in rises in many languages, including Bengali (Hayes and Lahiri 1991), Chickasaw (Gordon 2005), and many varieties of English (e.g. Grabe et al. 2000; Fletcher et al. 2005). Overall then, cross-linguistic research has confirmed the observation made by many scholars over the last century that the relationship between meaning and melody (as in global contour shape) is arbitrary and many-to-many. In turn, these results support the view that no useful generalizations about either intonational meaning or intonational form can be made on the basis of contour shape and its relationship to broad functional effects.

2.3

Intonation as a composite of rises and falls

If it is accepted that melodies are composed of independent elements, the question that arises is the nature of these elements. In this respect, most researchers have favored approaches in which the primitives are dynamic tones or movements, such as rises and falls, though the possibility of level tones (monotones) within such systems is also acknowledged (Bolinger 1964, 1986; Crystal 1972; Ladd 1980; ’t Hart et al. 1990). Perhaps the most thoroughly tested of these models is that of the IPO (Institute for Perception Research, ’t Hart et al. 1990, and references therein). The IPO model was developed on the basis of Dutch, but it has also been used for the description of intonation in other languages, such as English, German, and Russian (see ’t Hart et al. 1990: ch. 4, and references therein). In this system, the main elements are rises and falls, a choice justified on perceptual grounds: the IPO researchers noticed that pitch changes were realized more slowly than possible by laryngeal control (as determined by the studies of Ohala and Ewan 1973 and Sundberg 1979), and concluded that the purpose of this slow execution must be to give listeners the perception of “pitch movement” rather than of a jump in pitch (’t Hart et al. 1990: 71). Rises and falls are composed of four “perceptual features”: pitch direction, timing relative to syllable boundaries, rate of pitch change, and excursion size. In addition, rises and falls combine into larger configurations or contours; e.g. a rise–plateau–rise creates a “hat pattern” while a rise–fall creates a “pointed hat.” This architecture makes the system configurational in two ways, since both the primitives and their combinations are configurations. An important feature of the IPO approach is that meaning does not play a part in establishing either the primitives or the contours on the grounds that “intonation features have no intrinsic meaning” (’t Hart et al. 1990: 110). Instead, decisions as to the number and nature of melodies are based on experimental evidence derived from the close-copy technique, in which listeners are asked whether stylized versions of various contours sound the same or different. By using experiments like these, the IPO researchers determined the limits within which

The Representation of Intonation

9

contours may vary phonetically; in turn, variants that are considered different by listeners are used to establish the elements of the intonational system. The IPO stance toward the role of meaning in the analysis of intonation is the exact opposite of that taken by the British school (among many, Halliday 1967, 1970; Crystal 1972; O’Connor and Arnold 1973), where differences in meaning are crucial for establishing the existence of both primitives and entire tunes. In this system, tone units (or tunes or tone groups) can span entire utterances, but are also decomposed into smaller parts, the pre-head, head, nucleus, and tail (this is the division proposed by Crystal 1972 and O’Connor and Arnold 1973; for a review of additional analyses, see Ladd 1980: 16). The nucleus, defined as the pitch movement on the stressed syllable of the most important word of the utterance, is the only required element of a tune. The F0 of any unstressed syllables following it is the tail, while the F0 stretch covering all syllables from the first stressed syllable to the nucleus is the head; the F0 of any unstressed syllables preceding the head forms the pre-head. As can be surmised, not all melodies include all four primitives, while particular primitives can span arbitrary lengths of an utterance. For instance, the short utterance in Figure 32.1 consists only of a nucleus, while the two utterances in Figure 32.3 include a pre-head (that’s), a high head (twenty/really), a low-falling nucleus (the stressed syllable of dollars/awesome), and a tail (the unstressed syllable of dollars/awesome). These elements do not combine entirely freely. As an example, O’Connor and Arnold (1973) distinguish seven nuclear tones, four types of heads, and two types of pre-heads, but only 20 types of tone groups, instead of the 56 that all possible combinations of primitives would produce (tails follow the movement of the nucleus, so they do not enter into the calculation). The use of primitives of unconstrained length forces analyses of the British school into an artificial distinction between simple and compound rise–falls and fall–rises (e.g. Halliday 1970; O’Connor and Arnold 1973). The compound tunes fit uneasily into the system, since they are said to contain two nuclear tones. Furthermore, most practitioners admit that it is hard to distinguish simple tunes from their compound counterparts on the basis of meaning, a serious drawback for a system in which the role of meaning is central (e.g. Crystal 1972; O’Connor and Arnold 1973: ch. 1). This problem is illustrated by the contours in Figure 32.1: the tune of Me?! is more plausibly analyzed as a simple rise–fall, but that of A ballgown designer?! can only be treated as a compound tune, though their meaning affinity is evident. Finally, like gestalt models, the British school analyses face problems with their treatment of intonational meaning, which is hard to pinpoint, yet must be determined if the elements of the system are to be defined. As a result, the analyses of different authors disagree on the meaning and number of tunes, including the number and shape of nuclei; e.g. Halliday and Crystal recognize one type of fall, while O’Connor and Arnold distinguish between a high fall and a low fall. To the extent that meanings can be determined, they tend to be vague and occasionally contradictory, and to a large extent dependent on grammatical aspects of the utterance. For example, O’Connor and Arnold (1973: 78–79) describe the “Jackknife” (simple rise–fall) as showing that the speaker is “impressed, perhaps awed,” but they also add that the speaker can use it to sound “complacent, self-satisfied, even smug” or as “shrugging aside any involvement.” These meanings apply if the Jackknife is used with statements, but change if it is used with wh-questions, polar questions, or commands (for discussions of the problems of the British school approach to intonational meaning see Liberman 1975: 132ff. and Pierrehumbert and Hirschberg 1990).

10

2.4

Amalia Arvaniti

Superpositional models

In a number of configurational models, contours are said to be composed of two elements, a general trend and local perturbations which “ride” on this overall movement. This conception of intonation was also espoused by Bolinger, who distinguished accentuation from intonation, using accentuation to refer to pitch movements (accents) on stressed syllables, and intonation to refer to the general course of F0, “the rise and fall of pitch as it occurs along the speech chain” (Bolinger 1986: 194). The IPO system is one such superpositional model, in that its primitives and contours are seen as localized movements superposed on a larger declination component, which is taken to be largely automatic and due to the drop in subglottal pressure during the course of an utterance (declination reset and local movements, on the other hand, are seen as actively controlled by the speaker; ’t Hart et al. 1990: ch. 5). The exact role of declination and its physiology are still a matter of debate (see e.g. Pierrehumbert and Beckman 1988: ch. 3; Gussenhoven 2004: ch. 6), though evidence such as that provided for Japanese downtrends by Pierrehumbert and Beckman (1988) does not support the idea of declination playing as important a role as the IPO scholars envisioned. Superpositional models have also been presented by Fujisaki (1983, 2004), Gårding (1983, 1987), and Thorsen (1980, 1985, 1986). Simplifying somewhat, in Fujisaki’s system a phrase command results in the rising–falling course of F0 throughout an utterance (or a part thereof), with accent commands being responsible for more localized perturbations. Gårding (1983, 1987) posits grids (quasi parallel lines) within which most local F0 minima and maxima can be fitted; the overall range and direction of the grid (rising or falling or a combination thereof) reflect functional differences between utterances, such as the distinction between statements and questions. Similarly, Thorsen (1980: 1022) suggests that the rate of F0 drop in Danish is directly related to utterance function: “falling intonation contours are associated with declarative, intermediate contours with nonfinal, and flat contours with interrogative sentences.” Due to the connection between communicative functions and overall F0 trends, the models of Thorsen and Gårding face similar issues to gestalt models with respect to meaning. On the other hand, Fujisaki’s model, which does not rely on meaning distinctions, must resort to counterintuitive solutions – such as negative accent commands and phase commands that span linguistically arbitrary stretches – in order to adequately describe the course of F0 in languages other than Japanese (e.g. Fujisaki et al. 1997 on Greek; Fujisaki et al. 2005 on Mandarin; Gu et al. 2007 on Cantonese; for a discussion of these problems, see Ladd 2008: 23ff.; Arvaniti and Ladd 2009).

3 3.1

Pitch levels as primitives Early level-based models

Descriptions of intonation by means of level tones date from the American structuralists (Pike 1945; Trager and Smith 1951; Hockett 1955; Trager 1961). In these systems, intonation is analyzed by means of four levels, extra-high, high,

The Representation of Intonation

11

mid, and low. The level tones of these analyses are meant to be phonological abstractions equivalent to phonemes (chapter 11: the phoneme); as such, they are said to be defined relative to each other, rather than each representing a specific pitch range. These early analyses were heavily criticized by configurationalists, most notably Bolinger (1951), who questioned the claim that the four levels are relative, pointing out that if this assertion is taken at face value, it is not possible to distinguish combinations such as 123 from 234, although, theoretically, such combinations should be distinct. Thus, Bolinger concluded that level tones cannot be relative but must “rove each in its own bailiwick” (1951: 200), and set out to test this hypothesis by recording utterances in various ways (as Bolinger termed them, i.e. melodies differing in various aspects) and having listeners judge them for similarity or appropriateness for a given purpose, such as appeasing a child. His results showed that contours analyzed as contrastive in the system of Trager and Smith (whom he particularly targeted) are perceived as similar by listeners, while others, analyzed as allophones of the same basic melody, are considered by listeners to be contrastive. Bolinger used his results to argue that a system with four levels can be at the same time too powerful and not powerful enough to capture contrastive and allophonic variations in the intonational system of English. His results led him to reject level tone analyses as untenable and to propose instead that melodies are gestalts. Bolinger’s critique reflects the assumptions of concreteness and bi-uniqueness prevalent at the time. It is clear, for example, that Bolinger expected the different levels to faithfully represent the entire course of an utterance’s F0, and to do so in such a way that the pitch range of each level did not overlap with that of others at any point in the utterance. Further, his comment that F0 forms “a continuous line that can be traced on a piece of paper” (Bolinger 1951: 206), coupled with his distinction between monotones, which he accepts, and level tones, which he does not (e.g. Bolinger 1986: 29), suggest that he expected a level-based representation to be phonetically realized as a series of sustained pitch levels. It is obvious that if these assumptions are adopted, a level-based analysis is unworkable on both phonological and phonetic grounds (on the latter, see Xu and Sun 2002; Dilley and Brown 2007). Although Bolinger’s critique was well accepted, it is fair to note that many of the assumptions he attributes to the structuralists are not found in their works. Pike (1945), Hockett (1955), and Trager (1961) all note that levels represent “only those points in the contour crucial to the establishment of its characteristic rises and falls” (Pike 1945: 26); with the exception of terminal junctures, these points associate with stressed syllables. Similarly, the structuralists noted that absolute pitch levels are not significant as such, and recognized the existence of both level tones and contours (e.g. Trager and Smith’s terminal juncture phonemes). Further, Pike (1945) discusses at length the fact that level tones need not be realized as a series of sustained pitches but can be realized as glides, especially when they are found close to each other, as happens in short utterances, for example. Nevertheless, as a result of configurationalist critiques, level-based analyses were largely abandoned in the following decades. Research within the generative framework focused primarily on the description of tone languages and no theoretical position was strongly taken in favor either of levels or of configurations with respect to intonation.

12

Amalia Arvaniti

Intonational analyses by means of level tones adopting many of the principles of the early structuralist accounts appear again in the late 1970s, in Liberman (1975), Goldsmith (1976), and Leben (1976). Based on the idea of Leben (1973), who first conceptualized tones as distinct tonal segments rather than features of particular tone-bearing units (typically syllables or moras), Goldsmith represented tones as autosegments residing on a separate tier, and analyzed English intonational melodies as sequences of H and L level tones (chapter 14: autosegments). Liberman (1975), on the other hand, proposed an analysis of English intonation with the traditional four levels represented by means of two features, [±high] and [±low], though, informally, he also used autosegmental representations closer to those of Leben (note that [+high, +low] is possible in Liberman’s system, and represents a high-mid level). In addition to the obvious differences in terms of formalism and the overall conception of phonology, the early autosegmental models depart from those of the early structuralists in that they assume that all syllables in an utterance are associated with some tone (something that may be accomplished, e.g. by tone copying or spreading). However, as discussed in more detail in §3.2, the assumption that contours should be fully specified leads back to the problems that Bolinger (1951) first noted.

3.2

The autosegmental-metrical model of intonational phonology

A major breakthrough in our understanding of intonation came from Bruce’s (1977) dissertation, a phonetic investigation of Swedish tonal structure. Bruce showed that the difference between the two lexical pitch accents of Swedish, accent I and accent II, does not lie in the shape of the accent, which is a fall in both cases, but in its timing with respect to the accented syllable: for accent I the fall is timed early, and for accent II it is timed late. A corollary of this difference is that a large part of the fall is truncated when a syllable with accent I is utterance-initial, while it is fully present in words with accent II (giving rise to a peak preceding the fall). Furthermore, Bruce provided an explanation for the second peak of accent II words, the seemingly erratic presence of which had been a longstanding puzzle: he showed that this peak is not part of the lexical accent at all, but part of the utterance’s intonation, which he analyzed as a sequence of a sentence accent and a terminal juncture. Several key points emerged from Bruce’s work. First, it demonstrated the importance of turning points, F0 minima and maxima which temporally align with particular elements of the segmental string. Bruce showed that these timing relations are regular in production and salient in perception. Crucially, they are also sufficient for modeling a contour without specifying the F0 in between (that is, by interpolating between salient points). Second, Bruce showed that an F0 contour can be composed of elements of different origins within the grammar: in Swedish, some parts of the contour, the pitch accents, are lexically specified, while others, the sentence accents and terminal junctures, are phrasal elements, the result of intonation and phrasing. Similar distinctions were hinted at in other models – e.g. in Trager and Smith’s terminal junctures, and in the prominence lending vs. non-prominence lending distinction between pitch movements

The Representation of Intonation

13

espoused by the IPO – but they had not been clearly demonstrated before. Finally, Bruce showed that phrasal and lexical tonal elements simply concatenate, rather than forming two distinct systems superimposed on each other, and that this concatenation can result in lawful, context-dependent variation in the realization of tones. The insights of the early autosegmentalists regarding the representation of tone, together with Bruce’s insights about the structure of tunes and their phonetic realization, were applied to the analysis of English intonation in Pierrehumbert’s dissertation (1980) (see chapter 116: sentential prominence in english for a more extensive discussion of intonational patterns in English). Her model was further developed, particularly in Beckman and Pierrehumbert (1986) and Pierrehumbert and Beckman (1988), into the model currently known as the autosegmental-metrical model of intonational phonology (a term coined by Ladd 1996). The autosegmental-metrical model (henceforth AM) has since been applied, with various modifications, to a series of unrelated languages with diverging prosodic systems (e.g. Jun 2005a; see also Arvaniti, forthcoming and D’Imperio, forthcoming for reviews). In Pierrehumbert (1980), melodies are represented as strings of H and L tones on an autosegmental tier. Crucially, the purpose of this string of tones is not to trace or transcribe the course of F0, but rather to represent the linguistically significant parts of the melody; that is, intonational representations are underspecified (chapter 7: feature specification and underspecification). The tones associate with the segmental string indirectly, via associations with the metrical tree (in Pierrehumbert and Beckman 1988, this is formalized as association between tones and prosodic trees, specifically between tones on the one hand and feet and phrasal boundaries on the other). Thus, like Bolinger and Bruce, Pierrehumbert adopts the distinction between tones that associate with stressed syllables (i.e. the heads of feet) and tones that associate with the edges of phrasal constituents. The former, following Bolinger, are pitch accents notated with an asterisk (e.g. H*, a notation first used by Goldsmith 1976); the latter, known as boundary tones, are notated with a percentage sign (e.g. H%). Pierrehumbert also noted that tunes included a pitch movement between the boundary tone and the preceding nuclear accent (by definition, the last pitch accent of an utterance, often referred to in other literature as sentence stress). Pierrehumbert analyzed these pitch movements between the nuclear accent and following boundary tone as floating tones, dubbed phrase accents. In Beckman and Pierrehumbert (1986), phrase accents are instead analyzed as phrasal tones that associate with the right edge of intermediate phrases (ips), a prosodic constituent larger than the prosodic word and smaller than the intonational phrase (IP), the nature of which is formalized in Pierrehumbert and Beckman (1988). In Pierrehumbert (1980), the H and L tones are said to be realized as tonal targets, typically peaks and dips respectively, although the relationship between phonological tones and phonetic realization is not always transparent. First, phonological tones may not be realized as F0 minima or maxima; e.g. in Pierrehumbert’s original analysis the H*+L- accent is not realized as a fall from a high to a low F0 level; rather the L tone is said to trigger downstep on a following H tone. Realizations are also context dependent; e.g. after a L- phrase accent, L% is realized as a drop to the bottom of the speaker’s range, but after a downstepped H- phrase accent it is realized as sustained mid-level pitch. The

14

Amalia Arvaniti

scaling of targets is computed on the fly, and is determined by metrical strength and context (see also Liberman and Pierrehumbert 1984; Pierrehumbert and Beckman 1988; Prieto et al. 1996). Furthermore, following Goldsmith (1976), Pierrehumbert assumed that tones co-occur (align) with the segmental material they are (indirectly) associated with: pitch accents align with associated stressed syllables, and boundary tones with phrase-final syllables; phrase accents, being floating tones, are realized in a less precise manner (for evidence that this view cannot fully account for tonal alignment cross-linguistically, see, inter alia, Arvaniti et al. 2000; Grice et al. 2000; Gussenhoven 2000). Since Bruce (1977) and Pierrehumbert (1980), several phonetic studies have demonstrated the crucial role of local F0 minima and maxima in the production of intonation. Such tonal targets have been shown to be consistently aligned with the segmental string and to show lawful variation, based on a number of factors, such as speaking rate (Fougeron and Jun 1998; Prieto and Torreira 2007), phonological weight (Ladd et al. 2000), tonal crowding (Silverman and Pierrehumbert 1990; Prieto 2005; Arvaniti and Ladd 2009), and dialectal differences (Arvaniti and Garding 2007; Ladd et al. 2009). Results attesting to the regularity of turning points have been reported for languages with very different prosodic systems, including not only languages without lexical tone but also tone languages, e.g. Mandarin (Xu 1999), Kinyarwanda (Myers 2003), and Thai (Morén and Zsiga 2006), pitch accent languages, e.g. Roermond Dutch (Gussenhoven 2000), Basque (Hualde et al. 2002; Elordieta and Hualde 2003), and Serbian (Smiljanio 2006), and languages with hybrid systems, e.g. Chickasaw (Gordon 2008) – for a comprehensive review, see Arvaniti (forthcoming) and D’Imperio (forthcoming).1 The search for highly localized tonal targets is in part a corollary of one of the most important tenets of the AM model, namely that contours are underspecified not only phonologically, but phonetically as well: while H and L tones are realized as targets with specific scaling and alignment, the F0 between them is determined by interpolation. Underspecification was empirically documented by Pierrehumbert and Beckman (1988: ch. 2), who showed that the course of F0 between the initial phrasal H of unaccented accentual phrases in Japanese and the final L% of the phrase is a fall that varies in steepness depending on the number of moras between the two tones. This type of realization is incompatible with fully specified representations, whether such specifications are present from the beginning or determined at a later stage through tone spreading.2 1

Despite the overwhelming evidence in favor of level tones being realized as highly localized minima and maxima, it is clear that other realizations are also possible. First, in some cases turning points may be just indirect reflexes of tones that are convenient to identify and measure in a signal that presents as a continuous curve; for example, Arvaniti and Ladd (2009) show that the turning points that define the low-level stretch in the contours of Figure 32.2 are not necessarily targets themselves, but quite possibly the easy-to-measure outcome of the tune’s requirement for a low-level stretch, the reflex of the L- phrase accent. Results like these suggest that tones should not be posited as phonological entities simply because a turning point in an F0 contour has been noted. Second, tones may not be realized as instantaneous events (among many, see Pierrehumbert 1980; Silverman 1987; D’Imperio 2000; Arvaniti et al. 2006a; Knight and Nolan 2006; Arvaniti and Garding 2007). Thus, although it is convenient to measure tonal targets as F0 minima and maxima, tones may also have duration, a type of realization that may be perceptually enhancing (e.g. D’Imperio et al. 2000; D’Imperio et al. 2010). 2 Recent results of work by Barnes et al. (2010) indicate that F0 transitions from one tone to another can be perceptually relevant, and aid in the identification of particular pitch accents (at least in experimental settings in which transitions are the only information listeners have for tonal identification).

The Representation of Intonation

15

The role of underspecification cannot be underestimated: it leads to a clear understanding that the number of tones need not match the number of tone-bearing units (TBUs), so that both strings of tonally unspecified TBUs and instances of several tones associating with the same TBU are possible. Thus, in the AM framework, the two English contours in Figure 32.1 are both analyzed as L*+H L- H%. By using the same representation for these two contours, AM captures the fact that they are instantiations of the same melody, thereby generalizing beyond surface form. AM can also account for the systematic differences between contours like those in Figure 32.1, which, as mentioned earlier, were particularly problematic for the British school. In Me?!, all three tonal events must co-occur with the only syllable of this utterance; hence the obvious lengthening of me (720 msec) and the swift movement from one tonal target to the next. In A ballgown designer?!, L*+H is associated with the metrically strongest syllable in the utterance, i.e. ball, and it co-occurs with it (showing the peak delay expected for this accent; e.g. Pierrehumbert and Steele 1989; Arvaniti and Garding 2007). The H% is realized on the last syllable, which is the one showing a rise. The L-, which is associated with the ip boundary, spreads between the L*+H and H%, accounting for the fall and low-level stretch of F0 (for details on the realization of the L- in such contours, see Grice et al. 2000; Barnes et al. 2006). A similar analysis applies to the Greek wh-questions shown in Figure 32.2, analyzed as L*+H L- !H% (where !H refers to a downstepped H tone; Grice et al. 2000; Arvaniti and Baltazani 2005; Arvaniti and Ladd 2009). Overall, the AM model avoids several pitfalls of previous analyses. First, by formally separating stress from intonation and providing a mechanism for their interaction, the AM model incorporates the insights of Bolinger about pitch accents without requiring distinct accentual and phrasal components to account for pitch contours. In addition, the use of only H and L tones avoids the problems noted by Bolinger (1951) with respect to level tones. At the same time, by treating the issue of pitch range as a matter of phonetic realization, AM avoids the problems that plagued the British analyses due to the confounding of linguistic and paralinguistic aspects of pitch range (cf. the disagreements regarding whether high falls and low falls are distinct entities). Further, by making explicit the separation between the phonetics and phonology of intonation, the AM model provides a principled account of the context-dependent variation of tones, a point that was not explicitly addressed in previous models, which mostly confounded contours and representations. Finally, the use of underspecification provides a parsimonious and elegant way of capturing both the similarities of melodies and the differences in phonetic realization that arise from the properties of the metrical structure with which a melody associates. In this way, the model can account for both local phonetic detail and abstract phonological form, something that configurational and full specification models cannot do (for extensive discussions of this point, see Pierrehumbert and Beckman 1988; Arvaniti et al. 2006a; Arvaniti and Ladd 2009). Finally, since the degree of underspecification can vary, AM can account for languages with dense tonal specifications, such as Mandarin, as well as for languages with more sparse specifications, such as English. In short, then, AM not only provides an answer regarding the nature of intonational primitives, but crucially addresses the even more fundamental question of what should be represented phonologically when it comes to intonation, an issue that most other theories have not tackled by focusing exclusively on faithful representations of entire F0 curves.

16

3.3

Amalia Arvaniti

AM and meaning

AM has not grappled systematically with meaning, though several models of information structure have relied on AM analyses to understand the role of intonation, particularly with respect to focus marking (Steedman 2000; Büring 2007; among others). Although research on intonational meaning is not extensive, it so far suggests that the principles of AM are more likely to lead to an understanding of intonational meaning than configurational approaches, since AM analyses are compositional and thus in principle more flexible than configurational approaches in dealing with the complex relationship between intonational meaning and form both within and across languages. Perhaps the best-known treatment of meaning within AM is Pierrehumbert and Hirschberg (1990), who developed a theory of intonational meaning specifically for English (the principles of which are, however, applicable to intonational systems at large). According to this model, each pitch accent, phrase accent, and boundary tone is seen as a morpheme that has its own pragmatic meaning; the meaning conveyed by an entire melody is therefore compositional and depends on contributions from all tones. Further, Pierrehumbert and Hirschberg suggest that each tone’s meaning is to be interpreted with respect to the phonological domain with which it is associated: specific lexical items in the case of pitch accents, ips in the case of phrase accents, and IPs in the case of boundary tones. The advantage of a system like this is that it can account for similarities in meaning conveyed by identical components of different melodies, such as the use of the same pitch accent with different phrasal tones, or the use of a given boundary tone with different pitch and phrase accents. In addition, since the system does not rely on a natural or biologically determined relationship between tones and meaning, it is more easily amenable to an understanding of cross-linguistic and cross-dialectal variation in the relationship between meaning and form. Crucially, Pierrehumbert and Hirschberg take pains to explain that the meaning of a tune is not to be directly interpreted; e.g. H* L- L%, often used with declaratives, is not to be interpreted as “S (the speaker) believes x”; rather, the speaker’s belief in x will be inferred from all the tones that make up a tune and the context in which they are used. This conception of intonational meaning could account for the different interpretation of the same tune in That’s twenty dollars and That’s really awesome shown in Figure 32.3. Simplifying somewhat, the downstepped !H* accent on the second word (dollars, awesome) does not impart salience on the accented item (note that in the analysis of Pierrehumbert 1980, and that of Pierrehumbert and Hirschberg 1990, this accent is analyzed as H*+L); rather, it implies that this item should be inferable by the hearer. This is expected for dollars in a context in which dollars are the only currency in which purchases can be made; but when used with awesome the inferred predictability implies that the speaker is either ritualistically using an expression associated with excitement (and therefore is not excited) or deliberately avoids the more plausible H* pitch accent in order to convey sarcasm. Despite the existence of this framework, it is fair to say that many aspects of intonational meaning remain unclear. For example, Pierrehumbert and Hirschberg themselves note that similarities between the meanings of bitonal accents (L*+H and L+H*, and H*+L and H+L*) are evident in English, but not easy to account for in their model. Further, models similar to that developed by Pierrehumbert

The Representation of Intonation

17

and Hirschberg (1990) have not been as developed for other languages, so the framework has not been extensively tested for cross-linguistic validity. Research has also shown that certain tonal combinations are more frequent than others both in English (e.g. Dainora 2001, 2006) and in other languages (e.g. Arvaniti and Baltazani 2005 for Greek). Observations of this sort abound and have given rise to hypotheses that perhaps intonational meaning depends on strings of intonational primitives rather than being strictly compositional; in this case, the importance of the melody from the nuclear accent onward (akin to the nucleus of the British school) is emphasized (for a review and discussion, see Gussenhoven 2004: ch. 7). More recently, Calhoun (2010) has provided a metrical analysis for the importance of the nuclear pitch accent, suggesting that the presence of prenuclear accents is metrically motivated and that they do not contribute to information structure, contra Pierrehumbert and Hirschberg’s position that all accents contribute to meaning (for arguments similar to Calhoun’s, see Büring 2007). As even this brief discussion indicates, much remains to be done before the relationship between melodies and their contribution to information structure is fully understood. It is clear, however, that crude distinctions such as statement vs. question or focus vs. lack thereof are not sufficient to explain intonational meaning and that a more sophisticated understanding, quite possibly following the main principles of the treatment provided in Pierrehumbert and Hirschberg (1990), is likely to be more successful.

4

The phonetic realization of tones: Level or dynamic tones?

With the noted exceptions of Bolinger and the IPO, the choice between level and dynamic tones was not always explicitly motivated in the various models discussed here. It is of course easy to consider this a mundane issue: after all, as Bolinger (1986: 225) put it, “it makes no difference, in describing a movement, whether one says ‘first you are going to be up and then you are going to be down’ or ‘you are going to go down.’” Choosing the right answer, however, is neither trivial nor a matter of taste, as the answer has empirically testable consequences. This issue was addressed in Pierrehumbert and Beckman (1988: chs 2–4), who discuss extensively how the context-related scaling differences of tones they observed in Japanese cannot be elegantly accounted for in models using dynamic tones as primitives. Arvaniti et al. (1998) specifically compared the predictions of IPO to those of AM, by examining the realization of the rising pitch accents found in prenuclear position in Greek declaratives. They found that the timing, duration, and speed (rate of change) of these rises were not invariable, as advocated by the IPO, but depended on syllable duration. On the other hand, the timing and scaling of the onset and offset of the rise were held constant: the initial dip – interpreted as the reflex of an L tone – coincided with the onset of the accented syllable, while the peak – interpreted as the reflex of an H tone – was reached about 10 msec after the onset of the first post-accentual vowel. At the very minimum, the results of Arvaniti et al. (1998) suggest that the IPO notion of dynamic tones does not apply to all languages. More generally, they pose the question of how dynamic primitives such as rises and falls can be

18

Amalia Arvaniti

determined if none of the properties that may define them is stable. Finally, it is not clear how the notion of an indivisible unit can be defended for the Greek accents at all, since the beginning and ending points of the rise do not behave as one. Their relative autonomy is demonstrated by the fact that they align independently of each other and are not similarly affected by tonal crowding, which typically results in the undershooting of the L, while the realization of the H remains largely unaltered (Arvaniti et al. 2000). This pattern is difficult to account for if the rise is a unit, in which case one would more plausibly expect a curtailment of the entire pitch movement. It is thus clear that dynamic tones cannot account for some of the attested patterns. On the other hand, level tones can be used not only for the representation of loosely defined rises and falls, as in Greek, but also for rises and falls that are more closely knit. Such units have been reported by Frota (2002) for European Portuguese. Specifically, Frota found that in the falling accent indicating broad focus, the H and L are timed with respect to distinct segments (similarly to the Greek case), but the fall of the accent indicating narrow focus shows a constant timing relationship between the H and L tones (similar to that discussed by Pierrehumbert 1980 for L*+H and L+H* in English). This difference between the two accents of European Portuguese can be represented by means of a hierarchical representation of tones shown in (1), as first proposed by Grice (1995a, 1995b) and adopted by Frota, or it can be treated as an issue of phonetic realization, as argued in Arvaniti et al. (2006b). Either way, it is clear that while level tones can adequately describe all attested tonal patterns, dynamic tones cannot. In short, then, both the empirical evidence and phonological considerations of parsimony and descriptive adequacy make a theory based on level tones preferable. (1)

PA

PA

s

s

w s H+L*

5

s w H*+L

Conclusion

The original debate about levels vs. configurations (Bolinger 1951) focused on the issue of whether melodies are gestalts or should be seen as composites of primitives. Independently of this distinction, Bolinger’s views, espoused by many before him and since, are based on the idea that intonational contours should be represented in their entirety, either as a series of primitives, or as a “line [. . .] on a piece of paper.” Current understanding suggests that couching the problem in these terms is misleading, as neither type of representation is likely to be correct: as shown, gestalt approaches cannot account in a satisfactory manner for either intonational meaning or intonational form; yet representations that fully specify the course of F0, either in terms of dynamic tones or in terms of levels, do not fare much better. Due to the particularities of intonation, especially the fact that its realization depends on the metrical structure of the utterance with

The Representation of Intonation

19

which it co-occurs, significant generalizations about melodies and their phonetic variation are best captured if it is recognized that only certain parts of F0 contours are linguistically relevant and should be represented phonologically and phonetically. Finally, empirical evidence as well as considerations of representational parsimony strongly suggest that these linguistically relevant aspects of F0 contours are best represented as levels rather than movements.

ACKNOWLEDGMENTS I thank Christina Lee and Tara Boswell for providing the American English examples in Figures 32.1 and 32.3, respectively.

REFERENCES Armstrong, Lilias E. & Ida Ward. 1926. Handbook of English intonation. Cambridge: Heffer. Arvaniti, Amalia. 2007. On the relationship between phonology and phonetics (or why phonetics is not phonology). In Jürgen Trouvain & William J. Barry (eds.) Proceedings of the 16th International Congress of Phonetic Sciences, 19–24. Saarbrücken: Saarland University. Arvaniti, Amalia. Forthcoming. Prosodic representations. (Part I: Segment-to-tone association). In Cohn et al. (forthcoming). Arvaniti, Amalia & Evangelia Adamou. Forthcoming. Focus expression in Romani. Proceedings of the West Coast Conference on Formal Linguistics 28. Arvaniti, Amalia & Mary Baltazani. 2005. Intonational analysis and prosodic annotation of Greek spoken corpora. In Jun (2005b), 84–117. Arvaniti, Amalia & Gina Garding. 2007. Dialectal variation in the rising accents of American English. In Jennifer Cole & José Ignacio Hualde (eds.) Laboratory phonology 9, 547–576. Berlin & New York: Mouton de Gruyter. Arvaniti, Amalia & D. Robert Ladd. 2009. Greek wh-questions and the phonology of intonation. Phonology 26. 43–74. Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 1998. Stability of tonal alignment: The case of Greek prenuclear accents. Journal of Phonetics 26. 3–25. Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 2000. What is a starred tone? Evidence from Greek. In Broe & Pierrehumbert (2000), 119–131. Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 2006a. Phonetic effects of focus and “tonal crowding” in intonation: Evidence from Greek polar questions. Speech Communication 48. 667–696. Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 2006b. Tonal association and tonal alignment: Evidence from Greek polar questions and contrastive statements. Language and Speech 49. 421–450. Baltazani, Mary. 2006. Focusing, prosodic phrasing, and hiatus resolution in Greek. In Goldstein et al. (2006), 473–493. Barnes, Jonathan, Stefanie Shattuck-Hufnagel, Alejna Brugos & Nanette Veilleux. 2006. The domain of realization of the L-phrase tone in American English. In Rüdiger Hoffmann & Hansjörg Mixdorff (eds.) Proceedings of Speech Prosody 2006. Dresden: TUDpress Verlag der Wissenschaften GmbH. Available (August 2010) at http:// aune.lpl.univ-aix.fr/~sprosig/sp2006/contents/papers/PS3-11_0163.pdf.

20

Amalia Arvaniti

Barnes, Jonathan, Nanette Veilleux, Alejna Brugos & Stefanie Shattuck-Hufnagel. 2010. The effect of global F0 contour shape on the perception of tonal timing contrasts in American English intonation. Proceedings of Speech Prosody 2010. Available (August 2010) at http://speechprosody2010.illinois.edu/papers/100445.pdf. Beckman, Mary E. & Janet B. Pierrehumbert. 1986. Intonational structure in Japanese and English. Phonology Yearbook 3. 255–309. Bishop, Judith & Janet Fletcher. 2005. Intonation in six dialects of Bininj Gun-wok. In Jun (2005b), 331–361. Bolinger, Dwight L. 1951. Intonation: Levels versus configurations. Word 7. 199–210. Bolinger, Dwight L. 1964. Around the edge of language: Intonation. Harvard Educational Review 34. 282–293. Reprinted in Bolinger (1972), 19–29. Bolinger, Dwight L. (ed.) 1972. Intonation: Selected readings. Harmondsworth: Penguin. Bolinger, Dwight L. 1986. Intonation and its parts: Melody in spoken English. London: Edward Arnold. Botinis, Antonis. 1989. Stress and prosodic structure in Greek: A phonological, acoustic, physiological and perceptual study. Lund: Lund University Press. Broe, Michael B. & Janet B. Pierrehumbert (eds.) 2000. Papers in laboratory phonology V: Acquisition and the lexicon. Cambridge: Cambridge University Press. Bruce, Gösta. 1977. Swedish word accents in sentence perspective. Lund: Gleerup. Bryant, Gregory A. & Jean E. Fox Tree. 2005. Is there an ironic tone of voice? Language and Speech 48. 257–277. Büring, Daniel. 2007. Intonation, semantics and information structure. In Gillian Ramchand & Charles Reiss (eds.) The Oxford handbook of linguistic interfaces, 445–473. Oxford: Oxford University Press. Calhoun, Sasha. 2010. The centrality of metrical structure in signaling information structure: A probabilistic perspective. Language 86. 1–42. Chen, Yiya. 2006. Durational adjustment under corrective focus in Standard Chinese. Journal of Phonetics 34. 176–201. Chen, Yiya. 2010. Post-focus F0 compression: Now you see it, now you don’t. Journal of Phonetics 38. 517–525. Cohen, Antonie & Johan ’t Hart. 1968. On the anatomy of intonation. Lingua 19. 177–192. Cohn, Abigail C., Cécile Fougeron & Marie Huffman (eds.) Forthcoming. The Oxford handbook of laboratory phonology. Oxford: Oxford University Press. Cooper, William E. & John Sorensen. 1981. Fundamental frequency in sentence production. Heidelberg: Springer. Crystal, David. 1972. The intonation system of English. In Bolinger (1972), 110–136. Dainora, Audra. 2001. An empirically based probabilistic model of intonation in American English. Ph.D. dissertation, University of Chicago. Dainora, Audra. 2006. Modeling intonation in English: A probabilistic approach to phonological competence. In Goldstein et al. (2006), 107–132. de Jong, Kenneth J. 1995. The supraglottal articulation of prominence in English: Linguistic stress as localized hyperarticulation. Journal of the Acoustical Society of America 97. 491–504. Dilley, Laura C. & Meredith Brown. 2007. Effects of pitch range variation on f0 extrema in an imitation task. Journal of Phonetics 35. 523–551. D’Imperio, Mariapaola. 2000. The role of perception in defining tonal targets and their alignment. Ph.D. dissertation, Ohio State University. D’Imperio, Mariapaola. Forthcoming. Prosodic representations. (Part II: Tonal alignment). In Cohn et al. (forthcoming). D’Imperio, Mariapaola & David House. 1997. Perception of questions and statements in Neapolitan Italian. Proceedings of the 5th European Conference on Speech Communication and Technology (Eurospeech 1997), vol. 1, 251–254. Rhodes, Greece.

The Representation of Intonation

21

D’Imperio, Mariapaola, Jacques Terken & Michel Pitermann. 2000. Perceived tone “targets” and pitch accent identification in Italian. In Michael Barlow (ed.) Proceedings of the 8th International Conference on Speech Science and Technology, 206–211. Canberra: Australian Speech Science and Technology Association. D’Imperio, Mariapaola, Barbara Gili Fivela & Oliver Niebuhr. 2010. Alignment perception of high intonational plateaux in Italian and German. Proceedings of Speech Prosody 2010. Available (August 2010) at http://speechprosody2010.illinois.edu/papers/ 100186.pdf. Elordieta, Gorka & José Ignacio Hualde. 2003. Tonal and durational correlates of accent in contexts of downstep in Lekeitio Basque. Journal of the International Phonetic Association 33. 195–209. Fletcher, Janet, Esther Grabe & Paul Warren. 2005. Intonational variation in four dialects of English: The high rising tune. In Jun (2005b), 390–409. Fougeron, Cécile & Sun-Ah Jun. 1998. Rate effects on French intonation: Prosodic organization and phonetic realization. Journal of Phonetics 26. 45–69. Frota, Sónia. 2002. Tonal association and target alignment in European Portuguese nuclear falls. In Gussenhoven & Warner (2002), 387–418. Fujisaki, Hiroya. 1983. Dynamic characteristics of voice fundamental frequency in speech and singing. In Peter F. MacNeilage (ed.) The production of speech, 39–55. New York: Springer. Fujisaki, Hiroya. 2004. Information, prosody and modeling – with emphasis on tonal features of speech. In Bernard Bel & Isabelle Marlien (eds.) Speech Prosody 2004. Available (August 2010) at www.isca-speech.org/archive/sp2004/sp04_001.pdf. Fujisaki, Hiroya, Sumio Ohno & Takashi Yagi. 1997. Analysis and modelling of fundamental frequency contours of Greek utterances. Proceedings of the 5th European Conference on Speech Communication and Technology (Eurospeech 1997), vol. 2, 465–468. Rhodes, Greece. Fujisaki, Hiroya, Changfu Wang, Sumio Ohno & Wentao Gu. 2005. Analysis and synthesis of fundamental frequency contours of Standard Chinese using the command–response model. Speech Communication 47. 59–70. Gårding, Eva. 1983. A generative model of intonation. In Anne Cutler & D. Robert Ladd (eds.) Prosody: Models and measurements, 11–25. Berlin: Springer. Gårding, Eva. 1987. Speech act and tonal pattern in standard Chinese: Constancy and variation. Phonetica 44. 13–29. Goldsmith, John A. 1976. Autosegmental phonology. Ph.D. dissertation, MIT. Published 1979, New York: Garland. Goldstein, Louis, Douglas Whalen & Catherine T. Best (eds.) 2006. Laboratory phonology 8. Berlin & New York: Mouton de Gruyter. Gooden, Shelome, Kath-Ann Drayton & Mary Beckman. 2009. Tone inventories and tune-text alignments: Prosodic variation in “hybrid” prosodic systems. Studies in Language 33. 354–394. Gordon, Matthew. 2005. Intonational phonology of Chickasaw. In Jun (2005b), 301–330. Gordon, Matthew. 2008. Pitch accent timing and scaling in Chickasaw. Journal of Phonetics 36. 521–535. Grabe, Esther, Brechtje Post, Francis Nolan & Kimberley Farrar. 2000. Pitch accent realization in four varieties of British English. Journal of Phonetics 28. 161–185. Grabe, Esther, Greg Kochanski & John Coleman. 2003. Quantitative modelling of intonational variation. Proceedings of Speech Analysis and Recognition in Technology, Linguistics and Medicine 2003. Available (August 2010) at www.phon.ox.ac.uk/oxigen/ publications.php. Grice, Martine. 1995a. The intonation of interrogation in Palermo Italian: Implications for intonation theory. Tübingen: Niemeyer. Grice, Martine. 1995b. Leading tones and downstep in English. Phonology 12. 183–233.

22

Amalia Arvaniti

Grice, Martine, D. Robert Ladd & Amalia Arvaniti. 2000. On the place of phrase accents in intonational phonology. Phonology 17. 143–185. Gu, Wentao, Keikichi Hirose & Hiroya Fujisaki. 2007. Analysis of tones in Cantonese speech based on the command–response model. Phonetica 64. 29–62. Gussenhoven, Carlos. 1999. Discreteness and gradience in intonational contrasts. Language and Speech 42. 283–305. Gussenhoven, Carlos. 2000. The boundary tones are coming: On the non-peripheral realization of boundary tones. In Broe & Pierrehumbert (2000), 132–151. Gussenhoven, Carlos. 2004. The phonology of tone and intonation. Cambridge: Cambridge University Press. Gussenhoven, Carlos & Natasha Warner (eds.) 2002. Laboratory phonology 7. Berlin & New York: Mouton de Gruyter. Halliday, M. A. K. 1967. Intonation and grammar in British English. The Hague: Mouton. Halliday, M. A. K. 1970. A course in spoken English: Intonation. Oxford: Oxford University Press. Harrington, Jonathan, Janet Fletcher & Mary E. Beckman. 2000. Manner and place conflicts in the articulation of accent in Australian English. In Broe & Pierrehumbert (2000), 40–51. Hart, Johan ’t & Antonie Cohen. 1973. Intonation by rule: A perceptual quest. Journal of Phonetics 1. 309–327. Hart, Johan ’t & René Collier. 1975. Integrating different levels of intonation analysis. Journal of Phonetics 3. 235–255. Hart, Johan ’t, René Collier & Antonie Cohen. 1990. A perceptual study of intonation: An experimental-phonetic approach to speech melody. Cambridge: Cambridge University Press. Hayes, Bruce & Aditi Lahiri. 1991. Bengali intonational phonology. Natural Language and Linguistic Theory 9. 47–96. Hirschberg, Julia & Gregory Ward. 1992. The influence of pitch range, duration, amplitude and spectral features on the interpretation of the rise–fall–rise intonation contour in English. Journal of Phonetics 20. 241–251. Hirst, Daniel & Albert Di Cristo. 1998. A survey of intonation systems. In Daniel Hirst & Albert Di Cristo (eds.) Intonation systems: A survey of twenty languages, 1–44. Cambridge: Cambridge University Press. Hirst, Daniel, Albert Di Cristo & Robert Espesser. 2000. Levels of representation and levels of analysis for the description of intonation systems. In Merle Horne (ed.) Prosody: Theory and experiment, 51–87. Dordrecht: Kluwer. Hockett, Charles F. 1955. A manual of phonology. Baltimore: Waverly Press. Hualde, José Ignacio, Gorka Elordieta, Inaki Gaminde & Rajka Smiljanio. 2002. From pitch accent to stress accent in Basque. In Gussenhoven & Warner (2002), 547–584. Jones, Daniel. 1972. An outline of English phonetics. 9th edn. Cambridge: Heffer & Sons. Jun, Sun-Ah. 2005a. Korean intonational phonology and prosodic transcription. In Jun (2005b), 201–229. Jun, Sun-Ah (ed.) 2005b. Prosodic typology: The phonology of intonation and phrasing. Oxford: Oxford University Press. Kingdon, Roger. 1958. The groundwork of English Intonation. London: Longmans, Green & Co. Knight, Rachael-Anne & Francis Nolan. 2006. The effect of pitch span on intonational plateau. Journal of the International Phonetic Association 36. 21–38. Ladd, D. Robert. 1980. The structure of intonational meaning: Evidence from English. Bloomington: Indiana University Press. Ladd, D. Robert. 1996. Intonational phonology. Cambridge: Cambridge University Press. Ladd, D. Robert. 2008. Intonational phonology. 2nd edn. Cambridge: Cambridge University Press.

The Representation of Intonation

23

Ladd, D. Robert, Ineke Mennen & Astrid Schepman. 2000. Phonological conditioning of peak alignment in rising pitch accents in Dutch. Journal of the Acoustical Society of America 107. 2685–2696. Ladd, D. Robert, Astrid Schepman, Laurence White, Louise May Quarmby & Rebekah Stackhouse. 2009. Structural and dialectal effects on pitch peak alignment in two varieties of British English. Journal of Phonetics 37. 145–161. Leben, William R. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT. Leben, William R. 1976. The tones of English intonation. Linguistic Analysis 2. 69–107. Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, MA: MIT Press. Liberman, Mark. 1975. The intonational system of English. Ph.D. dissertation, MIT. Liberman, Mark & Janet B. Pierrehumbert. 1984. Intonational invariance under changes in pitch range and length. In Mark Aronoff & Richard T. Oehrle (eds.) Language sound structure, 157–233. Cambridge, MA: MIT Press. Morén, Bruce & Elizabeth C. Zsiga. 2006. The lexical and post-lexical phonology of Thai tones. Natural Language and Linguistic Theory 24. 113–178. Myers, Scott. 2003. F0 timing in Kinyarwanda. Phonetica 60. 71–97. O’Connor, J. D. & G. F. Arnold. 1973. Intonation of colloquial English. London: Longman. Ohala, John J. 1983. Cross-language view of pitch: An ethological view. Phonetica 40. 1–18. Ohala, John J. & William G. Ewan. 1973. Speed of pitch change. Journal of the Acoustical Society of America 53. 345. Pan, Ho-Hsien. 2008. Focus and Taiwanese unchecked tones. In Chungmin Lee, Matthew Gordon & Daniel Büring (eds.) Topic and focus: Cross-linguistic perspectives on meaning and intonation, 195–213. Dordrecht: Springer. Pierrehumbert, Janet B. 1980. The phonology and phonetics of English intonation. Ph.D. dissertation, MIT. Pierrehumbert, Janet B. & Mary E. Beckman. 1988. Japanese tone structure. Cambridge, MA: MIT Press. Pierrehumbert, Janet B. & Julia Hirschberg. 1990. The meaning of intonational contours in the interpretation of discourse. In Philip R. Cohen, Jerry Morgan & Martha E. Pollack (eds.) Intentions in communication, 271–311. Cambridge MA: MIT Press. Pierrehumbert, Janet B. & Shirley Steele. 1989. Categories of tonal alignment in English. Phonetica 46. 181–196. Pike, Kenneth L. 1945. The intonation of American English. Ann Arbor: University of Michigan Press. Prieto, Pilar. 2005. Stability effects in tonal clash contexts in Catalan. Journal of Phonetics 33. 215 –242. Prieto, Pilar & Francisco Torreira. 2007. The segmental anchoring hypothesis revisited: Syllable structure and speech rate effects on peak timing in Spanish. Journal of Phonetics 35. 473–500. Prieto, Pilar, Chilin Shih & Holy Nibert. 1996. Pitch downtrend in Spanish. Journal of Phonetics 24. 445–473. Redi, Laura. 2003. Categorical effects in the production of pitch contours in English. In M. J. Solé, D. Recasens & J. Romero (eds.) Proceedings of the 15th International Congress of Phonetic Sciences, 2921–2924. Barcelona: Causal Productions. Rialland, Annie. 2007. Question prosody: An African perspective. In Carlos Gussenhoven & Tomas Riad (eds.) Tones and tunes, vol. 1: Typological studies in word and sentence prosody, 35–62. Berlin & New York: Mouton de Gruyter. Rietveld, Toni & Carlos Gussenhoven. 1995. Aligning pitch targets in speech synthesis: Effects of syllable structure. Journal of Phonetics 23. 375–385. Silverman, Kim E. A. 1987. The structure and processing of fundamental frequency contours. Ph.D. dissertation, Cambridge University.

24

Amalia Arvaniti

Silverman, Kim E. A. & Janet B. Pierrehumbert. 1990. The timing of prenuclear high accents in English. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 72–106. Cambridge: Cambridge University Press. Smiljanio, Rajka. 2006. Early vs. late focus: Pitch-peak alignment in two dialects of Serbian and Croatian. In Goldstein et al. (2006), 494–518. Steedman, Mark. 2000. Information structure and the syntax–phonology interface. Linguistic Inquiry 31. 649–689. Sundberg, Johan. 1979. Maximum speed of pitch changes in singers and untrained subjects. Journal of Phonetics 7. 71–79. Swerts, Marc, Emiel Krahmer & Cinzia Avesani. 2002. Prosodic marking of information status in Dutch and Italian: A comparative analysis. Journal of Phonetics 30. 629–654. Thorsen, Nina. 1980. A study of the perception of sentence intonation: Evidence from Danish. Journal of the Acoustical Society of America 67. 1014–1030. Thorsen, Nina. 1985. Intonation and text in Standard Danish. Journal of the Acoustical Society of America 77. 1205–1216. Thorsen, Nina. 1986. Sentence intonation in textual context: Supplementary data. Journal of the Acoustical Society of America 80. 1041–1047. Trager, George L. 1961. The intonation system of American English. In D. Abercrombie, D. B. Fry, P. A. D. MacCarthy, N. C. Scott & J. L. M. Trimm (eds.) In honour of Daniel Jones: Papers contributed on the occasion of his eightieth birthday, 266–270. London: Longmans, Green & Co. Reprinted in Bolinger (1972), 83–86. Trager, George L. & Henry L. Smith. 1951. An outline of English structure. Norman, OK: Battenburg Press. Venditti, Jennifer, Kikuo Maekawa & Mary E. Beckman. 2008. Prominence marking in the Japanese intonation system. In Shigeru Miyagawa & Mamoru Saito (eds.) The Oxford handbook of Japanese linguistics, 456–512. Oxford: Oxford University Press. Ward, Gregory & Julia Hirschberg. 1985. Implicating uncertainty: The pragmatics of fall–rise intonation. Language 61. 747–776. Xu, Yi. 1999. Effects of tone and focus on the formation and alignment of F0 contours. Journal of Phonetics 21. 55–105. Xu, Yi. 2005. Speech melody as articulatorily implemented communicative functions. Speech Communication 46. 220–251. Xu, Yi & Xuejing Sun. 2002. Maximum speed of pitch change and how it may relate to speech. Journal of the Acoustical Society of America 111. 1399–1413.

33

Syllable-internal Structure Anna R. K. Bosch

There is no simple discovery procedure for determining phonological syllable structure (which, like phonological representations in general, may not be in a one-to-one relationship with systematic phonetic syllabification, and which may not necessarily conform to native speaker intuitions about syllable division). The nature of the mechanism which assigns syllabification (defines possible syllables) for a given language is an empirical hypothesis, whose confirmation depends on the extent to which linguistically significant generalizations can be expressed under it (Feinstein 1979: 255).

1

Introduction

Although the syllable has been used by generations of linguists both in language description and in phonological theory, it is still surprising to see the variety of opinions and arguments on the topic. As Einar Haugen expostulated as early as 1956, “everyone talks about syllables, but no one seems to do anything about defining them” (Haugen 1956: 196). Since that time we have had fifty years or more of attempts to outline and define the syllable and its constituents, perhaps without coming much closer to solid agreement. Although the terms syllable, onset, rhyme, nucleus, and coda remain in common usage among phonologists, we cannot yet point to invariant acoustic or articulatory evidence for these constituents. As Haugen continues, “the only real basis for assuming their existence is that speakers of the language can utter them separately, dividing utterances into sequences that seem natural when pronounced alone.” We point to this evidence again and again as certain proof that there is “something” called the syllable; perhaps the everyday linguistic knowledge of the speaker is the single constant throughout phonological research on syllable structure. And yet, as Feinstein emphasizes in the quote appended above, we also draw a certain distinction between the “speaker’s syllable” and the “phonological syllable.” The phonological syllable is defined empirically by “linguistically significant generalizations,” while the speaker’s syllable is defined simply and automatically (when the decision is indeed simple and automatic) in careful speech, or by a number of – also empirical – experimental methods exploring external evidence, such as language games and other speaker behavior (chapter 96: experimental approaches in theoretical phonology). The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0033

2

Anna R. K. Bosch

As Goldsmith (1990) points out, there are at least two competing, or perhaps parallel, views of the syllable that have influenced phonological theory over the past century or more: just as light can behave as both wave and particle, the syllable too has been shown to demonstrate both a wave-like property based on sonority (chapter 49: sonority), and a piece-like, or chunk-like, division into smaller constituents, such as onset and rhyme. While there are still good reasons to hold on to a wave-shaped understanding of the syllable, defining the syllable and its properties with reference to the peaks and valleys of sonority shaping each one, the present chapter will nonetheless focus exclusively on theories of the syllable which specifically address the question of constituent structure. The question of syllable structure can be understood as a question about the nature of linguistic representation (cf. Anderson 1985); as Anderson claims, “most of the history of twentieth-century phonology is the history of theories of representations, devoted to questions such as ‘What is the nature of the phoneme, morphophoneme, morpheme, etc?’ ” (Anderson 1985: 9), to which we might add the question that concerns us here: “What is the nature of the syllable?” This chapter begins with a brief historical sketch tracing the early arguments in favor of different representations of syllable structure, followed by an overview of different models of the internal structure of the syllable. A final section reviews the conclusions of experimental studies as they adduce evidence for or against internal constituents of the syllable. (See also chapter 109: polish syllable structure; chapter 115: chinese syllable structure; chapter 56: sign syllables.)

2

Early twentieth-century discussions of syllable-internal structure

Early twentieth-century linguists asked themselves the same question, “what is the nature of the syllable?” Saussure proposes an impressionistic account of the syllable as composed of a succession of explosive and implosive articulatory movements; while all speech consists of an alternating series of these movements, the syllable boundary itself is marked by “the passage from an implosion to an explosion in a chain of sounds” (Saussure 1922). Arguing for the syllable as a unit of phonology, he reasons that “the regular coincidence of a mechanical principle and a definite acoustical effect assures the implosive-explosive combination of a right to existence in phonology” (1922: 57). Saussure goes so far as to claim that these opening and closing articulatory motions, which he distinguishes from acoustic sonority, are themselves the irreducible units of the syllable; further, a close examination of Saussure’s diagrams demonstrates that he considers the vocalic peak to form a part of the implosion, hinting at something like a rhyme: “Whenever a particular phoneme is more open than the following one, the impression of continuity persists; . . . an implosive link, like an explosive one, obviously can include more than two elements if each has wider aperture than the following one” (Saussure 1922: 56). Saussure’s footnote demonstrating the syllabification of the word particularly is noteworthy, presenting an early sketch of an onset-rhyme division within the syllable: a vowel and following tautosyllabic consonant are both “implosive,” according to Saussure’s terminology, while the prevocalic consonants are “explosive.” (Superscript arrow-heads indicating explosion [] are placed immediately above each alphabetic graph in Saussure’s text.)

Syllable-internal Structure (1)

3

< > > < >> < > < > > < > [par tik iu lar li]

KuryÓowicz develops a notion of syllable structure which was clearly influenced by Saussure, referring to “the initial (explosive) consonant group and the final (implosive) consonant group as they relate to the vocalic center”1 (KuryÓowicz 1949). Here KuryÓowicz quite explicitly associates the structure of both “semantic” and “phonic” systems, presenting a “table of correspondence,” or equivalence, which presages hierarchical structure within the syllable, as in (2), in parallel with the semantic functions of subject, predicate, etc. (2)

Table of correspondence (KuryÓowicz 1949, reprinted in Hamp et al. 1966: 230, translated A. R. K. Bosch) semantic system proposition predicate subject additional information

phonic system syllable vowel initial consonant group final consonant group, etc.

That is, just as a proposition consists of subject, predicate, and additional information, a syllable can be seen to consist of initial consonant group, vowel, and final consonant group. Peak and coda are grouped into a constituent in KuryÓowicz (1948), on the basis of co-occurrence restrictions which are found between peak and coda, but not between onset and peak. Although the term “vocalic peak” is already in use by the time of Saussure’s writing, Selkirk (1982) credits Hockett (1955) with the terms “onset” and “coda”; the use of “rhyme” to refer to the conjunction of peak and coda is attributed to Fudge (1969), although of course the informal usage of this term to describe poetic form dates from the seventeenth century.2

3

Evidence for constituents within the syllable

Reviewing arguments for syllable-internal structure from Saussure onward, we find that evidence for structure within the syllable is typically modeled on evidence for the syllable itself; in a comprehensive overview of syllable theories, Blevins (1995) outlines four traditional arguments in favor of the syllable itself as a constituent. She notes that (a) the syllable has been employed as the domain within which phonological processes or constraints may apply; (b) the syllable edge is identified as the locus for the application of processes or constraints; (c) the syllable itself may be picked out as a “target structure,” e.g. for the application of language games or for the assignment of stress or tone; and finally (d) field linguists recount that native speakers can express intuitions regarding

1

“le groupe consonantique initial (explosif) et le groupe consonantique final (implosif) par rapport au centre vocalique” [translation A. R. K. Bosch]. 2 The Oxford English Dictionary cites Samuel Butler (1663), “For Rhime the Rudder is of Verses, With which like Ships they stear their courses.”

4

Anna R. K. Bosch

the number of syllables per word or utterance. So, for example, nasalization may spread within a syllable (a); a syllable-final consonant may be devoiced (b); syllables may be independently manipulated in language games such as the French Verlan (c); finally, field linguists commonly report formal and informal studies of speakers who are easily able to count syllables, or who pause between syllables when exaggerating slow and careful speech (d). All these are common examples of the utility of the constituent “syllable”; however, not all of these arguments provide evidence for sub-syllabic constituents. Upon closer examination, only (a) and (c) usefully apply in evaluating syllableinternal constituents. First, as argued in (a), the constituents onset or rhyme have been argued to serve as phonological domains: Davis (1992) argues from Italian that the choice of the article (il or lo) depends on the constituent structure of the following onset (chapter 55: onsets; chapter 38: the representation of sc clusters). And the constituent structure of the rhyme – short vowel, long vowel, or vowel + consonant – may play a crucial role in stress assignment in quantitysensitive languages. In addition (as argued in (c)), the separate constituents onset and rhyme may be singled out as “target structures” for the application of language games. In the American children’s game “ubby dubby,” popularized by the 1970s television show Zoom!, the sequence [Hb] is inserted between each onset and rhyme: hello becomes [hHbelHboÁ]. Numerous language games play on the identification of onset and rhyme as target structures, and studies suggest that speech errors may operate on onset and rhyme sequences as single units (see discussion of experimental evidence, below). Thus evidence for sub-syllabic constituents derives primarily from data suggesting that onset and rhyme function as phonological domains or as target structures for other linguistic behavior. However, when we return to examine common evidence for the syllable as a constituent, parallel arguments for syllable-internal structure are not as convincing. Evidence such as (b) that refers to syllable-edges as targets (e.g. devoicing a syllable-final obstruent; chapter 69: final devoicing and final laryngeal neutralization) in fact says nothing about syllable-internal structure per se: a syllable boundary, without reference to constituent structure, could identify this position (see Steriade 1999, for example). And finally, evidence in (d) from slow or careful speech by native speakers may provide clues as to syllable count or syllable boundaries, but generally provides little insight into sub-syllabic constituents, without additional manipulations such as we find in studies of language games, etc. Nonetheless, we frequently uncover parallels between discussions of syllable structure and discussions of syllable-internal structure. This is made explicit within the framework of prosodic phonology (Selkirk 1982; Nespor and Vogel 1986): here the internal structure of the syllable is seen as a natural extension of the higherlevel prosodic structure, to which the syllable naturally belongs (see chapter 40: the foot; chapter 51: the phonological word; chapter 84: clitics; chapter 50: tonal alignment). After laying out arguments for the syllable as a constituent, Selkirk goes on to conclude that: The same three reasons leading to the postulation of the syllable can be shown to motivate the existence of privileged groupings of segments within the syllable which must be thought of as constituent-like linguistic units themselves . . . an internally-structured tree quite analogous to a tree representing syntactic structure (Selkirk 1982: 237).

Syllable-internal Structure

3.1

5

“Flat” models of syllable-internal structure

Various models of syllable-internal structure have been proposed over the past century of linguistic study, from an entirely flat structure consisting primarily of syllable boundary markers to a more highly articulated hierarchical structure. Kahn (1980), for example, proposes the simplest two-tier flat structure consisting of syllable nodes (q1, q2, etc.) on one tier, associated directly with the segments of phonetic (or phonological) representation, as in (3). (3)

No internal constituent structure (e.g. Kahn 1980) q1 æ

q2 t

l

s

H

atlas

For Kahn, the discrete segments are “associated” with the syllable node, and among his syllable-building principles is one akin to the no-crossing constraint of Goldsmith’s (1976) autosegmental phonology; “given the way the term ‘syllable’ is understood, it would seem nonsensical to speak of discontinuous syllables” (Kahn 1980: 36). Kahn explicitly cites Goldsmith manuscripts from 1974 and 1975, and in a footnote outlines his claim that he himself is working in an autosegmental framework, “because all theories of the syllable, including my own, are ‘autosegmental’ in that they involve parallel analyses of phonological material into (traditional) segments and syllables” (Kahn 1980: 61; see also chapter 14: autosegments). A contemporary version of Kahn’s flat structure is echoed in a recent textbook, Hayes (2009). Hayes takes a non-committal position on the internal structure of the syllable: while he prefers constituent (tree) structure to the simple use of boundary symbols to identify syllables, he makes no claim about constituency within the syllable. Introducing the terms onset, coda, and nucleus, he explains that In some theories, the onset, nucleus, and coda are described as constituents (they are daughters of the syllable node q, and dominate segments). This book will use “onset,” “nucleus,” and “coda” merely as useful descriptive terminology (Hayes 2009: 251).

In diagrams throughout this textbook, as in Kahn (1980), segments are dominated directly by the syllable node itself, without intervening structure. A related flat structure with an intervening CV tier is proposed by Clements and Keyser (1983) in (4) (see also chapter 54: the skeleton). (4)

Syllable with intervening CV tier (e.g. Clements and Keyser 1983) q C

V

C

k

æ

t

cat

Among the options that do incorporate some representation of internal structure, however, ternary branching structure represents perhaps the simplest

6

Anna R. K. Bosch

option, incorporating the nodes onset, peak, and coda, as in (5). The primary distinction between the examples in (3) and (5) lies in the use of the constituents onset, peak, and coda to function as hierarchical nodes of the syllable in the latter example; more than one segment may occur within a single constituent in (5), for example. (5)

Ternary branching, with internal structure (e.g. Hockett 1955) q Onset Peak Coda

While Kahn’s flat structure in (3) appears comparable to the ternary branching flat structure in (5), as employed by Hockett (1955) and others, the differences are considerable. Hockett presents an internal structure to the syllable, labeling the constituents onset, peak, and coda (he leaves the door open to other types of internal structure as well). In contrast to Hockett, Kahn simply refers to syllableinitial position, or syllable-final position, when stating phonotactic constraints defined by syllable positions. Thus, for Kahn, the [p] in support [sH.phort] is syllableinitial; we know this because the syllable-initial voiceless stop is aspirated [ph]. In contrast, the unaspirated [p] in asparagus is not syllable-initial, since the [s] is the first consonant of the syllable in the case of the vegetable: [H.spæ.rH.gHs] (Kahn 1980: 73). The aspiration of voiceless stops in English is precisely the type of evidence Kahn requires, as the syllable boundary alone can provide the environment for aspiration. An analysis which allows for internal constituent structure, such as that in (5), would place the [p] of asparagus in the syllable onset, of course, even if it is not syllable-initial (see also chapter 38: the representation of sc clusters). While Hockett proposes the syllable structure in (5), he also allows for a wide variety of different syllable models, suggesting in a “survey of syllable types” (Hockett 1955: 51ff.) that languages may employ different syllables and differing syllable “systems” according to the requirements of each system. He cites Bella Coola as an example of a language demonstrating the “onset type,” for example, since “while every (or almost every) syllable has a distinctive onset, many syllables contain no other syllable-element distinctively” (Hockett 1955: 57). As American linguists of the mid-twentieth century developed a taxonomic or distributional approach to language categories, the informal use of the syllable became commonplace as one means by which the distribution of phonemes could be expressed most economically;3 this was and remains today perhaps the primary use of the syllable as a unit in phonological theory and description. The constituents onset and coda played such a role for Pike and Pike’s (1947) analysis of 3

Compare Fischer-Jørgensen’s comment: “it may nevertheless be due to Bloomfield’s influence that most American linguists, even in short phonemic descriptions (such as the numerous descriptions of American Indian languages in the International Journal of American Linguistics), give a rather detailed statement of the syllabic structure of the language, and in this way present the material on the basis of which the phoneme categories may be established” (Fisher-Jørgensen 1952, reprinted in Hamp et al. 1966: 300).

Syllable-internal Structure

7

Mazateco, demonstrating what Selkirk calls the “immediate constituent” principle of phonotactics: that the constraints that hold between positions within the syllable are more tightly bound when those positions themselves form a constituent. Thus the constituents onset, peak, and coda – for Pike and Pike – are precisely those constituents within which we can describe such constraints. Similarly, Pike argues here and elsewhere (Pike and Pike 1947; Pike 1975a, 1975b) that phonological processes will refer to syllable constituents, rather than to individual segments; thus pitch and stress in Mazateco are defined with reference to the syllable peak. For example, in Mazateco contrastive tone is “limited exclusively” to the nucleus of the syllable; while tone will spread within the nucleus, it never spreads to the consonants in the “margins” of the syllable. Haugen’s (1956) analysis of the syllable in Kutenai details possible initial and final consonant clusters in that language, and elaborates whether medial clusters (the “interlude”) can properly be determined on the basis of permitted onset and coda clusters alone. Rand’s (1968) analysis of Alabaman syllables is specific in the disclaimer that “the syllable in Alabaman cannot be identified by any physical boundary feature.” Indeed, making use of both the wave model and hierarchical structure, he first defines the syllable in terms of sonority: there are as many syllables as there are peaks; “the determination of what occurs as peak is based on phonetics.” Later, however, he relies on hierarchical structure: “Stated another way, the syllable is an endocentric construction of two layers of immediate constituents, head (the vowel) and satellite (the consonant)” (1968: 97).

3.2

Hierarchical ordering of constituents within the syllable

Two different hierarchically structured binary-branching syllable types have been suggested, as in (6) and (7) below; by far the more conventional type is (6), in which the syllable consists of the constituents onset and rhyme, and the rhyme is formed by nucleus and coda. Arguments in favor of (7) have been proposed for languages such as Korean and Mandarin Chinese (McCarthy 1979; Yoon and Derwing 2001; Wang and Cheng 2008); in these examples the syllable is seen to consist of body + coda; the body forms a unit composed of onset and nucleus together.4 (6)

Binary branching with rhyme (e.g. from Pike and Pike 1947 onwards) Syllable Onset

Rhyme

Nucleus

4

Coda

A reviewer points out that this appears to contradict the traditional description of Chinese syllables in terms of “initials” and “finals” (cf. also Blevins 1995: 212; chapter 115: chinese syllable structure).

Anna R. K. Bosch

8 (7)

Binary branching with body (e.g. Yoon and Derwing 2001) Syllable Body

Coda

Nucleus

Onset

Levin (1985) proposed a variation of the branching structure in (6), a metrical theory of syllabic structure in which the syllable is a projection of the category “nucleus,” or N, illustrated in (8). Here the coda is represented as the complement, or right sister, of N, dominated by the first projection N′. The onset is the specifier of the syllable, dominated by the second projection N″. (8)

Metrical theory of syllable structure (Levin 1985) N″ Specifier

N′ N

Complement

Nucleus Finally, it should be noted that government phonology, as described by Kaye et al. (1990) and employed in Kaye (1990), Harris (1994), Botma et al. (2008), and various chapters in van der Hulst and Ritter (1999), makes use of syllable structure without the syllable, so to speak. The “syllabic constituents” onset (O) and rhyme (R) are not united into a single constituent (the syllable) in this analysis. The nucleus (N) is a constituent of the rhyme, but neither the coda nor the syllable is recognized as a constituent in government phonology. (Some versions of government phonology require CV syllables throughout, and posit an empty V where needed; see e.g. Scheer 2004.) (9)

Syllable structure without the syllable (Kaye et al. 1985) R O N x

3.3

x

x

Development of models of internal structure, from Fudge through Selkirk

In his influential article on the syllable, Fudge (1969) argues that the syllable has two primary functions as a linguistic universal: first, the syllable plays a role in

Syllable-internal Structure

9

the location of suprasegmental phenomena such as stress and tone; and second, it serves as the most appropriate unit for the formulation of phonotactic constraints. For Fudge, the internal structure of the syllable plays a key role. Fudge presents an analysis of the syllable in RP English, based in part on prior work on Chinese syllable structure (Firth and Rogers 1937; Hockett 1947). According to Fudge, the internal structure of the syllable, along with a detailed set of “collocational restrictions,” “clearly accounts in an appropriate way for the majority of the systematic restrictions on sound-sequences” (1969: 266ff.), or surface phonotactic constraints. The syllable consists of a hierarchical branching structure; below the labeled nodes onset, peak, and coda are numbered positions which play a role in the formulation of co-occurrence restrictions, as in (10). The phoneme inventory is specified according to what is possible, or permitted, for each syllable position. Thus, onset position 1 may include sC clusters as well as any of the non-syllabic phonemes of English; position 2 only allows the sonorants [w l r m n]. In addition to the sub-syllabic constituents onset and rhyme, Fudge argues for a “termination” node, which forms a part of word-final syllables only, and which permits only a small subset of segments, primarily morphemes such as past tense or nominative plural, and [-st] or [h].5 Later phonologists have also employed some version of a termination, or appendix, at word edges (see various articles in Féry and van de Vijver 2003), as discussed further below. (10)

The English syllable (Fudge 1969: 268) Syllable Onset 1

2

Rhyme Peak 3

(Termination)

Coda 4

6

5

Like KuryÓowicz (1948), Fudge argues against a flat syllable “on the basis that there is no means of stating relations between peak and coda, which we wish to do, while there is no such constraint between onset and peak” (1969: 273). So, for example, he details a number of constraints holding between positions 1 and 2 (i.e. within the onset); 4 and 5 (within the coda); and 3 and 4 (holding across peak and coda, within the rhyme). Paraphrased examples of Fudge’s collocational restrictions are given below (from Fudge 1969). (11)

Examples of phonotactic rules governing onsets a. b.

5

Within the onset, if the second position consists of [m] or [n], the first position can consist only of [s]. If the second position consists of [l], the first position may not consist of an alveolar (no [tl-, dl-, stl-, hl-]), except [s-].

Note that Fudge does not require his collocational restrictions to apply without exception; for a certain rule he also adds a footnote: “exceptions to this rule are not lacking (. . .). This does not detract from the value of stating the rule – even an ‘80% rule’ is well worth stating, provided that the exceptions to it are indicated” (Fudge 1969: 271).

10 (12)

Anna R. K. Bosch Example of phonotactic rules governing final clusters (paraphrased) Nasals form final clusters only with plosives and voiceless fricatives.

Fudge’s detailed elaboration of the phonotactic constraints of RP, based on syllable constituents, provided a model for later phonologists such as Selkirk (1982) and Goldsmith (1990). Nonetheless, hierarchical structure within the syllable has gone in and out of favor over the past fifty years. A number of articles throughout the 1970s argue for the importance of the syllable as a unit, but are not concerned with internal syllable structure (e.g. Hoard 1971; Hooper 1972; Vennemann 1972; Kahn 1980). Feinstein (1979), presenting a syllable-based analysis of prenasalized consonants in Sinhalese, employs the terms onset and coda, and yet remains explicitly neutral on the subject of syllable-internal structure. Instead, his analysis relies on identifying syllable boundaries only, without any more elaborate constituent structure. He employs “a phonological syllabification mechanism in which syllable structure is defined in terms of a discrete boundary within the linear string of segments,” although he admits that “other structural definitions may be more appropriate” (1979: 246). A range of essays from this period do not generally present specific evidence to support syllable-internal structure, though they make frequent use of the terms onset, rhyme, and coda, at least informally. Any argument which identifies a phonological syllable requires some means of identifying syllable boundaries, however, and in so doing we also confront evidence for permitted onset or coda clusters, almost by necessity. Arguments that attempt to lay out principles of syllable division in a particular language are at least implicitly arguments in favor of a particular syllable shape, VC-CVC vs. V-CCVC, for example. These have been taken to be arguments demonstrating internal syllable constituencies, such as arguments for a particular structure of the rhyme. If a specific intervocalic sequence of consonants is not a permissible onset cluster, one consonant may be forced into the coda of the previous syllable. Syllable structure thus regularly plays a role in analyses of quantity-sensitive metrical structure (chapter 57: quantity-sensitivity). For example, Kenstowicz (1994) points to differences in the syllabification of certain internal consonant clusters to explain certain facts of English stress patterns. Secondary stress falls on the first syllable of the words in (13a), while the words in (13b) show no stress on the initial syllable: (13)

English stress pattern differences (Kenstowicz 1994: 251) a.

Montana arcade Atlantic arthritic

‘mon.’tana ‘ar.’cade ‘at.’lantic ‘ar.’thritic

b.

America Nebraska atrocious astronomy

a.’merica ne.’braska a.’trocious a.’stronomy

The difference in patterns of secondary stress between the words in (a) and (b) above depends crucially on the fact that the word-initial syllables in (a) all terminate in a coda consonant, and therefore receive some stress; the word-initial syllables in (b) are open and thus unstressed. The consonant cluster “interludes” in (a) cannot form permissible word-initial clusters, and therefore it is claimed a syllable boundary must divide them. Thus what can count as a “possible onset” is ultimately defined in terms of what can serve as a possible word-initial cluster.

Syllable-internal Structure

11

This is essentially what Venneman (1972) terms the “Law of Initials”: that “medial syllable-initial clusters should be possible word-initial clusters” (Venneman 1972: 11); or what Bell (1976) terms “the KuryÓowicz condition,” that “initial and final clusters of medial syllables conform to the same constraints as those in initial and final syllables” (Bell 1976: 255).6 On the other hand, there are languages which place more restrictions on word-internal consonant clusters, and allow extra consonants at word edges, such as Polish (Kenstowicz 1994: 262ff.). A number of chapters in Féry and van de Vijver (2003) tease out just these inconsistencies between word-internal clusters and clusters found in either word-initial or word-final position (Cho and King 2003; Green 2003; Kiparsky 2003; Wiltshire 2003). See also Dixon (1970) and many others, as well as chapter 36: final consonants.

3.4

Moraic phonology and syllable-internal structure

While the syllable itself has been in wide use throughout the past century or more, clear arguments providing evidence of the internal structure of the syllable are less common than one might expect. Many arguments which state a convincing case for the syllable as a phonological unit in fact fail to motivate syllable-internal structure. The constituent structure of onset, rhyme, nucleus, and coda intersects in complex ways with a moraic theory of syllable organization, as sketched in (14) below. While a moraic analysis often appears to supersede one employing syllable structure, in fact the notion of moraic weight is interwoven with an understanding of syllable structure, in particular the structure of the rhyme. (14)

The mora in the prosodic hierarchy Prosodic Word Foot Syllable Mora

The mora provides a useful means of representing syllable weight, in quantitysensitive languages where this is required. In languages such as English or Latin, a syllable with a short vowel is monomoraic, while syllables with a long vowel (VV), or vowel + coda consonant (VC), are bimoraic. We note, however, that only consonants in the rhyme may be moraic; onset consonants never contribute to 6

Recall that in this article Bell argues against “the distributional syllable.” Nevertheless, he does not entirely conclude that phonology can do without the syllable: “Let us, however, guard against too narrow a view, against confusing a tool with the problem. ‘Defining the syllable’ and ‘proving the existence of the syllable’ are pseudo-problems. Segment organization is the problem. If assumption of a syllabic unit leads to explanation of regularities of segment organization, so much the better. If not, we will be awaiting a more general theory of organization, and the syllable may enter the museum’s Hall of Scientific Constructs, taking its place beside ether, the noble savage, and the like” (Bell 1976: 261).

12

Anna R. K. Bosch

syllable weight (chapter 47: initial geminates; chapter 55 onsets). Thus we return to some notion of constituent structure within the syllable, if only to identify the domain in which moras are projected. Furthermore, whether the mora truly serves as a “constituent” is unclear; while the syllabic nucleus is typically affiliated with a mora, the affiliations of onset and non-moraic coda consonants are less clear. Non-moraic elements are sometimes associated with the syllable node directly, or sometimes argued to share the mora with the nuclear vowel. Essays in Ziolkowski et al. (1990) demonstrate the range of arguments regarding moraic structure within the syllable. Hyman (1985) originally suggested that a syllable-initial consonant links to the mora of the following vowel, creating what looks like a “body–coda” structure. More commonly, the syllable-initial consonant is assumed to associate directly to the syllable node, as in Hayes (1989) and many others. For a more recent position favoring the mora, Yip argues explicitly that the evidence in favor of the constituents onset and rhyme “is scanty and inconsistent” (Yip 2003: 779), and relies on a moraic model of the syllable to account for the behavior of pre-nuclear glides in English and Mandarin Chinese. I leave it to other contributors to this Companion to tease out the intricacies of moraic phonology in more detail (see chapter 39: stress: phonotactic and phonetic evidence; chapter 40: the foot; chapter 41: the representation of word stress).

4

Experimental studies

Experimental studies, both of acoustic properties of speech and of human behavioral responses to syllabification tasks, have also been constructed to explore this question of the nature of the syllable. The great majority of experimental work over the past fifty years involves studies of simple syllabification, with a view to accounting for syllable boundaries. Most of these studies place a particular focus on the syllabification of an intervocalic consonant or consonants; see for example studies on Dutch, Finnish, German, French, Japanese, and English (e.g. Fallows 1981; Gillis and DeSchutter 1996; Schiller et al. 1997; Berg and Niemi 2000; Content et al. 2001; Goslin and Frauenfelder 2001; Ishikawa 2002; Redford and Randall 2005). Still, a number of experimental studies have been adduced to test the validity of the internal constituents of the syllable; the majority of these studies focus on the primary constituents of onset and rhyme. Evidence regarding hierarchical structure within the syllable is mixed, with arguments drawn from language games (both “traditional” and invented/experimental), slips of the tongue, perceptual studies, investigations with children, and other experimental paradigms (chapter 96: experimental approaches in theoretical phonology) to investigate whether onset, rhyme, nucleus, and coda are or are not syllable constituents. A series of experiments by Treiman and co-authors argue in favor of an onset– rhyme structure, based on subjects’ performance on various word games (Treiman 1986; Fowler et al. 1993; Treiman et al. 1994; Treiman et al. 1995). Most recently, Kapatsinski (2009) claims to show from an experimental study that English speakers are able to learn rhyme–affix associations more easily than body–affix associations, basing this argument on the claim that associations should be easier to learn within rather than across constituents, given a hierarchical structure to the syllable. However, Pierrehumbert and Nair (1995) replicate Treiman’s word game paradigm

Syllable-internal Structure

13

(Treiman 1983) and conclude that flat models of the syllable are sufficient to account for the results; see Treiman and Kessler (1995) for a response. Although few experimental studies look for evidence for constituent structure within the rhyme, one study by Hindson and Byrne (1997) found that children had less difficulty learning a word game which kept the final consonant cluster intact compared to one which broke it up. They conclude that these results support a model “which attributes internal hierarchical structure to the rime, with the coda as a constituent” (Hindson and Byrne 1997). Also focusing on constituent structure within the rhyme, Hayes et al. (2009) examine the ability of 9-month-old infants to detect change in vowel, in coda, or in both, based on the head-turn procedure. One experiment suggests that infants detect a difference most easily when both vowel and coda are altered – perhaps an unsurprising conclusion. However, the authors argue that pre-verbal 9-month-olds do parse syllables into units smaller than just onset and rhyme, presumably those of nucleus and coda. An additional area of research focuses on the relation between syllables, syllable constituency, and literacy (e.g. Derwing 1992). It is argued variously that syllable constituents may assist in visual word recognition (Álvarez et al. 2004) and the development of literacy; and that literacy itself influences phonological judgments about syllable structure (Treiman et al. 2002). Orthography was found to influence onset–rhyme segmentation in Portuguese, for example, in a CVC blending task: “the C/VC segmentation of pseudo-words and homophones was much more frequent in a context of words spelled CVC than in a context of mute-ewords” (Ventura et al. 2001). Orthography was also found to influence Pig Latin production by adult English speakers, when speakers trained on singleton and true cluster onsets extended their production to Cj- and sC(C)- clusters (Barlow 2001). Handwriting production tasks by French schoolchildren also indicated the influence of orthography on syllabification, again in particular where “silent e” is orthographically employed (Kandel et al. 2009). Criticism has occasionally been voiced that some of the experimental work cited above is methodologically flawed. Davis (1989) points out that many of these studies are based on speech errors or word games that involve only monosyllabic words; when polysyllabic words are considered, we find evidence not for an onset–rhyme distinction, but for a distinction between onset and “the remainder of the word,” or onset vs. “everything else.” Davis aims this criticism in particular at studies by Treiman (1983, 1986), who investigates language game productivity, and Fudge (1987), who argues for internal syllable structure primarily drawing on evidence from monosyllabic speech-error blends. Other studies have not found relevant distinctions between onset and rhyme under experimental contexts. Geudens and Sandra (Geudens and Sandra 2003; Geudens et al. 2005) conducted four experiments with Dutch children, both pre-readers and young readers, concluding that their subjects “did not treat onsets and rimes as cohesive units of the syllable.” Recent articulatory studies based on gestural analyses of speech in fact call into question the evidence for syllable constituents based on speech error and slip-of-the-tongue data. These criticisms target methods of data collection for speech error studies – transcription-based methods that have changed little since studies in the 1970s specifically termed them “field studies” (e.g. Fromkin 1973 and others). Laboratory-based studies, such as the gestural studies of speech errors as described in Pouplier (2007, 2008) seem to indicate that many errors are not

14

Anna R. K. Bosch

merely substitutions of segmental units – errors in “selection” – but instead may be examples of gestural intrusion or mis-timing (Pouplier and Goldstein 2005). Various studies conducted on languages other than English appear to show that if there is an onset–rhyme distinction it may be a language-specific one; experiments with native speakers of Korean indicate that Korean syllables “contain a cohesive CV or body unit, in contrast to the VC or rhyme unit of English” (Yoon and Derwing 2001); one of the five experiments described here studied preliterate children, indicating that literacy could not be a confounding factor. A study involving Chinese–English bilingual children found “a preference for matching body over rime in Chinese, and for matching rime over body in English,” concluding that there must be cross-language differences in processing spoken syllables (Wang and Cheng 2008). Acoustic studies have examined timing relationships within the syllable to identify syllable constituents. Conducting an acoustic study on English disyllables and casual speech vowel reduction, in triplets of words such as support/sport/s’pport (reduced support) Fokes and Bond (1993) conclude that there were no invariant acoustic cues determining syllabicity. While the authors concede that sport and s’pport may in fact be phonetically distinct, the study found no invariant cues to distinguish them. Certainly, attempts to isolate acoustic or articulatory invariants of the syllable date from as early as Stetson’s (1928) “chest pulse” theory; however, there is no current consensus on either acoustic or articulatory definitions of the syllable, let alone of structure internal to the syllable. While Selkirk (1982: 340) set the stage for a good deal of ensuing research with her note that “other phonological, or shall we say phonetic, phenomena such as duration and closeness of transition between segments might also be taken as revealing of the immediate constituent structure of the syllable,” we still find very little clear evidence of any invariant property pointing to syllable-internal hierarchical structure.

5

Conclusion

Despite the lack of phonetic evidence for invariant acoustic or articulatory measures of syllable structure, research in this area too continues apace. As Ladefoged noted: There is no single muscular gesture marking each syllable . . . (but) there is evidence . . . that speakers organize the sequences of complex muscular events that make up utterances in terms of a hierarchy of units, one of which is the size of a syllable; and it is certainly true that speakers usually know how many syllables there are in an utterance. We will therefore assume that a neurophysiological definition is possible, even if one cannot at the moment state it in any way (Ladefoged 1971: 81).

Even those who argue against the use of syllable structure to account for phonotactics acknowledge the usefulness of the terms referring to syllable-internal structure: onset, peak, coda, and even rhyme. “Syllable structure, whether directly perceived or inferred, is an undeniable aspect of phonological representations,” claims Steriade (1999), although she goes on to argue against employing syllable constituents in a phonological analysis, concluding that syllable position “does not condition segment realization.” Steriade argues instead that knowledge of

Syllable-internal Structure

15

syllable structure, and syllable edges in particular, derives from or is founded on the speaker’s perception of word-based phonotactic regularities. Her claim is essentially that we have put the cart before the horse in arguing that phonotactic constraints are built upon syllable structure; instead, these phonotactic regularities may be precisely what allow us to identify syllable position. In any case, the labels we use to identify internal constituents of the syllable – onset, coda, and rhyme – remain convenient terminology, and seem likely to remain in common usage. Nevertheless, it also seems clear that a conservative view of linguistic structure – a view shaped by Occam’s razor, perhaps – would concede that these terms, while useful, may not be supported by empirical evidence. Acoustic and experimental studies offer only mixed results, while language-specific phonological studies continue to differ widely in their use of (and claims for) some particular organization of syllable-internal structure. How we use syllable structure to represent the patterns and organization of human language will differ depending on the questions we ask and the problems we confront in the specific languages we investigate. Syllable structure may turn out to be an organizational tool, rather than an object available for independent manipulation.

REFERENCES Álvarez, Carlos, Manuel Carreiras & Manuel Perea. 2004. Are syllables phonological units in visual word recognition? Language and Cognitive Processes 19. 427–452. Anderson, Stephen R. 1985. Phonology in the twentieth century. Chicago: University of Chicago Press. Barlow, Jessica. 2001. Individual differences in the production of initial consonant sequences in Pig Latin. Lingua 111. 667–696. Bell, Alan. 1976. The distributional syllable. In Alphonse Juilland (ed.) Linguistic studies offered to Joseph Greenberg, vol. 2, 249–262. Saratoga, CA: Anma Libri. Berg, Thomas & Jussi Niemi. 2000. Syllabification in Finnish and German: Onset filling vs. onset maximization. Journal of Phonetics 28. 187–216. Blevins, Juliette. 1995. The syllable in phonological theory. In John A. Goldsmith (ed.) The handbook of phonological theory, 206–244. Cambridge, MA & Oxford: Blackwell. Botma, Bert, Colin J. Ewen & Erik Jan van der Torre. 2008. The syllabic affiliation of postvocalic liquids: An onset-specifier approach. Lingua 118. 1250–1270. Cho, Young-mee Yu & Tracy Holloway King. 2003. Semisyllables and universal syllabification. In Féry & van de Vijver (2003), 183–212. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Content, Alain, Ruth K. Kearns & Uli H. Frauenfelder. 2001. Boundaries versus onsets in syllabic segmentation. Journal of Memory and Language 45. 177–199. Davis, Stuart. 1989. On a non-argument for the rhyme. Journal of Linguistics 25. 211–217. Davis, Stuart. 1992. The onset as a constituent of the syllable: Evidence from Italian. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26. 71–79. Derwing, Bruce L. 1992. A pause-break task for eliciting syllable boundary judgments from literate and illiterate speakers: Preliminary results for 5 diverse languages. Language and Speech 35. 219–235. Dixon, R. M. W. 1970. Olgolo syllable structure and what they are doing about it. Linguistic Inquiry 1. 273–276. Fallows, Deborah. 1981. Experimental evidence for English syllabification and syllable structure. Journal of Linguistics 17. 309–317.

16

Anna R. K. Bosch

Feinstein, Mark. 1979. Prenasalization and syllable structure. Linguistic Inquiry 10. 245–278. Féry, Caroline & Ruben van de Vijver (eds.) 2003. The syllable in Optimality Theory. Cambridge: Cambridge University Press. Firth, J. R. & B. B. Rogers 1937. The structure of the Chinese monosyllable in a Hunanese dialect. Bulletin of the School of Oriental Studies 8. 1055–1074. Fischer-Jørgensen, Eli. 1952. On the definition of phoneme categories on a distributional basis. Acta Linguistica 7. 8–39. Fokes, Joann & Z. S. Bond. 1993. The elusive/illusive syllable. Phonetica 50. 102–123. Fowler, Carol A., Rebecca Treiman & Jennifer Gross. 1993. The structure of English syllables and polysyllables. Journal of Memory and Language 32. 115–140. Fromkin, Victoria A. (ed.) 1973. Speech errors as linguistic evidence. The Hague: Mouton. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5. 253–286. Fudge, Erik C. 1987. Branching structure within the syllable. Journal of Linguistics 23. 359–377. Geudens, Astrid & Dominiek Sandra. 2003. Beyond implicit phonological knowledge: No support for an onset–rime structure in children’s explicit phonological awareness. Journal of Memory and Language 49. 157–182. Geudens, Astrid, Dominiek Sandra & Heike Martensen. 2005. Rhyming words and onset– rime constituents: An inquiry into structural breaking points and emergent boundaries in the syllable. Journal of Experimental Child Psychology 92. 366–387. Gillis, Steven & Georges DeSchutter. 1996. Intuitive syllabification: Universals and language specific constraints. Journal of Child Language 23. 487–514. Goldsmith, John A. 1976. Autosegmental phonology. Ph.D. dissertation, MIT. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Goslin, Jeremy & Uli H. Frauenfelder. 2001. A comparison of theoretical and human syllabification. Language and Speech 44. 409–436. Green, Antony Dubach. 2003. Extrasyllabic consonants and onset well-formedness. In Féry & van de Vijver (2003), 238–253. Hamp, Eric P., Fred W. Householder & Robert Austerlitz. 1966. Readings in Linguistics II. Chicago: University of Chicago Press. Harris, John. 1994. English sound structure. Oxford: Blackwell. Haugen, Einar. 1956. Syllabification in Kutenai. International Journal of American Linguistics 22. 196–201. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce. 2009. Introductory phonology. Malden, MA & Oxford: Wiley-Blackwell. Hayes, Rachel A., Alan M. Slater & Christopher A. Longmore. 2009. Rhyming abilities in 9-month-olds: The role of the vowel and coda explored. Cognitive Development 24. 106–112. Hindson, Barbara Anne & Brian Byrne. 1997. The status of final consonant clusters in English syllables: Evidence from children. Journal of Experimental Child Psychology 64. 119–136. Hoard, James E. 1971. Aspiration, tenseness, and syllabication in English. Language 47. 133–140. Hockett, Charles F. 1947. Peiping phonology. Journal of the American Oriental Society 67. 253–267. Hockett, Charles F. 1955. A manual of phonology. Baltimore: Waverly Press. Hooper, Joan B. 1972. The syllable in phonological theory. Language 48. 525–540. Hulst, Harry van der & Nancy Ritter (eds.) 1999. The syllable: Views and facts. Berlin & New York: Mouton de Gruyter. Hyman, Larry M. 1985. A theory of phonological weight. Dordrecht: Foris. Ishikawa, Keiichi. 2002. Syllabification of intervocalic consonants by English and Japanese speakers. Language and Speech 45. 355–385. Kahn, Daniel. 1980. Syllable-based generalizations in English phonology. New York: Garland.

Syllable-internal Structure

17

Kandel, Sonia, Lucie Hérault, Géraldine Grosjacques, Eric Lambert & Michel Fayol. 2009. Orthographic vs. phonologic syllables in handwriting production. Cognition 110. 440–444. Kapatsinski, Vsevolod. 2009. Testing theories of linguistic constituency with configural learning: The case of the English syllable. Language 85. 248–277. Kaye, Jonathan. 1990. “Coda” licensing. Phonology 7. 301–330. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1985. The internal structure of phonological elements: A theory of charm and government. Phonology Yearbook 2. 305–328. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7. 193–231. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Kiparsky, Paul. 2003. Syllables and moras in Arabic. In Féry & van de Vijver (2003), 147–182. KuryÓowicz, Jerzy. 1948. Contribution à la théorie de la syllabe. Bulletin de la Société Polonaise de Linguistique 8. 80–113. KuryÓowicz, Jerzy. 1949. Linguistique et théorie du signe. Journal de Psychologie 42. 170–180. Ladefoged, Peter. 1971. Preliminaries to linguistic phonetics. Chicago: Chicago University Press. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. McCarthy, John J. 1979. On stress and syllabification. Linguistic Inquiry 10. 443–465. Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Pierrehumbert, Janet B. & Rami Nair. 1995. Word games and syllable structure. Language and Speech 38. 77–114. Pike, Kenneth L. 1975a. Suprasegmentals in reference to phonemes of item, of process, and of relation. Bibliotheca Phonetica 11. 45–56. Pike, Kenneth L. 1975b. Tests for prosodic features of pitch, quantity, stress. Bibliotheca Phonetica 11. 4–5. Pike, Kenneth L. & Eunice V. Pike. 1947. Immediate constituents of Mazateco syllables. International Journal of American Linguistics 13. 78–91. Pouplier, Marianne. 2007. Tongue kinematics during utterances elicited with the SLIP technique. Language and Speech 50. 311–341. Pouplier, Marianne. 2008. The role of a coda consonant as error trigger in repetition tasks. Journal of Phonetics 36. 114–140. Pouplier, Marianne & Louis Goldstein. 2005. Asymmetries in the perception of speech production errors. Journal of Phonetics 33. 47–75. Rand, Earl. 1968. The structural phonology of Alabaman, a Muskogean language. International Journal of American Linguistics 34. 94–103. Redford, Melissa A. & Patrick Randall. 2005. The role of juncture cues and phonological knowledge in English syllabification judgments. Journal of Phonetics 33. 27–46. Saussure, Ferdinand de. 1922. A course in general linguistics. New York: Philosophical Library. Scheer, Tobias. 2004. A lateral theory of phonology, vol. 1: What is CVCV, and why should it be? Berlin & New York: Mouton de Gruyter. Schiller, Niels O., Antje S. Meyer & Willem J. Levelt. 1997. The syllabic structure of spoken words: Evidence from the syllabification of intervocalic consonants. Language and Speech 40. 103–140. Selkirk, Elisabeth. 1982. The syllable. In Harry van der Hulst & Norval Smith (eds.) The structure of phonological representations, part II, 337–383. Dordrecht: Foris. Steriade, Donca. 1999. Alternatives to syllable-based accounts of consonantal phonotactics. In Osamu Fujimura, Brian D. Joseph & Bohumil Palek (eds.) Item order in language and speech, 205–245. Prague: Karolinum Press. Stetson, Raymond H. 1928. Motor phonetics. Haarlem: Société Hollandaise de Science. Treiman, Rebecca. 1983. The structure of spoken syllables: Evidence from novel word games. Cognition 15. 49–74.

18

Anna R. K. Bosch

Treiman, Rebecca. 1986. The division between onsets and rimes in English syllables. Journal of Memory and Language 25. 476–491. Treiman, Rebecca & Brett Kessler. 1995. In defense of an onset–rime syllable structure for English. Language and Speech 38. 127–142. Treiman, Rebecca, Kathleen Straub & Patrick Lavery. 1994. Syllabication of bisyllabic nonwords: Evidence from short-term memory errors. Language and Speech 37. 45–59. Treiman, Rebecca, Carol A. Fowler, Jennifer Gross, Denise Berch & Sarah Weatherston. 1995. Syllable structure or word structure? Evidence for onset and rime units with disyllabic and trisyllabic stimuli. Journal of Memory and Language 34. 132–155. Treiman, Rebecca, Judith A. Bowey & Derrick Bourassa. 2002. Segmentation of spoken words into syllables by English-speaking children as compared to adults. Journal of Experimental Child Psychology 83. 213–238. Vennemann, Theo. 1972. On the theory of syllabic phonology. Linguistische Berichte 18. 1–18. Ventura, Paulo, Régine Kolinsky, Carlos Brito-Mendes & José Morais. 2001. Mental representations of the syllable internal structure are influenced by orthography. Language and Cognitive Processes 16. 393–418. Wang, Min & Chenxi Cheng. 2008. Subsyllabic unit preference in young Chinese children. Applied Psycholinguistics 29. 291–314. Wiltshire, Caroline. 2003. Beyond codas: Word and phrase-final alignment. In Féry & van de Vijver (2003), 254–268. Yip, Moira. 2003. Casting doubt on the onset-rime distinction. Lingua 113. 779–816. Yoon, Yeo Bom & Bruce L. Derwing. 2001. A language without a rhyme: Syllable structure experiments in Korean. Canadian Journal of Linguistics 46. 187–237. Ziolkowski, Michael, Manuela Noske & Karen Deaton (eds.) 1990. Papers from the 26th Annual Regional Meeting, Chicago Linguistic Society: Parasession on the syllable in phonetics and phonology.

34

Precedence Relations in Phonology Charles Cairns Eric Raimy

1

Introduction

“Precedence” in phonology refers to the fact that elements occur in ordered sequences. An understanding of precedence relations is key to explicating notions of locality, adjacency, and left–right asymmetries, which have played significant roles in the phonological literature, especially since the advent of autosegmental phonology (Goldsmith 1976; chapter 14: autosegments). McCarthy (1989: 71) writes “nonlinear phonology imposes strict requirements of locality on phonological rules,” and “locality . . . ensures that the elements referred to in phonological transformations and constraints are adjacent at some level of representation.” We must agree on a common set of questions in order to compare and contrast linguists’ notions of precedence relations, a requirement that presupposes a formal framework. Since the earliest days of phonology, the sequential nature of speech and the progression of letters across the printed page have been assumed to be sufficient to understand precedence. It is always salutary to explicate tacit assumptions, so this chapter explores the implications of a formally rigorous understanding of precedence. The formal rigor is supplied by graph theory, a branch of mathematics that we will use to unpack the question of what it means for phonemes to appear in an ordered sequence (Wilson 1996 is one of several good introductions). Phonology is concerned with the characteristics of precedence in human language, so graph theory itself can be no more than a useful tool. But because graph theory is an explicit mathematical model that provides specific and well-understood possible answers to questions of precedence, we explore its implications in this chapter. Once we have introduced the relevant aspects of graph theory, we go on to examine a sample of the claims and assumptions that have been made in the literature on phonological precedence, with varying degrees of explicitness. In particular, we examine the characterizations of various approaches to autosegmental phonology within graph theory. We will further explicate the nature of precedence in phonology by considering the basic operation of deletion. All phonologists must take as bedrock assumptions that phonemes appear in a sequence and that there exist phonological processes with the capacity to delete phonemes from a sequence. One interesting result of this exercise is that commonly assumed antagonistic theories of phonology converge on a common model of deletion. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0034

2

Charles Cairns & Eric Raimy

Before proceeding, we first ask if precedence relations are primitives of phonological representations or if they are derived. Van der Hulst (2008), for example, proposes that precedence relations can be derived from underlying syllables. Consider a highly articulated theory of the syllable with labeled nodes, e.g. Fudge (1969, 1987); Cairns (1988); see also chapter 33: syllable-internal structure. The idea is that if at the lexical level featural information such as [back] or [coronal] were stored in syllabic nodes like onset and rhyme, then perhaps the number and order of phonemes on the surface could be predicted from the inherent order of syllabic constituents. There are at least four natural limits to deriving precedence relations from syllable structure. First, syllabification is not always exhaustive; many languages are known to have sequences of unsyllabified consonants (Bagemihl 1991; Czaykowska-Higgins and Willett 1997; Vaux and Wolfe 2009). Second, it would in any case be necessary to specify the order among syllables. Consider the English loan from Tamil catamaran; hypothetical *matacaran and *tamacaran would serve as equally plausible loans into English, yet they differ from the existing word only by syllable order. The possibility of using foot structure to order syllables only moves this question higher in the prosodic hierarchy and requires more prosodic information to be stored in the lexicon. The necessity of stipulating explicit sequencing information in the lexicon cannot be escaped. Third, reference to precedence relations at the segmental level is necessary to properly account for phonotactic constraints. Blevins (2003), building on Steriade (1999), shows numerous compelling examples where phonological sequencing generalizations and cross-linguistic universal patterns refer to properties of the phonological string and not to syllable structure. Of course, as Blevins points out, there are many cases “where phonotactic constraints and syllable structure appear to converge.” She suggests that “this is because syllabifications are derivative of phonotactics, not vice versa” (2003: 393). Finally, we are encouraged in our focus on segmental precedence by the fact that resyllabification is rampant throughout phonology and phonetics; the effervescence of the syllable makes it a poor candidate for the bearer of lexical precedence relations. §2 sketches the basic elements of graph theory and their application to phonology. §3 presents the complications that arise in connection with considering precedence relations in autosegmental phonology and provides a plausible explanation of why the Obligatory Contour Principle (OCP) is so variably valid. §4 demonstrates how investigating the process of deletion in graph theory illuminates how different theoretical models of phonology converge on the same understanding of deletion in phonology. §5 demonstrates how theories of phonology that view segments as entities that occur in real time can benefit from the consideration of precedence in graph-theoretic terms. §6 shows how graph theory can illuminate issues in phonology such as the “no crossing constraint” and local vs. long-distance adjacency. §7 illustrates some extensions of graph theory approaches to different phonological and morphological phenomena. §8 concludes this chapter.

2

Precedence relations and graph theory

Phonologists differ about whether to view a phonological sequence as a string of discrete, point-like objects or as a series of possibly overlapping segments which exist in real time; Trubetzkoy (1939) described phonemes as essentially timeless

Precedence Relations in Phonology

3

(zeitlos), abstract entities like the dits and dots of Morse code, where each unit is unaffected by the producing process (although possibly influenced by proximal symbols). The alternative view, that phonological patterning is defined by the dynamic articulation of actual speech, goes back to Sweet (1877), Sievers (1881), and Saussure (1916). This serves as the fundamental assumption of such schools of thought as Articulatory Phonology (Browman and Goldstein 1986, 1989, 1990a, 1990b; chapter 5: the atoms of phonological representations), which views the elements that appear in phonological sequences to be events in real time. The consensus that phonetic sequences involve continuous, overlapping elements does not extend to more abstract levels of phonology. Theories of morphological operations such as reduplication and infixation and of many phonological processes such as morphophonemics, deletion, etc., generally operate on representations made up of discrete, point-like objects, and these theories have achieved considerable descriptive success; we will proceed on this basis and turn to the event-based outlook later. Consider the representation of the Margi word /tágú/ ‘horse’ (from Kenstowicz 1994: 312) in a generic version of autosegmental phonology in (1). This is a picture of what Coleman and Local (1991: 309) call “paddlewheel graphs.” It is drawn so as to induce the reader to visualize three half planes, each emanating from a common line. The three-dimensional metaphor is useful because it helps us break down general questions of precedence into smaller and hence more manageable ones. This imagery can be misleading, however, because, as Coleman and Local (1991) demonstrate, autosegmental representations are not necessarily threedimensional (i.e. non-planar, as defined in §6) in a mathematical sense. (1)

Paddlewheel graph of Margi /tágú/ syllable plane q #

x t

x a

x g

q x

%

u feature plane

H tone plane Lines are referred to as “tiers” in autosegmental phonology, and the line defined by the intersection of the three planes depicted in (1) is the “anchor tier.” The anchor tier contains the string of symbols # x x x x %, where # and % indicate the beginning and end, respectively, of the phoneme sequence (we ignore these now, but return to them below). Elements are also arrayed in tiers on the “feature plane,” the “syllable plane,” and the “tone plane” of (1). The feature plane is shown with the alphabetic symbols t, a, g, u in lieu of the familiar feature trees (we employ this notation throughout the remainder of the chapter); in §3 we will see that this is more complicated than

4

Charles Cairns & Eric Raimy

a simple plane. These elements are connected by “association lines” to elements on the anchor tier. Autosegmental phonology since Goldsmith (1976) has taken for granted that different distinctive features may be separated onto distinct tiers on the feature plane, and that all tiers are parallel to the anchor tier. Many phonologists posit a rich prosodic hierarchy on what we dub the syllable plane, but we eschew further discussion of this plane (see Cairns and Raimy 2009). A third plane represents tone; Kenstowicz (1994) gives the familiar arguments that the high tone in tágú is represented by a single element on the “tonal tier,” and this element is connected by association lines to two segments on the anchor tier. Phonologists differ on what elements inhabit each plane and tier, how they are organized with respect to each other on each plane, and how representations on different levels interact with one another. For example, the field has not agreed whether the elements on the anchor tier are root nodes, moras, or empty x-slots (see chapter 54: the skeleton). We will achieve our expository purposes by consistently representing anchor elements as x-slots under the belief that pretty much the same accounts we provide below could be supplied under mora theory, mutatis mutandis. The remainder of this section will introduce the formal logic of precedence relations with a focus only on the anchor tier; we will turn to representational implications of positing multiple tiers on several planes in the next section. In fact, if we were to analyze only the anchor tier and ignore the others we would be discussing only a sequence such as x x x x. Because this would make for awkward exposition, we indulge in the convenience of labeling the anchor tier elements with the familiar phoneme labels, suppressing information from the tone plane; we turn to how the features and tones are connected to their appropriate x-slots when we discuss precedence in the context of autosegmental representations in §3. The articulation of the phoneme sequence tagu obeys the principles of asymmetricity and irreflexivity. Asymmetricity means that if t precedes a then a does not precede t, assuming that there are single instances of t and a. The constraint of irreflexivity applies because no speech sound may precede or follow itself (see Bird and Klein 1990 for an alternative view). These principles follow from formal properties of ordered sequences of elements, as we will show later in this section. As will be explained below, whether transitivity holds of this utterance depends on whether we understand precedence to apply locally or non-locally. If precedence is local, by which we mean that only adjacent segments can be “seen,” then the phoneme sequence is not transitive; if t precedes a and a precedes g then t cannot precede g. If adjacency is non-local, so it can see beyond immediately adjacent segments, then precedence is transitive; if t precedes a and a precedes g then t must precede g. We see from this reasoning that the question of whether phoneme strings obey the principle of transitivity boils down to the question of locality. These principles are best understood in terms provided by graph theory, so we defer further exploration of asymmetry, irreflexivity, and transitivity until after a brief introduction to graph theory as it applies to phonology. Graphs are defined as sets of “vertices” and “edges,” as shown in (2) (see again Wilson 1996). Graph-theoretic representations of phoneme sequences consider vertices to stand for phonological segments and edges to symbolize the precedence relations among them; accordingly, whenever we use the term vertex, the reader can understand it as equivalent to a phonological segment. Consider the two different graph-theoretic representations for the word tagu in (2a) and (2b).

Precedence Relations in Phonology (2)

a. b.

t—a—g—u vertices: edges (unordered):

5

{t, a, g, u} {t a, a g, g u}

(2a) and (2b) contain the same information. (2a) depicts the graph in a visual manner, while (2b) defines the graph as a list of vertices and a list of edges. Each edge is defined by a pair of vertices, and these pairs are set off typographically from each other by commas in (2b). A list of edges suffices for many graphs, because the vertices can be determined from the list of edges. Note that the lists of vertices and edges are literally lists and not sets. Because the edges in (2b) are undirected, it is called an undirected graph, where the two vertices that define each edge are unordered with respect to each other; for example, the edge {t a} is equivalent to {a t}. Because this type of graph contains information only about adjacency of vertices and does not specify any order, it appears to be a poor candidate for a model to represent phoneme strings. For one thing, lexical representations must contain ordering information. Consider the existence of pairs of words like cat and tack in English. The adjacency pairs (undirected edges) for these two words are identical; both words have the edge set {t æ, æ k}. Beyond the obvious fact that ordering is distinctive is the observation made by de Lacy (2007) that if phonological graphs were undirected, we would predict that mirror-image rules or constraints like those that appeared in early versions of SPE-type phonology would be commonplace. An example would be rules of the form x → y / {z __ , __ z}, as suggested by Bach (1968), Langacker (1969), and Anderson (1974). This means that x is rewritten y if it either precedes or follows z. If we were to adopt undirected graphs as the representation for phonological forms we would predict, contrary to fact, that mirror operations should be the most common operations found in phonology. Because a theory containing only undirected edges cannot distinguish between cat and tack and makes false predictions about the directionality of phonological operations, we consider graphs where the vertices that specify each edge are ordered with respect to each other. Information about sequential order can be added to the graph in (2) by making it a “directed graph” (or “digraph”), as in (3). An edge in a digraph is an ordered pair of vertices; the vertex mentioned first in an edge specification precedes the one mentioned second. (3)

a. t → a → g → u b. vertices: edges (ordered): c. a u t g

{t, a, g, u} {t a, a g, g u}

(3a) is the graphic representation of the information in (3b). Because the edges in (3b) are ordered, they specify that t precedes a, a precedes g, and g precedes u. (3a) is not the only diagram consistent with (3b), however; for example, the diagram in (3c) is equally consistent with (3b), but not as convenient to read as (3a). They are formally equivalent; in fact, (3a), (3b), and (3c) all represent the same

6

Charles Cairns & Eric Raimy

graph. All essential characteristics of graphs are given by specifying the set of edges and the set of vertices, such as in (3b). For expedience, when we portray graphs we will continue to supplement the lists of vertices and edges with convenient diagrams and indicate that the graph is directed by using arrows as edges. The graph in (3) represents a kind of digraph that has properties that match widely held yet tacit assumptions about precedence in phonological representations. For example, it is a “connected graph.” In a connected graph, it is possible to construct a route following edges between each two vertices. In other words, a connected graph contains only vertices that share an edge with at least one other vertex. Determining whether a graph is connected or not consists of traversing the graph by moving from one vertex to another, following edges. The traversing of a graph is called a “walk,” and a connected graph is one where there is a possible walk from any vertex to any other vertex. The requirement that the phoneme strings that constitute the phonological representation of words be represented by a connected graph is an explication within graph theory of the usually tacit assumption that words be pronounceable from beginning to end. The graph in (3) exemplifies a “chain graph”; properties of chain graphs, like those of connected graphs in general, illuminate our understanding of asymmetry and irreflexivity as constraints on phonological sequences, as we now demonstrate. The definition of chain graphs involves a concept known as the degree of a vertex, which is the number of edges that it appears in. In (3), there are two vertices with a degree of one; all others have a degree of exactly two. Because the vertex t precedes only a and is not preceded by any other vertex, it appears in only one edge, {t a}; conversely, because u is preceded by only one vertex, g, and does not precede any other vertex, it also appears in only one edge, {g u}. Therefore, the two vertices t and u are of degree one. The remaining vertices, a and g, each have degree two; a follows t and precedes g while g follows a and precedes u, so each must appear in two edges. If a connected graph has two vertices of degree one, and if any other vertices have a degree of exactly two, it is a chain graph, and if it is also a digraph it is known as a “directed chain graph.” Directed chain graphs must obey the principle of asymmetry, because the requirement that edges are directed and the limits on the degrees of vertices together entail that if a graph like (3) has an edge such as {a g}, it cannot also have an edge {g a}; such a move would make the vertices {a} and {g} appear in three edges, violating a requirement of chain graphs. Directed chain graphs also obey the principle of irreflexivity for essentially the same reason. A reflexive edge is one where the same vertex appears twice in the definition of an edge; if g, for example, were to follow (or precede) itself, then there would be an edge with the specification {g g}. Adding this edge to the set in (3b) would make the degree of g three, because it would now appear in three edges. It follows from the preceding that the principles of asymmetry and irreflexivity follow from properties of directed chain graphs. The same is also true of the notion of transitivity, which, as mentioned above, can be considered locally as well as non-locally. Locally means that only edges (i.e. adjacent vertices) are considered, so precedence relations are not transitive from a local perspective. From a more long-distance perspective, of course, graphs like those in (3) must be considered to be transitive; there is a clear sense in which, for example, t precedes g and u. This sense of transitivity follows from the requirement that there be a walk through the graph; the walk through (3) traverses the vertex t before it reaches

Precedence Relations in Phonology

7

g or u, so in this sense it is true that if t precedes a and a precedes g, then t precedes g. Before concluding this section, note that it is convenient to add explicit beginning and end symbols, which we represent as # and %, respectively. Positing these symbols allows us to say that all vertices on the anchor tier that are of degree two are available for serving as phonological segments, and only the abstract terminal symbols are of degree one. These symbols are also convenient to define the environments of initial and final segments, needed by many phonological processes. Also, chain graphs supplemented with special beginning and end symbols allow for the definition of free vs. bound morphemes: a free morpheme like tagu is a chain graph with # and % at beginning and end; a bound morpheme would lack one or both of these symbols. The graph in (4) is the type we will use to explicate precedence issues when we consider deletion in §4. (4)

a. b.

#→t→a→g→u→% vertices: {#, t, a, g, u, %} edges (ordered): {# t, t a, a g, g u, u %}

This section has described the basic principles of graph theory as they apply to notions of phonological representations held by virtually all schools of thought in phonology. The overwhelming majority of phonologists operate on the assumptions, usually tacit, that sequences of phonemes have the properties of asymmetry, irreflexivity, and (in the qualified senses explicated above) transitivity, and the preceding paragraphs have shown that these are best understood in graph theory, where they derive from the properties of directed chain graphs. However, as stated at the beginning of this section, we have been considering only precedence relations on the anchor tier of (1). We now turn to a consideration of precedence relations in autosegmental phonology.

3

Precedence relations and autosegmental phonology

In the preceding section we adopted the expository convenience of depicting phonological content of segments on the anchor tier by means of phoneme symbols. This was a shorthand way of showing feature trees, sketched in (5), which is an elaboration of the diagram in (1). This section is devoted to analyzing the precedence relations among elements that are on different tiers, and we will see that this exercise provides insight into the OCP. A brief explanation of (5) is in order first. (5)

Schematized autosegmental representation of [tagu] # → x

→

x

→

x

→

x →%

laryngeal tier place tier

COR

DOR

VEL

DOR

feature tier

[ant]

[low]

[bk]

[hi]

anchor tier root tier

8

Charles Cairns & Eric Raimy

(5), like (1), is drawn so as to invite the reader to visualize a three dimensional representation. The anchor tier is the same as that in (1), except that the directed edges are shown. The branching below the root tier in (5) depicts the autosegmental nature of features. We assume the general proposals on feature geometry (see chapter 27: the organization of features) from Goldsmith (1990) and Archangeli and Pulleyblank (1994). The representation in (5) raises the question of whether precedence is encoded only on the anchor tier or whether each tier has its own precedence relations. The difference between these two options can be graphically seen by comparing (6) with (5). (5) contains precedence relations (indicated by arrows) only on the anchor tier, while (6) has precedence relations on every tier. Goldsmith (1976: 28) states that “each autosegmental level is a totally ordered sequence of elements,” so it appears that (6) is what is meant in the standard theory of autosegmental phonology. This conclusion requires some inferences on our part, because autosegmental representations in the literature universally assume that blank spaces between the printed symbols arrayed left to right across the printed page indicate sequential order. (6)

Within-tier precedence # → x

→ →

x

→ →

x

→

x →%

→

anchor tier root tier

laryngeal tier place tier

COR → DOR → VEL → DOR

feature tier

[ant] → [low] → [bk] → [hi]

The model in (6) raises a number of thorny issues. One is revealed in Archangeli and Pulleyblank’s discussion of adjacency and its role in the understanding of Obligatory Contour Principle (OCP) effects (Leben 1973; McCarthy 1986; Odden 1986, 1988). Recall that the OCP rules out adjacent, identical elements. The problem of course is to define “adjacency” in autosegmental terms; (7) is Archangeli and Pulleyblank’s definition of adjacency. (Note that we are now switching our example from Margi tagu to using the more abstract symbols employed by Archangeli and Pulleyblank, because we wish to hew as closely as possible to their statements.) (7)

Adjacency (Archangeli and Pulleyblank 1994: 35) a is structurally adjacent to b iff: a. at least one of the two is unassociated, both are on the same tier, and no element intervenes between the two on that tier; or, b. both a and b are associated to the same anchor tier and no anchor intervenes on that tier between the two anchors to which a and b are associated.

Archangeli and Pulleyblank argue that the definition in (7), when coupled with the OCP, rules out representations in (8a) and (8b) but allows (8c) and (8d). The diagrams in (8) are adapted from Archangeli and Pulleyblank and follow their

Precedence Relations in Phonology

9

convention of representing directed edges by means of the absence of any printed symbol between two segments on the same tier. In the disallowed (8a) and (8b), the two a elements are adjacent, so they are ruled out by the OCP. This is not the case in (8c) and (8d). (8)

a.

* x0 a

a

b. *x

x

a

a

c.

x

x

x

a

d. x

x

a

x a

The two occurrences of a elements are adjacent in (8a) because they are adjacent on the tier that houses them (recall that a blank space on the printed page indicates a directed edge), and they are adjacent in (8b) both on their own tier and through their links to adjacent elements on the anchor tier. (8c) has only a single a and therefore cannot violate the OCP even though it is associated to more than one x-slot on the timing tier. Finally, (8d) has an intervening x-slot that creates a gap between the two occurrences of a, thus supposedly preventing them from being adjacent. If we follow Goldsmith’s claim that precedence inheres on all tiers, as in (9), then we see that Archangeli and Pulleyblank’s representations and invocation of the OCP in fact yield quite different results. (9)

Precedence on all tiers a.

b. *x→ x

* x0 a→a

c.

x→x→x

d. x → x → x

a

a →

a→a

a

Once precedence is explicitly indicated in the representations, we see that (8a), (8b), and (8d) all violate the OCP, because there are two a elements that are adjacent (i.e. have an edge between them). The specific problem is that if precedence is encoded on every tier, then (8d) will have the two a elements adjacent to each other, regardless of whether an x-slot intervenes on the timing tier or not. The only representation that does not violate the OCP is (8c), because there is only a single a element. Let us now consider what representations violate the OCP if precedence is encoded only on the timing tier, as in (10). (10)

Precedence on only the timing tier a.

b. *x→ x

* x0 a

a

a

a

c.

x→x→x a

d. x → x → x a

a

These representations capture the contrast between (10b) and (10d) that Archangeli and Pulleyblank were getting at: the two a elements are adjacent in (10b), because the x-slots they are associated to are adjacent on the anchor tier, while the two a elements in (10d) are not adjacent, because of the intervening x-slot. What rules out (10a)? It follows from the fact that association lines are really undirected edges that there cannot be any floating elements because they would

10

Charles Cairns & Eric Raimy

violate the requirement that phonological representations be equivalent to connected graphs; floating elements would be represented by vertices that are not connected to the rest of the graph. Consequently, “floating” features must be connected to the rest of a phonological representation; this seems to be consistent with current thinking on this topic (see chapter 82: featural affixes). We will assume that precedence is encoded only on the anchor tier for the remainder of this chapter. This view improves our understanding of Odden’s (1986, 1988) critique of the OCP as a language universal: in general, the OCP appears to behave in a cross-linguistically arbitrary way, because the locality of distinctive features is mediated by the timing tier. Consequently, whether two elements are adjacent or not must be determined either directly from the list of edges or by calculating whether a walk exists between the two elements. The implementation of a walk appears to coincide with proposals about “searches” in phonology by Mailhot and Reiss (2007), Samuels (2009), and Nevins (2010) (see §7). This perspective has the advantage that it supplies one universal account of adjacency, which has two different specifications: two elements are adjacent if they either share an edge or if there is a walk connecting them. Each language-particular process must, of course, specify which definition of adjacency is required for its application. The existence of two ways of specifying adjacency is plausibly a major reason why the OCP appears superficially to be so variably valid.

4

Deletion and precedence

Deletion is a fundamental phonological phenomenon that must be accounted for by any phonological theory. The naive view of deletion is that segments can be simply eliminated from a representation without any complicating entailments; this is based, of course, on the conception of precedence portrayed by the left to right array of symbols across the printed page. Deletion is far more complicated when we consider how precedence structures are altered when a segment is deleted. For purposes of explicating how deletion affects precedence relations, we return to using our Margi example and suppose that a phonological process of Margi were to delete the g from tagu. It does not matter what triggers this operation, nor in what phonological theory this operation is described; we are interested here in specifying precisely what it means to delete a segment when it is considered as a vertex in a directed chain graph. In this section we describe the two characterizations of deletion (see chapter 68: deletion) that are possible within directed chain graphs. One consists of skipping over (or underparsing) the segment to be deleted, and the other involves merging two segments into one. Both appear to be empirically attested. A particularly significant part of this presentation is that two models of Optimality Theory (Containment and Correspondence) as well as the derivational model known as Precedence Based Phonology (Raimy 2000, 2009) converge on the representations and operations revealed by a graph-theoretic explication. Without considerations about the nature of precedence, (11a) presents the naive mapping between representations undergoing the deletion of g, where the symbol “>” means merely “becomes,” without reference to the nature of the operations involved. The question at hand is: once precedence relations are specified, how exactly does (11b) become (11c)?

Precedence Relations in Phonology (11)

a. b. c.

11

tagu > tau #→t→a→g→u→% #→t→a→u→%

The first-blush assumption might be that we can simply delete the g vertex from (11b). But this has the immediate effect of removing the two members of the edge set that mention the vertex g, because each directed edge is defined by an ordered pair of vertices; if g were to become unavailable to form edges, edges that contained that vertex prior to its removal cease to exist. Therefore, the result of understanding deletion as the simple removal of a segment (i.e. vertex) and its precedence relations (i.e. the edges it is associated with) is the production of a disconnected graph, as in (12). (12)

a. b.

#→t→au→% vertices: {#, t, a, u, %} edges (ordered): {# t, t a, u %}

The problem is that there is no walk from the beginning symbol # to the end symbol % and specifically there is no precedence relation from a to u, so (12) is not a connected graph; a new precedence relationship must be introduced to assure well-formedness after deletion. We will return to this problem below, but first let us observe that virtually all schools of thought in phonology ignore the fact that deletion must create a new precedence relationship. A clear way of seeing this aspect of deletion is to consider the explicit precedence structures that would support deletion as “underparsing” in the containment model of Optimality Theory proposed by Prince and Smolensky (1993). This version of OT claims that there is no loss of structure in deletion processes, but instead the deleted elements are unparsed in the output. A representation of precedence relations that meets this requirement for our hypothetical case of deletion is in (13). It is important to note that (13) is not a directed chain graph, because both a and u are mentioned in three edges; we return to this point below. (13)

a. #→t→a→g→u→% b.

vertices: edges (ordered):

{#, t, a, g, u, %} {# t, t a, a u, a g, g u, u %}

The operation of deletion in a containment model is achieved through the addition of the edge {a u} in (13b) to the input graph in (4), in effect creating a detour path around the g and rendering it underparsed. The “jump link” in (13a) illustrates the graph that results from adding this edge. As we show below, the only walk through (13) follows the added edge {a u} rather than {a g}, yielding the string tau. The representation in (13) contains all the information present in the input tagu and the output tau. It follows that in the containment model, GEN produces the structure in (13) as a candidate. Without considering the parochial constraint hierarchy that might select it as a candidate, the preceding suffices to show that the addition of a new edge is a viable way to achieve deletion in a containment model.

12

Charles Cairns & Eric Raimy

All graphs representing phonological structure must be interpreted by some sort of phonetic implementation mechanism, so the question now before us is how to assure that this mechanism follows the added “a to u” precedence relation, and not the old “a to g” precedence relation. There are two general solutions to the favoring of the newly added precedence relation and they differ in whether computation in phonology is “parallel” or “derivational.” “Parallel” computation is the type found in the containment model of OT, where there is a GEN function that creates a list of different candidates; there is also the EVAL function, which determines which candidate is the most harmonic given a language-specific ranking of CON. In this model, GEN would be free to create candidates that contain any number of novel precedence relations that do not exist in the input. Therefore, GEN will generate candidates with underparsed segments, and these candidates would indicate that phonetic implementation will follow the “deletion edge” and not the edges associated with the deleted g. Whether candidates with detours might emerge as the most harmonic will be the result of parochial constraint interactions. In short, preference for newly added precedence links that produce deletion can be implemented easily in the containment model of OT. This approach to deletion can be straightforwardly carried over to the Correspondence model of OT (McCarthy and Prince 1995). The difference between the implementation of deletion by underparsing in the Containment and Correspondence models of OT is in the status of a representation with conflicting precedence specifications as in (13). Whereas the Containment model can directly produce this type of representation as its output with the effect that segments that are “detoured” around are the underparsed segments, the Correspondence model can produce the chain graph in (11c) directly from (11b) by deleting the vertex g and by adding the new edge {a u}, operations freely performed by GEN. This computation requires the comparison of both (11c) and (11b) in order to determine the Max, Dep, and Contiguity violations. Thus it appears that the Containment and Correspondence models of OT converge on deletion as the addition of a new precedence link. Preference for newly added links and consequent deletion can be implemented in a derivational model of phonology by imposing an order on the list of edges. Recall that (13) is not a directed chain graph, a requirement for phonetic implementation. We return to this point in §7, but for now note that, as Idsardi and Raimy (forthcoming) point out, a derivational theory of phonology requires representations to have the characteristics of a directed chain graph only at the interface between modules. “Serialization” is a process that creates a walk from the beginning symbol # to the end symbol %; this walk creates a representation that is a directed chain graph. Idsardi and Shorey (2007) propose that serialization is a walk that uses the order of the list of edges to make decisions about which precedence link to follow when there is a choice. Thus, in (13b), when the walk reaches the a vertex, it must choose between following either the {a u} or the {a g} edge. This choice will be made by the order of the edge list, because one of these edges will be “first” and thus will be followed. In order for an added precedence link to produce deletion effects it must be added to the list of edges in a manner which places it before the precedence links associated with the segment to be deleted. We emphasize that the detour approach to deletion is virtually the same in three theories of phonology, i.e. the Containment and Correspondence models of OT

Precedence Relations in Phonology

13

and the derivational model. The differences among the phonological theories arise only from specific details on how the deletion process is actually computed in the different models. A second approach to deletion is also possible in graph-theoretic terms and can be understood as involving a type of coalescence, as suggested by de Lacy (2007) for OT. Another way to map a representation like (11b) to (11c) is to merge two of the vertices into one, without any addition of new precedence relations. We must first decide whether the “deleted” segment is merged with the preceding or following segment. For purposes of explication in our hypothetical example we merge the deleted g with the preceding a. The operation of vertex merger will be broken into steps in (14), so that important questions can be identified. (14)

Deletion as vertex merger a.

#→t→a→g→u→% vertices {#, t, a, g, u, %} edges (ordered) {# t, t a, a g, g u, u %}

b.

# → t → [ag] → u → %

c.

vertices edges (ordered) # → t → [a] → u vertices edges (ordered)

{#, t, [ag], u, %} {# t, t [ag], [ag] [ag], [ag] u, u %} →% {#, t, [a], u, %} {# t, t [a], [a] u, u %}

(14a) presents the representation for tagu as in (4). Vertex merger (14b) coalesces every occurrence of a and g into a new vertex ag. This produces the representation in (14b), where the merger of the a and g is indicated by the [ag] composite segment. The inevitable result of this is the production of the edge {ag ag}, or the edge that loops back onto itself, thus violating the constraint requiring irreflexivity. This reflexive edge will produce a geminate or long version of the composite [ag] segment. Although this is not the desired result for plain deletion, this situation produces compensatory lengthening effects (Hayes 1989; Sloan 1991: 80–87) in a straightforward manner without recourse to moras (see §7). The final steps that are required to produce (14c) are to eliminate the looping back arrow if it is not desired and to specify the phonetic interpretation of the vertex ag. There are a number of ways of accomplishing the former, one of which is to specify a parameter that allows languages to eliminate any “reflexive” edges (edges that are defined by two mentions of the same vertex) whenever they are formed. Any theory of phonology that has the resources to separate the melodic content of a segment from its timing slot easily handles the phonetic interpretation aspect of this process. Some languages retain features of both segments in the composite vertex; for example classic coalescence in Sanskrit /a/ + /i/ > /e/. In our example, we want the segment to simply be interpreted as a, which will require some extra statement. The preceding description of deletion in connection with compensatory lengthening and coalescence is theory-neutral, and will be implemented differently in different theories. The point is that considering explicit precedence relations

14

Charles Cairns & Eric Raimy

shows how deletion, coalescence, and compensatory lengthening are deeply connected regardless of how different theories focus on the various surface effects of deletion. One question that arises from this discussion is whether there is an embarrassment of wealth in the representational possibilities offered by graph theory to produce deletion. It appears that all the ways of accomplishing deletion are attested in different languages. For example, Tohono O’odham vowel syncope in reduplication is best characterized by the detour approach (Raimy 2000: 113–114). Chumash /l/-deletion (Raimy 1999: 82–83) exhibits deletion that is best characterized with the coalescence approach, as are examples of classic coalescence. Some languages appear to have both options available in competition with each other (Indonesian nasal assimilation; Raimy 2000: 99–112). Consequently, all of the analytical options based on different representational opportunities appear to be attested. Questions about deletion are thus typological in nature, where phonologists should ask which kind of deletion any particular instantiation in a particular language exhibits: does the type of deletion that occurs correlate with compensatory lengthening and/or coalescence processes, and what diagnostics are there to distinguish the different types of deletion? An important theme of this section is that making precedence relations explicit via graph theory is useful to all theories of phonology. This should not be misunderstood as suggesting that there are no differences among different theories of phonology; our point is that specific differences with respect to precedence can be identified explicitly through differences in the necessary graph structures. Put another way, graph theory provides a reasonably neutral lingua franca to discuss and explore the nature of precedence in phonology. Of course, up until this point we have been discussing phonology whose elements are discrete, timeless entities; we now turn to considering the role of precedence in theories that view segments as containing real-time articulatory gestures.

5

Precedence and discrete point theory

One assumption in the preceding sections of this chapter that can and should be called into question is that phonology operates strictly on abstract discrete elements. This is the standard assumption in the class of theories of phonology that we will call formal phonology. Not all theories of phonology make this assumption. According to the theory of Articulatory Phonology (Browman and Goldstein 1986, 1989, 1990a, 1990b), vertices represent “gestures,” i.e. abstract representations of dynamic articulatory events or actions. Thus the basic elements exist in real time, and are not abstract, timeless points. Such theories are also amenable to graph-theoretic representation, because they can be seen to have multiple precedence structures, one for each articulator in a segment. In fact, from a precedence structure point of view, it may be just the difference of whether there is a single abstract precedence structure or multiple more concrete articulator based precedence structures that produces an overall precedence structure that allows or disallows overlap. Gafos (2002) argues that it is the nature of precedence that is at issue in the formal vs. gestural models of phonology. Gafos (2002: 270) observes that “the phonologically relevant notion of time is overlap of dynamic units”; this is in contrast

Precedence Relations in Phonology

15

to “linear order of static units [being] the only relevant notion of time in phonology.” Both approaches still order phonological elements; they differ only in the nature of the precedence graphs. For formal phonology, the vertices are segments that are timeless points, and overlap is not definable. In gestural phonology, the vertices are gestures that have both spatial and temporal aspects, so there is a more complicated relation of overlap. Bird and Klein (1990) explicitly develop the model of precedence for phonological representations that can overlap. Another way of conceptualizing precedence in gestural phonology is to assign weight to the edges in a precedence graph to indicate the amount of overlap between each gesture; alternatively, the weights may indicate the actual time span between the gestures. Any model of phonology can be cast by precedence graphs. Different models of phonology simply argue for or assume different characteristics for the relevant precedence graph. The different approaches to phonology – formal and gestural – are only incompatible if one assumes that either model will account for all phonological phenomena. Both models are necessary in their own domains and provide insights into different aspects of phonology. Formal phonology with atemporal segments represents phenomena closer to the morphology–phonology interface, while gestural phonology represents phenomena closer to the phonetics– phonology interface.

6

Graph theory as a tool for phonology

One major advantage of adopting graph theory as the formal underpinning of precedence in phonology is that it supplies general and specific knowledge that can be applied to uniquely phonological questions. Two topics in phonology that directly benefit from general knowledge of the nature of graphs are the no crossing constraint and locality in phonology. The no crossing constraint originates in Goldsmith (1976: 27) as the statement “Association lines do not cross.” Archangeli and Pulleyblank (1994: 39) present this constraint as a ban on the representation in (15a), which we augment with explicit precedence relations in (15b). (15)

The no crossing constraint (Archangeli and Pulleyblank 1994: 39) a. [i [j

b. [i→[j

b

a→ b

a

The problem with (15) is that it appears to encode conflicting precedence relations. The association lines indicate that a and [j occur together, as do b and [i. The two tiers thus provide conflicting information about which feature precedes the other feature, because it encodes both a → b and b → a. Two assumptions dictate that (15b) is ineluctably derived from (15a): that the left-to-right array of printed symbols encode directed edges and that association lines must be straight. Coleman and Local (1991) argue that the no crossing constraint is an incoherent concept in autosegmental phonology, in part because there is no mathematical

16

Charles Cairns & Eric Raimy

justification for insisting on straight lines; that restriction is merely a convention for drawing diagrams and has no formal content. (15a) is formally equivalent to all the diagrams in (16), which do not portray a no crossing constraint violation. (16a) is a graphic trick making an association line curved; (16b) makes the diagram appear three-dimensional, and (16c) simply moves one of the elements to the other side of a tier. (16)

Graphic workarounds a.

[i [j a

c. [i

b. [i [j

b a

b

a

b [j

More importantly, Coleman and Local argue that the no crossing constraint implies that autosegmental representations must be “planar,” an important question of graph theory that bears directly on when lines cross or not. Kuratowski (1930) proves that two kinds of graph cannot be drawn within a single plane without lines crossing. One is the K3,3 graph in (17). This is a bipartite graph, which means that the vertices can be partitioned into two sets such that each edge of the graph has one end in one set and the other in the other. The presence of a K3,3 graph structure means that the graph which has it is “nonplanar” if the lines are not to cross. According to Kuratowski’s Theorem, a graph is non-planar if and only if it contains either the subgraph K3,3 or K5 (see Kuratowski 1930 for the K5 graph, which does not appear to be relevant to phonology). (17)

Kuratowski’s K3,3 graph

Coleman and Local (1991) establish that most supposedly three-dimensional (non-planar) representations in the literature are in fact planar; on the other hand, they also point out that phonology must countenance non-planar representations. The autosegmental representation of the word room in Guyanese English in (18), adapted from Coleman and Local (1991: 330), is an example of a non-planar graph (i.e. it contains a K3,3 graph); the three features [nasal], [round], and [back] are on independent tiers and each is associated with all three segments. The first x-slot represents the phoneme /r/, the second /u/, and the last /m/. Coleman and Local report that the three features depicted here spread to each of the three segments independently of one another, and therefore each should be on a separate tier. It is not possible to depict these associations in two-dimensional space without lines crossing.

Precedence Relations in Phonology (18)

[nasal]

17

[round]

x

x

x

[back] The import of Coleman and Local’s work is that autosegmental representations can be non-planar, which means that different distinctive features will not be able to cross lines by definition, because the association lines will be in different planes. The practical result is that phonologists must be explicit about the nature of precedence in phonological representations and distinguish between conventions on drawing convenient diagrams and formal properties of precedence. Restrictions on precedence relations in phonology do not necessarily reside in the representations themselves but may result from how they are implemented. Graphs are “abstract data structures” (Aho et al. 1985) that can be implemented in many different ways. The manner of implementation affects what operations are easier (or possible) to perform than others. For example, we have so far portrayed phonological representations as a list of vertices (x-slots and associated features) and edges (precedence relations) as the basis for implementation. Some scholars (e.g. Heinz 2007) use adjacency tables as the basis for implementation, as exemplified for tagu in Table 34.1 (which is technically a local adjacency table). An adjacency table is made by listing the vertices of the graph as the headers for columns and rows and indicating in each cell whether the relevant precedence relationship holds or not. The row headers in Table 34.1 indicate which segment precedes the column headers; Table 34.1 encodes that t is the word-initial segment, because no segment precedes it, indicated by the lack of any mark in the t column; t precedes a, as indicated by the mark in the t row’s a column, a precedes g, and so on. This type of precedence encoding makes information about immediate precedence easily accessible, reflecting the frequency of phonological operations that are strictly local. A drawback to this type of precedence encoding is that long-distance relationships required to account for phenomena like long-distance assimilation and vowel harmony (see chapter 77: long-distance assimilation of consonants; chapter 91: vowel harmony: opaque and transparent vowels; chapter 118: turkish vowel harmony; chapter 123: hungarian vowel harmony) must be calculated by a walk (see §2) through the representation. Adjacency tables can encode “transitive precedence.” Table 34.2 does this by marking in each row all of the other segments that a particular segment precedes Table 34.1 Immediate adjacency

Table 34.2 Transitive precedence

follows

t a g u

a

g

u

x x x

t precedes

precedes

t

follows

t a g u

a

g

u

x

x

x

x

x x

18

Charles Cairns & Eric Raimy

transitively. Table 34.2 differs from Table 34.1 in that for each row, a mark is put in the column for every segment that the row’s segment precedes transitively. Consequently, t is encoded as the word-initial segment because no other segment precedes it, and it precedes all other segments, as indicated by the marks in the a, g, and u columns in the t row; u is in word-final position because it precedes no other segment, and all segments precede it. The advantages of this type of precedence encoding is that long-distance relationships are directly encoded. For example, the fact that a has a mark in the u column indicates that there is a direct phonological relationship between these two segments, which would support processes like vowel harmony. The main disadvantage to this type of precedence encoding is that local adjacency has to be calculated by determining that there are no segments “between” the segments in question. Adjacency tables in general have two more drawbacks. The first one is that a new adjacency table will have to be created when a new segment is added to a representation, because the table is defined by the segments in the representation. This is computationally expensive. The second drawback is that adjacency tables are not very economical for the type of graphs that would be common for phonological representations. The point to understand here is that if we look at Table 34.1, there are a total of 16 cells that make up the table, but only three of them encode precedence information relevant to local operations. All cells in a table impose a representational cost, so if there is any pressure to keep phonological representation economical, adjacency tables have a lot of empty costs involved. Phonological representations in general are going to be “sparse graphs,” favoring the use of lists of vertices and edges (Aho et al. 1985). We approach the question of how to encode precedence information in phonological representations by considering the computations that are required to implement them. The type of graph introduced in (3) requires a walk through it to ensure that it is connected. The local adjacency Table 34.1 requires a walk through it to calculate long-distance relationships, and the temporal precedence Table 34.2 requires computation to determine local adjacency. This general conclusion coincides with recent work in both derivational and parallel models of phonology which argues that phonological environments need to be calculated in some manner. See Mailhot and Reiss (2007), Samuels (2009), and Nevins (2010) for the derivational model of searching, and work along the lines of Rose and Walker (2004) and Krämer (1999) for parallel Correspondence Theory approaches.

7

Extensions of precedence graphs

Two general phonological phenomena are directly derivable from the previous discussion of how deletion is accomplished. First, it is natural to recognize that epenthesis (see chapter 67: vowel epenthesis) should be the mirror image of deletion and in fact this is the case. (19) presents the hypothetical case of the form tagu undergoing epenthesis of r after the a. (19)

a. b.

#→t→a→g→u→% #→t→a→r→g→u→%

Precedence Relations in Phonology

19

Just as there are two distinct ways to produce the deletion of a segment, there are two distinct ways to epenthesize a segment. The first way is to explicitly add precedence relations to and from a new segment from the relevant points in the representation. This would take (19a) and produce the representation in (20). (20)

# → tt → →aa→ →gg→→uu→→%% r

The new structure added to (19a) to produce (20) is offset by having the segment r below the underlying segments. Notice the parallels between (19) and (13a), and recall that the actual graphic layout does not matter. The difference is that deletion is detouring around an old segment, while epenthesis is detouring through a new segment. Another parallel between the epenthesis and deletion representations is that both need to be “serialized” to resolve the precedence conflict present in them. As earlier in the chapter, all models of phonology have the relevant resources to produce the proper output to this form. “Fission” is to epenthesis as coalescence is to deletion. To see this, consider the explicit representation of x-slots and melodies in (21), which shows the splitting of an x-slot to produce a new segment. (21)

a.

t

a

g

u

# → x → xi → x → x → % b.

t

a

g

u

# → x → xi → xi → x → x → % This type of epenthesis raises the question how the new x-slot is formed without taking the associated melody along with it; reasons of space preclude a full exposition of this question here beyond pointing out that the interpretation of a bare x-slot will be determined on a language specific basis (see chapter 67: vowel epenthesis; chapter 58: the emergence of the unmarked). The differences between these two types of epenthesis suggest differences in the type of epenthetic segment. The approach in (20) supports a prespecified type of epenthetic segment, which cannot be derived from markedness conditions, while the approach in (21) suggests an “Emergence of the Unmarked” (McCarthy and Prince 1994) type of epenthetic segment, because an empty x-slot needs to be interpreted. A final note is that just as the coalescence approach to deletion creates an intermediate representation that can account for compensatory lengthening effects, so this fission approach can also account for simple segment lengthening effect by fissioning the melody along with the x-slot. If segmental length is encoded on the x-tier as opposed to being marked by a mora (see Ringen and Vago 2010), there are two representations for true geminates (Hayes 1986; Schein and Steriade 1986) that follow naturally. The structure in (22a) is the traditional multiply linked melody representation, while (22b) is a novel looped segment representation for a geminate.

20 (22)

Charles Cairns & Eric Raimy Two types of geminates a. traditional m x→x

b. looped m x

There appears to be evidence from Tohono O’odham (Raimy 2000: 116) that shows that both types of geminate segments can exist in a single language. The presence or absence of geminate inalterability effects (Hayes 1986; Schein and Steriade 1986) may also be seen as a difference in representations. The representations for deletion, epenthesis, and geminates given above reveal that there are natural graph representations that do not follow the constraints that define a chain-graph representation. The representations for deletion and epenthesis involve the addition of at least one precedence relation that causes the graphs to no longer be chain graphs. It appears that phonology tolerates graphs that do not follow the strict conditions that define a chain graph, at least for a portion of the phonological computation, however it is conceived. This suggests that other (morpho)phonological phenomena will have natural analyses based on different modifications to precedence graphs. Hypocoristic formation is one such process. Alber (2007, 2009) presents data showing hypocoristic formation in northern Italian dialects. The data in (23) show that hypocoristics are formed in these northern Italian dialects by truncating all phonological material after the first vowel of the form. (23)

Northern Italian hypocoristics (Alber 2007, 2009) Truncated Fra Cri Lu Ste

Source name Francesca Cristina Luisa Stefana

Hypocoristics are truncated forms, so we should expect truncation to parallel deletion. This can be implemented by allowing the morphological rule of hypocoristic formation to add a precedence link from the “first vowel” to the end symbol, %, as in (24). (24)

Hypocoristic formation in Italian via truncation

#→f→r→a→n→c→e→s→c→a→% The representation in (24) is not a chain graph, so it will undergo serialization and the truncated form will result. The main advantage to this approach is that the deletion involved in truncation is derived from the addition of phonological material, namely a new precedence link.

Precedence Relations in Phonology

21

A minimal extension of the epenthesis as addition of a precedence link produces infixing phenomena. Toba Batak infixation, taken from Halle (2001), is presented in (25) (stress is suppressed for expository purposes). (25)

Nominalizer -al- in Toba Batak batuk ogo

b-al-atuk al-ogo

‘ladder’ ‘wind’

This infixing pattern can be described as the affix al preceding the first vowel of the form. Whether al appears as an infix or a prefix is derived from whether the first vowel of the form is in word-initial position. This variation does not change the positional generalization about where this affix is concatenated to the base. (26) provides the precedence graphs for the examples in (25). (26)

a.

a→ l #→b→a→t→u→k→%

b. a → l #→o→g→o→% As with previous examples, the addition of the affix al produces a representation that violates the characteristics of a chain graph, which forces the representation to be serialized at some point in time. See Yu (2007) for further discussion of infixation. Our last example of (morpho)phonological processes that are illuminated by explicit precedence graphs is from Spokane, in (27). Repetitive morphology in Spokane is interesting because it combines elements of infixation and reduplication. The content of the repetitive morpheme is /e/; it appears to infix if the root begins with two consonants (27a), but causes reduplication if the base begins with a single consonant (27b). (27)

Repetitive morphology in Spokane (simplified, from Bates and Carlson 1992) a.

repetitive petax qesip

base ptax qsip

‘spit’ ‘long ago’

b.

repetitive sesil kekul

base sil kul

‘chop’ ‘make’

The precedence graphs that capture this behavior are shown in (28). (28)

a.

e #→p→t→a→x→%

b.

e #→s→i→l→%

22

Charles Cairns & Eric Raimy

Although the representations in (28) appear to be fundamentally distinct, they can be constructed with a single generalization on how the /e/ is concatenated to the base. The repetitive morpheme /e/ follows the first segment of the base and precedes the consonant before the first vowel. In (28a) and bases with two consonants in general, this description causes infixing of the /e/, because the two parts of the description are distinct, while in bases that begin with a single consonant, the /e/ loops back to this single consonant. The standard analysis of serializing precedence graphs that do not conform to the conditions for a chain graph produces a single repetition when there is a “loop” in a precedence graph like that in (28b). This observation forms the basis of the approach to reduplication proposed by Raimy (2000, 2009) (see chapter 100: reduplication).

8

Conclusions

Adjacency and linearity have been important topics in phonology for many years. As de Lacy (2007) noted, all rules/constraints seem to operate on elements that are in some sense adjacent; McCarthy (1989) spoke of “strict locality” as a central topic in phonology. Our understanding of adjacency is that all questions of adjacency are questions of precedence. Graph theory provides a theoryneutral framework in which formal accounts of precedence can be proposed and evaluated. By having a formal model of precedence, murkier questions of how adjacency operates in different models of phonology can be addressed straightforwardly. An explicit understanding of the logic of phonological sequences allows us to better understand phonological “action,” especially “action at a distance.” Mailhot and Reiss (2007), Samuels (2009), and Nevins (2010) propose that fundamental phonological operations involve search-and-copy algorithms. The immediate question raised by these proposals is what it is that these search procedures operate on. We suggest the precedence graphs discussed in this chapter as the answer to this question. This chapter has presented a survey of potentially useful graphs matched with fundamental phonological and morphophonological phenomena. All models of phonology require a theory of representations, because computation is inherently connected to the representations (and vice versa) (McCarthy 1988: 84). Thus, by increasing our precision and attention to the representation of precedence in phonology, the precision of our knowledge about computation is advanced. Questions of locality in phonology are long-standing, but the adoption of graph theory as a representational tool for precedence in phonology revolutionizes the questions that can be asked about phonology in the same way that syllabic representation (Kahn 1976), autosegmental representation (Goldsmith 1976), and prosodic representation (McCarthy 1981) have done in the past for phonology.

Precedence Relations in Phonology

23

REFERENCES Aho, Alfred, John Hopcroft & Jeffrey Ullman. 1985. The design and analysis of computer algorithms. Reading, MA: Addison-Wesley Publishing Company. Alber, Birgit. 2007. Deutsche und Italienische Kurzwörter im Vergleich. In Claudio Di Meola, Livio Gaeta, Antonie Hornung & Lorenza Rega (eds.) Perspektiven Zwei, 101–112. Rome: Istituto Italiano di Studi Germanici. Alber, Birgit. 2009. The foot in truncation. Paper presented at the CUNY Conference on the Foot, January 2009. Available (May 2010) at http://www.cunyphonologyforum.net/ footconf.php. Anderson, Stephen R. 1974. The organization of phonology. New York: Academic Press. Archangeli, Diana & Douglas Pulleyblank. 1994. Grounded phonology. Cambridge, MA: MIT Press. Bach, Emmon. 1968. Two proposals concerning the simplicity metric in phonology. Glossa 2. 128–149. Bagemihl, Bruce. 1991. Syllable structure in Bella Coola. Linguistic Inquiry 22. 589–646. Bates, Dawn & Barry F. Carlson. 1992. Simple syllables in Spokane Salish. Linguistic Inquiry 23. 653–659. Bird, Steven & Ewan Klein. 1990. Phonological events. Journal of Linguistics 26. 33–56. Blevins, Juliette. 2003. The independent nature of phonotactic constraints: An alternative to syllable-based approaches. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 375–403. Cambridge: Cambridge University Press. Browman, Catherine P. & Louis Goldstein. 1986. Towards an articulatory phonology. Phonology Yearbook 3. 219–252. Browman, Catherine P. & Louis Goldstein. 1989. Articulatory gestures as phonological units. Phonology 6. 201–251. Browman, Catherine P. & Louis Goldstein. 1990a. Gestural specification using dynamically defined articulatory structures. Journal of Phonetics 18. 299–320. Browman, Catherine P. & Louis Goldstein. 1990b. Representation and reality: Physical systems and phonological structure. Journal of Phonetics 18. 411–424. Cairns, Charles. 1988. Phonotactics, markedness and lexical representation. Phonology 5. 209–236. Cairns, Charles & Eric Raimy. 2009. Architecture and representations in phonology. In Raimy & Cairns (2009), 1–16. Coleman, John & John Local. 1991. The “No Crossing Constraint” in autosegmental phonology. Linguistics and Philosophy 14. 295–338. Czaykowska-Higgins, Ewa & Marie Louise Willet. 1997. Simple syllables in Nxa’amxcin. International Journal of American Linguistics 63. 385–411. de Lacy, Paul. 2007. The formal properties of phonological precedence. Paper presented at the CUNY Conference on Precedence Relations, January 2007. Available (May 2010) at http://www.cunyphonologyforum.net/forum.php. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5. 253–286. Fudge, Erik C. 1987. Branching structure within the syllable. Journal of Linguistics 23. 359–377. Gafos, Adamantios I. 2002. A grammar of gestural coordination. Natural Language and Linguistic Theory 20. 269–337. Goldsmith, John A. 1976. Autosegmental phonology. Ph.D. dissertation, MIT. Published 1979, New York: Garland. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Halle, Morris. 2001. Infixation versus onset metathesis in Tagalog, Chamorro, and Toba Batak. In Michael Kenstowicz (ed.) Ken Hale: A life in language, 153–168. Cambridge, MA: MIT Press.

24

Charles Cairns & Eric Raimy

Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62. 321–351. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Heinz, Jeffrey. 2007. The inductive learning of phonotactic patterns. Ph.D. dissertation, University of California, Los Angeles. Hulst, Harry van der. 2008. The syllable in RCVP: Structure and licensing. Paper presented at the CUNY Conference on the Syllable, January 2008. Available (May 2010) at http://www.cunyphonologyforum.net/syllconf.php. Idsardi, William J. & Eric Raimy. Forthcoming. Three types of linearization and the temporal aspects of speech. In Theresa Biberauer & Ian Roberts (eds.) Challenges to linearization. Berlin & New York: Mouton de Gruyter. Idsardi, William J. & Rachel Shorey. 2007. Unwinding morphology. Paper presented at the CUNY Phonology Forum Conference on Precedence Relations, January 2007. Available (May 2010) at http://www.cunyphonologyforum.net/forum.php. Kahn, Daniel. 1976. Syllable-based generalizations in English phonology. Ph.D. dissertation, MIT. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Krämer, Martin. 1999. A correspondence approach to vowel harmony and disharmony. (ROA-293.) Kuratowski, Kazimierz. 1930. Sur le problème des courbes gauches en topologie. Fundamenta Mathematicæ 15. 271–283. Langacker, Ronald. 1969. Mirror image rules II: Lexicon and phonology. Language 45. 844–862. Leben, William R. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT. Mailhot, Frédéric & Charles Reiss. 2007. Computing long-distance dependencies in vowel harmony. Biolinguistics 1. 28–48. McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12. 373–418. McCarthy, John J. 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 17. 207–263. McCarthy, John J. 1988. Feature geometry and dependency: A review. Phonetica 45. 84–108. McCarthy, John J. 1989. Linear order in phonological representation. Linguistic Inquiry 20. 71–99. McCarthy, John J. & Alan Prince. 1994. The emergence of the unmarked: Optimality in prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Rutgers University (ROA-13). McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 249–384. Amherst: GLSA. Nevins, Andrew. 2010. Locality in vowel harmony. Cambridge, MA: MIT Press. Odden, David. 1986. On the role of the Obligatory Contour Principle in phonological theory. Language 62. 353–383. Odden, David. 1988. Anti antigemination and the OCP. Linguistic Inquiry 19. 451–475. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Raimy, Eric. 1999. Representing reduplication. Ph.D. dissertation, University of Delaware. Raimy, Eric. 2000. The phonology and morphology of reduplication. Berlin & New York: Mouton de Gruyter. Raimy, Eric. 2009. Deriving reduplicative templates in a modular fashion. In Raimy & Cairns (2009), 383–404. Raimy, Eric & Charles Cairns (eds.) 2009. Contemporary views on architecture and representations in phonology. Cambridge, MA: MIT Press.

Precedence Relations in Phonology

25

Ringen, Catherine & Robert M. Vago. 2010. Geminates: Heavy or long? In Charles Cairns & Eric Raimy (eds.) Handbook of the syllable. Leiden: Brill. Rose, Sharon & Rachel Walker. 2004. A typology of consonant agreement as correspondence. Language 80. 475–531. Samuels, Bridget. 2009. The structure of phonological theory. Ph.D. dissertation, Harvard University. Saussure, Ferdinand de. 1916. Cours de linguistique générale. Lausanne & Paris: Payot. Schein, Barry & Donca Steriade. 1986. On geminates. Linguistic Inquiry 17. 691–744. Sievers, Eduard. 1881. Grundzüge der Phonetik zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. Leipzig: Breitkopf & Härtel. Sloan, Kelly D. 1991. Syllables and templates: Evidence from Southern Sierra Miwok. Ph.D. dissertation, MIT. Steriade, Donca. 1999. Alternatives to syllable-based accounts of consonantal phonotactics. In Osamu Fujimura, Brian D. Joseph & Bohumil Palek (eds.) Item order in language and speech, 205–242. Prague: Karolinum Press. Sweet, Henry. 1877. A handbook of phonetics. Oxford: Clarendon Press. Trubetzkoy, Nikolai S. 1939. Grundzüge der Phonologie. Göttingen: van der Hoeck & Ruprecht. Translated 1969 by Christiane A. M. Baltaxe as Principles of phonology. Berkeley & Los Angeles: University of California Press. Vaux, Bert & Andrew Wolfe. 2009. The appendix. In Raimy & Cairns (2009), 101–143. Wilson, Robin J. 1996. Introduction to graph theory. Longman: London. Yu, Alan C. L. 2007. A natural history of infixation. Oxford: Oxford University Press.

35

Downstep Bruce Connell

1

Introduction

Downstep is a pitch-lowering phenomenon that is widely recognized to occur in tone languages, particularly those of sub-Saharan Africa, in which it was first identified. It is also attested in several languages of the Americas, though only very rarely in Asia. The concept of downstep has also more recently been extended to account for phenomena associated with intonation in non-tonal languages,1 and is indeed perhaps better known to non-specialists in this context. Downstep is most commonly described as the lowering influence of a low tone (L) on a following high tone (H), such that a new, lower, “ceiling” is set for all subsequent Hs within a specifiable domain or prosodic unit. One of the early and still primary areas of debate in the study of downstep pertains to the sameness or otherwise of downstep as effected by a surface L as opposed to an underlying or floating L. Another area of interest has to do with the implications of its recurrent and cumulative nature for the analysis of contrastive tone levels; a language with only two apparent contrastive lexical levels (e.g. H, L) will manifest several intervening levels in actual speech, including situations in which a H occurring late in an utterance may be realized at a lower pitch than a L early in the same utterance. Another debate centers on the analysis of downstep applying to tones other than H; while the paradigm case is the lowering of H in a language with two tones (H, L), downstep does occur in languages with more than two tones, and in some such languages, cases of downstepped mid (M) and L tones are also attested. Whether such cases parallel downstepping of H remains moot. In addition, there are apparent cases of “upstep,” a tone raising phenomenon that ideally would be symmetrical to downstep in the details of its realization, though few if any of the attested cases are precisely so.

1

An anonymous reviewer suggests the term “non-tonal language” is a misnomer, given that “there is no essential qualitative difference between tones based on their origin in the lexicon or in phrasal phonology.” While I am sympathetic to the view that “non-tonal” is problematic, the claim that “there is no essential difference” is contentious. This discussion is outside the scope of the present chapter, but Hyman’s (2001: 1368) definition of a tone language serves to distinguish the two types of language, a tone language being one, “in which an indication of pitch enters into the lexical realization of at least some morphemes.” The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0035

Downstep

\ 2

Discussion as to how best to characterize downstep from a theoretical perspective has benefited from and contributed to general phonological theory, with proposals being shaped by phonetic and phonological approaches. This debate remains unresolved, with some scholars advocating the view that downstep is best accommodated in the phonetic implementation component (e.g. Poser 1984; Beckman and Pierrehumbert 1986; Pierrehumbert and Beckman 1988), and others arguing that it is phonological (e.g. Snider 1998, 1999). The discussion of downstep was particularly fruitful in the 1960s and then again through the 1980s and 1990s, with the advent of autosegmental phonology (chapter 14: autosegments) and feature geometry (chapter 27: the organization of features). Since then the debate has subsided somewhat, though two good general presentations of downstep have appeared in recent years, in Yip (2002) and Gussenhoven (2004). The most recent detailed argument for a particular theoretical approach to downstep is Snider (1999). This chapter characterizes downstep and related tonal phenomena, summarizing with representative data the key issues and current views. In doing so, I draw on those sources just mentioned, as well as several other published (and some unpublished) studies that have contributed to our understanding of downstep. The remainder of the chapter is divided as follows. §2 is an overview of work leading to the recognition and study of downstep in tone languages. This section defines the key terms used in the discussion, presents relevant illustrative data, and introduces some of the important theoretical issues involved. The third section then discuss downstep-related issues, including its distribution, downstepping of tones other than H and in languages with more than just H and L tones, and downstep in non-tonal languages. What triggers downstep is discussed in §4, phonetic aspects of downstep in §5, and issues pertaining to upstep and H-raising in §6. A particular personal concern has been the inconsistent and conflicting use of terminology found in the discussion of downstep and related pitch phenomena, and its importance for developing an adequate understanding of these phenomena; these issues are examined in §7. §8 presents instrumental evidence from the Bantoid language Mambila that bears on a resolution to the issues discussed in §7. Inevitably, some aspects of the topic receive fuller treatment than others; in particular, the substantial literature in which the notion of downstep is used in the analysis of non-tonal languages does not get the attention it deserves. This is in large part due to space constraints, but it is to some extent also a reflection of my own expertise and familiarity with the subject, as well as an attempt to address a perceived imbalance in the general theoretical phonology literature; while downstep is primarily a phenomenon of tone languages, it has perhaps received greater attention for its operation in non-tonal languages, and this aspect of the discussion is more accessible to the general reader. For related discussion, see also chapter 45: the representation of tone; chapter 114: bantu tone.

2 2.1

An overview of downstep in language study: Phonological issues Early studies

The phenomenon now known as downstep was first noted in print well over a century ago, by Christaller (1875: 15). In his discussion of Fante (Kwa, Ghana)

3

Bruce Connell

grammar, Christaller talks of three tones – high, middle, and low – but explains the middle tone as “high tones abating by one step or successive steps”; thereby recognizing the underlying sameness of the middle and high tones. Subsequent scholars appear to have followed Christaller, though the concept of downstep was not fully recognized until much later. Ward (1933), for example, describes Efik (Benue-Congo, Nigeria) as having “three well defined levels of tone . . . high, mid and low,” and the mid tone as “a subsidiary of the high group of tones” (1933: 34–35). Ward’s examples2 illustrate what are now recognized as “automatic downstep,” the lowering of a surface High tone under the influence of an immediately preceding surface Low, and “non-automatic downstep,” in which a H is lowered by an underlying or floating L3: e.g. in /eJe/ [ – – ] ‘he’, and /ubom/ [ – – ] ‘canoe’, the second tone is said to be lowered by the preceding L, since the sequence [ – – ] is unattested. Ward applies a similar analysis to words like /edep/ [ – – ] ‘he buys’ (“the second tone is a high tone lowered from the preceding high prefix to show a particular tense usage”; 1933: 35), despite the absence of a preceding L. She then argues, however, that there is a Mid, distinct from the lowered H, occurring in words such as /DbDI/ [ – – ] ‘chief’ contrasted with /DbDI/ [ – – ] ‘mosquito’, which “can in no way be considered as a high tone lowered as in edep.” Ward makes no attempt to account for this “mid” tone, and does not discuss why it cannot be identified with the lowered tone of /edep/. A brief unattributed description of Igbo (Benue-Congo, Nigeria) in The principles of the IPA (IPA 1949), moves a step closer to recognizing the underlying identity of H and what was called M for Efik, now referring to it (in Igbo) as “a lowered high tone.”4

2.2

Recognition of downstep

A closer understanding of the nature of such tone systems, however, was not offered until Winston (1960), who, like Ward, worked on Efik. Winston’s analysis expressly recognizes just two phonemic tones for Efik, H and L, together with a “downstep” phoneme, which effects a lowering of following H. Both the term “downstep” and the notational convention since widely adopted – a raised exclamation mark preceding the affected syllable – were introduced by Winston. (The exclamation mark is still current, though increasingly a downward pointing arrow ↓ has been adopted, following IPA conventions.) Winston also recognized that H tones subsequent to the downstepped H did not rise above the height of the downstepped H; i.e. a new “ceiling” for H is set by downstepping, creating a terracing effect. Although the details of this were not addressed explicitly by Winston, e.g. questions pertaining to the domain or extent of downstep remained unanswered, it is clear from his examples that downstep is cumulative and that there are limits to its application. Five sentences drawn from Winston’s data permit a simplified version of his arguments sufficient to illustrate the basic nature of downstep (Winston 1960: 185–186): 2

Throughout the chapter, transcriptions from the work of different authors have been standardized to follow IPA conventions. 3 The terms “automatic” and “non-automatic downstep” are due to Stewart (1965), discussed below. 4 Attributed to Ward in Welmers (1973: 85).

Downstep (1)

idi:(e) DbDI ke Ikut [–– ‘It isn’t a mosquito that I see.’ b. idi:(e) DbDI ke Ikut [–– ‘It isn’t a piece of cane that I see.’ c. idi:(e) DbDI ke Ikut [–– ‘It isn’t the chief that I see.’ d. ekpeJDI edi ufDk [–– ‘Ekpenyong came to the house.’ e. ekpeJDI emen inuen DJDI edi ufDk [–– ‘Ekpenyong picked up the bird and came home.’ a.

4

–––––] –

––––]

–––––] –––– ] – –––––––– –––]

The first three sentences, which differ only in the italicized words (i.e. the tone pattern of /DbDI/) show, in (1a) all Hs, in (1b) Hs where an intervening L has conditioned a lowering of following Hs, and in (1c) a similar lowering of the last four Hs, though without a preceding L to effect the lowering. Sentences (1d) and (1e) show successive lowering of what are analyzable as Hs, with only the initial and final tones being L. The “terracing” seen with the Hs in these two sentences shows the classic effect of downstep. These data reveal the unsatisfactory nature of an analysis that sees the lowered tones as M. First, such an analysis would require /ké I↓kút/ ‘that I see’ in sentences (1a)–(1c) to be identified first as HHH (1a) and later as MMM (1c); in (1b), while the tones of /ké I↓kút/ are phonetically similar to those in (1c), they are best treated as HHH, since they represent a conditioned lowering. Further, the successive lowering in (1d) and (1e) would require a number of mid tones (i.e. Ms of different heights), with /édí/ ‘came’ bearing a different, lower, M in (1e) than in (1d). Winston examines briefly a different solution, which contrasts non-low and low, with non-low tones divided into H and M, and with restrictions on the distribution of M, but finds this equally awkward. He instead proposes “two distinct systems of contrasting tonal units” (1960: 187): first, H vs. L, which accounts for sentence (1a) vs. (1b), and second, “a unit of ‘downstep’,” which operates only in the context of HH, and accounts for sentence (1a) vs. (1c). Winston draws attention to the fact that not only does downstep distinguish sentences, but it is phonemic in its own right, as the words /FbFI/ ‘mosquito’, /D↓bFI/ ‘chief’, and /GbGI/ ‘cane’ demonstrate. His analysis is also insightful in that it focuses attention not on the nature of the tones themselves – e.g. the last four tones of (1a)–(1c) – but on the relation between these tones and preceding tones; the drop is the realization of downstep.

2.3

Automatic and non-automatic downstep

As mentioned, the Efik data included instances of “non-automatic” downstep, in which a H is lowered by an underlying or floating L, and the lowering of a surface H tone under the influence of a preceding surface L, termed “automatic” downstep. These terms were introduced in Stewart (1965) and continue to be used, though many writers use (simply) downstep to refer to “non-automatic” downstep, and “downdrift” when referring to automatic downstep. As discussed below in §7, however, there are in fact different tonal processes grouped together as downdrift, and the use of downdrift in referring to automatic downstep as well

5

Bruce Connell

as other pitch-based phenomena such as declination has led to a lack of clarity in discussions of tonal phenomena. Stewart’s rationale for introducing the two terms was the recognition that in both cases – automatic and non-automatic downstep – the lowering is triggered by a L, in the former case a surface L and in the latter underlying. It should be pointed out that the underlying L was not simply an abstract postulate, but one that, at least in a great many cases, could be confidently established on either synchronic or diachronic evidence. The suggested essential unity of the two cases leads to the reasonable expectations that, first, at the phonetic level the degree of lowering introduced in both automatic and non-automatic downstep should be the same and, second, in languages in which a floating L triggers downstep, a surface L should do the same. While these expectations are indeed often met, this is not always the case, and this has been the subject of research and debate, returned to in §5 below.

2.4

Terracing and discrete level languages

The lowering of tones through downstep, with the new lower height continuing in subsequent tones giving a step-like F0 contour, has come to be known as “terracing.” Languages characterized by terracing were opposed to “discrete level” languages (e.g. Welmers 1959, 1965, 1973), in which “two or more contrastive levels of pitch are maintained from pause to pause with no intersection of actual pitch” (Welmers 1965: 50). Welmers preferred to see the difference in these terms (i.e. terracing vs. discrete) while others, e.g. Stewart (1965), in accord with Winston (1960), viewed terracing as a concomitant result of downstep. Welmers adds that in a discrete system there is generally no restriction on tone sequences; for example, in a three-tone language such as Jukun (Platoid, Nigeria), with H, M, and L, all possible combinations of these tones – i.e. HM, HL, LH, LM, MH, ML – are in principle permissible (Welmers 1973). In a terracing language with two tones (H, L), ↓H (as opposed to M) is recognizable by its distribution; there is no contrast between (phonetic) LH and LM; i.e. only ↓H follows L. A detailed debate between Stewart on the one hand and Welmers and Schachter on the other was published (Schachter 1965; Stewart 1965; Welmers 1965), with Stewart’s view ultimately holding sway. This view remained the basis of our understanding of downstep until the advent of autosegmental phonology (Leben 1973; Goldsmith 1976; chapter 14: autosegments) and feature geometry (Clements 1983; chapter 27: the organization of features), and although mechanisms available in these theories permitted an enhanced understanding of downstep, it is also accurate to say the existing understanding of downstep, together with other tonal phenomena in sub-Saharan African languages, laid the foundation for the development of autosegmental theory. Welmers’ strict dichotomy between terracing and discrete level languages is now seen as untenable, in that there seem to be clear cases of languages that are to some extent discrete level, but in which there is at least limited overlap of tones permitted. Terracing, however, is recognized as an integral aspect of downstep. Clements (1979) addresses the question of terracing in some detail, first examining what was one of the central issues in downstep-related discussion, the iterative application of phonological rules, i.e. whether a rule can apply to its own output (chapter 74: rule ordering). This was seen by some authors as the only way to account for the successive or cumulative lowering of Hs. Clements

Downstep

6

(1979) proposed an alternative view, which helped to lay the foundation for most current views of downstep. Clements’s proposal saw terracing as “the result of intonational processes applying to the tone level frame itself, rather than directly to individual tones” (1979: 358). The occurrence of actual downstepped tones was restricted to initial position in sequences in which pitch was lowered, but they have the effect of precipitating a (downward) register shift, re-establishing the levels at which subsequent tones within a given prosodic unit are realized. Whether this shift affects all tones (i.e. H and L, and M in a three-tone system) within a given prosodic unit, or just H, became an empirical and language specific question.

2.5

Downstep in autosegmental phonology

The development of autosegmental phonology (Leben 1973; Goldsmith 1976) provided the rest of the foundation for our current understanding of downstep and related tonal phenomena. Perhaps the key contribution of autosegmental phonology to our understanding of downstep was its ability to represent tone on a separate tier, and consequently different tone types on different tiers. Several authors have contributed to this debate and its development, exploiting this insight in different ways and to different degrees. Among them are Hyman (1979, 1993), Clements (1983, 1990), Stewart (1983, 1993), Pulleyblank (1986), Yip (1989, 1993), Snider (1990, 1999), and Clark (1993). The most recent and most detailed of these contributions is that of Snider (1999), who provides a proposal for an understanding of downstep that lays considerable emphasis on the incorporation of upstep as a phenomenon to be accounted for by the same means as downstep, as well as a critique of the related approaches, at least where they differ from his own. Snider’s (1999) proposal is presented within the theoretical framework of Register Tier Theory (RTT), which exploits the mechanisms of autosegmental phonology and feature geometry, and their tiered and hierarchical representations. RTT incorporates a Register tier admitting two features, h and l, a Tonal tier with two features, H and L, a Tonal Root Node tier (TRN), and a Tone-Bearing Unit (TBU) tier. Features on the register and tonal tiers are linked to a node on the TRN tier, and each TRN node is in turn linked to a mora on the TBU tier. This permits the specification of four level tones, which Snider labels Hi (= h, H), Mid2 (= h, L), Mid1 (= l, H), and Lo (= l, L). In a two-tone language (i.e. with H and L, as is frequently found in sub-Saharan Africa), the register feature associated with a particular TBU permits a register shift relative to the preceding TBU’s register: h = higher than the previous register setting, l = lower than the previous register setting. Automatic downstep, then, is represented by the spread of the register feature l, which effects a downward shift, realizing the following high tone a step lower than on the preceding register setting. Snider’s tonal features H and L equate well with Yip’s (1980) [±High], Pulleyblank’s (1986), and Yip’s (1989) [±Raised]. However, his h and l register features provide an advantage over Yip’s and Pulleyblank’s [±Upper], in that they are relative features, whereas [±Upper] is non-relative; it is either high or low. Both Snider’s system and the Yip/Pulleyblank system, then, account well for phonemic tone levels, but the relative nature of Snider’s register features permits description of the cumulative nature and the terracing associated with downstep.

7

3 3.1

Bruce Connell

Distributional characteristics of downstep Downstep type

Non-automatic downstep (often equated with phonologically distinctive downstep) is usually found only in systems that also have automatic (or non-distinctive) downstep, though the two need not co-occur. Automatic downstep is not uncommon in the absence of non-automatic downstep; however, the reverse – i.e. cases of non-automatic downstep in the absence of automatic downstep – is rare. Automatic downstep has been reported in languages such as Hausa (Chadic, Nigeria) by Leben (1984), Lindau (1986), Inkelas and Leben (1990), and Leben et al. (1989), and in Yoruba (Benue-Congo, Nigeria) by Connell and Ladd (1990), Laniran (1992), Akinlabi and Liberman (1995), and Laniran and Clements (2003). In neither of these languages is non-automatic downstep found, and downstep is neither lexically nor grammatically distinctive. Both automatic and non-automatic downstep occur in a great many languages, including not only the paradigm examples of Akan, Efik, and Igbo, but also languages such as Baule (Kwa, Ivory Coast; Ahoua 1996), Bimoba (Gur, Togo; Snider 1998), Chumburung (Kwa, Ghana; Snider 1999), Yala-Ikom (Benue-Congo, Nigeria; Armstrong 1968), and Zande (Adamawa-Ubangi, Democratic Republic of the Congo; Boyd 1981). For only very few languages has non-automatic downstep been reported in the absence of automatic downstep. Three such languages are Dschang (Grassfields Bantu, Cameroon), Ikaan (Benue-Congo, Nigeria; Salffner 2009), and Kikuyu (Bantu, Kenya; Clements and Ford 1980), and each of these present analytic complexities for which consensus has yet to be reached.

3.2

Geographical distribution

All of the above mentioned languages are geographically located in subSaharan Africa, the region best known for downstep, and representative of several different language families and phyla. Downstep is also found in the Americas, the best-known examples being in Central America. Isthmus Zapotec (Oto-Manguean, Mexico; Mock 1981, cited in Yip 2002) shows downstep functioning in a manner expected from research on African languages, i.e. it is triggered by a floating L. Other languages of this region show both downstep and upstep; varieties of Mixtec (Oto-Manguean, Mexico) are discussed in §6 below. Only very rarely has downstep been reported among Asian languages. One such language is Kuki-Thaadow (Tibeto-Burman; Hyman 2007), spoken in northeast India and Burma. Contrary to the well-reported differences between Asian and African tone languages, Kuki-Thaadow’s tone system behaves very much like those found in Africa. Hyman analyzes Kuki-Thaadow as having three underlying tones: HL, H, and L. Downstep occurs when HL precedes H, with L being realized as a downstep on the following H, as shown in (2) (Hyman 2007: 6): (2)

/mêeI vóm thúm hí/ → méeI ↓vóm thúm hí [ – – – – ] ‘these three black cats’

Downstep

8

According to Hyman’s report, downstep in Kuki-Thaadow is realized phonetically by raising of the preceding H, with the amount of raising being determined by the number of downsteps to follow; though no instrumental data is presented, this is reminiscent of Rialland’s (2001) findings for Dagara (see §6).

3.3

Downstep in languages with more than two tones and with tones other than H

The discussion thus far has centered on languages with just two tones, H and L, and downstepping of H tones. This is clearly the most commonly attested situation, though downstep does occur in languages with more than two tones, and with tones other than H, including in some languages with just two underlying tones. The first language with more than a basic two-tone system recognized as having downstep, and as downstepping tones other than H, was Yala-Ikom (Benue-Congo, Nigeria; Armstrong 1968). Yala-Ikom has three contrastive tones, H, M, and L. H is lowered after both L and M, and M is lowered following L; this occurs regardless of whether the tone causing the lowering is underlying or surface (i.e. floating or associated), and terracing results in both situations. The example below (Armstrong 1968: 53) illustrates downstepping of M following a M (Armstrong used the diacritic ’ to indicate downstep); note that the downstepped M is followed by H: (3)

F tabÖl↓ÖnÜ ní [ – – – – – – ] HMMMMH ‘It did not begin in the evening.’

Downstep begins with the third M, a result of the underlying (“latent” in Armstrong’s terms) floating L of /là/ ‘in’, which remains after vowel elision. Evidence for M triggering downstep of H is provided in examples such as the derived verbal noun, /òré↓ré/ ‘eating’; the presence of a floating M is confirmed by cross-dialectal comparison, where M surfaces in Yala-Ogoja /òródré/ ‘eating’. In Vute (Bantoid, Cameroon), also a three-tone language with H, M, and L, H apparently undergoes both automatic and non-automatic downstep (Guarisma 1978; Thwing and Watters 1987). Guarisma describes the second H of a HLH as being lowered, but does not comment on whether subsequent Hs are terraced, nor does she comment on the interesting situation of M, and the possibility of overlap of ↓H and M, or whether Ms and Ls lower correspondingly. However by Guarisma’s examples, non-automatic downstep does extend beyond the first tone of the sequence and so terracing exists. Guarisma describes non-automatic downstep as marking associative constructions, but her examples of both nonautomatic and automatic downstep are with associatives. (4)

Igwé sèhí +pr jémí

→ →

[–––] [–––]

‘head (of) abscess’ ‘spots (of) leopard’

Bamileke-Dschang (Grassfields Bantu, Cameroon) is a language with two tones, H and L. Dschang has received considerable attention in the literature on downstep (Tadajeu 1974; Hyman 1985; Pulleyblank 1986; Clark 1993; Manfredi 1993; Stewart 1993; Bird and Stegen 1995; Snider 1999) as the first language (and one of

9

Bruce Connell

very few languages) analyzed as having downstep affecting L as well as H; it has a four-way surface contrast, H, ↓H, L, ↓L. There is considerable divergence among the views of these authors as to the nature of downstep in Dschang; the downstepped H is attributable to a following floating L, with leftward spreading, and this seems largely agreed (Tadajeu 1974; Pulleyblank 1986; Snider 1999). Snider attributes the downstepped L to a floating H, which, when inserted between two Ls, results in downstepping of the second L. Clark (1993), however, prefers to see Dschang with a basic four-tone system and only relatively few occurrences of downstep. So, while there are languages like Yala-Ikom with more than two tones in which both H and M may be downstepped, and others like Vute in which only H is affected, there are no languages reported where (non-automatic) downstep affects M (or L) but not H. Similarly, like Dschang, there are no languages reported where ↓L occurs but not ↓H. Interestingly, for Yala-Ikom, discussed above, Armstrong (1968) reports that the effect of downstep triggered by M is indistinguishable from that triggered by L; i.e. ↓H lowers the same degree regardless of whether L or M is the trigger. On the other hand, in Yala-Ikom ↓H remains higher than M. This is contrary to reports for other three tone languages in which H is downstepped and for which it is claimed ↓H is indistinguishable from M (e.g. Supyire, Gur, Mali; Carlson 1983; Moba, Gur, Togo; Russell 1996, cited in Snider 1999; Bimoba, Gur, Ghana; Snider 1998). Snider (1998) provides instrumental evidence for the phonetic equivalence of ↓H and M in Bimoba.

3.4

Downstep in “non-tonal” languages

The possibility that downstep could account for pitch phenomena in non-tonal languages was first introduced in Pierrehumbert’s (1980) work on English, in which it was proposed that declination could, largely, be accounted for as the successive lowering of pitch accents (chapter 116: sentential prominence in english). The term “catathesis” rather than downstep was adopted for a time (Poser 1984; Beckman and Pierrehumbert 1986), in order to avoid the terminological conflicts inherent in the use of downdrift vs. downstep and automatic vs. non-automatic downstep (e.g. as mentioned briefly in §2.3, and more on which in §7.1.3); usage has since reverted to, and settled on, downstep. Pierrehumbert’s model has evolved considerably, both in her own work and that of others following, broadly speaking, the same tack. It has been applied to several other languages, e.g. Japanese (Japonic, Japan; Pierrehumbert and Beckman 1988; Kubozono 1989), and Dutch (Germanic, The Netherlands and Belgium; van den Berg et al. 1992), with discussion on the nature of downstep in these languages largely being separate from the debate on its functioning in African and other tone languages. One of the key issues has been the extent to which downstep is distinguishable from declination; in other words, can or does downstep account for all downward movement of pitch across a phrase or utterance? While there has been some effort to include both downstep and declination in models of intonation in such languages, resulting in a move away from the restrictive position in Pierrehumbert (1980), Ladd (2008) points out the methodological/empirical difficulty in separating the two. (In tone languages this is less problematic; see §7.1.2.) Ladd’s general view of downstep as part of intonation (e.g. 1992, 1993, 2008) sees (intonational)

Downstep

10

downstep as a result of metrical factors; for example (Ladd 2008: 77–78), the English phrase my mother’s diaries is realized either strong–weak (my mother’s diaries) or weak–strong (my mother’s diaries). The weak–strong version has two realizations, one of which has two accentual peaks, with that on diaries being the same as or higher than that on mother’s, and the other with unaccented diaries realized at a pitch lower than that of the peak on mother’s, but not as low as it is in the strong–weak realization; i.e. it is downstepped. So while phonetically similar to the strong–weak realization, the second weak–strong variant, that with downstep, is pragmatically distinct (i.e. focus on mother’s vs. focus on the entire phrase). This view of intonational downstep has become widely accepted.

4

The triggers of downstep

In the paradigm cases it is relatively straightforward to establish, on independent grounds, the existence of a floating L as the trigger of non-automatic downstep. Floating L, however, is not seen as the only possible trigger for downstep. First, as seen earlier in the discussion of Yala-Ikom, downstep (of H) may be triggered by M as well as L. Beyond this, in several languages where there is no independent evidence for floating L, an alternative analysis is preferable. For Kishambaa (Bantu, Kenya), Odden (1986) argues that the three-way surface contrast of HL, HH, and H↓H, is underlyingly HL, H, HH; that is, two surface Hs are treated as a single underlying H associated with two TBUs, while the second in a sequence of two underlying Hs is downstepped: (5)

HH ngoto → ngó↓tó

H ‘sheep’

njoka → njóká

‘snake’

Odden considers and rejects the option of inserting a floating L between the two Hs of /ngó↓tó/, arguing that there is no independent evidence for it, and its sole purpose would be to protect the assumed universality of the Obligatory Contour Principle (OCP), which places a constraint against sequences of identical tones in lexical entries. A similar analysis is proposed by Odden as being appropriate for other languages, including certain dialects of Shona (Bantu, Zimbabwe) and Temne (Atlantic, Sierra Leone).

5

Phonetic aspects of downstep

The description of downstep as the lowering of a H tone, with the establishment of a new, lower ceiling on the realization of subsequent Hs, leaves open the question of what happens to Ls within the same prosodic unit. That is, in sequences such as HLHHLHHLHL, in which the Hs following Ls are all subject to automatic downstep (and recalling that downstep is iterative), do the Ls remain at the same level, or do they also lower, in terrace-like fashion? And if so, do they lower at the same rate as do the Hs? It has been claimed (e.g. Welmers 1973; Hombert 1974) that L tones in such environments do not descend, or if they do,

11

Bruce Connell

it is at a slower rate than for Hs. This claim is called into question by e.g. Clements (1983), who notes that evidence from instrumental studies typically shows that low tones as well as highs fall significantly in pitch in such tone spans. Snider (1998) provides a detailed instrumental analysis of the phonetics of downstep in Bimoba (Gur, Ghana), finding in addition to ↓H being equivalent to M, as mentioned earlier, that in sequences involving downstepping of Hs, both Hs and Ls decline at a roughly similar rate (in percentage terms), and that this rate is substantially greater than what might be expected from the influence of declination alone. Laniran and Clements (2003), for Yoruba (Benue-Congo, Nigeria), also report that Ls lower, though at lesser absolute rate (as measured in Hertz) and with greater variability across speakers than is evident for H tones. Similarly, Urua (1996–97) reports a progressive lowering of Ls in LHLHLH sequences in Ibibio, again at a lower rate than for Hs. In contrast to these findings, Gibbon (1987: 293; see also Gibbon 2001) describes Tem (Gur, Togo) as having “downstep, downdrift, phonetically constant (non-terraced) low tone,” though instrumental data are not presented. In the early literature, downstepped H was typically identified as M, indicating a phonetic realization midway between H and L. This appears to be the typical case, and in some three-tone languages with M, such as Bimoba discussed above, they are phonetically equivalent. There are, however, languages in which a downstepped H is realized at the same level as a preceding L, giving rise to a distinction between partial and total downstep (the terminology apparently due to Meeussen 1970). Kikuyu (Bantu, Kenya; Clements and Ford 1980) and Chumburung (Kwa, Ghana; Snider 1998) are among languages reported to have total downstep.

6 6.1

Upstep and H-raising Upstep

A tonal process (or, more accurately, a collection of processes) termed “upstep” has been reported in a number of languages. It sometimes appears to be the converse of downstep, as the term seems to imply; Snider (1990, 1999), for example, refers to it as an upward register shift, but elsewhere any instance where H is raised has been labeled “upstep.” Snider also draws attention to the fact that while downstep occurs in languages in the absence of upstep, there is only one reported case of a language having upstep in the absence of downstep, viz. Acatlán Mixtec (Oto-Manguean, Mexico; Pike and Wistrand 1974; Snider 1999). In the ideal view, and as predicted by most models, upstep is the opposite of downstep, its symmetrical mirror image that is distinct from the “resetting” found at the beginning of a new prosodic unit following one that has been downstepped. That is, upstep should not only raise a H tone, but set a new, higher, ceiling for Hs subsequent to the upstepped H within the same prosodic unit, and its affect would be cumulative, creating upward terracing; in other words, an upward register shift occurs. The only languages reported as approaching this ideal are Acatlán Mixtec (Pike and Wistrand 1974), Yombe (Bantu, Democratic Republic of the Congo; Meeussen and Ndembe 1964, cited in Welmers 1973), and Mankon (Grassfields Bantu, Cameroon; Hyman 1993).

Downstep

12

In Acatlán Mixtec upstep occurs whenever a morpheme-initial H is followed by another H, including upstepped Hs, so upward terracing is possible and appears unlimited, as shown in (6) (data from Snider 1999: 106; ↑ indicates upstep): (6)

kó neg

Œí↑tú kisses

wá so

ní 2sg

méè baby

→

[kó ↑Œí↑tú ↑wá ↑ní ↑méè] – – –– – – –

Welmers (1973: 90), citing Meeussen and Ndembe (1964), reports Yombe as having upstep, after which “following highs continue on the same level until a low or downstep; the upstep simply shifts the entire sequences of terraces back up one level.” Whether successive upsteps are possible is not reported. In Mankon, as in Acatlán Mixtec and Yombe, H(s) following the upstepped H remain at the new level, but the number of Hs in such sequences appears restricted, and as in Yombe, the occurrence of two successive upstepped Hs is unattested; i.e. upward terracing does not occur. It might also be expected that tones other than H should be affected by upstep; just as Ls are lowered by the downward register shift associated with downstep, Ls should be raised by the upward register shift associated with upstep. This seems unattested, or is at least unreported, e.g. for Acatlán Mixtec. Krachi (Kwa, Ghana; Snider 1990) is another language argued to exhibit upstep of H following L (including floating L), and at a first glance it appears the L is also raised. However, Snider analyzes these cases of apparent L-raising as instances of H-spread and delinking. For upstepped Hs, Krachi shows no evidence of a cumulative effect; a H immediately following an upstepped H return to the level of a H preceding the upstep. Snider argues that this is due to a floating L, immediately downstepping the next H and thereby shifting the register downward, though it is not clear there is any independent evidence for the postulated insertion of a floating L. Ahoua (2002) and Leben and Ahoua (1997) report a raising of H tones in Baule (Kwa, Ivory Coast) which, on the surface appears to mirror declination: a sequence of Hs will start relatively low and increase in pitch throughout an utterance. It differs, however, in that it is appears to be restricted to no more than the first four syllables of an utterance, after which H plateaus, and it appears to be phonologically conditioned. A similar upward pitch movement is reported for Peñoles Mixtec (Oto-Manguean, Mexico; Daly 1993, cited in Yip 2002), and described as upstep, with each H following a floating H a step higher.

6.2

H-raising

Other instances that have sometimes been identified as “upstep” include the H-raising effect reported for several languages, including Yoruba (Connell and Ladd 1990; Laniran 1992; Akinlabi and Liberman 1995; Laniran and Clements 2003), Bimoba (Snider 1998), and Dagara (Gur, Burkina Faso; Rialland and Somé 2000; Rialland 2001), in which (typically) an initial H is raised before L. This can be interpreted in at least two different ways. On the one hand, the H-raising before L may be seen as dissimilation, a form of contrast enhancement; this may be supported by evidence from Yoruba that initial L is lowered before H (Akinlabi and Liberman 1995; Laniran and Clements 2003). An analysis such as this may prove to be appropriate in the case of languages such as Engenni (Benue-Congo, Nigeria;

13

Bruce Connell

Thomas 1974; Snider 1990, 1999), in which every H is raised when preceding a L tone. Alternatively, it can be considered an anticipatory raising, to provide the speaker with sufficient room for subsequent lowering resulting from downstep. Rialland (2001), on the basis of instrumental findings, argues in Dagara that the amount by which initial H is raised is determined by the number of downsteps that follow, although the precise strategy used may vary from speaker to speaker. This form of raising is frequently referred to as “reset,” and in its most usual manifestation involves the raising (or resetting) of H at the beginning of a new prosodic unit. There continues to be debate as to what determines both the height of the reset, and the left edge of the new prosodic unit. As far back as the original (impressionistic) work on Akan (Kwa, Ghana), Stewart (1965) and Schachter (1965) held opposing views as to whether it was determined by preplanning (i.e. the greater the number of subsequent downsteps, the higher the reset) or was the same regardless of the number of following downsteps. As for where in an utterance reset occurs, two general claims have been mooted, one being that speakers must reset when they have reached the bottom of their range, and the other, that prosodic boundaries are determined by higher order units in the overall hierarchical structure of an utterance. The raising of Hs in each of these latter cases discussed is quite different from upstep as reported for Acatlán, and should be treated as a distinct phenomenon. Whatever form the raising of H tones takes, it appears that there are several characteristics whose absence render problematic attempts to unify “upstep” and downstep within a single theoretical treatment. Interestingly, upstep is apparently restricted to H tones, as there are no attested instances of M or L being upstepped, and so far there are no clearly established instances of its being lexically contrastive; Snider (1990) claims that for Krachi it is, though no clear examples are offered.

7

Distinguishing downstep and related phenomena

The discussion thus far has focused on different aspects of downstep and how our understanding of downstep has developed. One problem that has been alluded to, but not addressed explicitly, is the confounding of similar or related phenomena with downstep, reflected in terminological conflicts in the downstep literature. This leads not only to difficulty in comparing findings and claims of different authors, but potentially a lack of clarity in the formulation of research questions; it is thus also related to more theoretical questions, such as whether the different effects are phonetic or phonological in nature, whether they are “local” or “global” phenomena, and what they mean for our understanding of the pitch register or span, or tonal space, and how speakers manipulate this space. In this section I look at the relation between downstep (both automatic and non-automatic), downdrift, and declination, in an effort to show how they are similar and/or different and with the hope of introducing a greater degree of consistency in how these different but related phenomena are labeled and hence integrated into phonological and phonetic theory. Results from experimental work on Hausa (Chadic; Lindau 1986) and Ibibio (Cross River, Benue-Congo, Nigeria; Urua 1996–97, 2002) are presented that illustrate the differences between these types of downtrend. In §8, experimental work on Mambila (Bantoid, Benue-Congo,

Downstep

14

Cameroon) is presented that sheds additional light on the nature of these downtrends, illustrating a particular local lowering that it is suggested could more appropriately be labeled downdrift.

7.1

Downtrends and labeling downtrends

Several different types of pitch downtrend have been identified, including downstep, downdrift, and declination.5 Some writers have distinguished two types of downstep – automatic and non-automatic – and, as presented above (§2.3), have argued for a relationship between the two; others, while not denying a possible connection between the two, have referred simply to downstep and downdrift, with the latter being equivalent to automatic downstep. Other writers (and indeed sometimes the same) have used the term downdrift as equivalent to what is now more commonly referred to as declination, either implicitly or explicitly attributing the same underlying mechanism to both (automatic) downstep and declination.

7.1.1 Declination Declination would seem to be the most basic of these, as it is generally considered to be a phonetic universal (Ladd 1984). The term is commonly used in work on intonation, particularly on European languages, though it is less well known or used in work on African tonal systems, where studies on downstep have been most prevalent. Declination refers to “a gradual modification (over the course of a phrase or utterance) of the phonetic backdrop against which the phonologically specified F0 targets are scaled” (Connell and Ladd 1990: 2). Despite its assumed universality, it may, however, be suspended in questions and other sorts of nondeclaratives (see e.g. Lindau 1986 for Hausa), or in situations where tonal contrasts might be endangered (Hombert 1974; Connell 1999). As mentioned earlier, Ladd (2008) points out the difficulty in distinguishing the effects of downstep from those of declination in languages like English, though he suggests “there is nothing in principle to rule out both downstep and true (i.e. time dependent) declination” (Ladd 2008: 80, fn. 12). He also draws attention to difficulties in devising methods that would permit separating the effects of each. This is clearly more true of nontonal languages than of tone languages. In the latter, the existence and influence of declination can most clearly be seen in phrases consisting of tones all of which have the same phonological value – e.g., all Hs, all Ms, or all Ls – and in a given language constructing test sentences comprised of such sequences may be possible (e.g. Lindau 1986; Connell and Ladd 1990; Connell 2002). In such sequences, any lowering of F0 over the course of an utterance may be attributed to this phonetic effect. (However, declination can occur regardless of tonal combination, and so it is possible that it may be problematic to distinguish declination from automatic downstep in a HLHLHL sequence.) Declination is illustrated in Figure 35.1, with data from Hausa (Lindau 1986), in a phrase that consists of all Hs, and has a decline of approximately 14 percent per second. In this case, declination would seem to be the only factor that contributes to the lowering of F0. 5

A fourth, final lowering, could perhaps also have been included in the present discussion, although it typically has been held separate from declination, and never in my experience been confounded with downstep. However, it is not always a straightforward matter to decide which effect, declination or final lowering, is responsible for an observed steep decline in F0 at the end of an utterance.

15

Bruce Connell 200

100 H H M u u di i

H ya a

H zoo

H H gi daa

Figure 35.1 Declination illustrated in a sequence of Hausa High tones in the sentence Muudii yaa zoo gidaa ‘Muudii came home’, adapted from Lindau (1986)

7.1.2 Downdrift Downdrift is somewhat more difficult to characterize, precisely because, as mentioned earlier, the term has been used in different senses. Most commonly, it has been used synonymously with Stewart’s automatic downstep. Hombert (1974: 171), for example, describes downdrift as “the progressive lowering of a high tone after a low tone,” and in a footnote explicitly equates it with automatic downstep. Similar views are expressed by a range of authors including, more recently, Snider and van der Hulst (1993) and Hyman (2001). However, Hombert also attributes an intonational element to downdrift, observing that Ls also descend, and then suggests that the term downdrift refer to “the lowering of like tones (consecutive or not)” (1974: 172, fn. 6). Given this proviso concerning like tones (or, when it applies to consecutive like tones), it appears to be the same phenomenon as that described earlier as declination; however as we have seen, declination is clearly something quite independent of automatic downstep. Both of these views of downdrift – i.e. that it involves a local assimilation between Ls and Hs, and that it is a phrase or sentence level effect – are found elsewhere in the literature, though it is most frequently characterized as being equivalent to automatic downstep. An illustration of downdrift of a presumably assimilatory nature, through the alternation of Hs and Ls, can again be taken from Lindau’s (1986) study of Hausa (Chadic, Nigeria), and is shown in Figure 35.2. In this figure, as in Figure 35.1, a trend line for the H tone only is represented (in Lindau’s representation, the second sentence also shows a downtrend affecting the Ls, and both map the actual F0 trace). What is important in comparing the two is that the slope in Figure 35.2 is steeper than that in Figure 35.1, with a decline of 33 percent per second; i.e. the slope shown in Figure 35.2 combines the declination of Figure 35.1 with the effect of downdrift, the local assimilation of Hs to Ls.6

7.1.3 Downstep An important, indeed defining, feature of downstep, in addition to its lowering of a H relative to a preceding H (or lowering of other tones relative to preceding 6

Gussenhoven (2004: 101) cites similar evidence for Japanese from Poser (1984).

Downstep

16

200

100 H L Ma a l a m

H ya a

L H auni

L H l eem oo

Figure 35.2 Downdrift in alternating H and L tones in the Hausa phrase Maalam yaa auni leemoo ‘The teacher weighed the oranges’, adapted from Lindau (1986)

340 320 300 280 260 240 220 200 180 160 Hz

é

ms

k

í

300 H

H

k

é r é

600 H H

j(è) ú

k

á

r(á) í

900 L H

H

d é

m

1200 H H

1500

Figure 35.3 Illustration of automatic downstep in the Ibibio phrase /ékíkéré jè úkárá ídém/ (HHHHLHHHH) ‘thought and self-rule’

tones of like phonological value) via a L (either surface or floating) that conditions the lowering is that, within specifiable bounds, the downstepped H sets a new ceiling for all subsequent Hs within a specifiable domain; i.e. these Hs do not rise above the height of the downstepped one, hence the descriptive label terracing. So downstep is generally accepted as a downward shift in register. A further characteristic of downstep, it will be remembered, is its cumulative nature: successive downsteps result in successively lower pitch levels. These characteristics are illustrated in Figures 35.3 and 35.4, showing automatic downstep and non-automatic downstep, respectively, in Ibibio.7 Figure 35.3 shows a sequence of Hs followed by a L, followed by another sequence of Hs; the downward shift of the register is noticeable: not only is the first H following the L realized at a lower F0, but the same applies to all subsequent Hs in the phrase. In Figure 35.4, there are no surface Ls, but the sequence of Hs is interrupted at three locations, indicated by ↓, and four distinct levels, or terraces, are in evidence in this phrase. 7

Thanks to Eno-Abasi Urua for contributing the recordings of the phrases used in Figs. 35.3 and 35.4.

17

340 320 300 280 260 240 220 200 180 160 Hz

Bruce Connell

ú b

ms H

F k

300 H

í

w

H

↓á

ú

600 !H H

b

↓DD I

900 !H

F

b ↓DD I

1200 H !H

1500

Figure 35.4 Illustration of non-automatic downstep in the Ibibio phrase /úbFk í↓wá ú↓bFFI F↓bFFI/ (HHH↓HH↓HH↓H) ‘hand of cassava of king’s kingship’

7.2

Summary and discussion

Declination, then, is generally agreed to be a phonetic effect, a backdrop lowering of F0 that has an extended, or global, domain, often the entire utterance. Downstep, whether automatic or non-automatic, appears to come about through the local interaction of adjacent tones; its terracing effect (the lowering of the ceiling) has frequently been described as key lowering (Stewart 1983) or register shift (Snider and van der Hulst 1993). And while downstep is in a sense “local,” in that a specific site – the adjacent tones – can be identified as to where it is triggered, it is typically global in its domain, as all subsequent tones within a given prosodic unit are affected. However, unlike declination, it is arguably a phonological effect. While some writers attribute the downward shift in register that comes about with downstep to phonetic implementation rules, much of the evidence for this comes from research on languages in which tone is not phonologically contrastive. The fact that in many, perhaps most, tone languages in which downstep occurs it (i.e. at least non-automatic downstep) is phonologically contrastive suggests rather strongly that it must be accounted for in the phonological component of the grammar. This and other arguments for a phonological view of downstep are developed in Snider (1990, 1999) and Snider and van der Hulst (1993). The question therefore arises as to whether each of these terms is identified with a distinct phenomenon, or whether one or more of them is in fact redundant. If some phenomena called downdrift are the same as automatic downstep, and other types of downdrift are the same as declination, why not abandon the term downdrift altogether and simply label things consistently as either automatic downstep or declination, as the case may be, and get rid of the terminological “confusion”? (Or, as has been the practice for some writers, eschew “automatic” and “non-automatic” in favor of “downdrift,” “downstep,” and “declination.”) In fact, the lack of terminological consistency is more than simply the existence of competing labels for the same thing; as shown in the following section, it actually masks to some extent our understanding of tone systems, and may therefore

Downstep

18

lead to the formulation of inappropriate research questions and to questionable theoretical conclusions, including the conflation of what are typologically different languages.

8

Downdrift in Mambila

Mambila (Bantoid, Cameroon) is a language with four level tones; it does not have either automatic or non-automatic downstep (Perrin 1974; Connell 1999, 2002, 2003), and the presence of declination appears to be variable, occurring consistently only with the lowest tone (T4), and never with the highest (T1). Connell (1999, 2002) discusses a pitch phenomenon in Mambila that resembles what has often been called downdrift: a successive lowering of F0 in sequences of alternating tones of different levels. The graphs in Figure 35.5 show averaged plots of pitch traces for a single male speaker for like tone sequences in Mambila. Measurements were taken at the apparent tonal target in each syllable. Since Mambila has essentially level tones, the tonal target was identified as the level portion of F0 in each syllable. This was typically straightforward and involved no more than excluding perturbations resulting from adjacent consonants. The graphs illustrate the absence of declination, or any downtrend, for all tones but T4. For each tone, sentences of different lengths (short and long, or short, medium, and long for T4, are superimposed) in the figure. Figure 35.6 shows averaged pitch plots of sentences comprised of alternating sequences of Hs and Ls for the same speaker, which show a slight downtrend; again, three sentences – short, medium, and long – are represented. Figure 35.7 shows alternating sequences of higher mid (T2) and lower mid tones (T3).

F0 (Hz)

180 160

T1

140

T2 T3

120

T4

100 80 0

2

4

6

8

10

12

Number of syllables Figure 35.5 Averaged pitch traces for like tone sequences in Mambila. Tone 1, long utterance = filled diamonds, short utterance = open squares; Tone 2, long utterance = open diamonds, short utterance = filled squares; Tone 3, long utterance = open squares, short utterance = filled squares; Tone 4: long utterance = open squares, medium utterance = open triangle, short utterance = filled triangle

Bruce Connell

19 180

F0 (Hz)

160 140 120 100 80 0

2

6

4

8

10

Number of syllables Figure 35.6 Averaged pitch traces for T1–T4 alternating tone sequences of three different lengths. Long utterance = filled squares; medium utterance = filled triangle; short utterance = open squares

160

F0 (Hz)

150

140

130

T2

120 0

T3 2

T2

T3 4

T2

T2 6

T3

T2 8

10

Number of syllables Figure 35.7 Averaged pitch traces for T2–T3 alternating tone sequences of two different lengths. Long utterance = filled diamonds; short utterance = open squares

Statistical analyses presented in Connell (2002) confirm the existence of a downtrend in Mambila: a progressive lowering of tones through the interaction of adjacent lower and higher tones; i.e. it appears to meet the criteria for, and can be termed, downdrift. Figure 35.8, on the other hand, shows pitch plots from two speakers, again averages of at least five repetitions, for a phrase with a tonal sequence of T1T4T1T1T1, i.e. HLHHH, and Figure 35.9 for a phrase with a tonal sequence of T4T4T1T1T1 (LLHHH). What is noticeable in these utterances is that the lowered H, i.e. the one immediately following the low, does not establish a new ceiling for subsequent Hs; these rise to the height of the initial H in Figure 35.8 (indeed slightly higher) and a similar rise is in evidence in Figure 35.9.

Downstep

20

180 170

Speaker 1 Speaker 2

160 150 F0 (Hz)

140 130 120 110 100 90 80 1

2

3

4

5

Number of syllables Figure 35.8 Pitch traces of T1T4T1T1T1 (HLHHH) for two Mambila speakers, showing the absence of automatic downstep 180 170

Speaker 1 Speaker 2

160 150 F0 (Hz)

140 130 120 110 100 90 80 1

2

3

4

5

Number of syllables Figure 35.9 Pitch traces of T4T4T1T1T1 (LLHHH) for two Mambila speakers, showing the absence of automatic downstep

Rather than downstep, Mambila has a local interaction, a lowering of a high(er) tone following a low(er) tone, that appears to be simply a phonetic effect that is corrected once the low tone is not involved, allowing the H to regain its former height. That the effect of this interaction is cumulative, i.e. in a

21

Bruce Connell

sequence of HLHLHL the downward trend continues, gives it the appearance of what has often been called automatic downstep, but without the register shift it can hardly be seen as the same phenomenon (viz. automatic downstep) that was illustrated in Figure 35.2 for Ibibio. One may wish to argue that Mambila has both downstep and upstep, i.e. the lowering seen is corrected by a floating H (cf. Stewart 1993 on Ebrié), but this analysis is found wanting on at least two grounds. First, the time it takes for T1 (H) to regain or attain its target is variable: typically one or two syllables, but sometimes within one, and at other times more than two syllables. This, impressionistically at least, appears to correlate with speech rate/style. Second, the tonal system of Mambila becomes unnecessarily complicated in such an analysis: the floating H would simply be a diacritic, with no independent evidence to motivate its existence; there would seem to be no reason to postulate such a device when the effect in question is well explained by other means. What happens in Mambila, then, is not the same as what happens in Ibibio, or other languages (Akan, Baule, Bimoba, Efik, Hausa, Igbo, etc.) that have automatic downstep; recognizing the difference between the two types of downtrend leads to the formulation of a different set of questions about the nature of tone systems and pitch realization. Mambila demonstrates tone co-articulation (e.g. Gandour et al. 1994; Xu 1994), a local, phonetic effect in pitch realization. The effect is cumulative in that lowering will continue in the event of a strict alternation (e.g. HLHL). Such a cumulative effect may appropriately be termed “downdrift.”

9

Conclusion

The paradigm case of downstep occurs in a language with two tones, H and L. H is downstepped before any other tone, regardless of whether the language has two or three (or more) tones, and its lowering introduces a terracing effect: a new ceiling is set for subsequent tones of the same phonological value. In this respect a clear connection exists between non-automatic and automatic downstep. Declination and downstep are distinct phenomena; there is, however, perhaps some question as to what constitutes downdrift, whether it is synonymous with automatic downstep, with declination, or is something separate. I have suggested here that it should be seen as a separate phenomenon. While there currently appears to be a consensus around some of the main issues involved in understanding downstep, several issues have not been discussed here in any depth; there remains a great deal which is unresolved and in need of further research.

REFERENCES Ahoua, Firmin. 1996. Prosodic aspects of Baule. Cologne: Rüdiger Köppe Verlag. Ahoua, Firmin. 2002. The phonology of upsweep. In Rainhard Rapp (ed.) Linguistics on the way into the third millennium, 697–704. Frankfurt: Peter Lang. Akinlabi, Akinbiyi & Mark Liberman. 1995. On the phonetic interpretation of the Yoruba tonal system. In Kjell Elenius & Peter Branderud (eds.) Proceedings of the 13th International Congress of the Phonetic Sciences, vol. 1, 42–45. Stockholm: KTH & Stockholm University.

Downstep

22

Armstrong, Robert G. 1968. Yala (Ikom): A terraced-level language with three tones. Journal of West African Languages 5. 49–58. Beckman, Mary E. & Janet B. Pierrehumbert. 1986. Intonational structure in Japanese and English. Phonology Yearbook 3. 255–309. Bendor-Samuel, John (ed.) 1974. Ten Nigerian tone systems. Jos: Institute of Linguistics. Berg, Rob van den, Carlos Gussenhoven & Toni Rietveld. 1992. Downstep in Dutch: Implications for a model. In Docherty & Ladd (1992), 335–359. Bird, Steven & Oliver Stegen. 1995. The Bamileke Dschang associative construction: Instrumental findings. Edinburgh: Centre for Cognitive Science, University of Edinburgh. Boyd, Raymond. 1981. À propos de la notion de faille tonale. In Gladys Guarisma (ed.) Tons et accents dans les langues africaines, 39–64. Paris: SELAF. Carlson, Robert. 1983. Downstep in Supyire. Studies in African Linguistics 14. 35–45. Christaller, J. G. 1875. A grammar of the Asante and Fante language. Basel: Basel Evangelical Missionary Society. Clark, Mary. 1993. Representation of downstep in Dschang Bamileke. In van der Hulst & Snider (1993), 29–73. Clements, G. N. 1979. The description of terraced-level tone languages. Language 55. 536–558. Clements, G. N. 1983. The hierarchical representation of tone features. In Ivan R. Dihoff (ed.) Current approaches to African linguistics, vol. 1, 145–176. Dordrecht: Foris. Clements, G. N. 1990. The status of register in intonation theory: Comments on the papers by Ladd and by Inkelas and Leben. In Kingston & Beckman (1990), 58–71. Clements, G. N. & Kevin C. Ford. 1980. On the phonological status of downstep in Kikuyu. In Didier Goyvaerts (ed.) Phonology in the 1980s, 309–357. Ghent: E. Story-Scientia. Connell, Bruce. 1999. Four tones and downtrend: a preliminary report on pitch realization in Mambila. In Paul Kotey (ed.) New dimensions in African linguistics and languages: Trends in African linguistics, vol. 3, 75–88. Trenton, NJ: Africa World Press. Connell, Bruce. 2002. Downdrift, downstep and declination. In Gut & Gibbon (2002), 3–12. Connell, Bruce. 2003. Pitch realization and the four tones of Mambila. In Shikeki Kaji (ed.) Cross-linguistic studies of tonal phenomena: Historical development, phonetics of tone, and descriptive studies, 181–197. Tokyo: ILCAA. Connell, Bruce & D. Robert Ladd. 1990. Aspects of pitch realization in Yoruba. Phonology 7. 1–29. Daly, John P. 1993. Representation of tone in Peñoles Mixtec. Dallas: Summer Institute of Linguistics. Docherty, Gerard J. & D. Robert Ladd (eds.) 1992. Papers in laboratory phonology II: Gesture, segment, prosody. Cambridge: Cambridge University Press. Gandour, Jackson, Siripong Potisuk & Sumalee Dechongkit. 1994. Tonal coarticulation in Thai. Journal of Phonetics 22. 477–492. Gibbon, Dafydd. 1987. Finite state processing of tone systems. In Proceedings of the 3rd Conference of the European Chapter of the Association for Computational Linguistics. 291–297. Gibbon, Dafydd. 2001. Finite state prosodic analysis of African corpus resources. In Paul Dalsgaard, Børge Lindberg, Henrik Benner & Zheng-hua Tan (eds.) Proceedings of Eurospeech 2001, 83–86. Aalborg: ISCA Archive. Goldsmith, John A. 1976. Autosegmental phonology. Ph.D. dissertation, MIT. Guarisma, Gladys. 1978. Etudes vouté (langue bantoïde du Cameroun). Paris: SELAF. Gussenhoven, Carlos. 2004. The phonology of tone and intonation. Cambridge: Cambridge University Press. Gut, Ulrike & Dafydd Gibbon (eds.) Typology of African prosodic systems. Bielefeld: University of Bielefeld. Hombert, Jean-Marie. 1974. Universals of downdrift: Their phonetic basis and significance for a theory of tone. Studies in African Linguistics. Supplement 5. 169–183. Hulst, Harry van der & Keith L. Snider (eds.) 1993. The phonology of tone: The representation of tonal register. Berlin & New York: Mouton de Gruyter.

23

Bruce Connell

Hyman, Larry M. 1979. A reanalysis of tonal downstep. Journal of African Languages and Linguistics 1. 9–29. Hyman, Larry M. 1985. Word domains and downstep in Bamileke-Dschang. Phonology Yearbook 2. 47–83. Hyman, Larry M. 1993. Register tones and tonal geometry. In van der Hulst & Snider (1993), 75–108. Hyman, Larry M. 2001. Tone systems. In Martin Haspelmath, Ekkehard König, Wulf Oesterreicher & Wolfgang Raible (eds.) Language typology and language universals: An international handbook, vol. 2, 1367–1380. Berlin & New York: Mouton de Gruyter. Hyman, Larry M. 2007. Kuki-Thaadow: An African tone system in Southeast Asia. University of California Berkeley Phonology Lab Annual Report. 1–19. Inkelas, Sharon & William R. Leben. 1990. Where phonology and phonetics intersect: The case of Hausa intonation. In Kingston & Beckman (1990), 17–34. IPA. 1949. The principles of the International Phonetic Association. London: University College. Kingston, John & Mary E. Beckman (eds.) 1990. Papers in laboratory phonology I: Between the grammar and physics of speech. Cambridge: Cambridge University Press. Kubozono, Haruo. 1989. Syntactic and rhythmic effects on downstep in Japanese. Phonology 6. 39–67. Ladd, D. Robert. 1984. Declination: A review and some issues. Phonology Yearbook 1. 53–74. Ladd, D. Robert. 1992. An introduction to intonational phonology. In Docherty & Ladd (1992), 321–334. Ladd, D. Robert. 1993. In defense of a metrical theory of intonational downstep. In van der Hulst & Snider (1993), 109–132. Ladd, D. Robert. 2008. Intonational phonology. 2nd edn. Cambridge: Cambridge University Press. Laniran, Yetunde. 1992. Intonation in tone languages: The phonetic implementation of tones in Yoruba. Ph.D. dissertation, Cornell University. Laniran, Yetunde & G. N. Clements. 2003. Downstep and high raising: Interacting factors in Yoruba tone production. Journal of Phonetics 31. 203–250. Leben, William R. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT. Leben, William R. 1984. Intonation in Chadic languages. Studies in African Linguistics. Supplement 9. 191–195. Leben, William R. & Firmin Ahoua. 1997. Prosodic domains in Baule. Phonology 14. 113–132. Leben, William R., Sharon Inkelas & Mark Cobler. 1989. Phrases and phrase tones in Hausa. In Paul Newman & Robert Botne (eds.) Current approaches to African linguistics, vol. 5, 45–61. Dordrecht: Foris. Lindau, Mona. 1986. Testing a model of intonation in a tone language. Journal of the Acoustical Society of America 80. 757–764. Manfredi, Victor. 1993. Spreading and downstep: Prosodic government in tone languages. In van der Hulst & Snider (1993), 133–184. Meeussen, A. E. 1970. Tone typologies for West African languages. African Language Studies 11. 266–271. Meeussen, A. E. & D. Ndembe. 1964. Principes de tonologie yombe (Kongo occidental). Journal of African Languages 3. 135–161. Mock, Carol C. 1981. Tone sandhi in Isthmus Zapotec: An autosegmental account. Paper presented at the Ithaca Symposium of PILEI, Cornell University. Odden, David. 1986. On the role of the Obligatory Contour Principle in phonological theory. Language 62. 353–383. Perrin, Mona J. 1974. Mambila. In Bendor-Samuel (1974), 93–108. Pierrehumbert, Janet B. 1980. The phonology and phonetics of English intonation. PhD. dissertation, MIT. Pierrehumbert, Janet B. & Mary E. Beckman. 1988. Japanese tone structure. Cambridge, MA: MIT Press.

Downstep

24

Pike, Eunice V. & Kent Wistrand. 1974. Step-up terrace tone in Acatlán Mixtec (Mexico). In Ruth M. Brend (ed.) Advances in tagmemics, 81–104. Amsterdam: North Holland. Poser, William J. 1984. The phonetics and phonology of tone and intonation in Japanese. Ph.D. dissertation, MIT. Pulleyblank, Douglas. 1986. Tone in Lexical Phonology. Dordrecht: Reidel. Rialland, Annie. 2001. Anticipatory raising in downstep realization: Evidence for preplanning in tone production. In Shigeki Kaji (ed.) Cross-linguistic studies of tonal phenomenon: Tonogenesis, typology, and related topics, 301–322. Tokyo: ILCAA. Rialland, Annie & P. A. Somé. 2000. Dagara downstep: How speakers get started. In Vicki Carstens & Fredrick Parkinson (eds.) Advances in African linguistics: Trends in African linguistics, 251–263. Trenton, NJ: Africa World Press. Russell, Jann. 1996. Some tonal sandhi rules in Moba. Togo: Summer Institute of Linguistics. Salffner, Sophie. 2009. Tone in the phonology, lexicon and grammar of Ikaan. Ph.D. dissertation, University of London. Schachter, Paul. 1965. Some comments on John M. Stewart’s “The typology of the Twi tone system.” Bulletin of the Institute of African Studies 1. 28–42. Snider, Keith L. 1990. Tonal upstep in Krachi: Evidence for a register tier. Language 66. 453–474. Snider, Keith L. 1998. Phonetic realization of downstep in Bimoba. Phonology 15. 77–101. Snider, Keith L. 1999. The geometry and features of tone. Dallas: Summer Institute of Linguistics & University of Texas at Arlington. Snider, Keith L. & Harry van der Hulst. 1993. Issues in the representation of tonal register. In van der Hulst & Snider (1993), 1–27. Stewart, John M. 1965. The typology of the Twi tone system. Preprint from the Bulletin of the Institute of African Studies 1. 1–27. Stewart, John M. 1983. Downstep and floating low tones in Adioukrou. Journal of African Languages and Linguistics 5. 57–78. Stewart, John M. 1993. Dschang and Ebrie as Akan-type total downstep languages. In van der Hulst & Snider (1993), 185–244. Tadajeu, Maurice. 1974. Floating tones, shifting rules, and downstep in DschangBamileke. Studies in African Linguistics. Supplement 5. 283–290. Thomas, Elaine. 1974. Engenni. In Bendor-Samuel (1974), 13–26. Thwing, Rhonda & John Watters. 1987. Focus in Vute. Journal of African Languages and Linguistics 9. 95–121. Urua, Eno-Abasi E. 1996–97. A phonetic analysis of Ibibio tones: A preliminary investigation. Journal of West African Languages 26. 15–25. Urua, Eno-Abasi E. 2002. The tone system of Ibibio. In Gut & Gibbon (2002), 65–85. Ward, Ida C. 1933. The phonetic and tonal structure of Efik. Cambridge: Heffer. Welmers, William E. 1959. Tonemics, morphotonemics, and tonal morphemes. General Linguistics 4. 1–9. Welmers, William E. 1965. Some comments on John M. Stewart’s The typology of the Twi tone system. Preprint from the Bulletin of the Institute of African Studies 1. 28–42. Welmers, William E. 1973. African language structures. Berkeley: University of California Press. Winston, F. Dennis. 1960. The “mid” tone in Efik. African Language Studies 1. 185–192. Xu, Yi. 1994. Production and perception of coarticulated tones. Journal of the Acoustical Society of America 95. 2240–2253. Yip, Moira. 1980. The tonal phonology of Chinese. Ph.D. dissertation, MIT. Yip, Moira. 1989. Contour tones. Phonology 6. 149–174. Yip, Moira. 1993. Tonal register in East Asian Languages. In van der Hulst & Snider (1993), 245–268. Yip, Moira. 2002. Tone. Cambridge: Cambridge University Press.

36

Final Consonants Marie-Hélène Côté

1

Introduction

Final consonants, in the stem, the word, or the phrase, often display properties that set them apart from consonants in other positions. Basic principles of syllabification predict that final consonants are codas (see also chapter 33: syllable-internal structure) and, as such, are expected to pattern like non-final codas. Final consonants thus pose an analytical challenge when this expectation is not fulfilled. Languages in which final consonants simply mirror internal codas are referred to as “symmetrical.” In Manam, only nasals appear in both positions, and stress is regularly attracted to closed syllables, internal and final (1a); default stress is penultimate in the absence of closed syllables (1b). So final consonants in Manam display the same segmental profile and stress-attracting power as internal codas (Buckley 1998; Piggott 1999). (1)

a.

b.

[’embegi] [?u’laI] [ura’pundi] [wa’bubu]

‘sacred flute’ ‘desire’ ‘I waited for them’ ‘night’

Spanish and Selayarese offer other illustrations of the correspondence between internal codas and final consonants. As shown by Harris’s (1983: 14–15) list of word-medial and final rimes in Spanish, the set of permissible codas is the same in both positions and includes any consonantal category, possibly followed by [s]. Final syllables closed by consonants also attract stress, which is consistent with their contributing weight to the final syllable, as coda consonants regularly do cross-linguistically (chapter 57: quantity-sensitivity). In Selayarese (Mithun and Basri 1986), medial codas are restricted to homorganic nasals, the first parts of geminates, and [?]; final consonants are limited to [?] and [I]. Assuming that [?] and [I] lack place specification (e.g. Paradis and Prunet 1993; Lombardi 2002), all codas can be characterized by the absence of independent place features (chapter 7: feature specification and underspecification; chapter 22: consonantal place of articulation): they are placeless or acquire the place of the following onset. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0036

Final Consonants

2

In many other languages, final consonants pattern differently from internal codas. First, the right edge of constituents regularly hosts more consonants than internal codas may accommodate. Second, final consonants may be ignored in the application of metrical processes, while internal codas cannot be. These two tendencies, formulated in (2), define final consonant exceptionality. (2)

a. b.

Segmental immunity Final consonants escape segmental constraints that apply to internal codas. Metrical invisibility Final consonants are ignored in the application of metrical processes.

Both patterns occur in Cairene Arabic. While only one consonant is allowed in phrase-internal codas, two may appear phrase-finally (Wiltshire 2003); the additional final consonant is said to escape the coda conditions applicable elsewhere in the phrase. In addition, word stress is attracted to non-final CVC syllables and final CVCC/CVVC, as opposed to CV and final CVC, as if the final consonant were invisible to the stress assignment algorithm (Hayes 1995). Final consonant exceptionality has attracted considerable attention in the development of modern phonological theory. Segmental immunity (2a), clearly the most widely discussed aspect of final exceptionality, is treated in §2–§4, starting in §2 with a review of various representative patterns (generalizations). Analyses of the special behavior of final consonants have almost exclusively relied on special accommodations to syllable structure in final position; different representational devices are examined and compared in §3 (representations). Representations, however, provide only part of the story: they offer a formal frame for the expression of the specificity of final consonants, but no explanation for it. §4 (motivations) is concerned with the formal, grammatical, or functional factors that have been called upon to account for the freedom of occurrence of right edge consonants. Metrical invisibility (2b) and its relationship to segmental immunity are addressed in §5. Final consonants are also implicated in other processes, which are not reviewed in this chapter, since they fall under different topics. First, if final consonants appear with greater freedom, they are also regularly subject to deletion processes (chapter 68: deletion). Final clusters variably simplify in many languages. The factors that govern simplification, however, appear to be relevant to final and non-final clusters alike (e.g. Côté 2004), and I have chosen not to address this topic. Single final consonants also delete, giving rise to various types of C/Ø alternations. Examples include French liaison (see chapter 112: french liaison), linking [r] in non-rhotic dialects of English (Hay and Sudbury 2005; among many others), and Maori verbal forms (Blevins 1994). Interestingly, such cases may involve a re-analysis of historically word-final consonants as epenthetic consonants (Vennemann 1972). Finally, final consonants are subject to resyllabification with a following initial segment. Both C/Ø alternations and resyllabification fall under the scope of external sandhi phenomena.

2

Segmental immunity: Generalizations

The immunity of final consonants emerges in static segmental distributions in the lexicon (§2.1) and in the asymmetrical application of segmental processes (§2.2).

Marie-Hélène Côté

3

More consonants are licensed in final than in internal coda position, allowing additional segmental slots (size effects) or a wider range of place, manner, or laryngeal contrasts (feature effects).

2.1

In the lexicon

Eastern Ojibwa (Piggott 1991, 1999) and Tojolabal (Supple and Douglass 1949; Lombardi 1995) exemplify increased licensing possibilities in final position in manner and laryngeal features. In Ojibwa, while nasals and fricatives are permissible codas in all positions in the word (3a), stops are only allowed word-finally (3b). In Tojolabal, the contrast between plain/aspirated and laryngeal stops and affricates is neutralized in word-internal codas, where only plain segments appear (4a), but remains active in onsets and word-finally (4b). (3)

a.

b.

(4)

a. b.

[baIgisin] [mo(œkine(] [wi(ja(s] [nindib] [ninik]

‘it falls’ ‘it is full’ ‘meat’ ‘my head’ ‘my arm’

[hutp’in-] [?atnija] [potot’] [k’ak]

‘to push’ ‘you bathed’ ‘class of plant’ ‘flea’

French illustrates the presence of additional consonantal slots word-finally. While it admits a large variety of final clusters of up to four consonants (5), all morphemeinternal clusters may be analyzed with codas limited to one consonant (Dell 1995). Final clusters include sequences of rising sonority, in violation of the Sonority Sequencing Principle (e.g. Clements 1990; chapter 49: sonority). This can be taken as a further indication that final consonants are not regular codas: they exceed the possibilities offered by the syllable template applicable elsewhere in the word not only in terms of the number of segments, but also in their relative autonomy with respect to general syllabic principles. (5)

[adDpt] [serkl] [√bidekstr]

‘adopt’ ‘circle’ ‘ambidextrous’

English offers different kinds of final exceptionality effects. As in French, more consonants are found finally than in internal codas: up to three in monomorphemic words (e.g. next) and four with the addition of word-level suffixes (e.g. thousandths) vs. only one internally (exceptionally two, as in empty; see Borowsky 1986). Unlike French, however, English does not tolerate word-final sequences of an obstruent followed by a non-sonorant. In addition, English displays asymmetries in vowel + consonant combinations. Word-finally, long vowels are followed by any consonant (6a); morpheme-internally, long vowels in closed syllables appear in restricted contexts: before fricative + stop (6b) or a sonorant homorganic with the following onset (6c), often with additional combinatorial constraints. Coronal obstruents enjoy

Final Consonants

4

a special status in final position (chapter 12: coronals). For example, long vowels are not followed by clusters except coronal ones (6d), [d] is the only voiced stop allowed after nasal consonants (6e), and vowel reduction is more likely to apply before coronals than non-coronals (6f) (e.g. Borowsky 1986; Burzio 2007). (6)

2.2

a. b. c. d. e. f.

soap, reach pastry, auspices, after1 council, chamber, example wild, paint blind, bond (vs. bomb, long) Unreduced final vowel: Adirondack, insect, chipmunk Reduced final vowel: Connecticut, Everest, elephant

In segmental processes

The asymmetrical application of segmental processes may also give rise to final exceptionality effects. For example, word-internal complex codas may be simplified by consonant deletion or vowel epenthesis, while final clusters are left intact; final vowel deletion may create configurations that are not tolerated in medial syllables. Various feature-changing processes may also result in additional contrasts being tolerated in final consonants. Although the majority of cases of final immunity appear to involve the word, exceptionality effects have also been reported at the stem and phrase levels. Interestingly, both morphosyntactic (stem) and prosodic (phrase) constituents appear to be targeted. However, the morphosyntactic or prosodic status of the word is usually unclear, being either left unspecified or assumed without argumentation for its status. Note that certain processes described at the word level may actually involve the phrase, since words are often considered in isolation. (See chapter 51: the phonological word.) Numerous patterns can be identified, depending on the specific configuration that is asymmetrically tolerated in final position, the process (deletion, epenthesis, or other) subject to asymmetrical application, and the level (stem, word, phrase) at which it applies. Some combinations are illustrated below (see Côté 2000 for additional cases). Kamaiurá allows codas only word-finally (Everett and Seki 1985). This language has a reduplication process that copies to the right the last two syllables of the base. When the base ends in a consonant, this consonant is lost word-medially, and surfaces only in the reduplicant (7). (7)

/o-mo-kon-mo-kon/ /je-umirik-mirik/

[omokomokon] [jeumirimirik]

‘he swallowed it frequently’ ‘I tie up repeatedly’

Kayardild (Evans 1995) displays a similar effect involving vowel deletion at the phrase level. Word-final [a] deletes phrase-finally but is kept before another word. This is illustrated in (8) with the two words [cirkuç-uI-ka] ‘from the north’ and [}a(-}a] ‘he returned’ pronounced in either order: only the second, phrase-final word loses its final [a]. See Piggott (1991) for other cases of apocope. 1

Long vowels before [ft] (6b) and [mp] (6c) are restricted to [A(], and only occur in some dialects, e.g. Southern British English.

Marie-Hélène Côté

5 (8)

[cirkuçuIka }a(-}] [}a(-}a cirkuçuIk]

Cairene Arabic allows CVCC syllables phrase-finally (9a), but enforces a CVC template phrase-internally. Vowel epenthesis prevents complex codas when clusters of more than two consonants are created through suffixation (9b) and word concatenation (9c) (e.g. Broselow 1980; Wiltshire 2003). (9)

a. b. c.

/katab-t/ /bint/ /katab-t-l-u/ /katabt gawaab/ /bint nabiiha/

[katabt] [bint] [katabtilu] [katabtigawaab] [bintinabiiha]

‘you wrote’ ‘girl’ ‘I/you wrote to him’ ‘you (masc) wrote a letter’ ‘an intelligent girl’

Ondarroa Basque exemplifies a more complex case, in which exceptionality effects apply simultaneously at the word and phrase levels. Three different processes conspire to prevent the appearance of stops and affricates in coda position: vowel epenthesis, stop deletion, and affricate simplification. The choice of the repair strategy depends on a complex interplay of lexical and syntactic factors; see Côté (2000) for details. Root-final stops and affricates are excluded before consonantinitial suffixes inside words (10a), are optionally retained before consonant-initial words inside phrases (10b), and remain intact phrase-finally (10c). (10)

a.

/kiœket-tsat/ /lapits-Œo/

b.

/i7u kiœket bota dot/ /semat mutil/ /eskats bat/

c.

/lau silbot/ /bost okots/

[kiœketatsat] [lapitsaŒo] [lapisŒo] [i7ukiœket(a)bota7ot] [sema(t)mutil] [eskatsbat] [eskasbat] [eskatsabat] [lausilbot] [bostokots]

‘lock-prolative’ ‘pencil+dim’ ‘I have thrown three locks’ ‘how many boys’ ‘a/one kitchen’

‘four prominent bellies’ ‘five chins’

Balantak displays different feature restrictions, involving manner and place of articulation at the stem level (Broselow 2003). Codas are limited to homorganic nasals inside morphemes (11a) and at the prefix–root juncture, where impossible codas are avoided by place assimilation (11b), deletion (11c), or epenthesis (11d). In contrast, all consonants other than voiced stops and glides may appear rootfinally (12a), and root-final nasals fail to assimilate in place to the following consonant (12b), pointing to the privileged status of the root-final position. (11)

a.

gampal uIgak b. /niI-borek/ /miI-sapit/ c. /saI-loloon/ /mo?-tokol/ d. /mVI-roIor/

[nimborek] [minsapit] [saloloon] [motokol] [moIoroIor]

‘underlayer’ ‘hornbill bird’ ‘lied’ ‘hidden’ ‘one thousand’ ‘to lie down’ ‘to hear’

Final Consonants (12)

a. b.

/siok-ta/ /bantil-kon/ /laigan-ku/ /wuruI-ta/

[siokta] [bantilkon] [laiganku] [wuruIta]

6

‘our (incl) chicken’ ‘inform (benefactive)’ ‘my house’ ‘our (incl) language’

Patterns so far have been described in terms of more consonants being allowed at the right edge. Another language type has been put forward, which requires constituents to end in a consonant. This possibility is instantiated in Yapese, the case most commonly discussed (Piggott 1991, 1999; Broselow 2003; Wiltshire 2003). This language has no internal codas, but a generalized final short vowel deletion process, which results in words ending in a consonant on the surface. (Final long vowels shorten but do not delete.) The status of Yapese as a distinct type is questionable. As in Kayardild (8), vowel deletion applies finally but not internally (at the word level rather than the phrase), leading to the same generalization as other cases of final consonant immunity: consonants are more easily tolerated in final position. The Yapese pattern may be interpreted as favoring vowel deletion to the extent that it results in phonotactically acceptable forms, rather than actively requiring words to end in a consonant. Menominee is another language in which words end in a consonant, lexically or as a result of final vowel deletion (Bloomfield 1962).2

3

Segmental immunity: Representations

If final consonants escape the conditions applying to codas in other positions, their identity as codas is called into question or must be qualified. At least four directions have been explored to account for the internal–final asymmetry. (i) One consists in admitting position-specific syllable well-formedness conditions, for example by defining different coda constraints for final and non-final syllables. This approach is often taken at a descriptive level but it has not been favored in analytical work. (ii) Uniform syllabic conditions may be maintained across positions but violated at edges under pressure from independent constraints. Recent Optimality Theory (OT) analyses have often relied on this type of reasoning; see §4 for a discussion of some relevant factors. (iii) Another line of research has explored the idea that syllabic structure is irrelevant in all or some of the final immunity effects, which arise through sequential generalizations. It has been argued, for instance, that final clusters in English and other languages are accounted for with a constraint limiting sequences of consonants to only one place of articulation (with coronals unspecified for place in English) (Iverson 1990; Yip 1991; Lamontagne 1993; see also Burzio 2007). Such a sequential generalization allows a unified account of consonant clusters in all positions in the word. Côté (2000) takes a more radically non-syllabic approach to consonant phonotactics in general, and final edge effects in particular, which are defined in terms of segment sequencing and adjacency to 2

If words may be required to end in a consonant, we should find cases of systematic final consonant epenthesis, instead of vowel deletion. Interestingly, I am aware of no such cases at the word level. However, many examples of phrase-final consonant epenthesis are reported (see some examples in Trigo 1988). This asymmetry between the word and the phrase needs to be investigated further, but it suggests that phrase-final epenthesis corresponds to an articulatory closure effect that is not relevant word-finally.

7

Marie-Hélène Côté

constituent boundaries (see §4.4). (iv) However, the most widespread approach in the last 40 years has consisted in elaborating specific syllabic representations that distinguish final consonants from internal codas. The main proposals are presented and compared below: appendices, defective syllables, attachment to higher prosodic constituents, extraprosodicity, and non-moraic consonants. These structural possibilities are illustrated in (13b)–(13f) for the English word wild. Each configuration allows final consonants to escape the conditions that apply to internal codas. The straightforward complex coda approach, which does not involve a specific final representation, is given for comparison in (13a). Notice that these devices are not mutually exclusive and have occasionally been combined. For example, Borowsky (1986) uses both extraprosodic and appendix consonants, Iverson (1990) appendices and empty-headed syllables.3 (13)

a. Coda

b. Appendix

PWd

PWd

q

q

O

N

w

a>

C l

d

c. Defective syllable PWd

O N

C App

w a>

l

d. Attachment to PWd PWd

q

q

O

N

C

w

a>

l

q

d

O

N

C

w

a>

l

d

f. Non-moraic “coda” PWd

e. Extraprosodicity PWd q

3

d

q

O

N

C (Ex)

w

a>

l

d

w

[

[

[

a

>

l

d

Another (partial) solution to the free occurrence of final consonants exploits the idea that certain combinations of consonants form complex phonemes and count as a single unit. It has been applied in particular to [s] + obstruent sequences in English (Fudge 1969; Fujimura and Lovins 1982; Selkirk 1982; Wiese 1996 for German; see also chapter 38: the representation of sc clusters). A word like wild, on this view, contains only one post-vocalic segment. Duanmu (2008) develops a richer theory of complex sounds, which allows him to maintain a simple CVX syllable template; see §4.5.

Final Consonants

8

Appendix (13b): It has been argued that final consonants belong to a separate constituent that hosts consonants that do not fit into the coda. This constituent has been variably called “appendix”4 (e.g. Halle and Vergnaud 1980; Mohanan 1982; Charette 1984; Borowsky 1986; Goldsmith 1990; Iverson 1990; Wiltshire 1994; Booij 1995; Kraehenmann 2001), “affix” (Fujimura and Lovins 1982), and “termination” (Fudge 1969). By stipulation, the appendix is available only in wordfinal position. This constituent is usually attached to the syllable node, as in (13b); it is alternatively part of the word structure, as a sister to the syllable. Two types of affixes may be distinguished, for non-suffixal and suffixal consonants (Goldsmith 1990; Duanmu 2008). Defective syllables (13c): Final consonants are by default taken to be part of the syllable headed by the closest preceding vowel. This assumption is regularly challenged by claims that these consonants in fact belong to a separate syllable, one without a pronounced nucleus (e.g. McCarthy 1979; Selkirk 1981; Iverson 1990; Burzio 1994; Dell 1995; Bye and de Lacy 2000; Cho and King 2003). Representational and terminological details abound here. These special syllables have been termed “degenerate,” “empty-headed,” “minor,” “defective,” “semi-syllables,” and “catalectic.” They may or may not contain a nucleus position; the consonants may be onsets, rimes, or segments attached directly to the syllable node. Different types of degenerate syllables may even be distinguished, for example moraic vs. non-moraic (Nair 1999), or syllables whose nucleus position is empty vs. those whose nucleus is occupied by segmental material shared by the onset (Goad 2002; Goad and Brannen 2003). Final consonants have been considered to be universally onsets, notably in the model of Government Phonology (Kaye 1990; Harris and Gussmann 2002). Others advocate a mixed coda vs. onset approach to final consonants, depending on their segmental profile and behavior, and determined on a language-specific basis or even varying within the same language (Piggott 1991, 1999; Goad 2002; Rice 2003). Attachment to higher prosodic constituents (13d): Final consonants may attach directly to prosodic constituents higher than the syllable, usually the prosodic word (PWd), but also phrasal constituents (e.g. Rubach and Booij 1990; Rialland 1994; Rubach 1997; Wiltshire 1998, 2003; Auger 2000; Spaelti 2002). As a variation on this theme, Piggott (1999) considers that final consonants are codas or onsets of empty-headed syllables licensed by, rather than attached to, the prosodic word. Attachment to higher prosodic constituents implies that the relevant domains for final consonant exceptionality are prosodic in nature. This proposal does not directly account for additional contrasts or slots at the end of morphosyntactic constituents, such as the stem (Broselow 2003). Extraprosodicity (13e): The most prevalent approach to final consonant exceptionality involves the concept of extraprosodicity (or extrametricality; see chapter 43: extrametricality and non-finality). Originally designed to exclude final syllables from stress assignment algorithms (Liberman and Prince 1977), extraprosodicity has been extended to final consonants by Hayes (1980 (citing a presentation by K. P. Mohanan 1982)) for stress and Steriade (1982) for syllabification. Designating final consonants extraprosodic makes them invisible for the 4

The term “appendix” has also been used to refer to non-moraic “coda” consonants (13f) (Sherer 1994; Zec 2007) or consonants attached directly to the syllable or prosodic word nodes (13d) (Rosenthall and van der Hulst 1999).

9

Marie-Hélène Côté

purposes of syllabification, stress assignment, and other processes. The consonants are later adjoined to prosodic structure, in conformity with the principle of Prosodic Licensing (Itô 1986), at a stage in the derivation when syllabic constraints are no longer applicable and metrical structure has already been built. Extraprosodicity is subject to the Peripherality Condition, which restricts it to edges of constituents. Here again, the theme of extraprosodicity allows for numerous variations, regarding its universality and the level at which it operates. Final consonants are claimed to be universally extraprosodic at the lexical level (Borowsky 1986; Itô 1986). At the word level, extraprosodicity is parametrized (Itô 1986) or universal (Piggott 1991). Itô argues that it is turned off post-lexically, but cases of final consonant exceptionality at the phrasal level motivate its possible extension to post-lexical phonology (Rice 1990). Non-moraic “coda” consonants (13f): In languages with moraic codas, additional final consonants may be represented as non-moraic (Lamontagne 1993; Sherer 1994; Hall 2002; Kiparsky 2003). The merits and disadvantages of each of these approaches depend in large part on theory-internal considerations. Each must give up on at least one established principle or generalization of phonological theory. The idea of enriching syllable structure with an appendix constituent has encountered some resistance, since it involves position-specific syllable architecture. Little evidence has been adduced to motivate the appendix as a constituent, which would be expected to act as a trigger or target of some phonological processes; the only case known to me is Mohanan’s (1982) suggestion that [r] depalatalizes in the appendix position in Malayalam. Attachment to higher prosodic constituents violates the principle of exhaustive syllabification, as well as strict layering of prosodic constituents. Extraprosodicity requires multiple levels of syllabification, in itself a contentious issue, and may be interpreted as a weakening of the principles of prosodic phonology (Piggott 1999). On the other hand, it avoids syllabic constituents that are otherwise unnecessary (Steriade 1982). Degenerate syllables imply a higher level of abstractness; empty syllabic positions are either viewed as going against the “uncontroversial assumption that syllables must have nuclei” (Rubach 1997: 570– 571) or, more positively, as a natural consequence of the phonological architecture in which the segmental and suprasegmental structures are independent of each other (Harris and Gussmann 2002). Beyond conceptual considerations, at least three issues must be addressed by all approaches relying on specific representations for final consonants. One issue concerns the featural or combinatorial restrictions that additional consonants allowed at the right edge may themselves be subject to, and how these should be expressed. Final obstruent + sonorant sequences occur in French, but not in English. Germanic languages are also well known for allowing word-final strings of voiceless coronal obstruents. The representations in (13b)–(13f) do not make explicit predictions as to the range of consonants they may host, with the exception of the onset approach, according to which final consonants are expected to display onset-like properties. Final consonants or clusters in many languages do have an onset or coda–onset profile (e.g. French; Dell 1995). But many other patterns appear more challenging for the onset approach, as final consonants are regularly much more limited than onsets. In particular, the claim that final consonants are universally onsets is not readily compatible with languages in which

Final Consonants

10

final consonants share the same segmental profile as internal codas (but see Harris and Gussmann 2002 for discussion). A related problem concerns the coda-like behavior of final consonants, even when they do not have a coda segmental profile. In Québec French, for example, high vowels have a lax variant that surfaces variably before internal codas and categorically before final consonants (14). This process has naturally been characterized as applying in closed syllables. If final consonants are not codas, can they be said to “close” the preceding syllable? Or should the laxing context “in closed syllables” be reformulated without reference to the closed syllables? (14)

[b>I.go] [pÁtr] ~ [pÁt] [kyb]

‘bingo’ ‘beam’ ‘cube’

Finally, why and how are such special representations restricted to final or edge positions – if indeed they are? Appendices, attachment to higher prosodic constituents, degenerate syllables, and extraprosodicity have been excluded inside constituents by stipulation. But in fact, some of these devices have been extended to non-peripheral positions as well (e.g. Rubach and Booij 1990; Rubach 1997). Relevant factors in answering this question are explored in the next section.

4

Segmental immunity: Motivations

Beyond representations, what factors possibly underlie segmental immunity? In particular, why do the exceptional structures described above occur at the right edge of constituents, or what motivates violations of coda conditions in final position? Five factors are discussed here: alignment, positional faithfulness, licensing parameters, perceptual factors, and morphology. Some of these proposals specifically address final immunity effects, others are integrated into a larger typological perspective and cannot be properly evaluated without considering the full range of phonotactic possibilities in internal and final syllables. As a step toward this objective, let us look at the idealized patterns in (15), which specify the number of consonantal slots available in internal codas vs. final position. Internal codas

Final position

Examples

a.

Ø

C

b. c. d. e. f.

C Ø C C CC

CC Ø C Ø C

Kamaiurá (7) Yapese (§2.2) Cairene Arabic (9) see Blevins (1995: 219), e.g. Fijian Manam (1) Warlpiri (Nash 1980) (unclear)

(15)

i. ii.

optional obligatory

Three categories of languages can be identified: languages that allow more consonants finally (15a) and (15b), symmetrical languages (15c) and (15d), and languages that allow more consonants internally (15e) and (15f). Languages with final immunity effects and symmetrical languages have already been discussed,

11

Marie-Hélène Côté

and require few additional comments. Within type (15a), a further division can be made based on whether final consonants are merely allowed (Kamaiurá) or required (Yapese). As mentioned in §2.2, the status of Yapese as a distinct case of final exceptionality requiring a separate analysis remains unclear. Apart from the Yapese case, all consonantal slots in (15) are optional. Symmetrical languages include those with simple codas or open syllables across the board; languages where complex codas are allowed internally and finally are not included in (15). Types (15e) and (15f), the opposite of (15a) and (15b), illustrate “internal immunity” effects. The pattern in (15e) is not uncommon; it applies to languages that allow internal codas but require words to end in a vowel, including Hixkaryana (Derbyshire 1979) and a number of Australian languages (e.g. Warlpiri; Nash 1980). Type (15f), the mirror image of Cairene Arabic, where complex codas are tolerated only in internal syllables, is more controversial, and its existence not well established. More complex internal codas may relate to factors that are independent from the internal–final distinction: they may be allowed in stressed or initial syllables, at morpheme boundaries, or as a result of coda–onset linking that is not available in final position. Languages in which internal codas in any position might be productively allowed to be more complex than at the right edge need to be further investigated. In any case, pattern (15f) does not appear to be as prominent as the opposite type (15b).

4.1

Alignment

Analyses of final consonant immunity have explored a range of alignment constraints, requiring coincidence between the edges of two prosodic or morphological constituents. First, constraints may force marked structures to appear at an edge. Contour tones, for instance, are limited to peripheral positions in many languages (chapter 45: the representation of tone). Similar restrictions have been proposed for non-canonical syllables used to accommodate additional final consonants: semisyllables (Cho and King 2003) and trimoraic syllables (Hall 2002). For example, Align([[[[]q-R, PWd-R) aligns the right edge of a trimoraic syllable with the right edge of a phonological word (Hall 2002). Alternatively, marked structures may be prevented to appear in internal positions. Thus, Bye and de Lacy (2000) and Clements (1997) exclude non-peripheral moraless syllables and extrasyllabic consonants, respectively, with constraints enforcing mora and syllable adjacency.5 Alignment constraints have also contributed to account for final consonant immunity in an indirect fashion. In her analysis of Kamaiurá (7), which allows no internal codas but one final consonant, Wiltshire (1998, 2003) keeps NoCoda unviolated and attaches the final consonant directly to the prosodic word. This disrupts alignment between the right edge of the word and that of a syllable, in violation of Align-R(PWd, q). Ranking this constraint below NoCoda and

5

These adjacency constraints are defined in terms of alignment or contiguity. But contiguity here involves only output forms and no comparison with inputs, as in other formulations of Contiguity, following McCarthy and Prince (1995).

Final Consonants

12

NoDeletion yields the desired candidate, with medial deletion and final extrasyllabicity, as shown in (16). Cairene Arabic, which allows extra consonants only phrase-finally (9), is analyzed in the same way, with NoComplexCoda and Align-R(Phrase, PWd) in place of NoCoda and Align-R(PWd, q); Clements’s (1997) analysis of Berber uses a similar constraint Align-L(Phrase, q).6 (16)

o-mo-kon-mo-kon NoCoda NoDeletion a. .o.mo.kon.mo.kon. b. .o.mo.ko.mo.ko.

Align-R(PWd,q)

*!*

☞ c. .o.mo.ko.mo.ko.n

**! *

*

Not considered here, however, are candidates with internal consonants linked to the PWd. Internal extrasyllabicity violates none of the constraints in (16), so the candidate [.o.mo.ko.n.mo.ko.n], with each [n] attached to the PWd, should win over that in (16c), since it incurs no violation of NoDeletion. With the addition of another alignment constraint banning internal extrasyllabic segments (Clements 1997), candidate (16c) could emerge as optimal, but so could [.o.mo.ko.n.mo.ko], with an internal extrasyllabic [n] and deletion of the final [n] (under the ranking NoCoda >> NoDeletion and the constraint against internal extrasyllabicity at the bottom). This could be argued to correspond to type (15e), the mirror image of (15a); likewise, types (15f) and (15b) are generated with equal likelihood. In this approach, then, final consonants are no more likely to surface than non-final ones, and final and internal immunity effects are treated symmetrically. Yet another use of alignment is found in McCarthy’s (1993) constraint FinalC, corresponding to Align-R(PWd, C), which requires words to end in a consonant. This constraint has served to account for the Yapese pattern (15a.ii) (Broselow 2003; Wiltshire 2003), but it cannot generate type (15b), in which consonant clusters are allowed finally but eliminated in internal codas, since FinalC does not distinguish between one or two final consonants. Lombardi (2002) and Wiltshire (2003) extend FinalC to the phrasal level; Wiltshire also uses Align-R(PWd, V), the vocalic equivalent of FinalC, to derive languages of type (15e).

4.2

Positional faithfulness

Another approach invokes constraints protecting the right edge of constituents, ensuring that final segments or syllables make it to the surface. Syllable wellformedness is obeyed inside constituents, but violated finally under pressure from right-edge faithfulness (chapter 63: markedness and faithfulness constraints). Such faithfulness constraints have taken different formulations, with slightly different effects and predictions; two are offered in (17).

6

Cairene Arabic also excludes internal CVVC syllables, which cannot be accomplished with NoComplexCoda. Wiltshire’s analysis does not address this issue.

13 (17)

Marie-Hélène Côté a.

b.

Anchor-R(GWd) A segment at the right edge of the grammatical word in the output has a correspondent at the right edge of the grammatical word in the input (Broselow 2003).7 Faith-R The rightmost syllable constituent in the word is faithful to its underlying form (Krämer 2003).

Ranked above the constraints banning codas and consonant deletion, any of these constraints straightforwardly derives the Kamaiurá deletion pattern (7). As shown in (18), the surface form [.o.mo.ko.mo.kon.] emerges as optimal: the final consonant is protected from deletion by the undominated Anchor/Faith-R constraint, while the internal [n] is eliminated by NoCoda outranking NoDeletion. (18)

o-mo-kon-mo-kon Anchor-R(GWd) NoCoda Faith-R a. .o.mo.kon.mo.kon. b. .o.mo.ko.mo.ko. ☞ c. .o.mo.ko.mo.ko.n

NoDeletion

**! *!

** *

*

This approach follows a line of analysis that has been developed for other positions: syllable onsets (vs. codas), roots (vs. affixes), stressed syllables (vs. unstressed ones), initial syllables (vs. non-initial ones), and long vowels (vs. short ones). These positions are protected by specific faithfulness constraints that reflect their privileged psycholinguistic or phonetic status (Beckman 1998). A similar treatment for final segments or syllables (vs. non-final ones) is conceivable, although it has not been functionally motivated in the way other privileged positions have. From a more formal perspective, right-oriented constraints, as in (17) and the alignment constraints in §4.1, go against recent claims that constraints may only refer to the left edge, not the right one. This has been argued for Anchor by Nelson (2003) and for all constraints by Bye and de Lacy (2000), who re-analyze right-edge effects, including final consonant exceptionality, without reference to the right edge. Beyond such conceptual questions, right-edge faithfulness constraints fail to account for cases where vowel deletion applies finally but not medially, as in Kayardild (8), since Anchor excludes final deletion. The constraint FinalC, requiring every word to end in a consonant, has been invoked to counter the effect of Anchor and generate the Yapese pattern. As a result, cases of final consonant immunity end up being motivated by distinct constraints, FinalC and Anchor, depending on whether they arise from final vowel deletion or processes applying internally. This analytical distinction seems questionable.

7

The Anchor constraint has a predecessor in McCarthy and Prince’s (1993) Align-R(stem, q), which, applied in the context of the early Parse-Fill approach to faithfulness in OT, has a faithfulness effect similar to that of Anchor in terms of protecting the final segment. This alignment approach is applied to the Cairene Arabic pattern by Wiltshire (1994) and extended to laryngeal contrasts by Lombardi (1995). It must be distinguished from later alignment constraints, discussed in the preceding section, which have no faithfulness effects.

Final Consonants

14

One difference between the Anchor (17a) and Faith-R (17b) approaches is that the former only targets the last segment, while the latter considers the entire final syllable. The last segment formulation cannot derive cases where clusters are maintained finally but eliminated internally (15b), as in Cairene Arabic. Consider the hypothetical example /arspont/ and the three constraints in (19). The only possible winners are the faithful [arspont], if NoDeletion outranks NoComplexCoda, and [arpot], under the opposite ranking. There is no ranking that yields the desired candidate [arpont].8 (19)

arspont ☞ a. .ars.pont. b. .ar.pon.

Anchor-R(GrWd) NoComplexCoda NoDeletion ** *

c. .ar.pont ☞ d. .ar.pot.

** *

* **

Since Krämer’s Faith-R considers the entire final syllable, it correctly derives the desired output [arpont]. However, by evaluating identity, and not only correspondence, between input and output, it also predicts that contrasts may be maintained only in final syllables, including in their onset. Could distinctive voicing, for instance, be found only in the onset of final syllables? Beckman (1998) describes patterns in which initial syllables accommodate more complex codas than non-initial ones. Languages with more complex onsets in final than in non-final syllables remain to be reported, however.9

4.3

Licensing parameters

Final consonant freedom of occurrence has been generated by licensing parameters that allow or even favor final consonants on a language-specific basis. Kaye (1990) proposes a Coda Licensing Principle, according to which all codas must be licensed by a following onset; Harris and Gussmann (2002) provide further arguments for this approach (see also Scheer 2008). It follows that wordfinal consonants cannot be codas; they are onsets to an empty-headed syllable. The occurrence of final consonants is independent from that of internal codas, and depends on whether languages allow final onsets to be licensed by an empty nucleus. A four-way typology emerges from the combination of two binary parameters: whether or not (internal) codas are permitted and final empty nuclei are licensed. This yields four categories of languages (20).

8

Broselow (2003) includes I-O Contiguity, which bans medial deletion and epenthesis, in her constraint set. This does not allow [arpont] to emerge as optimal, and it has the unexpected consequence of generating type (15f), whose status remains unclear. I-O Contiguity serves to derive languages that require words to end in vowels (15c), but this could also be accomplished with Align-R(PWd, V). 9 Faith-R also suffers from a problematic formulation. In the output [arspon], with deletion of the final [t], the last syllable [pon] is faithful to its corresponding string in the input. Without syllabification in the input, it is not clear how final deletion is penalized.

15 (20)

Marie-Hélène Côté a. b. c. d.

No codas, no final empty nuclei No codas, final empty nuclei Codas, no final empty nuclei Codas, final empty nuclei

Type Type Type Type

(15c) (15a) (15e) (15b)

Crucially, this typology excludes the other symmetrical pattern (15d), in which internal codas and final consonants obey the same constraints. This appears too restrictive, as argued by Piggott (1991). In response, Piggott (1999) proposes another parameter based on the notion of remote licensing. All segments must be licensed by a higher prosodic category, either directly by the syllable, or indirectly (remotely) by the PWd or a phrasal constituent. Piggott, unlike Kaye, allows final consonants to be either codas or onsets, depending on their segmental profile; final onsets, which escape coda restrictions, are always licensed remotely, final codas may be licensed by the syllable or a higher constituent, and vowels must be licensed directly by the syllable. Languages vary in whether remote licensing is excluded (all final segments are either vowels or codas), possible (final segments are vowels, codas, or onsets), or obligatory (final segments must be consonants). This last option, reminiscent of McCarthy’s (1993) FinalC, derives Yapese generalized apocope. Unlike Coda Licensing, Piggott’s parametric approach does not provide for languages of type (20c); it also predicts that final consonants that exceed the coda template display onset-like properties, which is not always the case, for example with final coronal obstruents in Germanic languages. An OT account similar in spirit to that of Piggott has been proposed by Spaelti (2002). According to his WeakEdge family of constraints, the right edge of a constituent should contain as little prosodic structure as possible. These constraints favor the attachment of final segments to constituents higher than the syllable in the prosodic hierarchy. Since only consonants may be so attached, WeakEdge establishes consonants as the preferred segment type in final position. Piggott’s and Spaelti’s proposals rely on the idea that constituents should end in a consonant, echoing the constraint Final-C mentioned above, and specifically in a non-coda consonant. Goad (2002) argues that final non-codas are indeed advantageous from a processing viewpoint. Final consonants that are not possible internal codas signal the right edge of words more clearly than final vowels or codas do, since they cannot appear syllable-finally inside words. Likewise, codas signal the right edge of syllables better than vowels do. This parsing argument needs to be tested; for now, two questions arise. First, if it is a desirable thing for words to end in onsets, one should expect to find more cases of generalized wordfinal vowel deletion or epenthesis of a non-coda consonant. As noted above, many languages actually require that words, but not syllables, end in a vowel. Second, if codas are the best indicators of the right edge of syllables, why are they considered marked in syllable typology?

4.4

Perception and adjacency to prosodic boundaries

The syllabic basis of consonant phonotactics has been questioned in the last decade or so (Steriade 1999a, 1999b; Blevins 2003), in particular by proponents of the “licensing by cue” approach, according to which the likelihood that a feature or segment occurs in a given context is a function of its relative perceptibility in that context (chapter 98: speech perception and phonology). Côté (2000) applies

Final Consonants

16

this idea to segmental immunity at the right edge, arguing that the additional licensing possibilities in peripheral positions are motivated by perceptual factors. More consonants are tolerated at edges, because their perceptibility is enhanced by a number of phonetic processes: lengthening, articulatory strengthening, and reduction of the amount of overlap with adjacent segments. The formal architecture is based on two constraint families that require consonants to be followed by a vowel (21a) or adjacent to a vowel (21b), contexts where they benefit from optimal transitional cues. But consonants with stronger internal or contextual cues are less dependent on vocalic transitions and vowel adjacency. This includes final consonants, which are subject to the more specific constraints in (21c) and (21d), where i ranges over the set of prosodic boundaries at the word level and above, including Ø for word-internal consonants (which are not adjacent to any boundary). (21)

a. b. c.

C→V C↔V C]i → V

A consonant A consonant A consonant by a vowel. d. C]i ↔ V A consonant to a vowel.

is followed by a vowel. is adjacent to a vowel. followed by a prosodic boundary i is followed followed by a prosodic boundary i is adjacent

It is assumed that the higher the prosodic boundary a consonant is adjacent to, the more easily it surfaces without the support of an adjacent vowel; consonants not adjacent to any prosodic boundary are the weakest. This is expressed in the rankings in (22), which follows the three-way distinction between phrase-final, word-final, and (word-)internal consonants established in §2.2. Syllable wellformedness and extraprosodicity are irrelevant concepts in this framework, but C → V and C ↔ V obviously bear similarity to, but are not equivalent to, NoCoda and NoComplexCoda/Onset, respectively. (22)

a. b.

C]Ø → V >> C]PW → V >> C]Ph → V C]Ø ↔ V >> C]PW ↔ V >> C]Ph ↔ V

This approach derives the Kamaiurá and Cairene Arabic patterns. In Kamaiurá, NoDeletion is ranked between C]Ø → V and C]PW → V; word-internal consonants have to be followed by a vowel, but word-final ones survive (23). It also directly accounts for cumulative immunity effects, as in Basque (10), with the rankings in (22). Note that C in these constraints may be restricted to specific categories or features (e.g. stops or [coronal] consonants). (23)

o-mo-kon-mo-kon C]Ø → V NoDeletion a. omokonmokon b. omokomoko ☞ c. omokomokon

*!

C]PW→V *

**! *

*

Broselow (2003) interprets the existence of exceptionality effects involving the stem – a morphosyntactic constituent – as contradicting the prosodic basis of the perceptual account. Moreover, enhancement effects are strongest at the phrase

17

Marie-Hélène Côté

level but tend to be weak or inconsistent word-finally (phrase-internally). This may appear at odds with final exceptionality effects being most often reported at the word level. One possibility is that word-final effects arise through a generalization of phrase-final effects, which stabilizes word forms across prosodic contexts (see Hyman 1977 and Gordon 2001 for similar ideas regarding stress). The Balantak case involving the stem investigated by Broselow is consistent with this idea, since stems in this language may surface unsuffixed, in word- and phrase-final position. It remains to be seen whether there are stem-level cases that are incompatible with a phrase-final generalization account. In any case, if a prosodic/perceptual explanation is justified at some level, it would need to be enriched with other grammatical factors to account for the generalization effects. Burzio (2007) pursues a related but complementary perceptual approach to final consonant immunity effects in English. As noted above, additional word-final consonants in English are most often coronals, and unstressed vowels preceding coronal consonants tend to reduce, unlike vowels preceding non-coronals (6f). Burzio argues that coronals being unmarked (“pre-neutralized”), they need not be cued by a full vowel and may survive with the weaker cues provided by a reduced vowel. Non-coronals, especially obstruents, need the cues of a full vowel. In other words, the special status of coronals in English (and presumably other languages) stems from their weaker perceptual demands.

4.5

Morphology

Morphology is obviously involved in some of the segmental immunity effects in final position. Additional final consonants in English (and many other languages) correspond in large part to word-level morphemes (plural and 3rd person [s z], past [t d], ordinal [h]), which are productively added to relevant lexical forms. Final consonants, then, are partly motivated by “the morphology,” this idea being implemented in various ways, depending on one’s view of phonology and its interaction with the morphosyntactic component (chapter 103: phonological sensitivity to morphological structure). Suffixes may be excluded from syllable well-formedness conditions (Selkirk 1982; Harris and Gussmann 2002) or added into morphological constituents separate from the syllable (Goldsmith 1990; Duanmu 2008). The role of morphology may also extend beyond consonantal suffixes, as argued by Duanmu (2008). First, he includes in a final suffix constituent not only true suffixes but also affix-like consonants, which correspond in English to all occurrences of [t d s z h] that exceed the syllable template. In bond, for instance, the final [d] is licensed by some analogical principle to the extent that it could potentially correspond to the past tense affix (of some verb bon). The bomb–bond contrast, with deletion and retention, respectively, of the final stop, relates here to the pseudo-suffixal status of [d], and not to its coronality, as others have argued. Second, any final consonant that is potentially followed by a vowel-initial suffix is supported by a paradigm uniformity or anti-allomorphy principle. For example, the [p] in help is protected by its prevocalic position in helping and helper; its stability in help ensures a uniform expression of the morpheme in all contexts. The exclusion from syllable constituency of all morphologically motivated consonants (plus cases of segmental fusion, not discussed here) allows Duanmu to maintain a simple and cross-linguistically invariant CVX syllable template.

Final Consonants

4.6

18

Summary and discussion

The proposals in §4.1–§4.5 can be compared along two empirical dimensions: (i) do they adequately account for all types of final immunity?; (ii) are they compatible with other phonotactic patterns involving the internal–final relationship? Concerning the first question, alignment constraints, due to their variety, are capable of deriving the full range of immunity effects; parameters such as Coda Licensing and Remote Licensing also account for additional consonantal slots, but their onset status appears at odds with the limited range of final consonants often tolerated beyond syllabic possibilities. Positional faithfulness struggles with final consonant immunity obtained though final vowel deletion; Anchor-based analyses also fail to protect final clusters by targeting only the last segment. Prosodic approaches raise the issue of immunity effects at the end of morphosyntactic constituents. Finally, it remains to be seen how reasonably morphology can embrace all cases of final consonant immunity. Beyond final immunity effects, Coda and Remote Licensing are integrated into parametric systems that claim to account for the full typology of internal–final phonotactic patterns, but the former fails to provide for symmetric languages of type (15d), where final consonants display a typical coda profile, and the latter ignores languages that require words to end in a vowel (15e). Alignment constraints probably offer the most flexible framework and derive the full range of patterns in (15). In fact, the flexibility of alignment constraints is such that final immunity effects enjoy no special status: consonant sequences are as likely to be more complex internally as to be more complex finally. This position might be argued to lack restrictiveness or explanatory power or, conversely, better reflect the range of formally possible patterns, depending in part on the status of type (15f), with complex codas productively allowed only inside constituents. Approaches based on positional faithfulness, perception, or morphology make no specific claims with regard to phonotactic patterns other than final immunity, especially those involving more complex internal codas, which need to be derived by independent constraints or factors. The requirement that words end in vowels might be interpreted as a morphological constraint, but type (15f) would seem to be more challenging. Progress in the analysis of final (and internal) immunity effects rests on a deeper understanding of the patterns in (15), how they arise diachronically, and what factors they are sensitive to.

5

Metrical invisibility

As noted in (2b), final exceptionality also manifests itself prosodically, final CVC syllables patterning like CV ones in stress assignment and vowel length alternations. While regularly noted, the metrical invisibility of final consonants has not given rise to the same analytical diversity as phonotactic immunity. Whether or not metrical invisibility and phonotactic immunity are amenable to a unified approach is also unclear: despite some attempts at a common analysis, there is evidence that the segmental and metrical manifestations of final exceptionality should be kept separate. §5.1 presents the relevant generalizations underlying metrical invisibility and its relationship with segmental immunity; §5.2 addresses its functional motivation.

19

5.1

Marie-Hélène Côté

Generalizations

In quantity-sensitive stress systems, syllable weight is normally determined by the segmental make-up of the rime. In some cases weight assignment also depends on the position of the rime in the word. Of interest here is the pattern where CVC syllables count as light in final position but as heavy elsewhere. See Hayes (1995: 57), Lunden (2006: 1), and Gordon et al. (2010: 142) for lists of languages in which final CVC patterns as light. Various dialects of Arabic illustrate this effect, among them Cairene (Hayes 1995: 67–71; see also chapter 124: word stress in arabic). Cairene has three syllable types: light CV, heavy CVC and CV(, and superheavy CVCC and CV(C, found only in final position. Stress falls on final superheavy syllables (24a), otherwise on heavy penults (24b), otherwise on the penult or antepenult, according to a complex algorithm not described here. This pattern reveals an equivalence between final CVCC/CV(C and penultimate CVC/CV(, which regularly attract stress, and between final CVC and penultimate CV, which do not. This is straightforwardly accounted for if word-final consonants are ignored in the computation of weight and stress. (24)

a. b.

[ka’tabt] [ha–’–a(t] [ka’tabta] [ha(’Ïa(ni] [mu’darris]

‘I wrote’ ‘pilgrimages’ ‘you (masc sg) wrote’ ‘these (masc dual)’ ‘teacher’

As shown in (9) and (24), final consonants are simultaneously invisible to syllabic restrictions and stress assignment, which makes Cairene Arabic a textbook example of final consonant exceptionality, regularly discussed since Broselow (1976) and McCarthy (1979). Such data make it tempting to interpret segmental immunity and metrical invisibility as two effects of a single phenomenon. Yet the two must be distinguished. In Cairene, stress assignment applies at the word level, but the CVC syllable template is enforced at the phrasal level, after resyllabification of word-final consonants with a following vowel, as in (25a). Crucially, stress remains on the second syllable in [œi’ribt] and the first in [’katab] (25b), even though on the surface the second syllable is assumed to be CVC in both cases. (25)

a.

/œiribt ahwa/

[œi.’rib.’tah.wa]

b.

/katab lilmalik/

[’ka.tab.lil.’ma.lik]

‘you drank coffee’ (Selkirk 1981: 222) ‘he wrote to the king’ (Broselow 1976: 16)

Other evidence of the independence of segmental immunity and metrical invisibility can be found in the stress pattern of Hindi-Urdu (Hayes 1980; Hussain 1997). The relevant facts resemble those of Cairene Arabic, with stress falling on final superheavy syllables, otherwise on the rightmost heavy. Unlike Cairene, however, superheavy syllables are not restricted to the final position. Final consonants are metrically invisible, since CVC attracts stress only in nonfinal syllables. But segmental immunity is not involved, since complex codas occur in all positions.

Final Consonants

20

Like stress, vowel length is sensitive to syllable shape, vowel shortening typically occurring in closed syllables and lengthening in open syllables. Again, final consonants appear to be ignored in some languages, with lengthening applying in final CVC syllables or shortening applying only in CVCC ones. Icelandic (Gussmann 2002) regularly stresses the initial syllable of the word. The stressed vowel lengthens in open syllables (26a) and in monosyllables closed by only one consonant (26b), but no lengthening is observed in non-final closed syllables (26c) and in monosyllables closed by two or more consonants (26d). This is a straightforward case of final consonant invisibility: final CVC patterns like non-final CV, and final CVCC like non-final CVC. Similar length alternations are observed in Swiss German (Spaelti 2002) and Menominee (Milligan 2005). (26)

a.

[’pu(] [’sta(ra] b. [’pru(n] [’ha(kh] c. [’senta] [’flaska] d. [’t h jalt] [’rixs]

‘estate’ ‘stare’ ‘edge’ ‘roof’ ‘send’ ‘bottle’ ‘tent’ ‘rich (gen sg masc)’

English shows both stress and length effects of final metrical invisibility. In verbs, final CVCC attracts stress (u’surp, tor’ment), but CVC does not (’edit, de’velop). The stress-attracting power of internal CVC is, however, visible in nouns (a’genda, a’malgam) (Hayes 1982). With respect to length, long vowels in final CV(C syllables regularly correspond to short vowels in final CVCC or non-final CVC, after the addition of a consonantal or syllable-size suffix (keep–kept, wide–width, five–fifth–fifty, wise–wisdom, intervene–intervention). This suggests that shortening applies in closed syllables (internal CVC and final CVCC) but spares final CVC, treated as open by virtue of final consonant invisibility. The representations specific to final consonants discussed in §3 have also been used to derive their metrical invisibility, in particular extraprosodicity (13e) and non-moraic coda consonants (13f). The latter directly accounts for invisibility in the context of stress assignment, if stress depends on syllable weight and weight on moraic structure. Extraprosodicity for metrical purposes is generally kept distinct from phonotactically motivated extrametricality (e.g. Hayes 1995: 106), echoing the remarks above on the non-equivalence between metrically invisible and segmentally immune consonants. Iverson (1990) in fact argues that extraprosodicity should be restricted to stress and excluded from the segmental domain, with cases of segmental immunity and vowel length alternations re-analyzed as involving some of the other devices mentioned in §3: appendices, empty-headed syllables, and sequential cluster constraints (see also Lamontagne 1993).

5.2

Motivation

The lightness of final CVC syllables is motivated by the avoidance of final stress, embodied in OT by the constraint Non-Finality (Prince and Smolensky 2004), which excludes final stressed syllables or head feet (see chapter 43: extrametricality and non-finality). Non-Finality, in conjunction with other constraints, correctly

21

Marie-Hélène Côté

derives stresslessness on final CVC and stress on final CVCC and CVVC in Arabic dialects (Rosenthall and van der Hulst 1999). If Non-Finality generates the facts, the question remains why final stress should be avoided or why final CVC is treated as light. A number of proposals functionally related to the special status of final CVC have recently been put forward. Ahn (2000), Lunden (2006), Hyde (2009), and Gordon et al. (2010) offer explanations that, although distinct, are all related to final lengthening.10 Ahn suggests that the increased vowel duration resulting from final lengthening jeopardizes the contrast between short and long vowels by making final short vowels comparable in duration to non-final long vowels. Stressing the final vowel would weaken the length contrast even further. Lunden and Gordon et al. develop a duration-based account of syllable weight, according to which syllables count as heavy (and attract stress) if their rime is sufficiently longer than that of light syllables. The relative difference in duration between internal CVC and CV is sufficient for CVC to be categorized as heavy, but this may not be the case finally, where final lengthening reduces significantly the durational ratio of CVC to CV. Gordon et al. (2010) also reveal that the languages that asymmetrically treat final CVC as light lack vowel length contrasts in final syllables. It is proposed that phonetic final lengthening tends to be more pronounced when no length contrasts need to be maintained, making it more likely that CVC be interpreted as light. Hyde (2009) focuses on certain properties of final lengthening rather than on duration itself. He notes that, unlike initial lengthening, final lengthening is typically associated with tempo deceleration and declining intensity. These characteristics make final position less compatible with stress, either because diminished intensity makes stress more difficult to perceive, or because the intensity that typically accompanies stress makes it more difficult to decelerate. See Hyman (1977) and Gordon (2001) for related ideas. Gordon invokes intonational factors, final stress being avoided because it would result in the high tone associated with stress and the low final boundary tone being realized on the same syllable. Note that these different factors – duration, final lengthening, length contrasts, tones, deceleration, and intensity – are potentially complementary rather than contradictory in explaining the distinction between final stressless CVC and stressed CVCC/CVVC.

6

Conclusions

Final consonants are implicated in a multiplicity of data and analytical approaches, involving a variety of representations, constraints, and parameters. Among the relevant empirical domains, consonant phonotactics has largely dominated the debates, and the stresslessness of final CVC has drawn some attention, while vowel length alternations have been relatively neglected. Analytically, no unified conception of final consonant exceptionality has really emerged, despite attempts based on extraprosodicity, which have been challenged by evidence for the independence of metrical invisibility and segmental immunity. In a changing theoretical landscape, discussions have tended to shift over time from issues of representation to motivations, ranging from abstract parameters (coda or remote 10

Final lengthening may also relate to length alternations that treat final CVC as open, as in Icelandic (26). I leave this issue open.

Final Consonants

22

licensing) and OT constraints (alignment, positional faithfulness) to the role of morphology and functional explanations (perceptual factors, final lengthening). These approaches are probably to some extent complementary. Focusing on phonotactic patterns, several issues remain to be clarified, concerning the typology of final immunity effects and their relationship to internal immunity effects, whereby more consonants are allowed in internal codas than in final position, and the prosodic or morphosyntactic nature of the constituents involved (phrases, words, stems). To what extent can patterns that require constituents to end in a consonant be analyzed along the same lines as those that merely allow additional consonants? What is the status of patterns displaying more complex codas inside constituents? Should internal and final immunity effects be accounted for in a unified and symmetrical fashion or do they involve distinct factors? One difficulty in answering such questions stems from the fact that positional asymmetries in the complexity of consonant sequences may relate to many factors other than the internal–final contrast, including stressed vs. unstressed syllables, initial vs. non-initial syllables, morpheme boundaries, and coda–onset linking.

ACKNOWLEDGMENTS Thanks to two anonymous reviewers and the Companion editors for helpful comments. I also acknowledge the financial support of the Social Sciences and Humanities Research Council of Canada.

REFERENCES Ahn, Mee-Jin. 2000. Phonetic and functional bases of syllable weight for stress assignment. Ph.D. dissertation, University of Illinois, Urbana. Auger, Julie. 2000. Phonology, variation, and prosodic structure: Word-final epenthesis in Vimeu Picard. In Josep M. Fontana, Louise McNally, M. Teresa Turell & Enric Vallduví (eds.) Proceedings of the 1st International Conference on Language Variation in Europe, 14–24. Barcelona: Universitat Pompeu Fabra. Beckman, Jill N. 1998. Positional faithfulness. Ph.D. dissertation, University of Massachusetts, Amherst. Blevins, Juliette. 1994. A phonological and morphological reanalysis of the Maori passive. Te Reo 37. 29–53. Blevins, Juliette. 1995. The syllable in phonological theory. In John A. Goldsmith (ed.) The handbook of phonological theory, 206–244. Cambridge, MA & Oxford: Blackwell. Blevins, Juliette. 2003. The independent nature of phonotactic constraints: An alternative to syllable-based approaches. In Féry & van de Vijver (2003), 375–403. Bloomfield, Leonard. 1962. The Menomini language. New Haven: Yale University Press. Booij, Geert. 1995. The phonology of Dutch. Oxford: Clarendon Press. Borowsky, Toni. 1986. Topics in the lexical phonology of English. Ph.D. dissertation, University of Massachusetts, Amherst. Broselow, Ellen. 1976. The phonology of Egyptian Arabic. Ph.D. dissertation, University of Massachusetts, Amherst. Broselow, Ellen. 1980. Syllable structure in two Arabic dialects. Studies in the Linguistic Sciences 10. 13–24. Broselow, Ellen. 2003. Marginal phonology: Phonotactics on the edge. The Linguistic Review 20. 159–193.

23

Marie-Hélène Côté

Buckley, Eugene. 1998. Alignment in Manam stress. Linguistic Inquiry 29. 475–496. Burzio, Luigi. 1994. Principles of English stress. Cambridge: Cambridge University Press. Burzio, Luigi. 2007. Phonology and phonetics of English stress and vowel reduction. Language Sciences 29. 154–176. Bye, Patrik & Paul de Lacy. 2000. Edge asymmetries in phonology and morphology. Papers from the Annual Meeting of the North East Linguistic Society 30. 121–135. Charette, Monik. 1984. The appendix in parametric phonology. Studies in African Linguistics. Supplement 9. 49–53. Cho, Young-mee Yu & Tracy Holloway King. 2003. Semisyllables and universal syllabification. In Féry & van de Vijver (2003), 183–212. Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Clements, G. N. 1997. Berber syllabification: Derivations or constraints? In Roca (1997), 289–330. Côté, Marie-Hélène. 2000. Consonant cluster phonotactics: A perceptual approach. Ph.D. dissertation, MIT. Côté, Marie-Hélène. 2004. Syntagmatic distinctness in consonant deletion. Phonology 21. 1–41. Dell, François. 1995. Consonant clusters and phonological syllables in French. Lingua 95. 5–26. Derbyshire, Desmond C. 1979. Hixkaryana. Amsterdam: North-Holland. Duanmu, San. 2008. Syllable structure: The limits of variation. Oxford: Oxford University Press. Evans, Nicholas D. 1995. A grammar of Kayardild. Berlin: Mouton de Gruyter. Everett, Daniel L. & Lucy Seki. 1985. Reduplication and CV skeleta in Kamaiurá. Linguistic Inquiry 16. 326–330. Féry, Caroline & Ruben van de Vijver (eds.) 2003. The syllable in Optimality Theory. Cambridge: Cambridge University Press. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5. 253–286. Fujimura, Osamu & Julie B. Lovins. 1982. Syllables as concatenative phonetic units. Bloomington: Indiana University Linguistics Club. Goad, Heather. 2002. Markedness in right-edge syllabification: Parallels across populations. Canadian Journal of Linguistics 47. 151–186. Goad, Heather & Kathleen Brannen. 2003. Phonetic evidence for phonological structure in syllabification. In Jeroen van de Weijer, Vincent J. van Heuven & Harry van der Hulst (eds.) The phonological spectrum, vol. 2: Suprasegmental structure, 3–30. Amsterdam & Philadelphia: John Benjamins. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Gordon, Matthew. 2001. The tonal basis of weight criteria in final position. Papers from the Annual Regional Meeting, Chicago Linguistic Society 36. 141–156. Gordon, Matthew, Carmen Jany, Carlos Nash & Nobutaka Takara. 2010. Syllable structure and extrametricality: A typological and phonetic study. Studies in Language 34. 131–166. Gussmann, Edmund. 2002. Phonology: Analysis and theory. Cambridge: Cambridge University Press. Hall, T. A. 2002. Against extrasyllabic consonants in German and English. Phonology 19. 33–75. Halle, Morris & Jean-Roger Vergnaud. 1980. Three-dimensional phonology. Journal of Linguistic Research 1. 83–105. Harris, James W. 1983. Syllable structure and stress in Spanish: A nonlinear analysis. Cambridge, MA: MIT Press. Harris, John & Edmund Gussmann. 2002. Word-final onsets. UCL Working Papers in Linguistics 14. 1–42. Hay, Jennifer & Andrea Sudbury. 2005. How rhoticity became /r/-sandhi. Language 81. 799–823.

Final Consonants

24

Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Published 1985, New York: Garland. Hayes, Bruce. 1982. Extrametricality and English stress. Linguistic Inquiry 13. 227–276. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hussain, Sarmad. 1997. Phonetic correlates of lexical stress in Urdu. Ph.D. dissertation, Northwestern University. Hyde, Brett. 2009. The rhythmic foundations of Initial Gridmark and Nonfinality. Papers from the Annual Meeting of the North East Linguistic Society 38(1). 397–410. Hyman, Larry M. 1977. On the nature of linguistic stress. In Larry M. Hyman (ed.) Studies in stress and accent, 37–82. Los Angeles: Department of Linguistics, University of Southern California. Itô, Junko. 1986. Syllable theory in prosodic phonology. Ph.D. dissertation, University of Massachusetts, Amherst. Published 1988, New York: Garland. Iverson, Gregory K. 1990. The stipulation of extraprosodicity in syllabic phonology. Language Research 26. 515–552. Kaye, Jonathan. 1990. “Coda” licensing. Phonology 7. 301–330. Kiparsky, Paul. 2003. Syllables and moras in Arabic. In Féry & van de Vijver (2003), 147–182. Kraehenmann, Astrid. 2001. Swiss German stops: Geminates all over the word. Phonology 18. 109–145. Krämer, Martin. 2003. What is wrong with the right side? Edge (a)symmetries in phonology and morphology. Unpublished ms., University of Tromsø (ROA-576). Lamontagne, Greg. 1993. Syllabification and consonant cooccurrence conditions. Ph.D. dissertation, University of Massachusetts, Amherst. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. Lombardi, Linda. 1995. Laryngeal neutralization and Alignment. University of Massachusetts Occasional Papers in Linguistics 18. 225–247. Lombardi, Linda. 2002. Coronal epenthesis and markedness. Phonology 19. 219–251. Lunden, Anya. 2006. Weight, final lengthening and stress: A phonetic and phonological case study of Norwegian. Ph.D. dissertation, University of California, Santa Cruz (ROA-833). McCarthy, John J. 1979. On stress and syllabification. Linguistic Inquiry 10. 443–465. McCarthy, John J. 1993. A case of surface constraint violation. Canadian Journal of Linguistics 38. 169–195. McCarthy, John J. & Alan Prince. 1993. Generalized alignment. Yearbook of Morphology 1993. 79–153. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 249–384. Amherst: GLSA. Milligan, Marianne. 2005. Menominee prosodic structure. Ph.D. dissertation, University of Wisconsin, Madison. Mithun, Marianne & Hasan Basri. 1986. The phonology of Selayarese. Oceanic Linguistics 25. 210–254. Mohanan, K. P. 1982. Lexical Phonology. Ph.D. dissertation, MIT. Distributed by Indiana University Linguistics Club. Nair, Rami. 1999. Syllables and word-edges. Ph.D. dissertation, Northwestern University. Nash, David. 1980. Topics in Warlpiri grammar. Ph.D. dissertation MIT. Published 1986, New York: Garland. Nelson, Nicole. 2003. Asymmetric anchoring. Ph.D. dissertation, Rutgers University. Paradis, Carole & Jean-François Prunet. 1993. A note on velar nasals: The case of Uradhi. Canadian Journal of Linguistics 38. 425–439. Piggott, Glyne L. 1991. Apocope and the licensing of empty-headed syllables. The Linguistic Review 8. 287–318.

25

Marie-Hélène Côté

Piggott, Glyne L. 1999. At the right edge of words. The Linguistic Review 16. 143–185. Prince, Alan & Paul Smolensky. 2004. Optimality Theory: Constraint interaction in generative grammar. Malden, MA & Oxford: Blackwell. Rialland, Annie. 1994. The phonology and phonetics of extrasyllabicity in French. In Patricia Keating (ed.) Phonological structure and phonetic form, 136–159. Cambridge: Cambridge University Press. Rice, Keren. 1990. Predicting rule domains in the phrasal phonology. In Sharon Inkelas & Draga Zec (eds.) The syntax–phonology connection, 289–312. Chicago: University of Chicago Press. Rice, Keren. 2003. On the syllabification of right-edge consonants: Evidence from Ahtna (Athapaskan). In Stefan Ploch (ed.) Living on the edge, 427–448. Berlin & New York: Mouton de Gruyter. Roca, Iggy (ed.) 1997. Derivations and constraints in phonology. Oxford: Clarendon Press. Rosenthall, Sam & Harry van der Hulst. 1999. Weight-by-position by position. Natural Language and Linguistic Theory 17. 499–540. Rubach, Jerzy. 1997. Extrasyllabic consonants in Polish: Derivational Optimality Theory. In Roca (1997), 551–581. Rubach, Jerzy & Geert Booij. 1990. Edge of constituent effects in Polish. Natural Language and Linguistic Theory 8. 427–463. Scheer, Tobias. 2008. Why the prosodic hierarchy is a diacritic and why the interface must be direct. In Jutta M. Hartman, Veronika Hegedüs & Henk van Riemsdijk (eds.) Sounds of silence: Empty elements in syntax and phonology, 145–192. Amsterdam: Elsevier. Selkirk, Elisabeth. 1981. Epenthesis and degenerate syllables in Cairene Arabic. MIT Working Papers in Linguistics 3. 209 –232. Selkirk, Elisabeth. 1982. The syllable. In Harry van der Hulst & Norval Smith (eds.) The structure of phonological representations, part II, 337–383. Dordrecht: Foris. Sherer, Timothy. 1994. Prosodic phonotactics. Ph.D. dissertation, University of Massachusetts, Amherst. Spaelti, Philip. 2002. Weak edges and final geminates in Swiss German. Unpublished ms., University of California, Santa Cruz (ROA-18). Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Ph.D. dissertation, MIT. Steriade, Donca. 1999a. Alternatives to syllable-based accounts of consonantal phonotactics. In Osamu Fujimura, Brian D. Joseph & Bohumil Palek (eds.) Item order in language and speech, 205–245. Prague: Karolinum Press. Steriade, Donca. 1999b. Phonetics in phonology: The case of laryngeal neutralization. UCLA Working Papers in Linguistics 2: Papers in Phonology 3. 25–146. Supple, Julia & Celia M. Douglass. 1949. Tojolabal (Mayan): Phonemes and verb morphology. International Journal of American Linguistics 15. 168–174. Trigo, Loren. 1988. On the phonological behavior and derivation of nasal glides. Ph.D. dissertation, MIT. Vennemann, Theo. 1972. Rule inversion. Lingua 29. 209–242. Wiese, Richard. 1996. The phonology of German. Oxford: Clarendon Press. Wiltshire, Caroline. 1994. Alignment in Cairene Arabic. Proceedings of the West Coast Conference on Formal Linguistics 13. 138–153. Wiltshire, Caroline. 1998. Extending ALIGN constraints to new domains. Linguistics 36. 423–467. Wiltshire, Caroline. 2003. Beyond codas: Word and phrase-final alignment. In Féry & van de Vijver (2003), 254–268. Yip, Moira. 1991. Coronals, consonant clusters, and the coda condition. In Carole Paradis & Jean-François Prunet (eds.) The special status of coronals: Internal and external evidence, 61–78. San Diego: Academic Press. Zec, Draga. 2007. The syllable. In Paul de Lacy (ed.) The Cambridge handbook of phonology, 161–194. Cambridge: Cambridge University Press.

37

Geminates Stuart Davis

1

Introduction

The term “geminate” in phonology normally refers to a long or “doubled” consonant that contrasts phonemically with its shorter or “singleton” counterpart (see also chapter 47: initial geminates). Such contrasts are found in languages like Japanese and Italian, as exemplified by the minimal pairs in (1) and (2), respectively.1 Languages such as English and Spanish do not have geminates. (1)

Japanese geminate contrast (Tsujimura 2007)2 a. b.

(2)

[saka] ‘hill’ [sakka] ‘author’

Italian geminate contrast a. b.

[fato] [fatto]

‘fate’ ‘fact’

The issue of the phonological representation of geminates has engendered much controversy over the past thirty years. The main issue revolves around how to distinguish formally a geminate consonant from its singleton counterpart in a way that captures the cross-linguistic phonological patterning of geminate consonants. The featural representation of geminate consonants posited in Chomsky and Halle 1

Languages with geminates vary considerably with respect to the durational difference between the geminate and its singleton counterpart. Idemaru and Guion (2008) report a 3:1 ratio in the duration of geminates to singletons in Japanese but only a 1.8:1 ratio for Italian. They further note that there may be other phonetic cues to geminates besides consonantal duration. These include pitch and intensity differences that may provide secondary acoustic cues to a geminate. However, this chapter will not focus on the phonetic properties of geminates, nor on the issue of which types of consonants are more likely to be geminated (but see Pycha 2007, 2009 and Kawahara 2007 for discussion on these issues). Instead, this article will focus on the phonological behavior of geminates and the matter of their representation in phonology. 2 In this chapter, geminate consonants are transcribed by a sequence of two identical letters; long vowels are represented either as a sequence of two identical vowel symbols or with the IPA length mark. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0037

Stuart Davis

2

(1968) as being a single consonant possessing the distinctive feature [+long] has long been considered insufficient, since, as noted by researchers such as Leben (1980), long consonants can behave like a sequence of two consonants for certain phenomena. Leben posited an autosegmental representation of geminates in which a single phoneme is linked to two slots on a skeletal tier that encodes the prosody of the word. This skeletal tier is also referred to as a CV-tier, an X-tier, or a length tier, depending on the specific conception of the researcher. Important earlier works that incorporate a CV-tier include McCarthy (1979, 1981), Halle and Vergnaud (1980), Clements and Keyser (1983) and Hayes (1986), while Levin (1985) posited that the tier consisted of X-slots (see chapter 54: the skeleton). Geminate representation on this view is exemplified by the geminate [kk] of the Japanese word in (1b), as is illustrated in (3). (3)

a. CV-tier representation C

V

s

a

C

C k

b.

X-tier representation

V

X

X

a

s

a

X

X k

X a

As seen in (3), a geminate consonant has one set of features indicated by the single consonant “k” on the phoneme (or melody) tier, whereas it is linked to two slots on a prosodic tier. In (4), we make clear the distinction between a geminate and a singleton using an X-tier that encodes prosody. (4)

Prosodic length analysis of geminates a.

X

X

b.

k (geminate)

X k (singleton)

While the proposals for the representation of geminates in (4) go back thirty years, this representation is specifically argued for by Ringen and Vago (2010), who refer to (4) as the segmental length analysis of geminates. A different representation of geminates from that in (4) is the two-root node analysis of geminates posited by Selkirk (1990) shown in (5). The root node in a feature-geometric framework indicates the major class features of a sound (McCarthy 1988) and it dominates the rest of the specified features. Every phoneme has a root node, but a geminate under this view has two root nodes (RN = root node, c = consonant). (5)

Prosodic length analysis of geminates a.

RN RN c (geminate)

b.

RN c (singleton)

There are at least two main differences between the two-root node analysis of geminates in (5) and the segmental length analysis in (4). First, unlike the X-slots

Geminates

3

in (4), a root node is not considered to be a prosodic unit. Second, the two-root node analysis can more readily capture certain phenomena whereby a single geminate splits into two phonemes, as in the case of Icelandic preaspiration: for instance, underlying /kappi/ ‘hero’ is realized as [kahpi]. (See Selkirk 1990 for a detailed discussion on how the two-root node theory captures this process, but also Keer 1998 for an optimality-theoretic analysis of Icelandic preaspiration that argues against the two-root node analysis.) Probably the standard view of geminate representation in current phonological work is the moraic representation of geminates posited by Hayes (1989) and argued for in Davis (1994, 1999a, 2003), as well as by Topintzi (2008) (see also chapter 55: onsets). On this view, geminates are represented as underlyingly moraic or heavy, as shown in (6) (where UR = underlying representation, and [ = mora): a geminate does not have double linking, be it to two slots on the prosodic tier, as in (4a), or to two root nodes, as in (5a). (6)

Moraic (weight) representation of geminates (Hayes 1989) a. Geminate in UR

b. Single consonant in UR

c (geminate)

c (singleton)

This inherent weight approach to geminates is couched within the theory of moraic phonology as developed in Hayes (1989), which characterizes the prosodic tier as being moraic rather than segmental, as in (3). Specifically, in Hayes’s theory of moraic phonology, a short vowel is underlyingly monomoraic while a long vowel is bimoraic; a geminate consonant differs from a short consonant in that the former is underlyingly moraic while the latter is non-moraic. Sample moraic representations are given in (7), where (7a) shows a short vowel, (7b) a long vowel, (7c) a singleton consonant and (7d) a geminate. We refer to the representation in (7d) as the weight analysis of geminates. (7) Underlying moraic weight representation (Hayes 1989) b.

a. a = /a/

c. a

= /a(/

d. t

= /t/

t = /tt/

The moraic weight representation of geminates in (7d), where a single phoneme is linked underlyingly to a mora on the prosodic tier, is quite different from the length representation shown in (3), in which a single phoneme is linked to two C-slots (or X-slots) on the prosodic tier. These two different views of geminate representation make different predictions with regard to the patterning of geminates in phonology. For example, as noted by Ringen and Vago (2010), if epenthesis is triggered by a word-final consonant cluster (i.e. a word ending in two C-slots), epenthesis would be predicted to occur in a word that ends in a final geminate since the word would end in two C-slots under the geminate representation in (3a). Ringen and Vago discuss Hungarian as a language with this

4

Stuart Davis

epenthesis pattern. On the other hand, if a geminate is represented as moraic, as in (6a) and (7d), epenthesis might not be predicted to occur with a word ending in a geminate, since the consonantal length of a geminate is not segmentally encoded. That is, there would not be two C-slots or two consonantal elements at the end of the word to trigger the epenthesis. Ringen and Vago point out that the Hungarian epenthesis pattern poses a problem for the moraic view. Further, given the weight analysis of geminates in (7d), geminate consonants are predicted to play a role in processes that are sensitive to syllable weight even when singleton (coda) consonants do not. Much of the recent research on geminates has focused on whether geminates display weight properties that are independent of other consonants. This will be discussed shortly. Over the past twenty years, a wide variety of phonological evidence has been brought to bear on the correct representation of geminates. The issue is still controversial.3 All three views of geminate representation presented in this section, namely the prosodic length view in (4), the two-root node view in (5) and the moraic weight view in (6), have been argued for on the basis of the phonological patterning of geminates. Some composite views have even been proposed that combine aspects of the above representations, such as those of Schmidt (1992), Hume et al. (1997) and Curtis (2003). In §2 we will present specific evidence from a variety of phenomena to argue for the inherent weight representation of geminates. In §3, we will examine the behavior of geminates with respect to stress processes, cross-linguistically. In these sections, we will try to maintain a consistent view for the weight analysis in (6a) even when the data presented seem problematic for such a view. In §4 we will reconsider the representational issue and suggest that a composite view of the representation of geminates under a constraint-based approach can account for the patterning of geminates in the world’s languages.

2

The weight analysis of geminates

The underlying weight analysis of geminate consonants, as proposed in Hayes (1989), views a geminate consonant as being underlyingly moraic, as shown in (6a), whereas a non-geminate consonant is underlyingly non-moraic, as in (6b). The weight representation of geminates in (6a) has a number of implications, which will be discussed in this section. One such implication is that if geminates are inherently moraic, they should count as moraic in considering minimal word effects: that is, the cross-linguistically common requirement that content words be at least bimoraic. In §2.1 we show that this is the case for Trukese. A specific structural aspect of the weight representation in (6a) is that geminates do not entail a double linking to two C-slots as in the length representation. This implies that there should be cases in which geminates do not pattern with a sequence of consonants. §2.2 discusses cases of the asymmetrical patterning of geminates and consonant 3

The controversy over geminates has fostered a number of dissertations with a focus on the phonology of geminates. Some of the important ones include Sherer (1994), Ham (1998), Keer (1999), Morén (1999), Kraehenmann (2001), Muller (2001), Curtis (2003), and Topintzi (2006). Although space does not allow me to discuss the wide variety of interesting issues and proposals that are raised in these dissertations and the different positions that are taken, some issues raised in these dissertations will be brought up in the course of this chapter.

Geminates

5

sequences. A third implication that emerges from the weight representation in (6a) is the prediction that there should be languages that treat syllables closed by a geminate (CVG) as heavy but do not otherwise treat syllables closed by a (coda) consonant (CVC) as heavy. In §2.3, we will provide evidence for this prediction by discussing languages that avoid long vowels in syllables closed by a geminate (CVVG), but do not generally avoid long vowels in closed syllables (CVVC). We hold off until §3 the discussion of geminate behavior in weight-sensitive stress systems.

2.1

Trukese initial geminates

One type of evidence for the underlying moraic nature of geminates as in (7d) comes from the bimoraic minimal word requirement in Trukese (also called Chuukese) and the behavior of word-initial geminates with respect to it. Although word-initial geminates are rare, they are attested in a number of languages. (Indeed, the dissertations of Muller 2001 and Topintzi 2006, 2010 are exclusively on initial geminates; see also chapter 47: initial geminates.)4 Muller (2001), whose study incorporates acoustic analyses of word-initial geminates in a variety of languages, including Trukese, concludes that initial geminates are moraic in some languages but not in others, while Topintzi (2006, 2008, 2010), focusing on languages where initial geminates pattern as moraic, argues that such geminates constitute moraic onsets, thus providing examples in which onsets carry weight.5 Trukese provides a clear example of a language where a word-initial geminate patterns as moraic. Consider the data in (8) and (9), which reflect a minimal word constraint on Trukese nouns. The data here are cited from Davis (1999b) and Davis and Torretta (1998), and are mainly taken from Dyen (1965) and Goodenough and Sugita (1980). The relevance of Trukese geminates for moraic phonology has previously been observed by Hart (1991) and Churchyard (1991). 4

It is clear from typological surveys of geminate consonants such as Thurgood (1993) and from the discussion in Pajak (2009) that geminates are most commonly found in intervocalic position and least commonly found when not adjacent to any vowel (e.g. between two consonants). Languages that allow for geminates that are only adjacent to one vowel (e.g. word-initial or word-final geminates), although not common, are not as rare as languages that allow for geminates to occur not adjacent to any vowel. As noted by Pajak (2009), the typological facts correspond to perceptual saliency in that the contrast between a singleton and a geminate consonant is most perceptually salient in intervocalic position and least salient in a position not adjacent to any vowel. 5 Following a suggestion in Hayes (1989), Davis (1999b) proposes that word-initial geminates are moraic but that the mora is not part of the syllable onset. His representation is in (i), while Topintzi’s moraic onset representation is given in (ii) (where the vowel of the syllable is also shown). (i)

(ii)

q

[ c

q

[

[

[

v

c(

v

One difference between (i) and (ii) is that the latter predicts that onset geminates could occur word-internally, not just at the beginning of the word. In support of (ii), Topintzi (2008) provides interesting evidence from Marshallese that word-internal geminates are syllabified as onsets and are not heterosyllabic as commonly assumed in Moraic Theory.

Stuart Davis

6 (8) a. b. c. (9) a. b. c. d. e.

Underlying representation

Output form

/maa/ /tHH/ /oo/

[maa] [tHH] [oo]

Underlying representation

Output form

/etiruu/ /ttoo/ /ŒŒaa/ /ssDD/ /ffHne/

[etiru] [tto] [ŒŒa] [ssD] [ffHn]

Unattested output ‘behavior’ ‘islet’ ‘omen’

*[ma] *[tH] *[o] Suffixed form [n] = relational

‘coconut mat’ ‘clam sp.’ ‘blood’ ‘thwart of a canoe’ ‘advice’

[etiruu-n] [ttoo-n] [ŒŒaa-n] [ssDD-n] [ffHne-n]

Trukese has a general process whereby a word-final long vowel shortens, as seen in (9a)–(9d). This is part of a more general process of final mora deletion, as evidenced in (9e). However, as (8) shows, final mora deletion does not apply if the result would be monomoraic, because Trukese has a minimal word constraint that requires nouns to be at least bimoraic. The fact that the word-final vowel does shorten in (9b)–(9d) strongly suggests that the initial geminate is moraic. That is, an output such as [tto] in (9b) is bimoraic, with a mora being contributed by both the vowel and the geminate. This is supportive of the underlying weight analysis of geminates.6 6

While discussion on Trukese in works such as Muller (2001), Curtis (2003), Topintzi (2006, 2010), and Ringen and Vago (2010) recognizes the inherent weight of Trukese geminates even though they do not all incorporate the underlying weight analysis of geminates (7d), these researchers have often contrasted the moraic behavior of initial geminates in Trukese with the clearly non-moraic behavior of initial geminates in Leti, as originally discussed in Hume et al. (1997). For example, although Leti has initial geminates. it lacks words consisting of an initial geminate followed by a short vowel such as [ppe]. Hume et al. maintain that the lack of such words argues against the moraicity of geminates, given the presence of a bimoraic minimal word condition. Following Davis (1999b, 2003), I maintain in this chapter that Leti is different from Trukese, because the initial geminates of Leti (but not Trukese) are extraprosodic, and that this is supported by the phonotactics of Leti. To see this, it is insightful to compare Leti geminates and word-initial clusters with those of Trukese. In Leti, underlying geminates occur only in word-initial position (Jennifer Muller, personal communication). In Trukese, in contrast, they occur in both word-initial and word-internal positions. Moreover, in Trukese, word-initial clusters other than geminates do not occur (with the exception of a few loanwords). On the other hand, word-initial clusters are pervasive in Leti, allowing for almost any possible sequence of two consonants at the beginning of the lexical word. There are no sonority restrictions on what these two consonants can be. The two consonants in a word-initial sequence can be an obstruent + sonorant, such as [pn pl pr tm tl tr vn vl vr], a sonorant + obstruent, such as [mb ms mv ns rs rv], a sonorant + sonorant, such as [mr nr rm rn rl], or two obstruents, such as [pt tp pk kp tk kt]. Given this patterning, one could realistically analyze the first consonant of a word-initial cluster in Leti as being extraprosodic. The initial consonant of such a cluster is unrestricted and can be identical to the following consonant. This means that the word-initial geminate of Leti consists of a sequence of identical consonants; the first consonant of the sequence would be extraprosodic just like the first consonant of any other word initial cluster. Such an analysis would explain the absence of Leti words like [ppe] or any other word of the shape CCV. With initial extraprosodicity, these forms would not comply with the bimoraic minimum. Given that underlying geminates only occur word-initially in Leti, and given the general phonotactics of word-initial clusters in Leti discussed above, I conclude that Leti presents a very different type of situation from Trukese, where the geminate phonology is tightly integrated with the rest of the phonology (see Davis and Torretta 1998). I suggest that the Leti case has no bearing on the issue of the underlying representation of geminates. Leti allows initial extraprosodic consonants, and the apparent geminate is just a coincidental case where the extraprosodic consonant has the same quality as the following prevocalic consonant.

Geminates

2.2

7

Asymmetrical cases of the patterning of geminates and consonant clusters

In the weight representation of geminates in (6a), repeated below in (10a), a consonant is underlyingly linked to a mora. This is contrasted with the length representation of geminates, in which a consonant is linked to two X-slots or C-slots, as in (10b). (10)

Contrasting representations for geminates a. Geminate in UR: weight representation

b. Geminate in UR: length representation C

c (geminate)

C

c (singleton)

An important difference between the two representations is that the length representation in (10b) tacitly assumes that geminates should pattern similarly to a sequence of two consonants for rules or constraints that reference the CV-tier. Crucially, the weight representation in (10a) does not make such an assumption. There can be cases where geminates will pattern differently than a sequence of two consonants. An example where there is a parallel in patterning between a geminate and a sequence of two consonants is the case of Hungarian epenthesis discussed by Ringen and Vago (2010) and noted earlier in the chapter. In Hungarian, in some verb stems that end in two consonants, an epenthetic vowel occurs after the two consonants when a consonantal suffix is added (e.g. /önt-s/ → [öntes] ‘pour-2sg’). No epenthesis occurs if the verb stem ends in a single consonant when the suffix is added (e.g. /kAp-s/ → [kaps] ‘receive-2sg’). This suggests a constraint for these forms that disallows a word-final sequence of three C-slots. The two representations of geminates in (10) seem to make different predictions for verb stems that end in a geminate when the consonantal suffix is added. On the length view (10b), epenthesis would be predicted since there would be three consecutive C-slots, but such a prediction is not clear, given the weight representation in (10a). As Ringen and Vago observe, epenthesis does occur in a verb stem ending in a geminate (e.g. /fygg-s/ → [fygges] ‘depend-2sg’), which clearly shows that a geminate patterns like a sequence of two consonants with respect to the CV-tier. While this is not conclusive evidence for a length representation, since one could express the epenthesis rule as being sensitive to mora structure, given certain other assumptions about Hungarian, the analysis is more straightforward under the length representation of geminates. Nonetheless, there exist striking cases where geminates do not pattern like a sequence of C-slots. One such example comes from Trukese geminates, shown in (9) above. As seen before, Trukese has word-initial geminates, but does not have word-initial consonant clusters. One would think that if a language allows for a word-initial geminate, it should also allow for a word-initial sequence of two consonants under the length representation of geminates in (10b). Moreover, Trukese has word-internal geminates that are intervocalic (e.g. [tikka] ‘coconut oil’), but Trukese does not generally allow for intervocalic consonant clusters. These

8

Stuart Davis

observations would be hard to account for under the length representation of geminates. One would expect that if two C-slots could occur at the beginning of the word or intervocalically, those two C-slots should not be restricted to just geminates. Furthermore, Trukese geminates do not pattern exactly like single consonants either. This is clearly seen in the observation that words in this language can end in a single consonant (as in the suffixed forms in (9)), but cannot end in a geminate. This contrastive behavior of geminates and consonant clusters on the one hand and geminates with singleton consonants on the other is consistent with and reflective of the weight analysis of geminates in (10a), especially in light of the fact that a word like [tto] in (9b), comprising an initial geminate followed by a short vowel, meets the bimoraic word-minimality requirement. The presence of word-final singleton consonants and the absence of word-final geminates can be seen as a reflection of a high-ranked constraint that disallows words to end in a moraic consonant. A final singleton consonant would not be considered moraic in Trukese. In this regard, it is important to note that Trukese lacks CVC words. This is expected, given bimoraic word minimality, if a word-final consonant is not moraic. Thus the patterning of Trukese geminates provides evidence for the weight representation in (10a) and against the length representation in (10b). (For further details and arguments for the weight representation of geminates in Trukese, see Davis and Torretta 1998.) Another example in which geminates pattern differently from consonant clusters concerns final geminates in Arabic. In some Arabic dialects, such as the Hadhrami dialect as spoken in the town of Ghayl Bawazir near the south coast of Yemen (Bamakhramah 2009, personal communication), consonant clusters are avoided in word-final position (e.g. [’girid] ‘a monkey’ from underlying /gird/, [’binit] ‘a girl’ from underlying /bint/); yet word-final geminates are allowed (e.g. [’rabb] ‘Lord’, [?a’xaff] ‘lighter’). Moreover, word-final geminates in Hadhrami Arabic are different from singleton consonants in that a word-final geminate attracts stress onto the last syllable of the word, but a word that ends in a singleton does not have such impact on stress (e.g. [?a’xaff] ‘lighter’ vs. [’?akbar] ‘greater’) (see also chapter 124: word stress in arabic). Under the length representation of geminates as in (10b) it would be difficult to explain why word-final geminates are allowed when word-final consonantal sequences are avoided. Moreover, the attraction of stress onto the final syllable of a word ending in a geminate is consistent with the weight representation, given that, as observed by Bamakhramah (2009), primary stress typically falls on the rightmost bimoraic syllable. Consequently, the patterning of geminates in languages like Trukese and Hadhrami Arabic calls the length representation into question.7 7

One matter that is historically important to note, but not focused on in the current chapter, concerns properties of geminates that were discussed in the literature in the 1980s in works such as Hayes (1986) and Schein and Steriade (1986) regarding geminate integrity and geminate inalterability. Geminate integrity is the observation that rules of epenthesis tend not to split up a geminate consonant (at least not a “true” geminate consonant, i.e. one that is monomorphemic and non-derived); geminate inalterability refers to the tendency that geminates are resistant to certain rules of segmental phonology (e.g. spirantization in Tiberian Hebrew) that a priori should apply to them. Hayes (1986), in particular, argued that the assumption of a CV-tier can account for integrity and inalterability effects. Integrity effects were accounted for by a length representation of geminates as in (10b), repeated below in (i), because the insertion of a specific vowel into a geminate as in (ii) would violate the prohibition on crossing association lines.

Geminates

2.3

9

Avoidance of CVVG syllables

Hayes’s theory of underlying moraic representation was presented in (7): geminate consonants (7d) differ from singleton consonants (7c) in being underlyingly moraic. Furthermore, Hayes distinguishes long vowels from short vowels by representing the former as underlyingly bimoraic and the latter as monomoraic. Moraic Theory has an implication for the patterning of geminates with respect to weight-sensitive processes – an implication not discussed by Hayes (1989), but taken up by other researchers such as Selkirk (1990), Tranel (1991), and Davis (1999a, 2003). That is, there should be languages in which syllables with a long vowel (CVV) and those closed by a geminate consonant (CVG) count as heavy since they would be bimoraic, while CV syllables and CVC syllables (i.e. syllable closed by a non-geminate consonant) would be considered light or monomoraic. This weight distinction is shown in (11) (G = geminate consonant, C = non-geminate consonant). Syllable weight distinction based on geminates being underlyingly moraic

(11)

heavy CVV CVG

light CV CVC

The system in (11) is predicted to occur under Hayes’s theory in any language with long vowels and geminate consonants that do not regard coda consonants as moraic. The moraic representation for syllables with the structure of (11) is given in (12). (i)

Geminate in UR: Length representation C

C c

(ii) Epenthesis into a geminate C

V

C

c

v

Geminate inalterability effects were handled by a condition on interpretation in segmental rules that association lines in structural description of rules had to be interpreted as exhaustive. A rule like spirantization in Tiberian Hebrew, which only applied to singleton consonants and not to geminates, would include in its rule environment a single C-slot linked to the phoneme. Since the rule environment did not explicitly show double linking as in (i), the rule would fail to apply to geminates. Kenstowicz (1994: 410–416) summarizes important criticisms of the CV account of both geminate integrity and inalterability. He points out that geminate integrity could be called into question if epenthesis is viewed as a two-stage process of inserting a V-slot followed by a late default spell-out rule. However, it is worth noting that geminate integrity effects follow automatically from the weight representation of geminates as in (10a). With respect to geminate inalterability, Kenstowicz specifically calls attention to work by Selkirk (1991), who noted that rules of inalterability tended to always involve spirantization processes, thus suggesting a more general explanation that does not involve the length representation of geminates. Along these lines, Kirchner (2000) approaches the issue of geminate inalterability from a general theory of lenition within a functionally based optimality-theoretic framework.

10 (12)

Stuart Davis Surface syllabification of the division in (11) Heavy (bimoraic) b.

a.

t

a = [ta(]

t

a

t

a = [tat.ta] (the first syllable is bimoraic)

Light (monomoraic) c.

d.

t

a

= [ta]

t

a

t

= [tat]

As seen in (12b), if a geminate is underlyingly moraic, the syllable closed by a geminate (i.e. the first syllable of 12b) will be bimoraic, just like a syllable containing a long vowel (12a). This should be contrasted with a syllable closed by a non-geminate, as in (12d). The syllable in (12d) reflects the syllabification in a language where the rule (or constraint) Weight-by-Position does not apply. Weight-by-Position is the language-specific rule posited by Hayes (1989) that makes a surface coda consonant moraic (see chapter 57: quantity-sensitivity). If the rule applies in a language, any closed syllable (CVC) in that language will behave as bimoraic. If it does not apply, closed syllables should pattern as light or monomoraic. However, because geminates are underlyingly moraic within Hayes’s theory in (7), then a syllable closed by a geminate (CVG) will always be bimoraic, as in (12b), even if Weight-by-Position generally does not apply to codas. Under the assumption that the syllable weight distinction in (11) exists, as Hayes’s theory predicts, a specific prediction that emerges is that we would expect to find languages where a long vowel in a syllable closed by a geminate (CVVG, where G = the first part of a geminate) is avoided while a long vowel in a syllable closed by a non-geminate is not (i.e. CVVC is allowed). Such avoidance of CVVG syllables can be seen as a particular instance of a cross-linguistically common phenomenon of avoiding trimoraic syllables (Prince 1990). When we have a language that treats coda consonants as moraic or heavy, trimoraic syllables are often avoided, as is the case with the dialectal Arabic examples in (13). (See Broselow 1992 and Kiparsky 2003 for overviews on Arabic syllables.) (13)

Avoidance of trimoraic syllables in Arabic dialects a. b.

Cairene Arabic /baab + na/ → [’bab.na] Meccan Arabic /baab + na/ → [’baa.ba.na]

‘our gate’ ‘our gate’

Both Cairene and Meccan Arabic avoid the potentially trimoraic parse of the first syllable of /baab + na/ as [’baab.na]. The dialects, however, differ in how they avoid the trimoraic syllable. While Cairene Arabic favors closed-syllable shortening, Meccan Arabic preserves the underlying vowel length by having vowel

Geminates

11

epenthesis apply between the two consonants to create open syllables, thereby avoiding any trimoraic syllable. In the Arabic dialects, the rule (or constraint) of Weight-by-Position applies, resulting in a moraic coda. However, if we consider the weight division in (11) in which Weight-by-Position does not apply, we would expect to find a language where CVVG syllables but not CVVC syllables are avoided. A potential CVVG syllable would not surface, since it is trimoraic, while a CVVC could still occur, as it would only be bimoraic. Kiparsky (2008a) discusses Swedish dialects where vowel shortening occurs before a geminate but not before a single coda consonant. For example in West Swedish (Kiparsky 2008a: 191) /ruu-dde/ ‘rowed’ surfaces as [rudde], with its underlying long vowel shortened, but no shortening occurs when a long vowel is before a singleton coda consonant. The shortening before a geminate changes the potential trimoraic CVVG syllable to bimoraic CVG. There is no need for shortening in a CVVC syllable since it would be bimoraic, given that Weight-by-Position does not apply. Kiparsky specifically uses the Swedish data to argue for the moraic representation of geminate consonants. Another language displaying this pattern of shortening is the Dravidian language Koya, brought up by Sherer (1994), based on Tyler (1969), and discussed in Davis (1999a, forthcoming). Koya has long vowels, coda consonants and geminate consonants. Sherer notes that there are words in Koya like those in (14a)–(14c), with a long vowel before a coda consonant. Crucially, as Tyler (1969: 6) observes, there are no words that contain a long vowel before a geminate. They are always short, as in (14d). All Koya data are cited from Tyler (1969), with the page numbers provided. (The transcription of the vowel quality is phonemicized and does not reflect the precise allophonic variant.) (14)

a. c.

le(Iga ne(rs

‘calf’ (p. 11) ‘learn’ (p. 76)

b. a(KÕa ‘female’ (p. 8) d. ett ‘lift’ (p. 76)

Sherer additionally notes cases where a stem-final long vowel shortens before a geminate-initial suffix, as the examples in (15) show. (15)

a. b.

ke( + tt + o(KÕu o( + tt + o(KÕu

[ketto(KÕu] [otto(KÕu]

‘he told’ (p. 39) ‘he bought’ (p. 38)

This shortening can be viewed as a way of avoiding trimoraic syllables. Shortening does not occur before a non-geminate consonant, as the examples in (16) illustrate. (16)

a. b.

na(l + ke tuIg + ana( + n + ki

[na(lke] ‘tongue’ (p. 47) [tuIgana(Iki] ‘for the doing’ (p. 90)

In (16), a long vowel surfaces before a syllable-final singleton coda consonant. Since vowel shortening occurs before a geminate in (15), the Koya data in (14)–(16) are consistent with the weight system in (11), in which CVV and CVG syllables are bimoraic whereas CVC syllables are light.8 8

Curtis (2003: 169–170) suggests that the lack of word-internal CVVG syllables in Koya may be due to a shortening effect that geminate consonants have on preceding vowels, since the perceptual cues for vowel length can be blurred in CVVG syllables; thus, Curtis maintains that vowel shortening before geminates is independent of the issue of the moraic status of geminates. However, this does not explain cases like Fula in (19) where avoidance of CVVG is achieved by degemination rather than vowel shortening.

12

Stuart Davis

While the above examples of Koya and Swedish are cases where vowel shortening occurs in syllables closed by a geminate, there are other languages where vowel-lengthening processes are prevented in CVG syllables but not in CVC syllables. This suggests that in such languages geminates are underlyingly moraic, although coda consonants in general are not. Vowel lengthening then does not apply before a geminate, since that would create a trimoraic syllable. This is illustrated by Seto (Southeastern Estonian), discussed by Kiparsky (2008b). According to Kiparsky, Seto has feet that are required to be trimoraic and such a restriction is normally implemented by foot-final vowel lengthening. As a result, a foot with the underlying sequence CV.CVC surfaces as CV.CVVC. However, given an input structure where the final consonant of the foot is part of a geminate, i.e. CV.CVG, no vowel lengthening occurs. This provides evidence that the geminate is underlyingly moraic: that is, foot-final vowel lengthening need not occur in CV.CVG since the foot is already trimoraic. A different case that avoids the surfacing of CVVG syllables can be found in the West African language Fula as discussed by Paradis (1988) and Sherer (1994). Fula avoids CVVG syllables by degemination of the consonant but, importantly, it allows for CVVC syllables, as seen in (17). (17) CVVC syllables in Fula (Sherer 1994: 176) a.

kaakt-e

‘spittle’

b.

caak-ri ‘couscous’

This language has a suffixation process that triggers the gemination of a root-final consonant. Consider the singular/plural alternations in (18). Because of an active constraint that requires geminates to be [−continuant] in Fula, a root-final continuant segment changes to a stop when it geminates. (I thank Abbie Hantgan for help with the Fula data.) (18)

Fula morphological gemination (Paradis 1988: 78) a. b.

stem (sg) lew lef

suffixed form (pl) lebb-i lepp-i

‘month’ ‘ribbon’

Of relevance here is that when a long vowel precedes the stem-final consonant, gemination fails to occur, but the stem-final consonant nonetheless is realized as a stop. This is illustrated by the singular/plural alternations in (19). (19)

Lack of gemination after a long vowel (Paradis 1988: 80) a. b.

stem (sg) laaw lees

suffixed form (pl) laab-i leec-e

expected form *laabb-i *leecc-e

‘road’ ‘bed’

Given that gemination is part of this suffixing process, we note that the expected forms in (19), where the initial syllable would be CVVG, fail to surface as such. Rather, the nature of the occurring suffixed forms in (19) makes it appear that degemination has occurred. This can be understood as the avoidance of a trimoraic CVVG syllable. Since CVVC syllables are allowed as in (17), Fula seems to

Geminates

13

instantiate a language with the weight system of (11), where CVG syllables are heavy but not other CVC syllables. Thus, we see that a variety of languages have strategies to avoid CVVG syllables but not CVVC syllables. These languages can be understood as providing support for the inherent weight analysis of geminates.

3

The patterning of geminates in stress systems

One of the criticisms of the weight analysis of geminates proposed by Hayes (1989), discussed by Tranel (1991), comes from the observation that there do not seem to be quantity-sensitive stress systems that support the weight division in (11), repeated below as (20), where stress would be attracted onto a syllable with a long vowel or closed by a geminate consonant, but not on a syllable closed by a non-geminate. (20)

Syllable weight distinction based on geminates being underlyingly moraic heavy CVV CVG

light CV CVC

The system in (20) is predicted to occur under Hayes’s theory in any language that allows long vowels and geminate consonants, but in which Weight-byPosition does not generally apply to coda consonants. According to the division in (20), CVV and CVG syllables would syllabify as bimoraic, while CV and CVC syllables would syllabify as monomoraic, as was shown in (12). Since quantitysensitive stress systems single out bimoraic syllables, it would be expected that at least some quantity-sensitive stress systems would reflect the weight division in (20) if the moraic representation of geminates were correct. Tranel suggests that weight systems like (20) do not exist and instead proposes a principle of equal weight for codas. Specifically, in languages in which codas pattern as moraic, geminates will be moraic; but in languages in which codas are not moraic, geminates would not be moraic. While our observation in §2.3 above – that in a variety of languages CVVG syllables are avoided but CVVC are not – can be taken as evidence for the weight division in (20), Tranel’s observation is of importance. In this section, we will overview the behavior of geminates in quantity-sensitive stress systems. In §3.1, we will provide stress data from various languages which indeed support the division in (20) whereby CVV and CVG syllables pattern together, thus supporting the moraic weight analysis of geminates. These are languages whose stress patterns Tranel predicts not to occur. In §3.2, I will review the type of case mentioned by Tranel in which quantity-sensitive stress treats all closed syllables in the same manner whether they be CVG or CVC. These are the languages that motivated Tranel’s principle of equal weight for codas and can be considered somewhat problematic for the weight analysis of geminates. In §3.3, I will present the case of the Australian language Ngalakgan (variably spelled Ngalagkan and Ngalakan; Baker 1997, 2008), in which CVC syllables can attract stress but apparently not CVG syllables. This seems to suggest that somehow geminates are resistant to carrying a mora. Thus, this section will identify three different types of geminate behavior with respect to quantity-sensitive

14

Stuart Davis

stress systems. In §4, we will try to reconcile this variability of behavior with respect to the representation of geminates.

3.1

Languages that uniquely treat CVG closed syllables as heavy with respect to stress

In this subsection I discuss two languages cited in the literature where syllables closed by a geminate (CVG) function as heavy, like a syllable with a long vowel in attracting stress, even when other closed syllables (CVC) may not act as heavy. Such languages provide stress evidence for the syllable weight division in (20). The two languages to be discussed are the Uto-Aztecan language Cahuilla and the San’ani (Yemen) dialect of Arabic. Hayes (1995) noted that stress assignment in the Uto-Aztecan language Cahuilla distinguishes CVG syllables from CVC syllables: CVG syllables behave like CVV syllables in attracting stress. Consider the data in (21), taken from Hayes (1995, and see sources cited therein). (21)

Cahuilla stress a. b. c. d. e. f.

’ta.ka.‘li.Œem ’Œe.xi.‘wen ’tax.mu.‘?at ’he.?i.’ka.kaw.‘la(.‘qa ’qa(n.‘ki.Œem ’Œex.‘xi.wen

‘one-eyed ones’ ‘it is clear’ ‘song’ ‘his legs are bow-shaped’ ‘palo verde (pl)’ ‘it is very clear’

The data items in (21a)–(21d) help to establish Cahuilla as having iterative left-to-right stress on alternating syllables (i.e. trochaic foot structure), although CVC syllables are not distinguished from CV syllables.9 (See also chapter 40: the foot and chapter 44: the iambic-trochaic law for more general discussion of trochaic foot structure.) The quantity-insensitive nature of CVC syllables is probably most clearly seen in the third and fourth syllables of the sequence in (21d), the trochaic foot (’ka.kaw). The second syllable of this foot is treated as light despite the presence of a coda consonant, which indicates the monomoraic nature of CVC syllables. On the other hand, the quantity-sensitive nature of the stress system can be seen not only in the stress-attracting nature of syllables with long vowels, but also in the observation that the syllable immediately after a long vowel receives stress. This is witnessed by the last two syllables in (21d), both of which have stress and comprise a foot, (la()(qa), and can be understood if Cahuilla foot structure is maximally bimoraic. Consequently, the last two syllables in (21d) cannot form a single foot, since such a foot would be trimoraic. The syllable with the long vowel (la() comprises a bimoraic foot on its own; the last syllable (qa) is forced to comprise a (monomoraic) foot on its own, due to a constraint in the 9

Hayes (1995) notes that Cahuilla CVC syllables closed by glottal stops, but not other CVC syllables, are also treated as bimoraic. I shall not discuss this, other than to note that in the Ngalakgan stress pattern presented in (25) CVC syllables closed by a glottal stop exceptionally act as monomoraic. The variable behavior of coda glottal consonants with respect to syllable weight is known but infrequently discussed (but see Churma and Shih 1996).

Geminates

15

language that requires exhaustive footing. The data item in (21e) is similar to the last two syllables in (21d) in that the syllable with a long vowel forms a foot on its own and the syllable immediately after it is at the beginning of a new trochaic foot, thereby receiving stress. We see then that bimoraic CVV syllables in Cahuilla are distinguished from monomoraic syllables, not only in bearing stress, but also by the presence of stress on the syllable immediately after them. CV and CVC syllables lack these two characteristics and function as monomoraic. It is interesting in this light to observe the patterning of syllables closed by a geminate, as in (21f), where the first syllable is CVG. This CVG syllable functions as bimoraic. It has stress, as would be expected of any initial syllable, but crucially it patterns exactly like a CVV syllable in that the syllable immediately after it also carries stress. This provides evidence that the CVG syllable comprises a bimoraic (trochaic) foot on its own. This constrasts with the initial CVC syllable in (21c) that forms a trochaic foot with the following syllable, suggesting the monomoraic nature of the CVC syllable. Cahuilla thus serves as a clear illustration for the weight distinction in (20) in which stress treats syllables with long vowels and those closed by geminates as bimoraic but not other types of syllables, be they CV or CVC. San’ani Arabic (Yemen) presents a very interesting case, in which CVV and CVG syllables pattern together with respect to stress. Watson (2002: 81–82) specifically notes that they pattern together as opposed to CVC syllables (see also chapter 135: word stress in arabic). Consider the data in (22), which illustrate the stress pattern in words without geminates. (22)

San’ani Arabic (Watson 2002: 81–82) a. b. c. d. e. f. g. h. i. j.

mak.’tu(b da.’rast ’sa(.fa.rat miú.’sa(.lih mi.’gam.bar ’mad.ra.sih mak.’ta.ba.ti( ’li.bi.sat ’ka.tab ’ra.ga.ba.tih

‘office’ ‘I/you (masc sg) learnt’ ‘she travelled’ ‘launderette’ ‘sitting’ ‘school’ ‘my library’ ‘she wore/put on’ ‘he wrote’ ‘his neck’

Stress normally falls on one of the last three syllables of the word: it falls on a final superheavy syllable (CVVC or CVCC) if there is one, as in (22a)–(22b); it falls on the rightmost non-final heavy syllable (CVC or CVV) up to the antepenultimate, as in (22c)–(22f); otherwise, stress falls on the leftmost CV syllable, as in (22g)–(22j). The data in general show that the word-final segment does not play a role in the computation of weight so that the final syllable can only be stressed if it is superheavy. The word in (22g) illustrates two important aspects of the stress system. It shows that a word-final syllable ending in a long vowel does not attract the stress; it also indicates that a CVC syllable in pre-antepenultimate position fails to attract stress. The latter point is significant, since it suggests that Weightby-Position, which assigns a mora to a coda consonant, is restricted to one of the last three syllables of the word. Now let us consider the data in (23) with words possessing geminate consonants.

16 (23)

Stuart Davis San’ani Arabic (Watson 2002: 81–82): stress on words with geminate consonants a. b. c. d. e. f.

ji.’pib.bu mit.’?ax.xi.ra(t mu.’sa–.–i.la.ti ’ha(.ka.Ïa.ha( ’daw.wart ’sa(.fart

‘they (masc) love/like’ ‘late (fem pl)’ ‘my recorder’ ‘like this’ ‘I/you (masc sg) looked for’ ‘I/you (masc sg) travelled’

The comparison between (22) and (23) indicates the priority of CVG and CVV syllables for stress assignment, in that CVG and (non-final) CVV syllables always attract stress even when in pre-antepenultimate position as in (23c) and (22d). The word in (22g), in contrast, shows that a CVC syllable does not receive stress in pre-antepenultimate position. The difference between CVG and CVC syllables can be readily understood on the inherent weight analysis of geminates. If a geminate is underlyingly moraic, it contributes weight to the syllable regardless of its location in the word. Recall that Weight-by-Position does not apply here, because it is restricted to one of the last three syllables in San’ani Arabic. In pre-antepenultimate position, only CVV and CVG act as bimoraic. Moreover, (23e) shows that CVG syllables have a priority of stress over a final superheavy syllable and should be compared with (22a) where a regular CVC syllable is devoid of such priority. It could be argued that Weight-by-Position in San’ani Arabic only applies to words that would not otherwise have bimoraic syllables (CVV or CVG). That is, there is no necessity for Weight-by-Position to apply in (23e) or (23f). While we do not pursue a full analysis here (but see Watson 2002), the priority given to both CVV and CVG syllables in stress assignment, especially as seen by the comparison of (23c) and (23d) with (22g), provides an interesting argument for the underlying moraic weight analysis of geminates, and, in turn, against Tranel’s (1991) claim of equal weight for codas. We have detailed above two cases where CVV and CVG syllables pattern together with respect to stress systems as predicted by the inherent weight analysis of geminates. Further support for the weight analysis of geminates is found in other languages. For example, Gupta (1987) discusses a Hindi dialect in which stress is attracted to the leftmost heaviest syllable in the word. The dialect treats both CVV and CVG syllables as bimoraic, while CVC syllables behave as light, although, as noted by Curtis (2003), such a pattern appears unusual among Hindi dialects. Additional support may come from the stress system of Pattani Malay, discussed by Topintzi (2006, 2008, 2010) and references cited therein. Pattani Malay has geminates that are restricted to word-initial position and the language lacks long vowels. Although primary stress typically falls on the final syllable of a word, stress occurs on the initial syllable in words that begin with a geminate consonant. This can be taken as evidence for the moraification in (10a) where a geminate is underlyingly moraic. That is, stress is attracted onto a syllable that is bimoraic. Despite the range of examples presented in this section, it remains rare to see languages that display the weight system in (20), grouping CVV and CVG syllables together as heavy. It is possible, on the other hand, that the rarity is due to the infrequent occurrence of the specific set of properties that is required for CVV and CVG to pattern together in stress assignment; namely, the language would

Geminates

17

have to have quantity-sensitive stress, long vowels, coda consonants, and geminates. Perhaps when such languages, San’ani Arabic for one, are examined closely, more instances of the special properties of CVG syllables will emerge. In this connection, it is worth noting that in most Arabic dialects CVG syllables are special: they always attract stress when in word-final position. This property separates them from other CVC syllables, which do not attract stress in word-final position. The difference thus finds a logical explanation in the underlying weight analysis of geminate consonants (especially in those dialects, such as Hadhrami Arabic, discussed earlier, which disallow final consonant clusters).10

3.2

Languages in which stress treats all codas equally

There are two types of languages in which stress assignment treats all codas equally. In the first type, stress is quantity-sensitive and is attracted to a heavy syllable, be it CVV, CVC, or CVG. Latin belongs to this group: any coda consonant makes a syllable heavy, so both CVC and CVG syllables are bimoraic. In the second type, which is more relevant for the representation of geminate consonants, both CVC and CVG syllables behave as light in a quantity-sensitive stress system. In such a language, stress is attracted to a CVV syllable, but both CVC and CVG syllables seem to pattern as monomoraic, treating CVG as light, just like other CVC syllables. As an illustration, consider the stress data from the Uralic language Selkup in (24). The data in (24a)–(24f) come from Halle and Clements (1983). The data items in (24g)–(24h) are reported in Ringen and Vago (2010) from the Selkup language scholar Eugene Helimski, and reflect the Taz Selkup dialect, which seems to have the same stress pattern as that in Halle and Clements (1983). (24)

Selkup stress a. b. c. d. e. f. g. h.

10

qu’mo(qi ’u(cqqo u(’cD(mqt ’qumqnqk ’amqrna ’u(cqkkak ’esykka es’s/(qo

‘two human beings’ ‘to work’ ‘we work’ ‘human being (dat)’ ‘eats’ ‘I am working’ ‘(it) happens (occasionally)’ ‘to happen (already)’

In many Arabic dialects, word-final CVC syllables behave as extrametrical. Ham (1998) puts forward the very intriguing observation that final CVC syllables are always extrametrical in languages that possess word-final geminates. This is because a word-final geminate is moraic and would need to be distinguished in final position from a potential moraic coda. With the underlying moraic weight representation of geminates as in (10a), final extrametricality of CVC syllables is able to preserve the contrast between an underlying final geminate and the corresponding final singleton consonant. The geminate of a final CVG syllable would surface as moraic while the singleton coda of the final CVC would be non-moraic. This difference is found in Arabic dialects where a final CVG syllable attracts stress, making it distinct from a final CVC syllable (i.e. bimoraic), which is light (monomoraic) and does not attract the stress. In a variety of other languages having word-final geminates examined by Ham (1998), the same distinction is made between final CVG and CVC syllables. If Ham’s observation holds up to further scrutiny, it constitutes an interesting argument for the underlying moraification of geminate consonants. (See also Topintzi 2008: 175 for discussion on this point.)

18

Stuart Davis

In Selkup, primary stress falls on the rightmost syllable with a long vowel (24a)–(24c) or on the initial syllable if there are no long vowels (24d). A CVC syllable does not count as heavy (24e), even if the CVC syllable is closed by a geminate, as seen in (24f) and (24g). As noted by Tranel (1991), if stress targets bimoraic syllables and geminates are underlyingly moraic, the second syllable in (24f) and (24g) would be the rightmost bimoraic syllable. Both the vowel and the geminate would contribute a mora to the second syllable. The fact that (24f) and (24g) do not receive stress on the second syllable, however, seems to provide evidence against geminates being underlying moraic, favoring a representation of geminates that is different from that in (10a). The stress pattern of Selkup does not appear to be unique in ignoring geminate consonants. Davis (1999a: 41) notes the Altaic language Chuvash (Krueger 1961), which exhibits an almost identical stress pattern to that of Selkup: stress is attracted to the rightmost syllable with a full vowel (interpreted as being bimoraic), but CVG syllables are ignored. Thus, in both Chuvash and Selkup, CVG syllables do not function as bimoraic CVV syllables but instead act like monomoraic CV and CVC syllables. Data from languages like Selkup have been used by Tranel (1991) and Ringen and Vago (2010) to argue against the underlying moraic weight representation of geminates. Ringen and Vago note that such languages are consistent with the length analysis of geminates as in (10b). In these languages, stress is sensitive to the presence of a long vowel, and ignores a coda consonant, whether the coda is part of a geminate or not. However, it is not that proponents of the weight representation are unaware of languages like Selkup. Topintzi (2008, 2010), who for the most part maintains the underlying moraic weight view of geminates, suggests that weightless geminates are represented by double consonants with two root nodes rather than as a single root node linked to a mora like (10a). But such a comment implies that there is language-specific variation in the representation of geminate consonants. On the other hand, Davis (2003) suggests that the stress pattern of languages like Selkup does not necessarily argue against the underlying moraic representation of geminates; viewed from an optimalitytheoretic perspective, the pattern can be a consequence of certain high-ranking stress constraints that have the effect of ignoring the bimoraicity of any CVC syllable. As suggested by Steriade (1990: 275), there may be reasons in some languages to restrict the set of stress-bearing segments to those that are also tone-bearing, “for reasons that are clearly related to the fact that pitch is one of the main realizations of metrical prominence.” Steriade’s suggestion can be incorporated into an optimality-theoretic approach as a constraint that restricts pitch realization to vocalic elements: the constraint prefers to place stress on any CVV syllable over any syllable closed by a consonant, even if that consonant is part of a geminate. Thus, the lack of second syllable stress in (24f) and (24g) of the Selkup data need not reflect on the underlying moraicity of geminate consonants.

3.3

Languages in which geminates repel stress

A third type of geminate behavior is witnessed in languages where stress is attracted to a closed syllable, but not to one with a geminate. The Australian language Ngalakgan, discussed by Baker (1997, 2008), serves as a major example. Consider

Geminates

19

the pattern of primary stress in (25), taken from Baker (2008) who notes that other geographically proximate languages have a similar stress pattern. (25)

Ngalakgan a. b. c. d. e. f. g. h. i. j.

pu’ruÍci ki’piÍkuluc mi’Èarppu? pu’Íolko? ‘maca’purka ’ÎaÈkurca ’calpurkic ’ciwi ’ceraÍa ’paÈa‘munu

‘water python’ ‘frogmouth (bird)’ ‘crab’ ‘brolga (bird)’ ‘plant sp.’ ‘vine sp.’ ‘fish sp.’ ‘liver’ ‘women’s ceremony’ ‘sand goanna’

k. l. m. n. o. p. q. r. s. t.

’calapir ’kupuj kaK’Íalppuru ’cakanta ’IuruKÍuc ’IoloIko? ’Iamuc‘culo ’capatta ’moÎoppoÎ ’IaKa?paj

‘red ant (species)’ ‘sweat (n)’ ‘plains kangaroo’ ‘macropod sp.’ ‘emu’ ‘eucalyptus’ ‘subsection term’ ‘tortoise sp.’ ‘catfish sp.’ ‘and moreover’

(25a)–(25l) show that primary stress in Ngalakgan falls on the leftmost (non-final) heavy syllable; if there is no heavy syllable, it falls on the initial syllable. From these data, it can be surmised that a coda consonant is moraic in making a syllable heavy, be it an obstruent (as in (25a) and (25b), or a sonorant (as in (25c)–(25g)). The data in (25m)–(25t) show that the leftmost closed syllable (underlined) fails to attract primary stress. Note that the coda in the leftmost closed syllable in (25m)–(25t) belongs to one of three types: in (25m)–(25p) the coda is a nasal homorganic with the following onset, in (25q)–(25s) it is the first part of a geminate consonant, and in (25t) the coda is a glottal stop. Key to our discussion is the fact that CVG syllables in (25q)–(25s) resist stress. However, the comparison of the stressresistant nature of CVG syllables with the other instances of stress-resistant closed syllables in (25m)–(25p) and (25t) points to the fact that a common property these syllables have is that they do not possess their own place features: either the place features are shared with the following onset, or, in the case of the glottal stop in (25t), there is a lack of place features altogether. Thus, Ngalakgan seems to divide closed syllables into two types: those in (25a)–(25l), in which the coda has independent place features and attracts stress, and those in which the coda does not have its own independent place features and fails to attract stress. This suggests that Ngalakgan is best analyzed as having a requirement that moraic elements have independent place features (i.e. not shared with a following onset), as advocated in Baker (1997). It follows that the stressed closed syllables in (25a)–(25l) would be bimoraic and attract stress, whereas the CVG syllables in (25m)–(25p) would be monomoraic and not attract stress. Languages like Ngalakgan seem to present a challenge for the underlying weight representation of geminates in (10a), since not only do syllables with geminates not attract stress, but they are not even equal in weight to CVC syllables, as in (25a)–(25l), which do attract stress. Baker (2008) offers an articulatory gestural analysis of the difference. He observes that the apparent CVC syllable attracts stress only if the postvocalic coda consonant has an articulatory gesture distinct from that of the following onset. That is, in a CVCCV sequence, the first syllable counts as heavy only if the two intervocalic consonants have distinct articulatory gestures. When the intervocalic sequence involves a geminate or a nasal that is homorganic with a following consonant (or a glottal stop, which does not have a distinct articulatory gesture) there is only one articulatory gesture, and stress

20

Stuart Davis

is not attracted onto CVG (or CVN where N is homorganic with a following consonant). Baker (2009) adopts a composite view of geminate representation that incorporates a gestural tier along with root nodes and moras. Under this view, stress in Ngalakgan is characterized as sensitive to the gestural tier. Ringen and Vago (2010), in contrast, take the Ngalakgan data as supporting the length representation of geminates where the stress rule treats linked structures like those in (10b) as light. Davis (2003), who maintains the underlying weight representation of geminates, suggests that the Ngalakgan stress pattern in (25) does not necessarily argue against the underlying moraic representation of geminates as in (10a); rather, the language has a high-ranked constraint that requires moraic elements to have their own place features. Thus, while geminates may be underlyingly moraic, they do not surface as moraic.11 To conclude this section, we have surveyed languages demonstrating three types of behavior of CVG syllables in stress systems: (i) cases where CVG and CVV pattern together; (ii) cases where CVG patterns with other CVC syllables; and (iii) cases where CVG syllables are specifically resistant to stress. We have tried to maintain the underlying moraic weight representation of geminates despite apparent evidence to the contrary. In the concluding section we will further discuss representational issues.

4

Representational issues and conclusion

In this overview, we have focused on the cross-linguistic patterning of geminate consonants while trying to maintain the representational view of geminates in (10a), in which geminates are marked as being underlyingly moraic, over the length representation in (10b). In §2, we provided evidence for the underlying moraic representation of geminate consonants by considering a variety of phonological patterning pertinent to geminates. This included the moraic analysis of initial geminates in Trukese in §2.1 and the cross-linguistic avoidance of CVVG syllables in §2.3. We made it clear in §2.2 that geminates do not always behave like a sequence of two C-slots in prosodic patterning, thereby contradicting the length representation of geminates in (10b). In §3, we surveyed geminate patterning in stress systems identifying three types of behavior. Despite the differences, we still argued for the underlying moraic view of geminate consonants. The issue of the representation of geminate consonants has been a controversial matter and will most likely remain so in future investigations. This is because geminates do not display uniform behavior, as we have illustrated. It seems that the very nature of the data under examination determines what type of representation must be appropriate. For example, the parallel patterning between final CC sequences and final geminates in Hungarian seems to be supportive of the length representation, while the difference between final CC sequences and final 11

It should be noted that Baker (2008) actually considers the intervocalic geminates in (25q)–(25s) and the intervocalic homorganic nasal clusters in (25m)–(25p) to be syllabified completely as onsets rather than as heterosyllabic. This differs from his earlier work, Baker (1997), where a heterosyllabic parse of geminates and homorganic nasal clusters is maintained. One shortcoming of Baker’s (2008) onset analysis is that geminates and homorganic nasal clusters do not occur word-initially. Nonetheless, even if geminates could be analyzed as syllabifying as onsets, they would not add weight to a syllable and thus would be different from the initial geminates of Trukese discussed earlier.

Geminates

21

geminates in Arabic dialects such as Hadhrami Arabic appears to be better accounted for by the weight analysis of geminates. Such differences lead to three possibilities for a representational view of geminates, in a general sense. The first possibility is that geminate representation is language-specific. In a language of the Trukese type, geminates are represented to be underlyingly moraic (10a), as argued for by Davis and Torretta (1998) and Davis (1999b), but in a language of the Hungarian type, geminates have a length analysis (10b), as argued for by Ringen and Vago (2010).12 The second possibility – one maintained in the body of this chapter – is that there is one specific universal representation of geminates (i.e. the underlying moraic weight representation), and apparent counterexamples to it are explained by constraints that act on surface forms. This strategy was emphasized especially in §§3.2 and 3.3. For example, in the Selkup-type stress pattern in (24), which ignores all CVC syllables, including those closed by a geminate, there is an independent constraint that restricts pitch realization to vocalic elements. The constraint would choose to place stress on any CVV syllable over any syllable closed by a consonant, even if that consonant were part of a geminate. Thus, the lack of CVG syllables that attract stress in Selkup need not reflect on the underlying moraicity of geminate consonants. Given that geminates show a varied degree of sensitivity to stress, as seen in §3, any single universal analysis of geminates would have to make use of such flexible strategies to account for the apparent problematic cases. A third possibility for a representational account of geminates is to maintain that there is a universal representation of geminates, but one that combines different aspects of the various representations discussed in §1 and elsewhere in this chapter. Hume et al. (1997) combine a length representation with moraic theory, and Curtis (2003) integrates the two-root node representation presented in (5) with a moraic view. (See Curtis 2003 for a detailed discussion and comparison of different representational views of geminate consonants.) A more recent composite view is implicit in Baker (2009), in which a geminate is represented on both a timing (length) tier and a gestural tier. It is also viewed as having a moraic representation if it functions as heavy. Baker’s (2009) representation of a geminate [k] is shown in (26). (26) [k] X

X

Gestural tier Timing (length) tier

Baker’s analysis is quite attractive as a universal, especially if we assume the underlying moraic nature of geminates. Languages display variable patterning for geminates because the phonology (i.e. the constraints or rules) makes reference 12

José and Auger (2005) argue that even within a single language not all geminates must have the same representation. According to them, in Vimeu Picard (phrase-)initial geminates differ as to whether they have a single set of features or two sets of identical features linked to two root nodes. From phonological patterning they argue that initial [ll] has the former representation while initial [nn] has the latter.

22

Stuart Davis

to different aspects of the representation. For example, in Trukese, the moraic aspect of the geminate representation is crucial, while it may or may not surface with two X-slots, depending on the constraints. In Hungarian, given the parallel behavior of word-final clusters and geminates, the X-tier representation is effective, while the geminate may or may not surface as moraic, depending on the constraints. In Ngalakgan, the behavior of geminates with respect to stress can be understood through the gestural tier, as proposed by Baker (2008, 2009). Geminates and homorganic nasal + consonant clusters have a single gesture. Consequently, they can pattern as single consonants despite having two X-slots, as long as there is a high-ranked constraint that requires a moraic element not to share place features. This would provide an explanation for the repulsion of geminates to stress. It follows that while geminates may have one underlying universal representation (as in (26)), its surface realization may vary cross-linguistically, e.g. as non-moraic in Ngalakgan, but moraic in Trukese. The composite view, or some version of it, may ultimately be the best universal representation for an underlying geminate. For example, one criticism of a purely weight account of geminates is that it cannot distinguish between a geminate that syllabifies entirely in a coda from a single coda consonant in a language in which Weight-by-Position applies. This matter comes up in the Palestinian Arabic dialect described by Abu-Salim (1980) and mentioned in Rose (2000), where a coda singleton in a word like [bit.na] ‘our house’ is representationally indistinguishable from a coda geminate in a word like [sitt.na] ‘our grandmother’ on a strictly moraic view of geminates as in (10a). Unless a length tier (or two root nodes) is assumed, there is no obvious way to distinguish the two cases. Although such examples are probably rare, the occurrence of this type of contrast indeed favors a composite analysis, especially given the language-specific evidence for Arabic moraic structure presented in various parts of this chapter. That said, it may still be possible to argue for the underlying moraic weight representation as universal, but with the understanding that the surface realization of geminates may vary across languages because of the interaction of relevant constraints. In conclusion, there is much about the phonology of geminates that remains to be investigated. Geminates do not all pattern the same way across languages. Consequently, geminate phonology will remain an area of theoretical controversy for the foreseeable future.

ACKNOWLEDGMENTS I thank Brett Baker, Ken de Jong, Daniel Dinnsen, Tracy Alan Hall, Abbie Hantgan, Brian José, Michael Marlo, Paul Newman, Anne Pycha, Nathan Sanders, Robert Vago and Islam Youssef for discussion on various aspects of this paper. I especially thank Natsuko Tsujimura for her detailed comments. The usual disclaimers apply.

REFERENCES Abu-Salim, Issam M. 1980. Epenthesis and geminate consonants in Palestinian Arabic. Studies in the Linguistic Sciences 10(2). 1–11. Baker, Brett. 1997. Edge crispness: Segment to mora isomorphism. Proceedings of the West Coast Conference on Formal Linguistics 16. 33 –47.

Geminates

23

Baker, Brett. 2008. Word structure in Ngalakgan. Stanford: CSLI. Baker, Brett. 2009. Monogestural clusters as onsets: The Australian evidence. Paper presented at the 83rd Annual Meeting of the Linguistic Society of America, San Francisco. Bamakhramah, Majdi. 2009. Syllable structure in Arabic varieties with a focus on superheavy syllables. Ph.D. dissertation, Indiana University. Broselow, Ellen. 1992. Parametric variation in Arabic dialect phonology. In Ellen Broselow, Mushira Eid & John J. McCarthy (eds.), Perspectives on Arabic linguistics IV, 7 –45. Amsterdam & Philadelphia: John Benjamins. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Churchyard, Henry. 1991. Compensatory lengthening and “gemination throwback” in Trukese and Puluwat as evidence for Rime in moraic phonology. Unpublished ms., University of Texas. Churma, Donald & Yili Shi. 1996. Glottal consonants and the “sonority” hierarchy. Proceedings of the Eastern States Conference on Linguistics (ESCOL) 1995. 25–37. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Curtis, Emily. 2003. Geminate weight: Case studies and formal models. Ph.D. dissertation, University of Washington. Davis, Stuart. 1994. Geminate consonants in moraic phonology. Proceedings of the West Coast Conference on Formal Linguistics 13. 32 –45. Davis, Stuart. 1999a. On the moraic representation of underlying geminates: Evidence from prosodic morphology. In René Kager, Harry van der Hulst & Wim Zonneveld (eds.), The prosody–morphology interface, 39 –61. Cambridge: Cambridge University Press. Davis, Stuart. 1999b. On the representation of initial geminates. Phonology 16. 93 –104. Davis, Stuart. 2003. The controversy over geminates and syllable weight. In Féry & van de Vijver (2003), 77 –98. Davis, Stuart. Forthcoming. Quantity. In John A. Goldsmith, Jason Riggle & Alan C. L. Yu (eds.) The handbook of phonological theory, 2nd edn. Malden, MA & Oxford: Wiley-Blackwell. Davis, Stuart & Gina Torretta. 1998. An optimality-theoretic account of compensatory lengthening and geminate throwback in Trukese. Papers from the Annual Meeting of the North East Linguistic Society 28. 111–125. Dyen, Isadore. 1965. A sketch of Trukese grammar. New Haven: American Oriental Society. Féry, Caroline & Ruben van de Vijver (eds.) 2003. The syllable in Optimality Theory. Cambridge: Cambridge University Press. Goodenough, Ward & Hiroshi Sugita. 1980. Trukese–English dictionary. Philadelphia: American Philosophical Society. Gupta, Abha. 1987. Hindi word stress and the obligatory branching parameter. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23. 134 –148. Halle, Morris & G. N. Clements. 1983. Problem book in phonology. Cambridge, MA: MIT Press. Halle, Morris & Jean-Roger Vergnaud. 1980. Three-dimensional phonology. Journal of Linguistic Research 1. 83 –105. Ham, William. 1998. Phonetic and phonological aspects of geminate timing. Ph.D. dissertation, Cornell University. Hart, Michele. 1991. The moraic status of initial geminates in Trukese. Proceedings of the Annual Meeting, Berkeley Linguistics Society 17. 107 –120. Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62. 321–351. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press.

24

Stuart Davis

Hume, Elizabeth, Jennifer Muller & Aone van Engelenhoven. 1997. Non-moraic geminates in Leti. Phonology 14. 371–402. Idemaru, Kaori & Susan Guion. 2008. Acoustic covariants of length contrast in Japanese stops. Journal of the International Phonetic Association 38. 167 –186. José, Brian & Julie Auger. 2005. Geminates and Picard pronominal clitic allomorphy. Catalan Journal of Linguistics 4. 127 –154. Kawahara, Shigeto. 2007. Sonorancy and geminacy. University of Massachusetts Occasional Papers 32. 145 –186. Keer, Edward. 1998. Icelandic preaspiration and the moraic theory of geminates. Proceedings of the 10th Conference of Nordic and General Linguistics, Reykjavik. (ROA-312.) Keer, Edward. 1999. Geminates, the OCP and the nature of Con. Ph.D. dissertation, Rutgers University. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Kiparsky, Paul. 2003. Syllables and moras in Arabic. In Féry & van de Vijver (2003), 147–182. Kiparsky, Paul. 2008a. Fenno-Swedish quantity: Contrast in Stratal OT. In Bert Vaux & Andrew Nevins (eds.) Rules, constraints, and phonological phenomena, 185–219. Oxford: Oxford University Press. Kiparsky, Paul. 2008b. Weight and length. Paper presented at the CUNY Conference on the Syllable, New York City. Kirchner, Robert. 2000. Geminate inalterability and lenition. Language 76. 509 –545. Kraehenmann, Astrid. 2001. Quantity and prosody asymmetries in Alemannic: Synchronic and diachronic perspectives. Ph.D. dissertation, University of Konstanz. Published 2003, Berlin & New York: Mouton de Gruyter. Krueger, John. 1961. Chuvash manual: Introduction, grammar, reader, and vocabulary. Bloomington: Indiana University Press. Leben, William R. 1980. A metrical analysis of length. Linguistic Inquiry 11. 497 –509. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. McCarthy, John J. 1979. Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT. McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12. 373 –418. McCarthy, John J. 1988. Feature geometry and dependency: A review. Phonetica 45. 84–108. Morén, Bruce. 1999. Distinctiveness, coercion and sonority: A unified theory of weight. Ph.D. dissertation, University of Maryland at College Park. Muller, Jennifer. 2001. The phonology and phonetics of word-initial geminates. Ph.D. dissertation, Ohio State University. Pajak, Bozena. Forthcoming. Contextual constraints on geminates: The case of Polish. Proceedings of the Annual Meeting, Berkeley Linguistics Society 35 (ROA-105). Paradis, Carole. 1988. On constraints and repair strategies. Linguistic Review 6. 71– 97. Prince, Alan. 1990. Quantitative consequences of rhythmic organization. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26(2). 355–398. Pycha, Anne. 2007. Phonetic vs. phonological lengthening in affricates. In Jürgen Trouvain & William J. Barry (eds.) Proceedings of the 16th International Congress of Phonetic Sciences, 1757 –1760. Saarbrücken: Saarland University. Pycha, Anne. 2009. Lengthened affricates as a test case for the phonetics–phonology interface. Journal of the International Phonetic Association 39. 1– 31. Ringen, Catherine & Robert M. Vago. 2010. Geminates: Heavy or long? In Charles Cairns & Eric Raimy (eds.) Handbook of the syllable. Leiden: Brill. Rose, Sharon. 2000. Rethinking geminates, long-distance geminates, and the OCP. Linguistic Inquiry 31. 85 –122. Schein, Barry & Donca Steriade. 1986. On geminates. Linguistic Inquiry 17. 691–744.

Geminates

25

Schmidt, Deborah. 1992. Compensatory lengthening in a segmental moraic theory of representation. Linguistics 30. 513 –534. Selkirk, Elisabeth. 1990. A two-root theory of length. University of Massachusetts Occasional Papers 14. 123 –171. Selkirk, Elisabeth. 1991. On the inalterability of geminates. In Pier Marco Bertinetto, Michael Kenstowicz & Michele Lopocaro (eds.) Certamen Phonologicum II: Papers from the 1990 Cortona Phonology Meeting, 187 –209. Turin: Rosenberg & Sellier. Sherer, Timothy. 1994. Prosodic phonotactics. Ph.D. dissertation, University of Massachusetts, Amherst. Steriade, Donca. 1990. Moras and other slots. Proceedings of the Formal Linguistics Society of Midamerica 1. 254 –280. Thurgood, Graham. 1993. Geminates: A cross-linguistic examination. In Joel Ashmore Nevis, Gerald McMenamin & Graham Thurgood (eds.) Papers in honor of Frederick H. Brengelman on the occasion of the twenty-fifth anniversary of the Department of Linguistics, CSU Fresno, 129 –139. Fresno: Department of Linguistics, California State University, Fresno. Topintzi, Nina. 2006. Moraic onsets. Ph.D. dissertation, University College, London. Topintzi, Nina. 2008. On the existence of moraic onsets. Natural Language and Linguistic Theory 26. 147 –184. Topintzi, Nina. 2010. Onsets: Suprasegmental and prosodic behaviour. Cambridge: Cambridge University Press. Tranel, Bernard. 1991. CVC light syllables, geminates and moraic theory. Phonology 8. 291–302. Tsujimura, Natsuko 2007. An introduction to Japanese linguistics. 2nd edn. Cambridge, MA: Blackwell. Tyler, Stephen. 1969. Koya: An outline grammar. Berkeley: University of California Press. Watson, Janet C. E. 2002. The phonology and morphology of Arabic. Oxford: Oxford University Press.

38

The Representation of sC Clusters Heather Goad

1

Introduction

Clusters of the shape s + consonant (sC), exemplified by stack, have posed a challenge for theories of syllabification, as they defy many of the constraints holding of true branching onsets, as found in, for example, track. Accordingly, some researchers have proposed that s is organized outside the onset constituent that contains the following consonant. Others have proposed that this sort of analysis holds only for a subset of sC clusters; those that rise in sonority, for example in slack, are represented in the same fashion as branching onsets. Yet others have argued that some sC clusters, those of the shape s + stop, form complex segments. This chapter will critique each of these proposals. An element shared by all of them is that phonological units are highly articulated: the burden of explanation is placed precisely on the structural relationships that adjacent segments enter into. Although this approach captures many peculiarities of sC clusters, there is little attempt to explain why the consonant that displays unorthodox behavior is typically s. Under the view that segments are ordered to maximize their perceptibility, the behavior of s becomes less puzzling: strident fricatives have robust internal cues, ensuring their perceptibility even in non-optimal contexts. If the acoustic properties of s are of central importance, it behoves us to ask whether the differences between sC clusters and branching onsets can be explained solely by perceptual considerations. This, of course, would challenge the view that a structural approach to cluster well-formedness is necessary. In the final section of the paper I will argue that this position is too strong. I will conclude that an adequate understanding of sC clusters requires consideration of both perceptual and structural factors. Much of the paper compares sC clusters with branching onsets. We will observe that they differ in several respects: phonotactic constraints, word-internal syllabification, allomorph selection, patterns of reduplication, options for cluster repair, etc. Although the general observations we will detail are likely to be accepted by most phonologists, there is little agreement on how these differences should be formally represented. In this context, there are three topics that will be addressed. The first involves critical assessment of various proposals in the literature concerning the representation of sC clusters – as a single class – in contrast to The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0038

The Representation of sC Clusters

2

obstruent + sonorant clusters. Henceforth, I will use the term “obstruent” to refer to obstruents other than s. “s” is itself a cover term for the sibilant(s) appearing in sC clusters; although this sibilant is usually /s/, in some languages, other sibilants pattern as s (e.g. German s is usually /œ/; in Russian, /s z œ Ú/ all pattern as s). Concerning obstruent + sonorant clusters, there will be nothing particularly special to say about their representation; they form branching onsets, and there is little controversy on this matter among those who accept a hierarchically organized syllable. For sC clusters, in contrast, several options will be considered. Most of these share the idea that s is an appendix, a segment which is not organized by any sub-syllabic constituent; one views s as a coda. We will see, in addition, that some researchers propose a single representation for sC in all languages; others argue for different representations across languages. The second topic that must be addressed is whether, in a given language, all sC clusters are represented in the same fashion. s + sonorant clusters are phonotactically ambiguous: like obstruent + sonorant clusters, they rise in sonority, yet like s + obstruent clusters, they do not respect the place constraints holding of obstruent + sonorant clusters. Depending on the weight assigned to each of these, different conclusions will be arrived at concerning the analysis of s + sonorant. For those researchers who place most weight on sonority profile, s + sonorant clusters form branching onsets. This research itself falls into two categories. One body of work aims to show that s + sonorant patterns with branching onsets while s + obstruent patterns differently, and is organized with some type of appendix. Another body of work considers s + sonorant clusters to be branching onsets, but focuses on arguing that s + stop clusters form complex segments. The proposals sketched above assume that the syllable is hierarchically organized. However, there is a growing literature that de-emphasizes the role of constituency and aims to provide phonetically grounded explanations for phonological behavior. The third topic therefore considers whether differences in the behavior of obstruent + sonorant, s + sonorant, and s + obstruent can be explained by perceptual considerations alone. This topic will be the focus of the final section of the paper. Until then, a structural approach will be assumed.

2

Cluster phonotactics

We begin by detailing the phonotactic constraints most commonly held of obstruent + sonorant clusters on the place and sonority dimensions, in turn, examining sC clusters on these same dimensions (see also chapter 33: syllableinternal structure; chapter 55: onsets). Our focus will be on the left word edge; other types of phonological behavior will be discussed in later sections, when we examine alternative representations for sC clusters. Consider the inventories of two-member clusters found in word-initial position in English and Dutch in (1) and (2).1 The data are organized by the place and manner values of C1 for obstruent + sonorant clusters and of C2 for sC clusters. (On cluster phonotactics for English, see Fudge 1969; Selkirk 1982; Clements and 1

We restrict discussion of obstruents in clusters to those that are voiceless; some languages display fewer options for voiced obstruents (e.g. */vl vr/ in English). We avoid consonant + glide clusters altogether, as there are more representational options available for glides than we have space to consider.

Heather Goad

3

Keyser 1983; Goldsmith 1990; Harris 1994; for Dutch, see Trommelen 1984; van der Hulst 1984; Fikkert 1994; Booij 1995; van der Torre 2003.) (1)

English a.

(2)

Obstruent + sonorant pl *tl kl pr tr kr fl *hl *œl fr hr œr

b. sC clusters sp st sk sm sn sl *sr

Dutch a.

Obstruent + sonorant *tn kn pl *tl kl pr tr kr fl øl fr ør

b. sC clusters sp st sø sm

sn sl *sr

We first consider place identity, which forbids the consonants in obstruent + sonorant clusters from having the same place (chapter 22: consonantal place of articulation). This captures the ill-formedness of */tl hl œl/ in English (1a), and */tn tl/ in Dutch (2a).2 Turning to (1b) and (2b), the well-formedness of /st sn sl/ indicates that sC clusters do not respect place identity, suggesting that sC clusters do not have the same representation as obstruent + sonorant clusters. However, before we can conclude this with certainty, we must consider place-sharing */sr/, which is ill-formed in English and Dutch.3 Importantly, */sr/ is illicit even in Dutch dialects with dorsal /r/, suggesting that the ill-formedness of this cluster has nothing to do with place identity. We return to */sr/ later in the chapter. A second, less commonly discussed, constraint on place concerns asymmetries that hold between C1 and C2 in obstruent + sonorant clusters (when C2 ≠ glide). English is not very revealing, because C2 is restricted to liquids, which are coronal. Dutch is potentially more illuminating, because it contains stop + nasal clusters, and nasals contrast for place.4 As (2a) shows, when a nasal is in C2 in an obstruent + sonorant cluster, it must be coronal: /kn/ is well-formed; */km/ is out. The broader generalization is thus that when C2 is a contoid, it must be coronal (except /r/; see note 2). 2

Place identity is not respected with /r/. In English and Dutch dialects with coronal /r/, coronal + /r/ clusters are well-formed, as are dorsal + /r/ clusters in Dutch dialects with dorsal /r/. Even in languages where /t/ and /r/ are articulated near-identically, the constraint is not respected (see Arvaniti 2007 on Greek). This may suggest that /r/ permanently lacks place (Rice 1992; Goad and Rose 2004; see also chapter 30: the representation of rhotics). 3 Concerning Dutch */sr/, some speakers realize /sør/ as [sr] (Waals 1999). If this represents a re-analysis of /sør/ (van der Torre 2003), then /sr/ is well-formed for these speakers. Concerning English /œr/, we have placed this cluster in the obstruent + sonorant category (Goad and Rose 2004), rather than treating it as an assimilated form of /sr/ (Clements and Keyser 1983; Goldsmith 1990). 4 I say “potentially,” because, as is undoubtedly evident, obstruent + sonorant clusters will be analyzed as branching onsets below. There is, however, dispute about the status of /kn/ in Dutch, as branching onset (Fikkert 1994; Booij 1995) or appendix-initial (van der Hulst 1984; Trommelen 1984; Kager and Zonneveld 1986). Notably, intervocalic /kn/ is syllabified as coda + onset, contra the branching onset analysis.

The Representation of sC Clusters

4

C2 in sC clusters has a different profile: it can have any place of articulation. Indeed, in s + obstruent and s + nasal clusters, C2 displays the same range of place contrasts attested for singleton onsets (e.g. /sp st sk/ alongside /p t k/ in English). Directly comparing nasal-final clusters in Dutch, we find the following: */km/, /kn/; /sm/, /sn/. The absence of */km/ alongside the presence of /sm/ is unexpected, if these clusters are represented identically. Because of the disputed status of Dutch /kn/ (note 4), we turn to Modern Greek to better examine differences in place profile between C2 in obstruent + sonorant vs. sC clusters. (On Modern Greek cluster phonotactics, see Joseph and PhilippakiWarburton 1987; Drachman 1990; Klepousniotou 1998; Morelli 1999; Tzakosta and Vis 2009; on Attic Greek, see Steriade 1982.) The data in (3) reveal that, although C2 in an obstruent + sonorant cluster can have any manner, leading to a wider range of cluster profiles than in Dutch or English, C2 must still be coronal (Klepousniotou 1998).5 (3b) shows that sC clusters are more restricted on the manner dimension than they are in Dutch, but among clusters with obstruents in C2, it can nevertheless be seen that C2 can have any place. (3)

Modern Greek a.

Obstruent + sonorant pt (pn) pl pr ft fh *fn fl fr (mn)

*tn *tl tr *ht (hn) (hl) hr

kt (kn) kl kr xt (xh) xn xl xr

b.

sC clusters sp sf (sm)

st (sh) (sn) *sl *sr

sk sx

In sum, we have observed that C2 in an obstruent + sonorant cluster does not parallel C2 in an sC cluster on the place dimension. On the contrary, a closer parallel is observed between C1 in an obstruent + sonorant cluster and C2 in an sC cluster, which we return to below. We consider finally the sonority constraints that hold between C1 and C2 in initial clusters.6 Greek is not very revealing here, as both obstruent + sonorant and sC clusters can have a falling, flat, or rising sonority profile (although (3b) indicates that the productivity of the latter for sC clusters is questionable; we return to this below). We therefore focus on English and Dutch. (1) and (2) show that sC clusters need not rise in sonority, in contrast to obstruent + sonorant clusters. Although C1 in an sC cluster is an obstruent and so the potential exists for these 5

Stop + stop and fricative + stop are often considered to be archaic in spoken Modern Greek (Joseph and Philippaki-Warburton 1987; Morelli 1999). They do occur in higher registers, which is why they are included here. /ps ts ks/ are absent from (3a); I assume they are complex segments (following Tzakosta and Vis 2009). Clusters in parentheses (as well as /tm/) are not productive, although the number of /sm/-initial roots is somewhat larger than for the others. Thanks to Jenny Dalalakis and Katerina Klepousniotou for help with the Greek data. 6 I assume the following sonority scale, which is roughly based on relative intensity: stop < fricative < nasal < liquid < vocoid. See also chapter 49: sonority.

Heather Goad

5

clusters to be limited to those that rise in sonority, this is not what is observed. Both languages contain falling sonority s + stop, and Dutch has flat sonority /sø/.

3

sC clusters ≠ branching onsets

Our examination of phonotactic constraints has revealed that obstruent + sonorant and sC clusters pattern differently on three dimensions: place identity, C2 place profile, and C1C2 sonority profile. Hereafter, we turn to other types of phonological behavior. Our first goal is to motivate the position that sC clusters are represented differently from obstruent + sonorant clusters. We then delve into the various options for sC clusters. For both of these topics, we will focus on phenomena where sC clusters pattern as a single class. The following section will then examine evidence that s + sonorant and s + obstruent clusters may be organized differently, s + sonorant clusters as branching onsets and/or s + stop as complex segments. In the preceding section, we observed that obstruent + sonorant clusters have a profile consistent with the prototypical syllable, one that rises in sonority toward the peak. This indicates that clusters of this shape (modulo language-specific constraints on place identity and sonority distance) are organized as branching onsets; see (4). (4)

Branching onset O p

l

Following Kaye et al. (1990), I assume that onsets are universally left-headed. Heads can host a range of segmental contrasts; dependents are segmentally restricted. Considering place contrasts, earlier discussion of C2 place profile revealed that there is more in common between C1 in a branching onset and C2 in an sC cluster. This suggests that sC clusters are right-headed. If onsets are left-headed, this de facto places s outside of this constituent, as reflected in the representations in (5). In (5), skeletal slots have been suppressed to facilitate comparison across theories; PWd abbreviates Prosodic Word (see chapter 51: the phonological word). (5)

sC cluster a. Extraprosodic (e.g. Steriade 1982)

b. Licensed by PWd (e.g. Goldsmith 1990)

O

PWd

p

q

O s

p

The Representation of sC Clusters c. Licensed by q (e.g. van der Hulst 1984)

d. Coda (e.g. Kaye 1992) O

q

R

O

N

O s

6

s

p

p

In (5a), s is extraprosodic. I am using this term in its narrowest sense, to refer only to the situation where an element is licensed but not organized into higher structure; compare (5b) and (5c). These two share with (5a) the idea that s is an appendix: s is not organized by any sub-syllabic constituent, in contrast to (5d), where s is a coda (technically in Kaye’s model a rhymal dependent). Two predictions follow from the difference in representation and headedness in (4) vs. (5). First, there should be languages that permit dependents in branching onsets but not sC clusters, and vice versa. This prediction holds true. Spanish is a language with branching onsets, but lacking initial sC clusters (Harris 1983). Acoma, spoken in New Mexico, has the opposite profile: initial clusters are restricted to sC (Miller 1965). Relatedly, Fikkert’s (1994) study on the acquisition of Dutch reveals that some children acquire branching onsets first, parallel to Spanish, while others acquire sC clusters first, parallel to Acoma. Second, languages permitting both types of structures should not prevent them from being combined. To my knowledge, this prediction always holds. In languages that have branching onsets and sC clusters, three-member clusters of the shape s + branching onset are also well-formed. However, these clusters may be restricted to a subset of what would be expected from a free combination of sC clusters and branching onsets in the particular language (e.g. Greek /sx/, /xr/, */sxr/; English /sk/, /kl/, */skl/ (loans aside)). Further, an explanation emerges under (5) for why the constraints against place identity and for rising sonority do not hold of sC clusters. It is not enough for two consonants to be adjacent; they must be sisters, as in (4). Finally, we demonstrate that the difference in headedness between branching onsets and sC clusters can account for a commonly attested pattern of cluster reduction in acquisition, illustrated in (6) from two learners of German and English respectively, Annalena (Elsen 1991) and Amahl (Smith 1973).7 (6)

a.

Annalena (age 1;4–1;9) obs + son s + obs s + son

7

output [daQbe] [f>kH] [p>gHl] [da>nH] [m>sHn] [lA(fH]

target [tr]aube [fl]iege [œp]iegel [œt]ein [œm]eißen [œl]afen

‘grape’ ‘fly’ ‘mirror’ ‘stone’ ‘to throw’ ‘to sleep’

[d Ò ı] indicate voiceless unaspirated lenis stops in Amahl’s data (Smith 1973).

Heather Goad

7 b.

Amahl (age 2;2–2;6) output obs + son [de(t] [ıi(m] s + obs [daidH] [ı>p] s + son [ni(d] [lZg]

‘plate’ ‘cream’ ‘spider’ ‘skipping’ ‘sneezed’ ‘slug’

In (6), the head of the cluster survives, regardless of its position in the string or its relative sonority. This suggests that some learners recognize that branching onsets are left-headed and sC clusters right-headed, even though left-edge clusters are altogether absent from their outputs (Goad and Rose 2004).

4

Alternative representations for sC clusters

Thus far, we have motivated different representations for branching onsets and sC clusters, based on evidence from phonotactics and cluster reduction in acquisition. We turn now to critique the various options for sC clusters in (5). In (5a), s is extraprosodic.8 Although it is unaffiliated to any syllable constituent (at least at early levels of the phonology), it is licensed and thus not subject to stray erasure. This structure is motivated by, for example, Steriade (1982) with data from Attic Greek, Sanskrit, Latin, and English; Vennemann (1982) and Wiese (1988) from German; Kager and Zonneveld (1986) from Dutch; Chierchia (1986) and Davis (1990) from Italian. Although (5a) captures the unique behavior of sC clusters, it cannot easily express the observation that some languages have different types of extraprosodic segments. In English, for example, the same status should not be assigned to /s/ in sC clusters as to inflectional /s/ in (7a), yet both fall outside of core syllabification. Further, on the face of it, some languages permit adjacent extraprosodic segments. English allows an extra morpheme-internal position at the right edge of words that is not permitted word-internally, which some researchers have analyzed as extrarhymal (e.g. Myers 1987). Since such words can also be inflected, (7b), the result would appear to be two adjacent extraprosodic positions (see also chapter 36: final consonants).9 (7)

English a. b.

træp straps tæk taxed træm

tramps

If the final two positions in (7b) are both licensed as extraprosodic at the same point, the result would violate the Peripherality Condition: extraprosodic status can only be assigned to elements in peripheral positions (Hayes 1980; Harris 1983). 8

(5a) disregards the fact that scholars working with this type of representation hold different views on whether the syllable is internally structured. 9 More troubling are words like [nekst] next, with two morpheme-internal extraprosodic positions; we return to such cases when we examine the possibility that s + stop clusters form complex segments.

The Representation of sC Clusters

8

The alternative that is typically adopted, therefore, is that these two types of extraprosodic segments are extraprosodic at different points in the derivation. Extraprosodic /s/ in taxed and /p/ in tramps are incorporated as codas in the post-lexical phonology before inflectional /t/ and /s/ are added (cf. Borowsky 1986). Initial /s/ in (7a) similarly loses extraprosodic status in the post-lexical phonology, where lexical constraints on sonority profile are no longer assumed to hold. The problem with this approach, however, is that it fails to show that extraprosodic elements ever function as members of the constituents that ultimately come to organize them, the onset in (7a) and the coda in (7b) (Piggott 1991). This problem is resolved once the Strict Layer Hypothesis is abandoned – the requirement that elements be dominated by the immediately higher category in the prosodic hierarchy (Nespor and Vogel 1986). Extraprosodic segments on this view are technically not extraprosodic (unaffiliated); instead, they are organized by some higher constituent in the prosodic hierarchy. One possible representation for /tæm

/, consistent with this approach, is in (8): initial /s/ is organized by the PWd, extrarhymal /p/ is organized by the syllable, and inflectional /s/ is adjoined to the PWd. This representation is consistent with the Peripherality Condition: extrarhymal /p/ and inflectional /s/ are each at the right edge of a separate PWd (Goad and White 2006). PWd

(8)

PWd q O

s

t

R N

C

æ

m

p

s

Returning specifically to sC clusters, the representation for /st/ in (8) involves s linking directly to the PWd (see e.g. Goldsmith 1990, drawing on evidence from English; Trommelen 1984 and Fikkert 1994 from Dutch; Goad and Rose 2004 from German). An alternative involves s linking directly to the syllable, the inverse of extrarhymal /p/ in (8) (see e.g. van der Hulst 1984 with evidence from Dutch; Levin 1985 and Kenstowicz 1994 from English; Barlow 1997 and Gierut 1999 from English in phonologically delayed children; Drachman 1990 and Tzakosta and Vis 2009 from Greek; Tzakosta 2009 from child Greek).10 These two proposals were provided earlier as (5b) and (5c).11 10

Note that Levin (1985) adjoins, rather than directly links, s to the syllable. To the list in the text we can add Giegerich (1992), Hall (1992), and Booij (1995), who analyze s as an onset-internal appendix, using data from English, German, and Dutch, respectively; and Ewen and Botma (2009), who organize s into the specifier position of the onset for Germanic. 11 Other less commonly proposed licensers for s will not be considered due to space constraints: e.g. the Foot (Green 2003 on Munster Irish) and the Phonological Phrase (Vaux 1998 on Armenian).

Heather Goad

9

The alternatives in (5b) and (5c) make different predictions concerning the distribution of sC clusters. I show that both options (or their equivalents) are needed, indicating that sC clusters cannot be represented identically across languages (Goad and Rose 2004; Vaux 2004; Ewen and Botma 2009). In languages with (5b), sC is only licensed PWd-initially. German has this profile (Goad and Rose 2004). (9) shows that sC clusters only occur stem-initially; word-internal tautosyllabic sC clusters are actually stem-initial, (9b). If stem-initial corresponds to PWd-initial, this restriction on sC distribution can be captured through (5b). (9)

German a. b.

[œp>nH]PWd [œte(Hn]PWd [bH[œte(Hn]PWd]PWd [gH[œte(Hn]PWd]PWd *[CV.œCV]PWd

‘spider’ ‘to stand’ ‘to insist’ ‘to confess’

Hall (1992) argues against (5b) for German on grounds that it incorrectly predicts aspiration in s + stop clusters, as the stop is syllable-initial in this representation. However, Iverson and Salmons (1995), following Kim (1970), offer an explanation for the absence of aspiration in s + stop that holds independently of how s is organized: because s is voiceless, the peak of glottal width that characterizes aspiration is internal to s, not the following stop. Thus, we do not see the absence of aspiration in s + stop as reason to reject (5b). In contrast to German, sC clusters in Dutch and English have a wider distribution, requiring (5c). Both languages contain monomorphemic examples where the rhyme preceding sC appears unable to accommodate s (e.g. Dutch [ekstHr] ‘magpie’, English [ekstrH] extra (van der Hulst 1984; Levin 1985)). (5c), however, freely permits violations of the Peripherality Condition. Accordingly, before we definitively conclude that it is required for morpheme-internal sC, we must examine the following alternative: PWd-initial sC clusters involve appendices organized as in (5b); in word-medial clusters, s is a coda. If this analysis could be supported, we could dispense with (5c). To show that (5c) is truly needed, we examine word-medial sC clusters in English in detail. Harris (1994) discusses the constraints governing three-position rhymes shaped VVC in this language. As (10a) reveals, coda sonorants in these superheavy syllables are confined to coronals which share place with the following onset (*[œowlbHr], *[mawmpHn]). PWd-internal VCC rhymes are not considered by Harris (they are not well-formed in Government Phonology, the framework in which he works). (10b) reveals that the onset is similarly constrained to coronal and the preceding consonants must be homorganic nasal + stop (*[v>ltnHr], *[–ZlkœHn]). (10)

English a.

VVC rhymes [œowldHr] shoulder [mawntHn] mountain [kawnsHl] council

b. VCC rhymes [æntlHr] antler [v>ntnHr] vintner [–ZI(k)œHn] junction

The Representation of sC Clusters

10

With these constraints in mind, we turn to cases where the consonant following VV/VC is s. Parallel to (10a), (11a) shows that only coronal s is permitted after VV (*[i(ftHr]) and the following consonant must be coronal (*[i(spHr]).12 This leads Harris to conclude that s in (11a) is syllabified as the coda of a three-position rhyme. The forms in (11b) are similarly parallel to those in (10b), seemingly leading to the same conclusion. (11)

a.

VVs rhymes [i(stHr] Easter [DjstHr] oyster

b.

VCs rhymes [mAnstHr] monster [m>nstrHl] minstrel

Problems arise, however, in (12). (12a) shows that VCs does not always respect the constraints holding of (11b): the consonant preceding s is not restricted to placesharing sonorants. (12b) reveals that the onset following s can be other than coronal, in contrast to (11a) and (11b). One could object on grounds that, extra aside, the words in (12) involve Latinate prefixes. However, these prefixes are not synchronically productive: they fall within the stress domain, and must therefore be contained inside the lower PWd ([’AbstHkHl]PWd, *[’Ab[stHkHl]PWd ]PWd). (12)

Appendix s a.

b.

Non-coronal codas [’ekstrH] extra [’AbstHkHl] obstacle Non-coronal onsets [‘ekspH’z>œHn] exposition [’kAnskr>pt] conscript

If the forms in (12) truly involve appendixal s, we expect to find s occurring after three-position rhymes of the shape in (10). (13) shows that such words are well-formed, albeit rare. Bolster and holster are monomorphemic, and while -ster in upholster is historically a class 2 suffix (seventeenth-century uphold-ster ‘small furniture dealer’), this analysis no longer holds, as revealed by the fact that upholster is now a verb and -y can attach outside (upholstery).13 (13)

VVC rhymes followed by s [bowlstHr] [howlstHr] [Zp(h)owlstHr]

bolster holster upholster

In short, while some instances of medial sC in English may involve s as coda, appendixal s is required to capture the data in (12) and (13). This thereby supports the postulation of (5c), where s is licensed by the syllable. Although languages like English appear to require (5c), this analysis cannot straightforwardly capture the fact that these same languages syllabify s as a coda 12

Neither constraint holds in dialects with lengthened [A(]/[æ(] in e.g. after, basket (Harris 1994). Hayes (2009: 210–211) considers the vowel in such words to be monophthongized (so presumably short) in some dialects. However, this does not hold of all dialects; witness, for example, RP [bHwlstH]. 13

11

Heather Goad

after short stressed vowels. If the appendix representation is available wordmedially, why do native speakers judge sC as heterosyllabic (’pes.ter) rather than as appendix + onset (’pe.ster)? It cannot be due to sonority profile, as will be seen shortly for Dutch. Heterosyllabicity of medial sC is handled more elegantly in Kaye’s (1992) Government Phonology approach to sC clusters, which we consider now. Kaye (1992) proposes that sC clusters are syllabified as coda + onset sequences, shown earlier in (5d).14 He provides support from Italian, Ancient Greek, European Portuguese, and British English; see also Brockhaus (1999) on German and Cyran and Gussmann (1999) on Polish. We consider first Italian, where the coda + onset pattern observed for ’pester-type words is illustrated more concretely. In Italian, rhymes of stressed syllables must branch (Chierchia 1986). When a stressed syllable lacks a coda, the vowel lengthens; see (14a) and (14b). (14c) shows that sC clusters do not trigger lengthening as do branching onsets; instead, they pattern with coda + onset sequences (14d), revealing that medial s is a coda. (14)

Medial sC in Italian a. b. c. d.

[’fa(to] [’ka(pra] [’pasta] [’parko]

‘fate’ ‘goat’ ‘pasta’ ‘park’

Turning to word-initial position, Kaye proposes that s in this position is also a coda; the difference between the initial and medial environments is that, in the former, s is the coda of an empty-headed syllable. Word-initial coda s follows from the Uniformity Principle in Government Phonology, which requires syllabification to be constant for a given string of segments, within and across languages. Kaye provides empirical support for Uniformity from Italian masculine definite article allomorphy and raddoppiamento sintattico. In the former, vowel- and sC-initial words pattern together, in contrast to words beginning with (branching) onsets; compare (15a) and (15b) with (15c). Since the representation for s in an sC cluster in (5d) includes a preceding nucleus, there is a structural parallel with vowel-initial words. (15)

Masculine definite article allomorphy (Davis 1990) a. b. c.

l’est /lo est/ lo studente il burro il clima

‘the ‘the ‘the ‘the

east’ student’ butter’ climate’

In raddoppiamento sintattico, the first consonant in an onset geminates when the preceding word ends in a stressed vowel, while the first consonant in an sC cluster resists gemination; see (16). The pattern in (16b) follows directly from the view that s is a coda: as coda + onset, sC already has precisely the structure that holds of geminates. 14

A precursor to this appears in Vennemann (1988). Vennemann proposes that initial s is “quasinuclear” in some languages, a type of degenerate syllable. He argues that this analysis of Latin s explains its development into a regular Vs syllable in some Romance languages, notably Spanish.

The Representation of sC Clusters (16)

12

Raddoppiamento sintattico (Chierchia 1986) a. b.

palto pulito citta triste citta straniera

[pal’toppu’lito] [Œit’tat’triste] [Œit’tastra’niera] *[Œit’tasstra’niera]

‘clean coat’ ‘sad city’ ‘foreign city’

Although it is evident from (15) and (16) that sC clusters in Italian cannot be analyzed in the same fashion as branching onsets, the proposal that they contain an initial appendix can also handle these data (see Chierchia 1986; Davis 1990). Of the cases Kaye considers, the construction that poses a particular challenge for appendixal s is European Portuguese vowel nasalization. We turn to this case now. In European Portuguese, nasal consonants cannot close syllables. While /n/ is realized intact before vowel-initial bases (17a), before onset-initial bases, nasality surfaces on the preceding vowel (17b), (17c). Interestingly, sC-initial bases pattern as vowel-initial (17d). (17)

European Portuguese a. b.

[in]admissivel [h]pureza [h]satisfeito c. [h]tratavel d. [inœk]apavel

‘inadmissable’ ‘impurity’ ‘dissatisfied’ ‘unsociable’ ‘inescapable’

(17d) can be straightforwardly expressed under Kaye’s view that sC clusters are coda + onset, because, for independent reasons, all syllables in Government Phonology contain an onset constituent. Consider the representations below.15 In (18a), /n/ associates to the onset of the first syllable in the base. In (18b), this position is occupied, so nasality is preserved on the preceding vowel. The right result obtains in (18c) precisely because the syllable containing s includes an empty onset. (18)

Coda analysis a.

O

R

+

O

N

c.

O

i

n

R

+

N i 15

O

R

b.

R

N

N

a

i O

R

+

O

R N

n

p

u

R N

N n

O

œ

k

a

No representations are provided by Kaye. (18) reflects my best guess (minus X-slots), based on his discussion.

13

Heather Goad

An appendix analysis of s, it seems, cannot formally capture (17d). See (19c), where bases with initial sC are incorrectly predicted to pattern with onset-initial bases because there is no empty constituent to host /n/. (19)

Appendix analysis a.

q R

+

O

N i c. *

n

q

b. q

R

R

N

N

a

i

O

+

N i

+

O

R N

n

p

u

q

q R

q

R N

n

œ

k

a

Kaye’s paper compares the coda + onset analysis of sC clusters to the alternative that they form branching onsets; the option that s is analyzed as an appendix is not discussed. We have seen that appendixal s cannot straightforwardly capture European Portuguese. It also goes against the Uniformity Principle: Italian s is an appendix in stra’niera but a coda in ’pasta; this would likely be considered a weakness by proponents of Government Phonology. However, there are contexts where word-internal s maintains its appendix status, in contrast to the pattern observed for Italian; sC can follow a rhyme that is already full, so s cannot be accommodated as an ordinary coda. We have already observed this for English, but we have also seen that English permits word-internal three-position rhymes under limited circumstances and most of the problematic sC data are in words that historically involve prefixes. To ensure that there is nothing unusual about English concerning the distribution of sC clusters, let us turn to Acoma (Miller 1965). Two-position rhymes in Acoma are limited to VV and seemingly Vs (loans aside); see (20a) and (20b). If word-internal s were always a regular coda, as it appears to be in (20b), we would expect it to be restricted to occurring after short vowels, as in Italian. (20c) reveals that this is not the case. (20)

Acoma a. b.

[spúuná] [ja?ái] [sust’á] [?éská]

‘pottery’ ‘sand’ ‘I took water’ ‘rawhide’

The Representation of sC Clusters c.

14

[?úuscúutsh i] ‘drum’ [ku’isc haÏa] ‘knot’ [w’i’isp’i] ‘cigarette’

It appears that s is an appendix to the syllable in Acoma, rather than a regular coda. Before we accept this, however, an alternative analysis must be considered, that sC clusters are not actually clusters in Acoma but are, instead, adjacent onsets interrupted by an empty nucleus. This analysis would, of course, allow s to occur after long vowels, and it could be motivated by the observation that the plain– aspirated contrast is maintained after s ([?úu.sØ.cúu.tsh i] vs. [ku’i.sØ.c ha.Ïa]). Evidence from the right word edge, however, reveals a potential challenge: word-final consonants, which are always onsets of empty-headed syllables in Government Phonology (Kaye 1990), are not permitted in Acoma: *[. . . CV.sØ]. In sum, we have critiqued four representations for sC clusters. While we can likely dispense with the extraprosodic representation for s in (5a) in favor of the alternatives in (5b)–(5d), choosing between these alternatives is no easy task. On the one hand, there are languages like Acoma and English, which are most compatible with the position that s is an appendix to the syllable; on the other, there are languages like European Portuguese and to a lesser extent Italian, which seem to require a coda analysis for s. When German is compared with English and Dutch, it becomes clear that languages also differ in where sC can occur – stem-initially only or also morpheme-internally – revealing that appendixal s must be licensed by the PWd in the former case and by the syllable in the latter.

5

Are all sC clusters represented in the same manner in a given language?

Thus far, we have treated all sC clusters as a single class. Given that these clusters fall into two groups concerning their sonority profile, several researchers have proposed different representations for each group within a given language. One body of work considers rising sonority sC clusters to be structured as branching onsets, while s + obstruent clusters are appendix-initial (see e.g. Hall 1992 with data from German; Fikkert 1994 and Booij 1995 from Dutch; Gierut 1999 from English in phonologically delayed children). Another body of work motivates a different analysis for s + stop clusters: they are not actually clusters but instead form complex segments (see for example van de Weijer 1996, drawing on data from several languages; Fudge 1969, Ewen 1982, Fujimura and Lovins 1978, and Selkirk 1982 from English; Broselow 1983 from L2 English; Wiese 1996 from German; Barlow 1997 from English in phonologically delayed children). Since all of this research draws a boundary between rising sonority clusters and s + stop clusters, much of the data effectively shows evidence for either position. The section on cluster phonotactics revealed that, when place of articulation is examined, end-state grammars challenge the view that s + sonorant clusters form branching onsets. Evidence from acquisition (chapter 101: the interpretation of phonological patterns in first language acquisition), however, suggests that some learners disregard place, relying solely on sonority in their analysis of left-edge clusters: all rising sonority clusters – both obstruent + sonorant and s + sonorant – pattern together. Consider the Dutch children in Fikkert’s (1994)

15

Heather Goad

Table 38.1 Jarmo’s cluster development Stage

Pattern

Examples

Age

1

stop

/’vl>ndHr/ /’sla(pH(n)/

[’k>nr] [’ta(pH]

‘butterfly’ ‘to sleep’

2;2.6 2;0.28

2

liquid

/’vl>ndHr/ /’sla(pH(n)/

[’l>Ir] [’la(pH]

‘butterfly’ ‘to sleep’

2;2.27 2;3.9

3

fricative

/’ølei‘ba(n/ /’sla(pH(n)/

[’øeiøa(n] [’sa(pH]

‘slide’ ‘to sleep’

2;4.1 2;3.9

4

stop + liquid

(skipped by Jarmo)

5

fricative/s + liquid

/’flesjH/ /slAk/

[’slesjH] [flAk]~[slAk]

‘bottle (dim)’ ‘snail’

2;4.1 2;4.1

study: all follow the same developmental path for fricative + liquid and s + lateral clusters. The stages through which learners pass are exemplified in Table 38.1 with data from Jarmo. Further, all children in Fikkert’s study master s + stop clusters at a different point in time from rising sonority clusters. As all rising sonority clusters pattern together in Dutch acquisition, in contrast to s + obstruent, Fikkert concludes that s + sonorant clusters are represented in the same manner as branching onsets. For her, s + obstruent clusters involve s licensed by the PWd, as in (5b) above. While sonority plays a decisive role in Fikkert’s data, this is not the case for all children. Indeed, we observed in (6) that when only one member of a cluster is produced by Annalena and Amahl, it is the cluster head that survives, regardless of its relative sonority. To show that this pattern extends past the deletion stage, consider the developmental path for Amahl in Table 38.2. Table 38.2 Amahl’s cluster development Stage

obstruent + liquid

1–8

reduction to head reduction to head

13–14

branching onset acquired

15–19

/sl/

/sm sn/

/sp sk/

/st/

Age

reduction to head

reduction to head

reduction to head

2.60–2.175 2.233–2.256

fusion

fusion

vacuous fusion

20–22 24 25 26–29

vacuous fusion

2.261–2.333

fusion

2.345–3.38

appendix acquired

3.78–3.96 appendix acquired

3.104–3.128 appendix acquired

appendix acquired

3.133–3.355

The Representation of sC Clusters

16

All clusters reduce to the head through stage 8. Branching onsets are acquired first, emerging at stage 9 and being fully mastered at stage 13. sC clusters are reduced to the head until stage 15, at which point the two consonants undergo fusion. Fusion is overt for s + sonorant (e.g. /sn/ → [n]), but not initially for s + stop since both consonants are voiceless (e.g. /st/ → [t]). At stage 20, when stridency is acquired, fusion becomes overt for /st/ (/st/ → [ts~s~Ú]). Importantly, at this stage, no target s + sonorant clusters are realized as [s]; thus, earlier reduction to the head cannot have been due to the unavailability of [s] in Amahl’s productions. Appendices are not acquired until 210 days later than branching onsets, first for /sl/. The remaining sC clusters are then acquired over a period of 55 days. The developmental profiles for Jarmo and Amahl are completely different for s + sonorant clusters. It truly appears that these clusters are analyzed as branching onsets in Jarmo’s grammar, and with an initial appendix/coda in Amahl’s grammar. We briefly consider the consequences of this. In the section on phonotactics we observed that place identity and C2 place profile suggest that, in end-state Dutch, all sC clusters are represented differently from branching onsets. If learners like Jarmo initially only consider sonority in assigning clusters to categories, then some unlearning must take place to arrive at the target representations. These children must discover that: (i) place identity is respected in branching onsets (*/tl/), but not in s + sonorant clusters (/sl/); (ii) s + sonorant is represented in the same fashion as s + obstruent. The implication is that, prior to this, sonority pattern learners do not attend to place identity. Fikkert provides some data consistent with this (e.g. Jarmo’s /’dr>IkH(n)/ → [’tl>IkA] ‘to drink’ (2;4.1)). Positive evidence for (i) and (ii) comes from medial clusters following stressed vowels: as (21) shows, /tl/, /sl/, and /st/ are syllabified as coda + onset, while /tr/ forms a branching onset (Trommelen 1984; Kager and Zonneveld 1986).16 (The initial vowel in (21d) is not underlyingly long, which could have prevented heterosyllabification of medial /tr/; it is lengthened because rhymes are minimally bipositional in Dutch.) (21)

Dutch a. b. c. d.

[’At¬.lAs] [’Ds.lo(] [’pAs.ta(] [’ma(.tr>ks]

‘atlas’ ‘Oslo’ ‘paste’ ‘matrix’

Although data such as these reveal that s + sonorant clusters do not pattern as branching onsets, words of the profile in (21a) and (21b) are largely restricted to borrowings and proper names (van der Torre 2003) and are likely infrequent in child-directed speech (Fikkert 1994). Thus, the analysis that s + sonorant clusters are not actually branching onsets may be unlearnable under the scenario that children begin assigning clusters to classes based solely on sonority. Does this mean that adult Dutch is a language where s + sonorant clusters are analyzed as branching onsets? This is Fikkert’s (1994) position, but it leaves unexplained the 16

Thanks to Janet Grijzenhout and Stephanie Schreven for help with the phonetic detail in (21).

Heather Goad

17

differences in place profile of true branching onsets and s + sonorant clusters, as well as the syllabification of s + sonorant with s + obstruent in word-medial position. We have just observed that some researchers consider sonority decisive in assigning clusters to categories with the result that rising sonority s-initial clusters are analyzed as branching onsets. Consequently, it is s + obstruent that is singled out as formally different. Another approach to the different behavior of s + stop has been to treat these strings as complex segments. This view is autosegmentally expressed in (22). (22)

s + stop as complex segment X s

p

Many of the arguments for a single segment analysis of s + stop are distributional. Indeed, most researchers who motivate this analysis begin with the observation that all left-edge three-member clusters in the languages under examination are s + stop-initial (e.g. Fudge 1969; Selkirk 1982; van de Weijer 1996; Wiese 1996). The position that branching onsets contain a maximum of two consonants can be maintained if s + stop clusters form complex segments (chapter 29: secondary and double articulation). As Selkirk (1982) asserts, similar problems can be resolved at the right edge if the single segment analysis holds independent of position (see earlier Fudge 1969). On the one hand, words like next violate the finding that, inflection aside, English codas are maximally binary. On the other, words such as next, wasp, and task violate the constraint that C2 in an obstruent coda cluster is always coronal. If these words are segmented as [n-ek-st], [w-A-sp], and [t-æ-sk], these problems are resolved. Following on this, Ewen (1982) observes for English that, unlike other clusters which appear in mirror-image order in syllable-initial vs. -final position (clip, silk), sC clusters occur in the same order in both positions (skip, task). He notes that similar observations were made earlier by Vogt (1942) for Norwegian and Sigurd (1965) for Swedish. A different sort of distributional argument is made by Wiese (1996) for German. He shows that s + stop and other complex segments, namely affricates, have similar distributions. Alongside [œpl>t] ‘chippings’ and [œpri(sHn] ‘to sprout’, we find words like [pfl>çt] ‘duty’ and [pfri(m] ‘awl’. Turning to other types of phonological behavior, most of the evidence that van de Weijer (1996) puts forward involves languages where s + sonorant behaves differently from s + stop and together with true branching onsets, thereby supporting a branching onset analysis for s + sonorant and another analysis for s + stop. Reduplication in Gothic provides the most compelling evidence that the appropriate analysis for s + stop is as a complex segment (see also Cairns and Feinstein 1982; Wiese 1996). Gothic has a class of verbs where the preterite is formed through reduplication. The data in (23a) and (23b) (from van de Weijer 1996) reveal that singleton onsets and the first member of branching onsets are copied. (23c) and (23d) show that sC clusters do not behave uniformly. Those with rising sonority pattern with branching onsets – only C1 is copied – while s + stop is copied as a unit.

The Representation of sC Clusters (23)

18

Gothic reduplication a.

haita hwopa b. fraisa c. sle(pan d. skaida

‘I am called’ ‘I boast’ ‘I try’ ‘to sleep’ ‘I sever’

hai-hait hwai-hwop fai-frais sai-sle(p skai-skaih

‘I ‘I ‘I ‘I ‘I

was called’ boasted’ tried’ slept’ severed’

The pattern in (23d) is entirely as expected if s + stop forms a complex segment; indeed, accounting for it through an appendix/coda + onset representation is challenging at best. Somehow, the analysis would have to copy the segments up to and including the cluster head. We turn now to examine some challenges for the complex segment approach to s + stop. Although we have provided distributional evidence for this proposal, another look reveals problems on this front. Van der Hulst (1984) rejects the analysis for Dutch, on the grounds that s + stop does not have the same distribution as singleton obstruents. Putative complex segments can follow a vowel but not a consonant, unlike non-complex obstruents: [sesp] ‘wasp’ but *[selsp], cf. [selp] ‘lion cub’. The same problem holds for English: [wAsp] wasp, [wArp] warp, *[wArsp]. If s + stop forms a two-consonant string, this observation follows straightforwardly. Consider next patterns of epenthesis in cluster repair in L2 acquisition and loanword adaptation (chapter 95: loanword phonology). (24) presents four patterns of repair for learners/borrowers whose native grammars lack both branching onsets and sC clusters (V = epenthetic vowel). Although there are languages where the position of the epenthetic vowel is constant (24a) and (24b), in languages where both prothesis and anaptyxis are observed, two general patterns are found (24c) and (24d): s + stop is never interrupted, obstruent + sonorant always is, and s + sonorant is variably interrupted.17 (24)

Patterns of cluster repair a. b. c. d.

Japanese (Lovins 1974) Iraqi Arabic (Broselow 1983) Egyptian Arabic (Broselow 1983) Farsi (Karimi 1987)

s-V-stop V-s-stop V-s-stop V-s-stop

obstr-V-son V-obstr-son obstr-V-son obstr-V-son

s-V-son V-s-son s-V-son V-s-son

Broselow (1983) proposes that the reluctance of L2 learners to epenthesize into s + stop motivates their analysis as complex segments (see also van de Weijer 1996). While this analysis helps to explain the patterns in (24b)–(24d), it cannot straightforwardly account for the Japanese pattern. Clearly, absence of strings of the shape VsTV (T = stop) in Japanese makes prothesis an impossible repair. Given this, under the view that s + stop forms a single segment, the expected pattern for Japanese is simplification of the complex segment, as in van der Weijer’s (1996) analysis of Sanskrit reduplication: C1 is normally copied ([snih]-[si-Ïnih] ‘to be sticky’ (root-perf), [dru]-[du-druv] ‘to run’), but complex segments are reduced to their head (the stop): [skand]-[ka-’skand-a], *[ska-’skand-a] ‘to leap’ (see chapter 119: reduplication in sanskrit). 17

As Fleischhacker (2001) shows, more divisions among s + sonorant are actually observed. We return to this below.

19

Heather Goad

Finally, the single segment hypothesis for s + stop is challenged by Byrd (1994) on articulatory grounds. Byrd experimentally examines /sk/ strings in English in three contexts: word-initially, cross-word, and word-finally. She observes that #sk involves less overlap than s#k (as well as sk#), and sk# less overlap than gd#. If s + stop forms a single segment, we would expect to find a higher degree of overlap than is observed in strings of consonants that clearly do not form single segments (s#k and gd#). To sum up, the goal of this section was to examine whether all sC clusters pattern as a class, in contrast to branching onsets. On the one hand, we observed that s + sonorant sometimes patterns with obstruent + sonorant, suggesting that they may have the same representation. Supporting evidence was provided from the developmental path for left-edge clusters in Dutch. On the other hand, we examined evidence in favor of the position that s + stop clusters form complex segments. The strongest evidence for this comes from Gothic reduplication. However, both positions were challenged as well, the analysis for child Dutch by the observation that end-state Dutch treats s + sonorant together with s + obstruent. Perhaps the most damaging evidence against the complex segment analysis is that the articulatory evidence available does not support it.

6

Perceptual considerations

Throughout the chapter, it has been assumed that a structural difference holds between true branching onsets and (at least some) sC clusters. We have observed, however, that languages do not always draw a clear line between these two cluster types: s + sonorant sometimes patterns with s + stop and at other times with obstruent + sonorant. When we consider Fleischhacker’s (2001) survey on epenthesis in L2 and loanword phonology, the problem becomes even more acute: some languages draw the boundary between cluster type internal to the s + sonorant class, as shown in (25), in contrast to what was observed in (24).18 (25)

more prothesis more anaptyxis s + stop < s + m < s + n < s + l < s + r < s + glide < stop + son (Wolof) Egyptian Hindi Kazakh Farsi2 (Catalan) Arabic

Results like these lead Fleischhacker to abandon a structural approach to cluster representation, putting her in the company of others who advocate eliminating the syllable (e.g. Steriade 1999).19 She argues instead for a perceptually motivated 18

Farsi2 refers to the data collected by Fleischhacker. Although Karimi (1987) reports that all s + sonorant clusters pattern together, as per (24d), she provides no s + rhotic or s + glide examples. The data collected by Fleischhacker show a different pattern for these clusters, as revealed in (25). Since this may reflect a dialect difference, I have labeled the language from which Fleischhacker collected data Farsi2. In addition, Catalan is in parentheses because, although it draws a division between s + rhotic and s + glide, only prothesis is attested; s + glide and stop + sonorant do not undergo epenthesis. 19 Steriade’s arguments include the absence of clear evidence for word-internal syllable boundaries. Some of the evidence concerns the syllabification of s + stop clusters in English when the following vowel is stressed (e.g. mysterious). Speakers’ judgments vary on where to draw the boundary in such words, because neither parse is sanctioned at word edge: V1.stV2 is problematic because V1 is illicit word-finally; V1s.tV2 is problematic because [t] is not aspirated, as it would be word-initially.

The Representation of sC Clusters

20

approach to cluster behavior. The epenthesis site, in particular, is chosen to maximize perceptual similarity between the target (non-epenthesized) form and the output. In view of this, we consider in this section whether the differences that hold between sC clusters and true branching onsets can be explained by perceptual considerations alone; this, of course, would challenge the claim that a structural approach to cluster behavior is necessary. We do not have space to examine Fleischhacker’s proposal in detail, but the predictions she motivates are as follows: (i) anaptyxis is preferred to prothesis in stop + sonorant; (ii) prothesis is preferred to anaptyxis in s + stop; (iii) among s + sonorant, more anaptyxis is expected as C2 increases in sonority; and (iv) more anaptyxis is expected in stop + sonorant than in fricative + sonorant. Concerning (iv), note that Fleischhacker’s account does not distinguish s from other fricatives; that is, no explanation is provided for the observation that fricatives other than s pattern with stops in preferring anaptyxis to prothesis ((24) above). We return to this shortly. First, let us examine the role of perception in sC well-formedness in more detail (chapter 98: speech perception and phonology). As alluded to earlier, the acoustic properties of s, unlike other obstruents, enable it to appear in positions where it is not followed by a sonorant: strident fricatives have robust internal cues for both place and manner, ensuring their perceptibility in all contexts, even before stops (Wright 1996, 2004). Clearly, then, the view that segments are ordered to yield a rise in sonority toward the peak does not extend to s. Indeed, in spite of the sonority reversal, (strident) fricative + stop is superior to both stop + stop and stop + fricative,20 even though the latter two contain a sonority plateau and minimal rise respectively, because (strident) fricatives are less dependent on formant transitions for their identification than stops (Wright 1996, 2004).21 However, while the acoustic properties of s explain why appendices are so often limited to s on the one hand and why these segments can be followed by stops on the other, they cannot, as far as I can tell, explain cross-linguistic preferences on sC profile. Table 38.3 shows that sC clusters have a rather unusual distribution across languages when viewed from the perspective of perceptual robustness. We focus on word-initial position. Since the perceptibility of all consonants in C2 position in an initial sC cluster will be partly compromised by the preceding s, Table 38.3 sC cluster profiles across languages Spanish

French, Acoma

Greek

English

Dutch

German

Russian

s + stop

*

✓

✓

✓

✓

✓

✓

s + fricative

*

*

✓

*

✓

*

✓

s + nasal

*

*

(*)

✓

✓

✓

✓

s + lateral

*

*

*

✓

✓

✓

✓

s + rhotic

*

*

*

*

(*)

✓

✓

20

Evidence that this observation is not restricted to strident fricative + stop comes from Greek: the stop + stop clusters in (3a) are often replaced by fricative + stop. 21 See Morelli (1999) for an alternative explanation of obstruent cluster well-formedness that appeals to markedness constraints on segment sequencing.

21

Heather Goad

we would expect consonants that are most perceptible to be positioned after s. Masking should not be too severe in this context; as mentioned earlier, Byrd (1994) observes that #sk clusters involve less overlap than s#k and sk#. The problem may rather be one of duration: Byrd finds that /s/ is longer in #sk than in both s#k and sk#, while /k/ is shorter in #sk than in both s#k and sk#. If the relatively short duration of C2 can be generalized to other #sC clusters, we would expect segments with robust internal cues to be favored in this position. Liquids should be optimal, since they have clear formant structure throughout. Nasals should be favored over stops, since their manner (and to a lesser extent their place) properties are present in the nasal spectrum. Stops, which have weak internal cues, should be the least optimal. What we observe in Table 38.3, by contrast, is that s + stop is favored. French and Acoma do not permit s + sonorant clusters at all (French has s + sonorant in loanwords), and depending on the status of marginal s + nasal clusters, Greek may fall into this category as well. Otherwise, it permits s + sonorant clusters of lower sonority than those of higher sonority. Although a larger typology of languages is required before firm conclusions can be drawn, Table 38.3 suggests that s + stop › s + nasal › s + lateral › s + rhotic (› = is more harmonic than). The favored profile in sC clusters is thus the opposite of that observed for branching onsets. This is not unexpected on a structural account if all sC clusters are head-final, in contrast to branching onsets. In sC clusters, C2 is the onset head; thus it should respect the preferences holding of singleton onsets. Since obstruents are the optimal onsets (e.g. Clements 1990), a parallel should be observed between obstruents in C1 position in branching onsets and stops in C2 position in sC clusters (not fricatives more generally, because of the preceding s; see Wright 2004: 51). While the C1C2 asymmetry in branching onsets vs. sC clusters follows from the status of s as an appendix, it is best captured, I suggest, under Kaye’s proposal that s is a coda. Recall from (14) and (21) that medial sC clusters in Italian and Dutch are heterosyllabic. If sC clusters are always syllabified as coda + onset clusters, then their profile should respect cross-linguistic preferences for optimal syllable contact. Syllable contact will favor C2 with lower sonority: Vs.TV › Vs.NV › Vs.lV › Vs.rV. As C2 increases in sonority, the cluster prefers to be syllabified as a branching onset, but if this option is never available for sC clusters, then higher-sonority sC clusters will be forbidden, regardless of their position in the word. The profile in Table 38.3 closely parallels Fleischhacker’s typology in (25) for preferred epenthesis sites, in sC clusters. Prothesis occurs more commonly when C2 has lower sonority. As the sonority of C2 increases, prothesis will result in poor syllable contact. Note as well that the proposed syllable contact account of sC well-formedness leads to a distinction between s + sonorant and fricative + sonorant, as only the latter can form branching onsets. Thus, the fact that fricative + sonorant patterns with stop + sonorant in epenthesis follows, in contrast to under Fleischhacker’s account (see (iv) above). In sum, I contend that both perceptual and structural considerations must be factored into our understanding of cluster well-formedness. While perceptual considerations can explain why appendices are so often limited to s and why s + stop is well-formed in spite of its sonority profile, it is the structural differences between sC clusters and branching onsets that explain the preference for sC profile on the sonority dimension as well as some observed differences in epenthesis site.

The Representation of sC Clusters

7

22

Conclusion

In this chapter, we have examined several alternative analyses for sC clusters. On the empirical front, we have seen that s + stop clusters reliably pattern differently from true branching onsets. Not surprisingly, then, the proposals we have examined for s + stop all share an important property: s + stop clusters are head-final, whether s is unaffiliated, an appendix organized by some prosodic constituent above the onset, the coda, or the first member of a complex segment. Branching onsets, by contrast, are head-initial. This difference in headedness helps to explain parallels on the place dimension between C2 in an s + stop cluster and C1 in a branching onset, as well as syllabification preferences in word-medial contexts. Beyond that, however, details of the proposals differ, and, following on this, each proposal is both supported and challenged by the available evidence. There are languages like Acoma, English, and Dutch where s can appear medially after rhymes that are “full,” thereby providing support for the analysis that s is linked to the syllable and potentially challenging the coda analysis. However, in some of these same languages, namely English and Dutch, as well as in languages like Italian, the observation that sC clusters are heterosyllabic after stressed vowels supports the coda analysis and thereby questions the proposal that s is licensed by the syllable or that s + stop form a complex segment. The complex segment analysis, in turn, is supported by the reduplication pattern in Gothic which both the appendix and coda analyses fail to elegantly capture. At present, then, it seems that multiple representations for sC clusters may be required. If the number of parametric options is limited and there is robust evidence available for learners to determine the appropriate representation for the language being acquired, this is far from problematic. For example, the fact that sC clusters have a more limited distribution in some languages (German) than in others (English, Dutch) can be captured if licensing by the PWd represents the least marked option and therefore the starting point for learners. There will then be positive evidence available to signal learners of some languages that sC clusters are licensed lower down, by the syllable. However, we have also seen that this type of scenario may not always work. If Dutch learners initially assume that s + sonorant clusters form branching onsets, the evidence available to undo this analysis in favor of one where all sC clusters pattern as a class is far from robust and may present a learnability challenge. The problem with s + sonorant clusters more generally is that they are phonotactically ambiguous and, following from this, they pattern ambiguously across languages. While one could fail to be surprised by this, on the grounds that these clusters are both s-initial and rise in sonority, exactly how their ambiguous behavior should be formally expressed is far from clear. They appear to be analyzed as branching onsets in some languages (e.g. Jarmo’s Dutch grammar) and as appendix/coda-initial in others (e.g. Amahl’s English grammar), but in this particular case, this finding is surprising, in view of the otherwise high degree of similarity between the two target languages. The solution that languages employ different analyses for s + sonorant is far from optimal and may lead some researchers to abandon a structural approach to the syllable altogether in favor of a perceptually grounded account of segmental

23

Heather Goad

contact. Indeed, the latter may find support in the observation that even within the class of s + sonorant, languages show different patterns of behavior; we have seen that the division between prothesis and anaptyxis can be drawn anywhere internal to this class. At the same time, however, a purely perceptually based account seems to be challenged by the finding that preferences for sC cluster profile are virtually the inverse of those observed for obstruent + sonorant clusters. While an appeal to syllable contact was made to capture both of these observations, the analysis follows most straightforwardly from the proposal that sC clusters are always syllabified as coda + onset strings. We have already seen that this proposal may be challenged by languages such as English, Dutch, and Acoma. In spite of the quantity of research that has been undertaken on sC clusters, it is perhaps most evident that more needs to be done before the issue of their representation can be resolved (if ever). A sampling of questions at the two extremes includes the following. At one end of the spectrum, can a more detailed examination of perceptual factors capture differences in the behavior of fricative + sonorant and s + sonorant clusters, thereby further questioning the need for a structurally based approach to segmental contact and syllabification behavior? At the other end, if a structural account of behavior based on syllable contact proves fruitful to pursue, with judicious use of abstract representations, can the coda + onset analysis be motivated for all languages? I leave these and many other questions in between to future research.

ACKNOWLEDGMENTS Many thanks to two anonymous reviewers and the editors for helpful comments. This work was supported by grants from SSHRC and FQRSC.

REFERENCES Arvaniti, Amalia. 2007. Greek phonetics: The state of the art. Journal of Greek Linguistics 8. 97–208. Barlow, Jessica. 1997. A constraint-based account of syllable onsets: Evidence from developing systems. Ph.D. dissertation, Indiana University. Booij, Geert. 1995. The phonology of Dutch. Oxford: Clarendon Press. Borowsky, Toni. 1986. Topics in the lexical phonology of English. Ph.D. dissertation, University of Massachusetts, Amherst. Brockhaus, Wiebke. 1999. The syllable in German: Exploring an alternative. In Harry van der Hulst & Nancy Ritter (eds.) The syllable: Views and facts, 169–218. Berlin: Mouton de Gruyter. Broselow, Ellen. 1983. Nonobvious transfer: On predicting epenthesis errors. In Susan Gass & Larry Selinker (eds.) Language transfer in language learning, 269–280. Rowley, MA: Newbury House. Byrd, Dani. 1994. Articulatory timing in English consonant sequences. Ph.D. dissertation, University of California, Los Angeles. Cairns, Charles & Mark Feinstein. 1982. Markedness and the theory of syllable structure. Linguistic Inquiry 13. 193–225. Chierchia, Gennaro. 1986. Length, syllabification and the phonological cycle in Italian. Journal of Italian Linguistics 8. 5–33.

The Representation of sC Clusters

24

Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Cyran, Eugeniusz & Edmund Gussmann. 1999. Consonantal clusters and governing relations: Polish initial consonant sequences. In Harry van der Hulst & Nancy Ritter (eds.) The syllable: Views and facts, 219–247. Berlin & New York: Mouton de Gruyter. Davis, Stuart. 1990. Italian onset structure and the distribution of il and lo. Linguistics 28. 43–55. Drachman, Gaberell. 1990. A remark on Greek clusters. In Joan Mascaró & Marina Nespor (eds.) Grammar in progress: GLOW essays for Henk van Riemsdijk. Dordrecht: Foris. Elsen, Hilke. 1991. Erstspracherwerb: Der Erwerb des deutschen Lautsystems. Wiesbaden: Deutscher Universitäts-Verlag. Ewen, Colin J. 1982. The internal structure of complex segments. In van der Hulst & Smith (1982: Part II), 27–67. Ewen, Colin J. & Bert Botma. 2009. Against rhymal adjuncts: The syllabic affiliation of English postvocalic consonants. In Kuniya Nasukawa & Phillip Backley (eds.) Strength relations in phonology, 221–250. Berlin & New York: Mouton de Gruyter. Fikkert, Paula. 1994. On the acquisition of prosodic structure. Ph.D. dissertation, University of Leiden. Fleischhacker, Heidi. 2001. Cluster-dependent epenthesis asymmetries. UCLA Working Papers in Linguistics 7: Papers in Phonology 5. 71–116. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5. 253–286. Fujimura, Osamu & Julie B. Lovins. 1978. Syllables as concatenative phonetic units. In Alan Bell & Joan B. Hooper (eds.) Syllables and segments, 107–120. Amsterdam: North-Holland. Giegerich, Heinz J. 1992. English phonology: An introduction. Cambridge: Cambridge University Press. Gierut, Judith. 1999. Syllable onsets: Clusters and adjuncts in acquisition. Journal of Speech, Language, and Hearing Research 42. 708–726. Goad, Heather & Yvan Rose. 2004. Input elaboration, head faithfulness and evidence for representation in the acquisition of left-edge clusters in West Germanic. In René Kager, Joe Pater & Wim Zonneveld (eds.) Constraints in phonological acquisition, 109–157. Cambridge: Cambridge University Press. Goad, Heather & Lydia White. 2006. Ultimate attainment in interlanguage grammars: A prosodic approach. Second Language Research 22. 243–268. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Green, Antony Dubach. 2003. Extrasyllabic consonants and onset well-formedness. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 238–253. Cambridge: Cambridge University Press. Hall, T. A. 1992. Syllable structure and syllable-related processes in German. Tübingen: Niemeyer. Harris, James W. 1983. Syllable structure and stress in Spanish: A nonlinear analysis. Cambridge, MA: MIT Press. Harris, John. 1994. English sound structure. Oxford: Blackwell. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Hayes, Bruce. 2009. Introductory phonology. Malden, MA & Oxford: Wiley-Blackwell. Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Hulst, Harry van der & Norval Smith (eds.) 1982. The structure of phonological representations. 2 parts. Dordrecht: Foris. Iverson, Gregory K. & Joseph C. Salmons. 1995. Aspiration and laryngeal representation in Germanic. Phonology 12. 369–396.

25

Heather Goad

Joseph, Brian D. & Irene Philippaki-Warburton. 1987. Modern Greek. London: Croom Helm. Kager, René & Wim Zonneveld. 1986. Schwa, syllables, and extrametricality in Dutch. The Linguistic Review 5. 197–221. Karimi, Simin. 1987. Farsi speakers and the initial consonant cluster in English. In Georgette Ioup & Steven Weinberger (eds.) Interlanguage phonology: The acquisition of a second language sound system, 305–318. Cambridge, MA: Newbury House. Kaye, Jonathan. 1990. “Coda” licensing. Phonology 7. 301–330. Kaye, Jonathan. 1992. Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics 2: 293–313. Reprinted 1996 in Henryk Kardela & Bogdan Szymanek (eds.) A Festschrift for Edmund Gussmann, 155–176. Lublin: University Press of the Catholic University of Lublin. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7. 193–231. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Kim, Chin-Wu. 1970. A theory of aspiration. Phonetica 21. 107–116. Klepousniotou, Katerina. 1998. Onset clusters in Modern Greek. Unpublished ms., McGill University. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. Lovins, Julie. 1974. Why loan phonology is natural phonology. Papers from the Annual Regional Meeting, Chicago Linguistic Society: Parasession on natural phonology, 240–250. Miller, Wick R. 1965. Acoma grammar and texts. Berkeley & Los Angeles: University of California Press. Morelli, Frida. 1999. The phonotactics and phonology of obstruent clusters in Optimality Theory. Ph.D. dissertation, University of Maryland at College Park. Myers, Scott. 1987. Vowel shortening in English. Natural Language and Linguistic Theory 5. 485–518. Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Piggott, Glyne L. 1991. Apocope and the licensing of empty-headed syllables. The Linguistic Review 8. 287–318. Rice, Keren. 1992. On deriving sonority: A structural account of sonority relationships. Phonology 9. 61–99. Selkirk, Elisabeth. 1982. The syllable. In van der Hulst & Smith (1982: Part II), 337–383. Sigurd, Bengt. 1965. Phonotactic structures in Swedish. Lund: Uniskol. Smith, Neil V. 1973. The acquisition of phonology: A case study. Cambridge: Cambridge University Press. Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Ph.D. dissertation, MIT. Steriade, Donca. 1999. Alternatives to syllable-based accounts of consonantal phonotactics. In Osamu Fujimura, Brian D. Joseph & Bohumil Palek (eds.) Item order in language and speech, 205–245. Prague: Karolinum Press. Torre, Erik Jan van der. 2003. Dutch sonorants: The role of place of articulation in phonotactics. Ph.D. dissertation, University of Leiden. Trommelen, Mieke. 1984. The syllable in Dutch: With special reference to diminutive formation. Dordrecht: Foris. Tzakosta, Marina. 2009. Asymmetries in /s/ cluster production and their implications for language learning and language teaching. In Anastasios Tsangalidis (ed.) Selected Papers from the 18th International Symposium of Theoretical and Applied Linguistics, 365–373. Thessaloniki: Monochromia. Tzakosta, Marina & Jeroen Vis. 2009. Phonological representations of consonant sequences: The case of affricates vs. “true” clusters. In Giorgos K. Giannakis, Mary Baltazani, Giorgos I. Xydopoulos & Anastasios Tsangalidis (eds.) Proceedings of the 8th International

The Representation of sC Clusters

26

Conference on Greek Linguistics, 558–573. Ioannina: Department of Linguistics, University of Ioannina. Vaux, Bert. 1998. The phonology of Armenian. Oxford: Clarendon Press. Vaux, Bert. 2004. The appendix. Paper presented at the Symposium on Phonological Theory: Representations and Architecture, City University of New York. Vennemann, Theo. 1982. Zur Silbenstruktur der deutschen Standardsprache. In Theo Vennemann (ed.) Silben, Segmente, Akzente, 261–305. Tübingen: Niemeyer. Vennemann, Theo. 1988. Preference laws for syllable structure and the explanation of sound change: With special reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. Vogt, Hans. 1942. The structure of the Norwegian monosyllable. Norsk Tidsskrift for Sprogvidenskap 12. 5–29. Waals, Juliette. 1999. An experimental view of the Dutch syllable. Ph.D. dissertation, Utrecht University. Weijer, Jeroen van de. 1996. Segmental structure and complex segments. Tübingen: Niemeyer. Wiese, Richard. 1988. Silbische und lexikalische Phonologie: Studien zum Chinesischen und Deutschen. Tübingen: Niemeyer. Wiese, Richard. 1996. The phonology of German. Oxford: Clarendon Press. Wright, Richard. 1996. Consonant clusters and cue preservation in Tsou. Ph.D. dissertation, University of California, Los Angeles. Wright, Richard. 2004. A review of perceptual cues and cue robustness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.) Phonetically based phonology, 34–57. Cambridge: Cambridge University Press.

39

Stress: Phonotactic and Phonetic Evidence Matthew Gordon

1

Introduction

Stress can be signaled through a number of different acoustic properties, including increased duration, greater intensity, and higher fundamental frequency. Stress may also affect segmental and syllable structure. Typically, stressed syllables trigger qualitative fortition and/or lengthening, whereas unstressed syllables are associated with lenition and/or shortening. To take an example of a stress-driven fortition process affecting syllable structure, Dutch (Booij 1995) inserts an intervocalic glottal stop as an onset to stressed vowels; epenthesis does not interrupt vowel sequences in which the second vowel is unstressed. We thus have pairs such as [’xa.Ds] ‘chaos’ and [a.’?Dr.ta] ‘aorta’, in which the presence of glottal stop is predictable from stress. American English provides well-described cases of lenition in unstressed syllables. For example, post-vocalic coronal stops weaken to taps before unstressed syllabic sounds, e.g. /s/ti/ → [’s/7i] ‘city’. Furthermore, most unstressed vowels reduce to schwa, e.g. [’kan‘tekst] context vs. [kHn’teksŒuHl] contextual, or may delete in certain contexts delete, e.g. [’tme/7oÁ] ~ [’tHme/toÁ] tomato, [’ksændPH] ~ [kH’sændPH] Cassandra. While most segmental effects of metrical structure can be transparently linked to stress, there are others that are not predictable from stress, despite displaying properties typically associated with stress-induced alternations. For example, Nganasan, a Uralic language (Tereshchenko 1979; Helimski 1998; Vaysman 2009), has an alternation between strong and weak intervocalic consonants, termed “consonant gradation,” whereby strong consonants, generally voiceless or prenasalized obstruents, alternate with weak consonants, typically voiced or not prenasalized. The appearance of strong and weak consonants is predictable from syllable count (1). In the onset of even-numbered non-initial syllables, the strong grade appears, while the weak grade appears in the onset of oddnumbered non-initial syllables. Long vowels interrupt the alternating syllable count and, as long as they are not word-initial, are always preceded by weak consonants.

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0039

Stress: Phonotactic and Phonetic Evidence (1)

2

Nganasan consonant gradation (Vaysman 2009: 43) ‘jama’Ïa-tu ‘Ioru’mu-tu su(’ÏH(-Ïu Iu’hu-Ïu

‘his/her/its ‘his/her/its ‘his/her/its ‘his/her/its

animal’ copper’ lung’ mitten’

As Vaysman shows, this pattern is explained if one assumes that words are parsed into binary feet starting at the left edge of words with long vowels forming monosyllabic feet and degenerate feet allowed word finally. Strong consonants occur foot medially and weak consonants occur in foot-initial syllables that are not also word-initial (2). (2)

Nganasan consonant gradation as a reflex of foot structure (Vaysman 2009: 43) (‘jama)(’Ïa-tu) (‘Ioru)(’mu-tu) (su()(’ÏH()-(Ïu) (Iu’hu)-(Ïu)

‘his/her/its ‘his/her/its ‘his/her/its ‘his/her/its

animal’ copper’ lung’ mitten’

The interesting feature of the Nganasan data is that stress does not always fall on syllables predicted to be stressed by the metrical structure diagnosed by consonant gradation. Primary stress in Nganasan falls on the penultimate syllable in all the words in (2), with a secondary stress occurring on initial syllables that are not adjacent to the primary stress. The monosyllabic foot in the last two words is thus completely unstressed, as is the first foot in the penultimate word. This chapter provides a typological overview of the phonetic correlates of stress and the various types of effects of stress and metrical structure on segmentlevel features, exploring how these effects can offer insight into the nature of stress and metrical structure and their formal representation (see chapter 40: the foot, chapter 41: the representation of word stress and chapter 57: quantity-sensitivity for related issues). The structure of the chapter is as follows. §2 examines supra-segmental correlates of stress including the phonetic parameters of duration, fundamental frequency, and intensity. §3 examines segmental alternations conditioned directly by the presence or absence of stress. §4 focuses on the role of foot structure in predicting fortition and lenition of vowels and consonants. §5 addresses segmental changes triggered by foot structures that conflict with metrical constituency as diagnosed by the stress system. §6 explores the role of history in shaping these mismatches between stress and the foot structure relevant for segmental alternations. Finally, §7 summarizes the chapter.

2

Suprasegmental phonetic correlates of stress

Fry (1955, 1958) pioneered research on the acoustic correlates of stress in his examination of the effect of stress in English on duration, intensity, and fundamental frequency. Focusing on the vowels in noun–verb minimal pairs such as ’convert (noun) vs. con’vert (verb) and ’import (noun) vs. im’port (verb), Fry found that stressed vowels were associated with greater duration, greater intensity, and higher

3

Matthew Gordon

fundamental frequency than their unstressed counterparts, with the last of these properties being most reliable as a cue to stress. Since Fry’s work, phoneticians have considerably broadened the typological database on stress correlates by examining other potential correlates of stress and by targeting a diverse set of languages for phonetic study. This research program has yielded many important results. For example, beyond the acoustic domain, stress is also associated with hyperarticulation of segments, which has ramifications for the segmental alternations discussed in §3. Furthermore, other potential acoustic correlates of stress have come to light, such as measurements of stress that are sensitive to spectral tilt (Sluijter and van Heuven 1996a, 1996b) or that integrate intensity over time (Lieberman 1960; M. Beckman 1986). Finally, typological study has shown that many languages are similar to English in using duration, intensity, and/or fundamental frequency to signal stress, e.g. Polish (Jassem et al. 1968), Tagalog (Gonzalez 1970), Mari (Baitschura 1976), Indonesian (Adisasmito-Smith and Cohn 1996), Pirahã (Everett 1998), Aleut (Taff et al. 2001), Chickasaw (Gordon 2004), Turkish (Levi 2005), and Kabardian (Gordon and Applebaum 2010). It has also become increasingly clear that the phonetic study of stress is a complicated matter for several reasons. Languages differ in their relative reliance on different cues to stress where the relevance of certain properties is functionally constrained in many languages by the extent to which potential stress correlates are used to mark phonemic contrasts other than stress. For example, lexical tone languages – e.g. Thai (Potisuk et al. 1996) and Pirahã (Everett 1998) – are less reliant on fundamental frequency to cue stress, and languages with phonemic length contrasts, e.g. Finnish, may have phonetically longer unstressed vowels than stressed vowels. There are also languages in which potential phonetic markers of stress do not converge on a single syllable but rather are shared between multiple, often, though not always, adjacent syllables. For example, in Welsh (Williams 1985) an unstressed final syllable often has higher fundamental frequency and longer vowel duration than an unstressed penultimate syllable in the same word. In such cases, lengthening of the consonant immediately following the stressed vowel seems to be the most reliable correlate of stress. A similar situation arises in Estonian, where the primary stressed initial syllable, if it contains a phonemic short vowel, will be shorter than the immediately following syllable and often have less intensity and lower fundamental frequency (Lehiste 1965; Eek 1975; Gordon 1995). Lengthening of the consonant in the onset of the stressed syllable serves as the most reliable cue to stress in Estonian (Lehiste 1966; Gordon 1997). Hyman (1989) discusses cases in Bantu of different diagnostics leading to different conclusions about the location of stress. For example, certain Eastern and Southern Bantu languages display evidence for metrical prominence on the penultimate syllable, such as vowel lengthening, attraction of high tone, and even phonetic stress. However, these properties may conflict with other properties that suggest stress on another syllable, e.g. high tone on the antepenult in Zulu, even though the penult conditions vowel lengthening. A similar pattern of high tone on the antepenult preceding a stressed penult is found in the Northern Iroquoian language Onondaga (Chafe 1970, 1977; Michelson 1988), the Polynesian language Tongan (Schütz 1985), and several Micronesian languages (Rehg 1993). In “split-cue” stress systems such as those described in this paragraph, determining the location of stress is potentially problematic.

Stress: Phonotactic and Phonetic Evidence

4

The separation of high tone and stress ties in with another problematic issue in the phonetic realization of stress. Since Fry’s work on stress correlates in English, it has become apparent that word-level stress must be distinguished from phrase-level intonational prominence, which is characteristically associated with a prominent fundamental frequency event or “pitch accent” (see also chapter 116: sentential prominence in english). Because a word uttered in isolation constitutes an entire phrase, this means that stress in isolated words is confounded with phrasal pitch accent. Many studies intending to examine correlates of wordlevel stress but targeting words in isolation are thus more accurately regarded as studies of phrasal rather than word-level stress, although there is potentially overlap between the two levels of prominence in their acoustic manifestations. Sluijter and van Heuven (1996a, 1996b) are important acoustic studies that show that a frequency-dependent measure of intensity skewed toward higher frequencies rather than either an overall measure of intensity or a measure of fundamental frequency acts as a reliable predictor of word-level stress disambiguated from phrasal pitch accent in Dutch and English. Unfortunately, cross-linguistic phonetic research aimed at disentangling word-level stress from higher-level pitch accent is in its relative infancy. A related issue involving the relationship between prominence at different prosodic levels is the interplay between phrasal tones falling at or near boundaries of intonational constituents larger than the word but smaller than the domain characterized by pitch accents. In certain languages, such as Korean (Jun 1993) and French (Jun and Fougeron 1995), the prominence traditionally regarded as stress has turned out to be attributed to fundamental frequency peaks assigned by a phrase-level intonational constituent termed an Accentual Phrase. For example, in French, the prominence associated with phrase-final syllables is due to a high tone aligned with the right edge of an Accentual Phrase (Jun and Fougeron 1995). It is conceivable that the stress in many, if not most, languages described as having a phrasal rather than a word-level distribution will turn out to be a tonal property attributed to the intonational system, as in French. Another important issue lurking in the typology of stress correlates is the relationship between the phonetic manifestations of stress and the taxonomy of prominence systems (see also chapter 45: the representation of tone and chapter 42: pitch accent systems). The prototypical stress language possesses a number of characteristics that differentiates it from a tone language, one of these differences lying in the phonetic realization of prominence as primarily a fundamental frequency phenomenon in a tone language but potentially distributed over multiple phonetic parameters – e.g. duration, intensity, and spectral tilt – in a stress language. M. Beckman (1986) demonstrates the phonetic validity of this distinction in her comparative study of English and Japanese, in which she shows that a measure of intensity integrated over time is an important correlate of prominence in English but not in Japanese, a language employing tone-based lexical contrasts. In practice, however, a purely phonetic characterization of prominence is unlikely to yield a perfect dichotomy of languages into prosodic prototypes, especially for languages possessing some traits of a stress system but other traits of an – albeit limited – tone system; these languages are often regarded as having a “pitch accent” system (Hyman 2006).

5

3

Matthew Gordon

The taxonomy of segmental correlates of metrical structure

Fortition and lenition effects associated with metrical structure may be broadly classified into three groups according to the property triggering these segmental alternations. The first type of segmental effect is well documented and involves stress (or lack thereof) directly as a trigger of fortition and/or lenition. A second type of segmental effect is predictable from constituent structure rather than stress, but the constituent structure motivating the segmental change accords with the metrical parse evinced by the stress system. A third type of segmental alternation, exemplified by Nganasan, is linked to metrical constituency, where the foot structure diagnosed by the segmental change is at odds with that suggested by the stress system. In the following sections we take a closer look at examples of each of these types of relationships between segmental properties and metrical structure.

3.1 Stress-driven segmental phenomena: Fortition and lenition Many languages display segmental changes that are conditioned by stress or lack of stress. The typical pattern is for sounds to strengthen in stressed contexts and to weaken in unstressed positions. Fortition and lenition can target either consonants or vowels. In the case of consonants, unstressed position is usually associated with decreased resistance to coarticulatory effects and hypo-articulation (de Jong 1995), resulting in reduced constriction either temporally or in magnitude. Kirchner (2001), Lavoie (2001), Bye and de Lacy (2008), and Vaysman (2009) summarize a number of segmental alternations conditioned by stress, of which I mention a few here (see chapter 66: lenition for an overview of the typology of lenition). Post-vocalic coronal stops in American English reduce to flaps before an unstressed syllabic sound, and stops become aspirated in the onset of stressed syllables. In Kupia (Christmas and Christmas 1975), the stops /p Í/ have lenited variants in the onset of unstressed syllables: /p/ is realized as a fricative and /Í/ as a tap. West Tarangan (Nivens 1992) displays fortition in the onset of stressed syllables: /j/ affricates to /–/, and /w/ occlusivizes to /g/, a change that also applies to word-initial consonants. In the development from Proto-Samurian to pre-Lezgian (Topuria 1974; Giginejshvili 1977; Yu 2004), voiced stops devoiced, a type of fortition, and geminated in the onset of stressed syllables. Stress often also triggers lengthening of consonants. Thus, in Urubu Kaapor (Kakumasu 1986) and optionally in Tukang Besi (Donohue 1999) oral stops lengthen in the onset of primary stressed syllables. Lengthening is also employed as a strategy to beef up the rime of stressed syllables. Hayes (1995: 83) discusses several cases of lengthening in order to enhance the weight of stressed syllables. For example, in Munsee (Goddard 1979), a consonant geminates after metrically prominent short vowels, thereby converting the stressed syllable from light (CV) to heavy (CVC). Vowels are also subject to fortition and lenition processes conditioned by stress. As in the case of consonantal alternations, vowels may be affected either qualitatively or quantitatively by the presence or absence of stress. Cross-linguistically, it is very common for vowels to lengthen in stressed syllables. Hayes (1995: 83)

Stress: Phonotactic and Phonetic Evidence

6

catalogs a number of cases of vowel lengthening under stress, where the lengthening effect is quite substantial, even potentially neutralizing an underlying phonemic length contrast. Vowel lengthening can also be associated with qualitative differences as well. The short low central vowel /ô/ in Kabardian lengthens and lowers to /a(/ under stress in the first syllable of disyllabic nouns and adjectives (Colarusso 1992). Revithiadou (2004) discusses the Livisi dialect of Greek, in which stressed high vowels lower to mid vowels, a shift which increases the sonority of the stressed vowel. Conversely, vowels often shorten and qualitatively reduce in unstressed syllables. Crosswhite (2001, 2004) presents a typology of vowel reduction, in which she broadly classifies reduction into two groups. One type involves centralization of unstressed vowels. For example, most vowels in English reduce to a schwa-like vowel when unstressed. This type of reduction is typically linked to articulatory factors. The smaller duration of unstressed vowels leaves relatively little time for the tongue to reach more peripheral targets in the articulatory space. Crosswhite also points out that this type of reduction has the advantage of reducing the intrinsic prominence of unstressed vowels. The other type of vowel reduction in Crosswhite’s taxonomy involves vowels becoming more, rather than less, peripheral in unstressed syllables. This increase in peripherality can be manifested as either vowel raising or lowering, depending on the language. For example, the phonemic mid vowels /e o/ raise to the high vowels [i u] when unstressed in Luiseño (Munro and Benson 1973), and unstressed /e D/ raise to [e o] in standard Italian (Maiden 1995). In Belarussian, on the other hand, the unstressed mid vowels /e o/ lower to [a] (Kryvitskij and Podluzhni 1994). Crosswhite attributes this superficially less intuitive raising type of reduction to the goal of maximizing the perceptibility of contrasts in unstressed contexts where they are perceptually more vulnerable. Either raising mid to high vowels or lowering mid to low vowels in unstressed syllables creates greater perceptual dispersion of different phonemic vowel qualities in the face of the shorter duration and reduced intensity associated with unstressed vowels. The most extreme manifestation of vowel reduction is deletion (see chapter 68: deletion), which often targets unstressed vowels. It is pervasive in casual speech in English, even creating syllable structures that are otherwise illicit underlyingly, e.g. [’tme/7oÁ] ~ [’tHme/7oÁ] tomato, [’pte/7oÁ] ~ [’pHte/7oÁ] potato. Vowel deletion in certain contexts has the advantage of increasing the weight of stressed syllables. For example, deletion of the medial vowel in [’pæP‘lel] parallel creates a heavy initial stressed syllable, thereby increasing its prominence (see Gordon 2001 and Gouskova 2003 on the relationship between vowel deletion and syllable weight).

3.2 Harmony The increased strength of stressed syllables can also manifest itself in harmony systems. Flemming (1994) identifies three types of harmony patterns that are sensitive to metrical structure. One pattern involves spreading of a feature from a stressed syllable to unstressed syllables. Eastern Mari (Vaysman 2009) provides an example of a language in which harmony propagates from a stressed vowel to a neighboring unstressed vowel. Stress in Eastern Mari falls on the rightmost full, i.e. non-schwa, vowel in monomorphemic roots (3a) and otherwise on the

Matthew Gordon

7

initial syllable in non-derived words containing only reduced vowels underlyingly (3b). (We abstract away from cases where an underlying schwa alternates with a full vowel on the surface.) (3)

Eastern Mari stress (Vaysman 2009: 62–64) a.

b.

koI’ga œer’ge køgør’Œen ’teIgHz ’olHk ’joIHlHœ pu’œaIgH ’ßHJHr ’HœkHl ’lHßH

‘oven’ ‘comb’ ‘dove’ ‘sea’ ‘meadow’ ‘mistake’ ‘tree’ ‘canvas’ ‘step’ ‘butterfly’

Rounding harmony is propagated rightward from the stressed vowel in a word. Thus, the 3rd person possessive suffix surfaces as [œe] when the stressed vowel is unrounded (4a) but as [œø] or [œo] when the stressed vowel is a rounded vowel (4b). (The backness of the rounded vowel is conditioned by a process of front–back harmony.) (4)

Eastern Mari rounding harmony (Vaysman 2009: 92) a.

b.

’ergHœe y’remHœe pykœer’me-œe ’œyrH-œø kHO’mo-œø ’œoœH-œo

‘his/her/its ‘his/her/its ‘his/her/its ‘his/her/its ‘his/her/its ‘his/her/its

boy’ street’ walnut tree’ soup’ shovel’ spring’

Another type of harmony that is sensitive to stress involves the propagation of a feature from an unstressed syllable up through a stressed syllable, which blocks further spreading of the harmonizing feature. Tudanca Spanish (Penny 1978) instantiates this type of harmony. In Tudanca, underlying final high vowels, which surface as more centralized than their non-final counterparts, induce centralization (in the front/back and/or height dimension depending on the vowel) of preceding vowels up to and including the stressed vowel (5). (Centralized vowels are marked by the R diacritic.) Stress is lexically governed and may fall on either the penultimate or antepenultimate syllable. (5)

Tudanca laxness harmony (Penny 1978: 54 –55) ’pÇntÑ ’sÅkÑ ’pÉrtÇkÑ ’pÑlpÇtÑ anti’gwÇsÇmÑ o’rÅgÄnÑ

‘male calf’ ‘dry (masc)’ ‘portico’ ‘pulpit’ ‘very old’ ‘oregano’

cf. ’pinta cf. ’seka

‘female calf’ ‘dry (fem)’

Stress: Phonotactic and Phonetic Evidence

8

Walker (2004, 2005) discusses several cases of metaphony in Romance languages involving harmonizing in height of a stressed vowel to a posttonic one (see also chapter 110: metaphony in romance for an overview of similar processes in Italian dialects). In some language varieties, as in Tudanca Spanish, harmony propagates left-wards from the triggering vowel to the stressed vowel through any intervening unstressed vowels. In other varieties, e.g. Asturian Lena Bable (Hualde 1989, 1998), the stressed vowel is transparent to the harmonizing feature, which propagates past the stressed vowel leftward to the pre-tonic vowel.

3.3 Exceptional lenition in prominent syllables Despite the cross-linguistic tendency for stressed syllables to be associated with increased segmental strength, this pattern is not universal. Mok:a Mordvin (Vaysman 2009) optionally lenites consonants in the onset of stressed syllables that are not word initial. Stress in Mok:a is sensitive to a distinction between the low-sonority vowels [i u H] and the high-sonority vowels [a æ o e]. In words containing vowels belonging to the same sonority class, stress falls on the first syllable of a word (6a). In words consisting vowels of different sonority classes, stress falls on the leftmost vowel belonging to the higher-sonority group (6b). (6)

Mok:a Mordvin a. ’b j HnHŒ ’mHkur ’kuœin ’juÚH ’aka ’lopæ ’pango ’s j eja b. ts j H’ræ vi’na Hz’na s j iãHk-’ka

stress (Vaysman 2009: 135–137) ‘boat’ ‘buttocks’ ‘jug’ ‘skin’ ‘older sister, aunt’ ‘leaf’ ‘mushroom’ ‘goat’ ‘son’ ‘alcohol’ ‘older sister’s husband’ ‘elm (prolative)’

Lenition in stressed onsets entails voicing of underlying voiceless obstruents, liquids, and glides, the spirantization of underlying voiced stops, the conversion of /m/ to /w/ and the deletion of /n/, with concomitant nasalization of the stressed vowel. Crucially, lenition does not target word-initial consonants, as the examples in (7) indicate. (7) Mok:a Mordvin lenition in stressed onsets (Vaysman 2009: 142–143) kur’ka ~ kur’ga ‘turkey’ j j s Hr’pe ~ ’s Hr’be ‘heart’ j j j j b Hn H’Œ-oze ~ b Hn H’–-oze ‘my boat’ pHjH’l j-oze ~ pHjH’l joze ‘my knife’ bu’jæ-ze ~ bu’jæ-ze ‘my end’ t jH’b j e-ze ~ t j e’ß j e-ze ‘my work’ pin’gæ-ze ~ pin’:æ-ze ‘my period of time’ p j i’ma ~ p j i’wa ‘large cup, mug’ kH’nak-oze ~ kH’ãk-oze ‘my guest’

9

Matthew Gordon

3.4 The phonetic basis for fortition and lenition in stressed syllables The tendency for consonants to undergo fortition in the onset of stressed syllables finds an explanation in terms of speech perception (see chapter 98: speech perception and phonology), as suggested by J. Smith (2000, 2004) and Gordon (2005). In the Smith and Gordon accounts, two key independently known aspects of auditory processing underlie the relationship between sonority and stress. First, the auditory system is most attuned to the beginning of a stimulus before a gradual decline in sensitivity sets in. This reduction in sensitivity, termed “adaptation,” has both physical and psycho-acoustic manifestations, including a reduction in auditory nerve firing rates (Delgutte 1982) and a lessening of perceived loudness (Plomp 1964; Wilson 1970; Viemeister 1980). On the other hand, after a phase of silence or reduced acoustic intensity, auditory nerve firing rates and perceived loud-ness increase during an immediately following sound characterized by greater intensity, reflecting a process of auditory “recovery” (R. Smith 1979; Viemeister 1980; Delgutte 1982, 1997; Delgutte and Kiang 1984). Given the auditory benefit afforded by a period of reduced intensity or silence, it is easy to see the perceptual advantage of fortition. By either devoicing a consonant in the onset of a stressed syllable or increasing its degree of constriction, the immediately following stressed rime – including the vowel and coda consonants, if any are present – receives a boost in auditory prominence. Under this auditorily driven account of fortition, prominence is not enhanced directly, by the increased strength of the onset consonant, but rather indirectly, through its effect on the following rime. A limitation of this approach is its apparent failure in predicting lenition in the onset of stressed syllables, as optionally occurs in Mok:a Mordvin. Since lenition increases the sonority, and thus the acoustic intensity, it would be expected under the Smith and Gordon account to actually decrease the auditory prominence of the immediately following rime by reducing the positive effects of recovery. The reconciliation of lenition with a phonetically driven approach to stress-sensitive fortition must await further research.

3.5 The formal analysis of stress-driven alternations in segments and syllables 3.5.1 Positional faithfulness The analysis of stress-driven effects on segmental properties and syllable structure has attracted substantial attention in the phonology literature, in particular within the optimality-theoretic framework. There are two basic approaches to stressinduced segmental phenomena. One approach (e.g. Casali 1997; Steriade 1997; J. Beckman 1998; Lombardi 2001) focuses on cases in which underlying contrasts are asymmetrically preserved in stressed syllables (as well as other prominent positions) but lost in unstressed contexts. In this type of analysis, termed “positional faithfulness,” input–output faithfulness constraints requiring preservation of an underlying property are divided into two classes: those that are context free and those that are only enforced in prosodically privileged positions such as stressed syllables. Positional faithfulness arises when a markedness constraint is prioritized above a generic faithfulness constraint but below a positional faithfulness constraint.

Stress: Phonotactic and Phonetic Evidence

10

This constraint interaction can be illustrated by considering the analysis of Guaraní vowel harmony developed in J. Beckman (1998). In Guaraní (Gregores and Suárez 1967), nasalized and oral vowels contrast in stressed syllables, but in unstressed syllables nasalized vowels may only surface before a nasal consonant. The [nasal] feature also spreads leftward from a prenasalized stop and from a phonemic nasalized vowel up to but not including a stressed vowel (8). Nasalization additionally spreads rightward (as the examples below indicate) although its phonetic properties are different, which has led certain researchers, e.g. Flemming (1994), to analyze it as phonetic rather than phonological. Beckman thus does not develop an analysis of rightward spreading of nasality. (8)

Guaraní nasal harmony (Gregores and Suárez 1967: 69) /amaa’porõre’ju/ → ?ãm ba?a’porõrgju ‘if I work you come’ /je’ qntena/ → je’qntgnã ‘just once more!’ /ija‘kãra’ku/ → ?hnã‘kãrã’ku ‘is hot-headed’ /rojotopa’pamarõro’xova‘rã/ → rojotopa’pamãXõXõ’xoºã‘Xã ‘if now we meet all of us, we will have to go’

Two faithfulness constraints play a pivotal role in Beckman’s analysis. First, a generic faithfulness constraint, Ident(nasal), requires that segments underlyingly associated with a [nasal] feature preserve that feature on the surface. The second constraint is the positionally defined analog to Ident(nasal), Ident-Ä(nasal), which requires that surface segments in stressed syllables preserve their underlying [nasal] specification. The existence of contrastive nasality on stressed vowels but not on unstressed vowels follows from the ranking of a markedness constraint banning nasalized vowels, *Vnasal, above generic Ident(nasal) but below positionspecific Ident-Ä(nasal). This ranking ensures that any underlyingly nasalized vowel will lose its nasality if it surfaces in an unstressed position. Critical to the analysis of nasal harmony in Beckman’s analysis is an alignment constraint, Align-L(nasal), requiring that all instances of the feature [nasal] be aligned with the left edge of a word. This constraint is honored in forms in which nasality either is underlyingly associated with a segment in the first syllable or has propagated to the first syllable through nasal spreading. One violation is incurred for each segment intervening between a nasal feature and the left edge of the word. By sandwiching Align-L(nasal) above Ident (nasal) but below Ident-Ä(nasal), nasality is correctly predicted to spread as far leftward as the stressed syllable, where it is blocked from spreading any further (9). (9)

/je’qntenã/ Ident-Ä(nas) Align-L(nas) Ident(nas) ‘just once more’ ☞ a. je’qntgnã b. ng’0ntgna c. je’qntenã

*** *!

* ***

***, *!*****

Lenition in unstressed syllables can be handled similarly in the positional faithfulness approach. A constraint banning non-lenited segments is ranked above a generic faithfulness constraint, but below a positional faithfulness constraint

11

Matthew Gordon

requiring featural identity between underlying forms and their surface correspondents in stressed syllables. The result is a typologically common type of system; cf. many of the vowel reduction patterns described by Crosswhite (2001, 2004), in which a contrast between lenited and non-lenited segments in stressed syllables is neutralized to the lenited variant in unstressed syllables.

3.5.2 Positional markedness Another approach to stress-induced segmental alternations employs positional markedness constraints (Zoll 1998; J. Smith 2000, 2004; de Lacy 2001; see also the positional licensing constraints of Walker 2004, 2005), targeting stressed and other prominent positions. A positional markedness approach has an advantage over the positional faithfulness approach in capturing fortition phenomena, since it explicitly predicts that faithfulness will be violated in strong positions. To illustrate the positional markedness approach, let us consider J. Smith’s (2000, 2004) analysis of glottal stop epenthesis before vowel-initial stressed syllables in Dutch. An anti-epenthesis constraint, Dep, is ranked above a generic markedness constraint requiring that syllables have an onset, Onset, but below a positional markedness constraint, Onset-Ä, mandating that stressed syllables have an onset. The result is insertion of a glottal stop before an otherwise onsetless stressed syllable but not before an unstressed syllable beginning with a vowel (10). (10)

/aDrta/ ‘aorta’ Onset-Ä Dep Onset ☞ a. a.’?Dr.ta b. a.’Dr.ta

* *!

* **

/xaDs/ ‘chaos’ ☞ a. ’xa.Ds b. ’xa.?Ds

* *!

The common process of stressed vowel lengthening is also standardly attributed to a positional markedness constraint requiring that stressed syllables be heavy, the Stress-to-Weight Principle (Crosswhite 1998), or its counterpart specific to primary stress, Main-to-Weight (Bye and de Lacy 2008). By ranking the Stressto-Weight Principle above the constraint banning surface moras that do not have a correspondent in the underlying representation, Dep-[, lengthening of the stressed syllable results. Whether the lengthening targets a vowel or consonant is attributed to the relative ranking of faithfulness constraints pertaining to the two types of segments. The blocking of lenition in stressed syllables can also be captured by assuming that the constraint driving lenition – e.g. Kirchner’s (2001, 2004) Lazy constraint banning articulatory effort – outranks a generic faithfulness constraint, but is ranked below a constraint blocking the type of segment resulting from lenition from surfacing in stressed contexts. For example, the lenition of coronal stops to flaps in post-tonic contexts in English can be attributed to a constraint banning the amount of articulatory effort required to produce a full-fledged stop in intervocalic position. This constraint is ranked above a generic faithfulness constraint requiring preservation of the features characterizing a stop, but below a constraint banning flaps in stressed syllables.

Stress: Phonotactic and Phonetic Evidence

12

Likewise, the blocking of feature spreading by stressed syllables, e.g. nasality in Guaraní, can be handled by assuming a positional markedness constraint banning that feature in stressed syllables. While this type of analysis is equipped to handle the harmony process, underlying featural restrictions holding of unstressed syllables are difficult to treat formally. For example, the ban on contrastive nasalized vowels in lexically unstressed syllables in Guaraní does not follow from the same markedness constraint blocking nasal spreading onto stressed syllables. Cases of lenition in stressed syllables, e.g. in Mok:a Mordvin, are problematic for principles of markedness in general, independent of the paradigm in which they are couched, since such processes contradict the scales of segmental strength observed in most languages (see chapter 4: markedness). It is possible to posit a positional markedness constraint requiring that stressed syllables be associated with lenited segments, but such a constraint would undermine the restrictiveness of the theory by admitting both lenition and fortition constraints referring to the same context. The proper formal treatment of apparently “unnatural” phenomena such as lenition in stressed syllables hinges on the understanding of the motivating factors driving them. It is conceivable that phonetic considerations offer an explanation for lenition in stressed syllables, but it is also possible that a confluence of historic events have conspired to produce a phonology whose synchronic naturalness has been rendered opaque (see §6).

4

Metrically harmonic constituency-driven segmental alternations

4.1 Foot-initial fortition A second type of effect of metrical structure on segmental properties requires reference to feet for an adequate characterization, but these feet correspond to the feet diagnosed by stress placement. One subclass of these effects involves foot-level manifestations of phonetic patterns characteristic of prosodic words and larger constituents. For example, the Alutiiq dialects of Yupik (Leer 1985) have a process of fortition affecting consonants in foot-initial position. Leer (1985: 84–85) describes: two major distinguishing characteristics of the fortis consonant: complete lack of voicing with voiceless consonants (stops and voiceless fricatives), and preclosure [which is a voiceless interval following the preceding segment], during which the mouth also assumes the configuration of the [fortis] consonant.

Leer further observes that the phonetic distinction between fortis and non-fortis consonants is most salient in the Prince William Sound dialect of Chugach Alutiiq, “where lenis stops may become so loosely articulated as to sound virtually like voiced fricatives,” which themselves also either weaken to an approximant or are deleted, thereby avoiding neutralization. Feet in Alutiiq are iambic, with long vowels and word-initial (but not word-medial) CVC counting as heavy. A key feature of stress, however, that has implications for footing and by extension the analysis of fortition is that stress adheres to a ternary

13

Matthew Gordon

pattern where there are consecutive light syllables word medially between the first and last foot of a word (see chapter 52: ternary rhythm). For example, stress (indicated here uniformly as primary stress, since it is unclear which stress is most prominent) falls on the first and fourth syllables in the four-syllable Chugach Alutiiq word /’anŒiqu’kut/ ‘we’ll go out’ (Leer 1985: 84). The strengthened consonant (in bold) occurs in the onset of the pretonic syllable, which is analyzed by Hayes (1995) as the first syllable in a disyllabic foot comprising the pretonic and the tonic syllable. Under his approach, which assumes that a post-tonic light syllable is skipped over in the metrical parse (a phenomenon termed “weak local parsing” by Hayes) in languages with ternary stress intervals, /’anŒiqu’kut/ is footed as (’an)Œi(qu’kut). A virtue of this analysis is that fortition in Alutiiq can be characterized as a foot-initial phenomenon, thereby bringing it into line with the general cross-linguistic pattern of strengthening associated with initial position of prosodic domains (e.g. Pierrehumbert and Talkin 1992; Byrd 1994; Dilley et al. 1996; Cho and Keating 2001; Keating et al. 2003). In contrast, the locus of fortition is not easily defined with reference only to stress; it would need to be described as fortition in the onset of pretonic syllables, which would not appear to be a natural phenomenon with parallels in other languages. chapter 41: the representation of word stress describes several other segmental alternations that require reference to foot structure as opposed to stress for an adequate characterization.

4.2 Foot-final lengthening The parallel between the foot and larger prosodic domains has also been proposed to extend to other phonetic properties. In his analysis of consonant gradation in Finnic languages, Gordon (1998) proposes that a phenomenon of foot-final lengthening accounts for certain lengthening phenomena found in the language family. According to Gordon’s account, foot-final lengthening played a crucial role in the proto-language in triggering consonant gradation. Gordon argues that vowels in absolute foot-final position, i.e. in foot-final open syllables, lengthened in the proto-language, thereby disrupting the prosodic profile of the trochaic feet characteristic of the proto-language. This disruption ultimately triggered lengthening in the consonant immediately preceding the lengthened foot-final vowel. Gordon’s analysis can be seen most clearly when cast in moraic terms. Assuming consonants were non-moraic and there were no phonemic long vowels in the proto-language (see Gordon 1998), the lengthening of a foot-final vowel created a light–heavy trochaic foot, a prosodically non-optimal foot in which the unstressed syllable is heavier than the stressed syllable. Lengthening the consonant preceding the lengthened foot-final vowel may thus be viewed as a foot optimization process, beefing up the light stressed syllable in a trochee. The result was a chain shift whereby single consonants became geminates and geminates became overlong before foot-final vowels (CVCV → CVC(V( and CVC(V → CVC((V(, but CVCVC → CVCVC and CVC(VC → CVC(VC), thus setting in motion the paradigmatic consonant gradation alternations that still exist, albeit in modified form, in many languages in the Finnic and Sámi language families. The reconstructed process of foot-final lengthening finds support from analogous phenomena found in certain modern Finnic languages, such as Ingrian. Broader cross-linguistic evidence for foot-final lengthening comes from a pattern reported by Teranishi (1980) of lengthening targeting the second vowel in a disyllabic sequence, analyzed by Poser

Stress: Phonotactic and Phonetic Evidence

14

(1990) as a foot, in recited Japanese verse. Finally, foot-final lengthening may be viewed as the foot-level manifestation of the well-attested phenomenon of final lengthening observed at levels above the foot, such as the word and phrase (Wightman et al. 1992). Foot-final lengthening has also been appealed to by Revithiadou (2004) as a factor in promoting the cross-linguistically pervasive phenomenon of iambic lengthening discussed in the next section.

4.3 Iambic/trochaic length asymmetries Certain asymmetries in quantitative alternations are also best explained with reference to foot structure rather than directly to stress. In particular, stressed syllables in iambic and trochaic feet appear to display fundamentally different length characteristics (see Hayes 1985, 1995 and chapter 44: the iambic–trochaic law). Stressed syllables in languages employing an iambic parse are often lengthened cross-linguistically, whereas those with trochaic feet characteristically either fail to lengthen the stressed syllable, or, in some cases, even shorten the stressed syllable. Chickasaw (Munro and Ulrich 1984; Munro and Willmond 1994, 2005; Munro 1996, 2005; Gordon et al. 2000; Gordon 2003, 2004; Gordon and Munro 2007) is an iambic language in which closed syllables and syllables containing long vowels are heavy, i.e. are parsed as monosyllabic feet word-initially or following a stressed syllable. The final syllable is also parsed as a foot even if it is light (CV). The rightmost stress, i.e. the one on the final syllable, is the primary one, except that a long (or lengthened) vowel in pre-final position attracts the primary stress from a final CV(C). As shown in (11), stressed vowels in open non-final syllables substantially lengthen, where the output of lengthening is a vowel that is either nearly neutralized or completely neutralized in length with a phonemic long vowel depending on the vowel and the speaker (see Gordon et al. 2000 for phonetic duration results). (11)

Iambic lengthening in Chickasaw (lengthened vowels indicated by )) (Œi‘pi))(sa’li))(‘tok) (Œi‘Œo))(‘koœ)(ko’mo))(‘Œi) (Œi‘ki))(si’li))(‘tok) (a‘sa))(bi’ka))(‘tok)

‘I looked at you.’ cf. (pi’sa))(li‘tok) ‘He looked at you.’ ‘He makes you play.’ cf. (Œo‘koœ)(ko’mo))(‘tok) ‘He played.’ ‘He bit you.’ cf. (ki’si))(li‘tok) ‘He bit it.’ ‘I was sick.’ cf. (a’bi))(ka‘tok) ‘He was sick.’

The process of iambic lengthening has an intuitive purpose, in that it enhances the prominence of the stressed syllable. In some languages, the beefing up of the stressed syllable in an iambic foot is achieved by lengthening a consonant rather than a vowel (see Hayes 1995: 82–83 for a survey of iambic lengthening; but see Bye and de Lacy 2008 for re-analyses of iambic consonant lengthening). This consonant can either be the coda consonant in a stressed CVC syllable, as in the Chevak dialect of Central Alaskan Yupik (Woodbury 1981) or the onset of the following syllable, the first half of which ends up closing the stressed syllable, as in Delaware (Goddard 1979, 1982).

15

Matthew Gordon

Interestingly, substantial lengthening of stressed syllables in trochaic feet appears to be less sparsely attested, and those cases that do exist differ in certain important respects from their iambic counterparts (Hayes 1985, 1995; Mellander 2001). Cases of trochaic lengthening are found in languages with quantity-insensitive stress systems unlike cases of iambic lengthening, which are limited to languages with quantity-sensitive stress systems. Particularly striking is the existence of trochaic languages that shorten stressed syllables. For example, in Fijian (Schütz 1985) a trochaic foot is constructed at the right edge of a word. Phrases that underlyingly end in a heavy penult (one containing a long vowel) followed by a light ultima shorten the long vowel in the penult; thus, underlying /mbu(Igu/ ‘my grandmother’ surfaces as (’mbuIgu) (Schütz 1985: 528). Mellander (2001) finds that languages with trochaic shortening differ from those reported to display trochaic lengthening in being quantity-sensitive.

5

Segmental alternations predicted by foot structure rather than stress

5.1 Nganasan consonant gradation In the introduction, we briefly examined the consonant alternations in Nganasan (Tereshchenko 1979; Helimski 1998; Vaysman 2009) that were shown to be predictable from foot structure but not from stress. Stress is confined to a threesyllable window at the right edge of a word, falling on the final syllable if it contains a long vowel or diphthong (12a) and otherwise on either the penult or antepenult depending on vowel quality in those syllables. If the penult contains a vowel other than the central vowels [q H], stress falls on the penult (12b). If the penult contains a central vowel and the antepenult contains a non-central vowel, stress falls on the antepenult (12c). Cases involving a central vowel in both the penult and the antepenult display a more complex and variable pattern that will be discussed along with secondary stress below. (12)

Nganasan stress (Vaysman 2009: 23, 52) a. b. c.

ky’ma( le’hua ba’kunu ’kaÏar ku’butHnu ’hwa(tHnu

‘knife’ ‘board’ ‘salmon’ ‘light’ ‘skin, fur (loc sg non-poss)’ ‘tree (loc sg non-poss)’

Consonant lenition is predictable if one assumes a left-to-right parse into binary feet, with stray syllables being parsed into a monosyllabic foot: lenition targets foot-initial consonants that are not also word-initial. Lenition entails a number of different alternations including the loss of prenasalization (/I h/ → [h], /n t/ → /t/, /Ik/ → [k], /ns/ → [s], /Jç/ → [ç], /Jc/ → [c]), voicing with or without a change in manner (/t/ → [Ï], /k/ → [g], /s/ → [Á], /ç/ → [Á], /c/ → [Á]), and a shift from /h/ to [b] and from /j/ to zero. Data illustrating the consonant alternations for two suffixes and their relationship to foot structure appear in (13).

Stress: Phonotactic and Phonetic Evidence (13)

16

Nganasan consonant alternation and foot structure (Vaysman 2009: 43, 52) (‘jama)(’Ïa-tu) (‘Ioru)(’mu-tu) (su()(’ÏH()-(Ïu) (Iu’hu)-(Ïu) (ba(r)(’pH-n tH)(nu) (hqa)(’ŒH-n tH)(nu) (ku’bu)(tHnu) (’hwa()(tHnu)

‘his/her/its animal’ ‘his/her/its copper’ ‘his/her/its lung’ ‘his/her/its mitten’ ‘master, chief (loc sg non-poss)’ ‘thumb (loc sg non-poss)’ ‘skin, fur (loc sg non-poss)’ ‘tree (loc sg non-poss)’

As these examples show, the strong variant of the 3rd person possessive suffix (beginning with [t]) and the locative singular non-possessive suffix (beginning with a prenasalized [n t]) surface foot medially, whereas the weak allophone (beginning with [Ï] and plain [t] in the two suffixes, respectively) occurs foot initially. The strong grade and the weak grade both occur in the onset of unstressed syllables, meaning that stress does not predict the alternation. Furthermore, the foot structure diagnosed by the consonant alternations does not accord with the foot structure that would be required to predict primary stress. It is not the case, however, that stress assignment in Nganasan is completely blind to foot structure. Secondary stress falls on odd-numbered syllables counting from the left edge of a word in keeping with the footing predicted by consonant gradation (14a). Two provisions to this generalization, however, make the relationship even between secondary stress and foot structure opaque. First, secondary stress may not clash with an immediately following stress (14b) and, second, secondary stress skips over a light (CV) syllable in favor of a heavy (CVV) syllable (14c). In both situations, a final syllable is potentially footed but unstressed, and a word may display a mix of iambic and trochaic feet. (14)

Nganasan secondary stress (Vaysman 2009: 24) a. b. c.

(‘baku)(‘numH)(’numH) (‘tiri)(‘mimH)(’numH) (‘kaÏar)(mH’nu)(mH) (‘ŒemJi)(mH’nu)(mH) (ky)(‘ma()(mH’nu)(mH) (le)(‘hua)(mH’nu)(mH)

‘my ‘my ‘my ‘my ‘my ‘my

salmon (prolative)’ caviar (prolative)’ light (prolative)’ salary (prolative)’ knife (prolative)’ board (prolative)’

A further context in which foot structure is relevant to stress arises in words in which both the penult and the antepenult contain a central vowel. We abstract away from cases in which the penult and antepenult contain different central vowels, a situation that gives rise to variability in the location of stress, and consider here only cases in which both the penultimate and antepenultimate syllables contain the same central vowel. In words of this profile, the penult attracts stress if it is footinitial (15a) but the antepenult carries the stress if the penult is foot-final (15b). (15)

Nganasan stress in words with the same central vowel in the penult and antepenult (Vaysman 2009: 36) a.

(‘bqÏq)(’tqrH) (‘Jqlq)(’tqrH)

‘you (sg) are drinking (intr)’ ‘you (sg) are living’

17

Matthew Gordon b.

(’bqnq)(mH) (‘bqÏqp)(’tqtq)(rH)

‘my rope’ ‘you (sg) are drinking (tr)’

The location of stress in these words is consistent with an analysis assuming a left-to-right parse into trochaic foot in keeping with the foot structure diagnosed by consonant gradation and secondary stress. As in the case of secondary stress assignment in many words, however, a stray final light syllable is predicted to be footed by consonant gradation but is unstressed if the penult and antepenult together form a foot.

5.2 Eastern Mari schwa fortition Vaysman (2009) presents another case of a foot-driven segmental alternation that is not predictable from stress in Eastern Mari, another Uralic language. As discussed in §3.2, stress in Eastern Mari falls on the rightmost full, i.e. non-schwa, vowel in monomorphemic roots (16a) and otherwise on the initial syllable in non-derived words containing only reduced vowels underlyingly (16b). (16)

Eastern Mari stress (Vaysman 2009: 62–64) a.

b.

koI’ga œer’ge køgør’Œen ’teIgHz ’olHk ’joIHlHœ pu’œaIgH ’ßHJHr ’HœkHl ’lHßH

‘oven’ ‘comb’ ‘dove’ ‘sea’ ‘meadow’ ‘mistake’ ‘tree’ ‘canvas’ ‘step’ ‘butterfly’

In addition to rounding harmony (see §3.2), Eastern Mari also has an alternation between reduced and full vowels. Underlying reduced vowels surface as full vowels in absolute word-final position, i.e. not in final closed syllables, under conditions that Vaysman argues are metrically governed. Schwa only strengthens to a full vowel in final position of a foot, where feet are parsed from left to right and a stray syllable left at the end of the word remains unparsed. Thus, unlike Nganasan, which allows degenerate feet at the end of the word, Eastern Mari does not. The quality of the full vowel resulting from fortition is determined by both front/back and rounding harmony. The examples in (17) illustrate the fortition of schwa foot finally in the suffix meaning ‘the one who/ that is’ (17a), and the preservation of schwa in metrically unparsed final syllables (17b). (17) Eastern Mari vowel fortition (Vaysman 2009: 83) a.

(’taI-se) (’my-sø) (’joIH)(lHœ-so) (pu’œaI)(gH-so)

‘the ‘the ‘the ‘the

one one one one

who is a friend’ that is honey’ that is a mistake’ that is a tree’

Stress: Phonotactic and Phonetic Evidence b.

(’jeIH)-sH (ok’sa)-sH (o’la)-sH (puœaI)(gH-’na)-sH (pare)(IH-’na)-sH

‘the ‘the ‘the ‘the ‘the

one one one one one

18

who is human’ that is money’ that is a city’ that is our tree’ that is our potato’

That we are dealing with fortition of schwa rather than lenition of full vowels to schwa is demonstrated by the patterning of alternating schwas as light syllables in the stress system. Thus, vowels that are underlyingly schwa pass stress on to a preceding full vowel even if they wind up as full vowels due to foot-final fortition (18). (18)

Eastern Mari rejection of stress by alternating schwa (Vaysman 2009: 79) ’susko ’muno ’jeIe

‘scoop’ ‘egg’ ‘human’

*sus’ko *mu’no *je’Ie

cf. ’suskHm ‘scoop (acc sg)’ cf. ’munH ‘egg (acc sg)’ cf. ’jeIHÚ ‘human (lat sg)’

The Nganasan consonant alternations and the Eastern Mari vowel alternations present problems both for theories that attempt to link fortition and lenition directly to stress and for theories that analyze stress patterns without recourse to foot structure (e.g. Prince 1983; Selkirk 1984; Gordon 2002). On the other hand, stress cannot be predicted from the type of metrical structure needed to account for segmental alternations in these languages.

6

The role of history in foot-driven segmental alternations

The mismatch between stress and metrical structure diagnosed through segmental alternations raises the question of whether such cases are the result of a confluence of historical events that have removed an originally transparent link between stress and the alternations. The literature contains a number of cases of seemingly unprincipled synchronic patterns being veiled by later diachronic changes that have obscured a process that at one time was clearly phonetically grounded (e.g. Buckley 2000; Hyman 2001; Yu 2004). In fact, there is evidence that the cases of disharmony between stress and segmental alternations in the Uralic languages Eastern Mari and Nganasan are the result of diachronic changes. Let us first consider the case of Eastern Mari schwa fortition (§5.2). Recall that this alternation involved the strengthening of wordfinal schwa in absolute foot-final position, e.g. /puœaIgH-sH/ → (pu’œaI)(gH-so) ‘the one that is a tree’ but (’jeIH)-sH ‘the one who is human’. As we have seen, this alternation falls out from an analysis positing binary feet constructed from left to right. These feet, however, do not accord with the stress system, which places a single stress on the rightmost full, i.e. non-schwa, vowel in monomorphemic roots and otherwise on the initial syllable in non-derived words containing only reduced vowels. It is likely the case that the binary feet evinced through schwa fortition at one time corresponded to those diagnosed through stress. The dominant stress pattern

19

Matthew Gordon

in the Uralic language family, of which Mari is a member, involves stress on the initial syllable of a word, with many languages displaying a binary trochaic pattern from left to right. In keeping with the synchronic evidence, Sammallahti (1988) thus reconstructs a binary trochaic stress system for the proto-language. Bereczki (1988) reconstructs initial stress for pre-Mari on the basis of vowel reduction shifts occurring in non-initial syllables. Initial stress is also the default pattern for Eastern Mari words with all light syllables. It is plausible that the binary pattern reconstructed for Proto-Uralic existed through the period during which schwa fortition developed as a productive alternation in Eastern Mari. Assuming this to be the case, the foot structure required to account for fortition was transparently linked to stress at one time. Furthermore, schwa strengthening becomes a phonetically motivated phenomenon driven by the same process of foot-final lengthening claimed by Gordon (1998) to drive consonant gradation in Proto-Saamic-Fennic (§4.2) and by Revithiadou (2004) to play a role in iambic lengthening cross-linguistically (§4.3). Schwa is phonetically much shorter than its non-reduced counterparts in Mari (Gruzov 1960; Vaysman 2009) and, more generally, in other languages with a weight distinction between central and peripheral vowels (Gordon 2002). Assuming that foot-final vowels are lengthened in Eastern Mari, the conversion from schwa to a peripheral vowel foot-finally plausibly follows from the incompatibility of schwa with increased duration. The reason this lengthening is likely confined to foot-final vowels that are also word final is due to the tendency for an enhanced lengthening effect in word-final position, which is independently predisposed to lengthening crosslinguistically. Support for this analysis comes from languages such as Yupik (Reed et al. 1977) and Javanese (Horne 1974), where underlying schwa in word-final position shifts to a peripheral vowel quality on the surface (see also chapter 26: schwa). It is also possible to pursue a diachronic approach to the mismatch between stress and the foot structure required to account for rhythmic consonant gradation in Nganasan (§5.1), although the complexity of the data complicates the discussion. Recall that consonants in foot-initial position are lenited in Nganasan, e.g. foot-initial (ku’bu)(tHnu) vs. foot-medial (hqa)(’ŒH-n tH)(nu), where footing adheres to a trochaic pattern from left to right. Stress, on the other hand, is limited to a three-syllable window at the right edge of a word. As mentioned above, a trochaic stress system is reconstructed for Proto-Uralic. Indeed this basic pattern is preserved in the placement of secondary stress in Nganasan itself subject to interruptions in the alternating count due to heavy syllables. If one assumes that a trochaic stress pattern existed at the time rhythmic gradation became entrenched as a process, the link between gradation and foot structure diagnosed through stress becomes transparent. Under this analysis one puzzling aspect of consonant gradation remains, however: the fact that segments in foot-initial, i.e. historically stressed, syllables undergo lenition rather than fortition. Regardless of the historical origins of mismatches between stress and the metrical structure driving segmental alternations, a synchronic analysis must admit the possibility of orthogonal representations of stress and foot structure. Furthermore, the data from languages like Nganasan and Eastern Mari suggest that stressless feet are a possibility even if typologically rare (see Hyde 2002 for a recent constraint-based theory explicitly allowing for stressless feet).

Stress: Phonotactic and Phonetic Evidence

7

20

Summary

Stress or lack of stress is associated with both suprasegmental and segmental properties. On a suprasegmental level, stress typically triggers lengthening, higher fundamental frequency, and greater intensity, although there are many languages in which these properties do not converge on a single syllable but rather are distributed over multiple syllables. On a segmental level, stress characteristically, although not always, triggers consonant fortition or blocks lenition targeting unstressed syllables. Some metrically conditioned segmental alternations, on the other hand, are better explained with reference to foot structure. For example, boundary-driven processes such as foot-initial fortition and foot-final lengthening are plausibly the foot-level analogs of well-documented phenomena applying at the word level. Furthermore, stressed vowel lengthening in many languages is explicable in terms of stress, but its cross-linguistic bias toward applying in iambic stress systems suggests that it is sensitive to foot structure. A final type of segmental alternation cannot be accounted for with reference to stress but rather suggests the relevance of foot structure that is orthogonal to the stress system in certain languages. Examination of historical data potentially provides insight into these mismatches between foot structure diagnosed by fortition and lenition and foot structure diagnosed by stress by showing that the segmental changes became entrenched at a chronologically earlier stage, when stress and foot structure coincided.

ACKNOWLEDGMENTS The author wishes to thank two anonymous reviewers and the editors for their many helpful comments and suggestions on an earlier draft of this chapter.

REFERENCES Adisasmito-Smith, Niken & Abigail C. Cohn. 1996. Phonetic correlates of primary and secondary stress in Indonesian: A preliminary study. Working Papers of the Cornell Phonetics Laboratory 11. 1–16. Baitschura, Uzbek. 1976. Instrumental phonetische Beiträge zur Untersuchung der Sprachmelodie und des Wortakzentes im Tscheremissischen. Études Finno-Ougriennes 3. 109 –122. Beckman, Jill N. 1998. Positional faithfulness. Ph.D. dissertation, University of Massachusetts, Amherst. Beckman, Mary E. 1986. Stress and non-stress accent. Dordrecht: Foris. Bereczki, Gábor. 1988. Geschichte der wolgafinnischen Sprachen. In Sinor (1988), 314 – 350. Booij, Geert. 1995. The phonology of Dutch. Oxford: Clarendon Press. Buckley, Eugene. 2000. On the naturalness of unnatural rules. Proceedings from the 2nd Workshop on American Indigenous Languages, 1–14. Santa Barbara: Department of Linguistics, University of California, Santa Barbara. Bye, Patrik & Paul de Lacy. 2008. Metrical influences on lenition and fortition. In Joaquim Brandão de Carvalho, Tobias Scheer & Philippe Ségéral (eds.) Lenition and fortition, 173–206. Berlin & New York: Mouton de Gruyter.

21

Matthew Gordon

Byrd, Dani. 1994. Articulatory timing in English consonant sequences. Ph.D. dissertation, University of California, Los Angeles. Casali, Roderic F. 1997. Vowel elision in hiatus contexts: Which vowel goes? Language 73. 493–533. Chafe, Wallace. 1970. A semantically based sketch of Onondaga. Baltimore: Waverly Press. Chafe, Wallace. 1977. Accent and related phenomena in the Five Nations Iroqouis languages. In Larry M. Hyman (ed.) Studies in stress and accent, 169–181. Los Angeles: Department of Linguistics, University of Southern California. Cho, Taehong & Patricia Keating. 2001. Articulatory and acoustic studies on domaininitial strengthening in Korean. Journal of Phonetics 29. 155 –190. Christmas, Raymond B. & J. Elisabeth Christmas. 1975. Kupia phonemic summary. Kathmandu: Tribhuvan University. Colarusso, John. 1992. A grammar of the Kabardian language. Calgary: University of Calgary Press. Crosswhite, Katherine. 1998. Segmental vs. prosodic correspondence in Chamorro. Phonology 15. 281–316. Crosswhite, Katherine. 2001. Vowel reduction in Optimality Theory. New York: Routledge. Crosswhite, Katherine. 2004. Vowel reduction. In Hayes et al. (2004), 191–231. de Jong, Kenneth J. 1995. The supraglottal articulation of prominence in English: Linguistic stress as localized hyperarticulation. Journal of the Acoustical Society of America 97. 491–504. de Lacy, Paul. 2001. Markedness in prominent positions. MIT Working Papers in Linguistics 40. 53–66. (ROA-282.) Delgutte, Bertrand 1982. Some correlates of phonetic distinctions at the level of the auditory nerve. In Rolf Carlson & Björn Granström (eds.) The representation of speech in the peripheral auditory system, 131–150. Amsterdam: Elsevier Biomedical. Delgutte, Bertrand. 1997. Auditory neural processing of speech. In W. J. Hardcastle & John Laver (eds.) The handbook of phonetic sciences, 507–538. Oxford & Cambridge, MA: Blackwell. Delgutte, Bertrand & Nelson Y. S. Kiang. 1984. Speech coding in the auditory nerve. I: Vowel-like sounds. Journal of the Acoustical Society of America 75. 879 –886. Dilley, Laura C., Stefanie Shattuck-Hufnagel & Mari Ostendorf. 1996. Glottalization of wordinitial vowels as a function of prosodic structure. Journal of Phonetics 24. 423–444. Donohue, Mark. 1999. A grammar of Tukang Besi. Berlin & New York: Mouton de Gruyter. Eek, Arvo. 1975. Observations on the duration of some word structures II. Estonian Papers in Phonetics 1975. 7–51. Everett, Keren. 1998. Acoustic correlates of stress in Pirahã. Journal of Amazonian Languages 1. 104 –162. Flemming, Edward. 1994. The role of metrical structure in segmental rules. Papers from the Annual Meeting of the North East Linguistic Society 24. 97–110. Fry, D. B. 1955. Duration and intensity as physical correlates of linguistic stress. Journal of the Acoustical Society of America 27. 765 –768. Fry, D. B. 1958. Experiments in the perception of stress. Language and Speech 1. 126 –152. Giginejshvili, B. K. 1977. Sravnitel′naja fonetika dagestanskix iazykov. Tbilisi: Tbilisi University. Goddard, Ives. 1979. Delaware verbal morphology: A descriptive and comparative study. New York: Garland. Goddard, Ives. 1982. The historical phonology of Munsee. International Journal of American Linguistics 48. 16 –48. Gonzales, A. 1970. Acoustic correlates of accent, rhythm, and intonation in Tagalog. Phonetica 22. 11– 44. Gordon, Matthew. 1995. Acoustic properties of primary and secondary word-level stress in Estonian. Poster presented at the 130th Meeting of the Acoustical Society of America, St Louis.

Stress: Phonotactic and Phonetic Evidence

22

Gordon, Matthew. 1997. Phonetic correlates of stress and the prosodic hierarchy in Estonian. In Jaan Ross & Ilse Lehiste. Estonian prosody: Papers from a symposium, 100–124. Tallinn: Institute of Estonian Language. Gordon, Matthew. 1998. A fortition-based approach to Balto-Fennic-Sámi consonant gradation. Folia Linguistica Historica 18. 49–79. Gordon, Matthew. 2001. Syncope induced metrical opacity as a weight effect. Proceedings of the West Coast Conference on Formal Linguistics 20. 206–219. Gordon, Matthew. 2002. A phonetically-driven account of syllable weight. Language 78. 51–80. Gordon, Matthew. 2003. The phonology of pitch accents in Chickasaw. Phonology 20. 173–218. Gordon, Matthew. 2004. A phonetic and phonological study of word-level stress in Chickasaw. International Journal of American Linguistics 70. 1–32. Gordon, Matthew. 2005. A perceptually-driven account of onset-sensitive stress. Natural Language and Linguistic Theory 23. 595 –653. Gordon, Matthew & Ayla Applebaum. 2010. Acoustic correlates of stress in Turkish Kabardian. Journal of the International Phonetic Association 40. 35 –58. Gordon, Matthew, Pamela Munro & Peter Ladefoged. 2000. Some phonetic structures of Chickasaw. Anthropological Linguistics 42. 366 –400. Gordon, Matthew & Pamela Munro. 2007. A phonetic study of final vowel lengthening in Chickasaw. International Journal of American Linguistics 73. 293–330. Gouskova, Maria. 2003. Deriving economy: Syncope in Optimality Theory. Ph.D. dissertation, University of Massachusetts, Amherst (ROA-610). Gregores, Emma & Jorge A. Suárez. 1967. A description of colloquial Guaraní. The Hague: Mouton. Gruzov, Leonid Petrovich. 1960. Sovremenyi mariiskii yazyk: Fonetika. Joshkar-ola: Mariiskoe Knizhnoe Izdatel′stvo. Hayes, Bruce. 1985. Iambic and trochaic rhythm in stress rules. Proceedings of the Annual Meeting, Berkeley Linguistics Society 11. 429 –446. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hayes, Bruce, Robert Kirchner & Donca Steriade (eds.) 2004. Phonetically based phonology. Cambridge: Cambridge University Press. Helimski, Eugene. 1998. Nganasan. In David Abondolo (ed.) The Uralic languages, 480 –515. London & New York: Routledge. Horne, Elinor Clark. 1974. Javanese–English dictionary. New Haven & London: Yale University Press. Hualde, José Ignacio. 1989. Autosegmental and metrical spreading in the vowel-harmony systems of northwestern Spain. Linguistics 27. 773–805. Hualde, José Ignacio. 1998. Asturian and Cantabrian metaphony. Rivista di Linguistica 10. 99–108. Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313–359. Hyman, Larry M. 1989. Accent in Bantu: An appraisal. Studies in the Linguistic Sciences 19. 115–134. Hyman, Larry M. 2001. The limits of phonetic determinism in phonology: *NC revisited. In Elizabeth Hume & Keith Johnson (eds.) The role of speech perception in phonology, 141–185. San Diego: Academic Press. Hyman, Larry M. 2006. Word-prosodic typology. Phonology 23. 225 –257. Jassem, Wiktor, John Morton & Maria Steffen-Batóg. 1968. The perception of stress in synthetic speech-like stimuli by Polish listeners. Speech Analysis and Synthesis 1. 289–308. Jun, Sun-Ah. 1993. The phonetics and phonology of Korean prosody. Ph.D. dissertation, Ohio State University. Jun, Sun-Ah & Cécile Fougeron. 1995. The accentual phrase and the prosodic structure of French. In Kjell Elenius & Peter Branderud (eds.) Proceedings of the 13th International Congress of the Phonetic Sciences, vol. 2, 722–725. Stockholm: KTH & Stockholm University.

23

Matthew Gordon

Kakumasu, James. 1986. Urubu-Kaapor. In Desmond C. Derbyshire & Geoffrey K. Pullum (eds.) Handbook of Amazonian languages, vol. 3, 326–406. Berlin & New York: Mouton de Gruyter. Keating, Patricia, Taehong Cho, Cécile Fougeron & Chai-Shune Hsu. 2003. Domain-initial articulatory strengthening in four languages. In John Local, Richard Ogden & Rosalind Temple (eds.) Phonetic interpretation: Papers in laboratory phonology VI, 145–163. Cambridge: Cambridge University Press. Kirchner, Robert. 2001. An effort-based approach to consonant lenition. New York & London: Routledge. Kirchner, Robert. 2004. Consonant lenition. In Hayes et al. (2004), 313–345. Kryvitskij, A. A. & A. I. Podluzhni. 1994. Uchebnik belorusskogo jazyka dlja samoobrazovanija. Minsk: Vysheishaia Shkola. Lavoie, Lisa. 2001. Consonant strength: Phonological patterns and phonetic manifestations. New York: Garland. Leer, Jeff. 1985. Prosody in Alutiiq (the Koniag and Chugach dialects of Alaskan Yupik). In Michael Krauss (ed.) Yupik Eskimo prosodic systems: Descriptive and comparative studies, 77–133. Fairbanks: Alaska Native Language Center. Lehiste, Ilse. 1965. Vowel quantity in word and utterance in Estonian. Congressus Secundus Internationalis Fenno-Ugristarum, 293–303. Helsinki: Societas Fenno-Ugrica. Lehiste, Ilse. 1966. Consonant quantity and phonological units in Estonian. Bloomington: Indiana University Press. Levi, Susannah V. 2005. Acoustic correlates of lexical accent in Turkish. Journal of the International Phonetic Association 35. 73–97. Lieberman, Philip. 1960. Some acoustic correlates of word stress in American English. Journal of the Acoustical Society of America 32. 451–454. Lombardi, Linda. 2001. Why Place and Voice are different: Constraint-specific alternations in Optimality Theory. In Linda Lombardi (ed.) Segmental phonology in Optimality Theory: Constraints and representations, 13–45. Cambridge: Cambridge University Press. Maiden, Martin. 1995. Evidence from the Italian dialects for the internal structure of prosodic domains. In John Charles Smith & Martin Maiden (eds.) Linguistic theory and the Romance languages, 115–131. Amsterdam & Philadelphia: John Benjamins. Mellander, Evan W. 2001. Quantitative processes in trochaic systems. Proceedings of the West Coast Conference on Formal Linguistics 20. 414 –427. Michelson, Karin. 1988. A comparative study of Lake-Iroquoian accent. Dordrecht: Kluwer. Munro, Pamela. 1996. The Chickasaw sound system. Unpublished ms., University of California, Los Angeles. Munro, Pamela 2005. Chickasaw. In Heather Hardy & Janine Scancarelli (eds.) Native languages of the southeastern United States, 114–156. Lincoln: University of Nebraska Press. Munro, Pamela & Peter John Benson. 1973. Reduplication and rule ordering in Luiseño. International Journal of American Linguistics 39. 15 –21. Munro, Pamela & Charles Ulrich. 1984. Structure-preservation and Western Muskogean rhythmic lengthening. Proceedings of the West Coast Conference on Formal Linguistics 3. 191–202. Munro, Pamela & Catherine Willmond. 1994. Chickasaw: An analytical dictionary. Norman, OK: University of Oklahoma Press. Munro, Pamela & Catherine Willmond. 2005. Chikashshanompa’ kilanompoli’. Los Angeles: UCLA Academic Publishing. Nivens, Richard. 1992. A lexical phonology of West Tarangan. In Donald Burquest & Wyn Laidig (eds.) Phonological studies in four languages of Maluku, 127–227. Arlington: Summer Institute of Linguistics. Penny, Ralph J. 1978. Estudio estructural del habla de Tudanca. Tübingen: Niemeyer.

Stress: Phonotactic and Phonetic Evidence

24

Pierrehumbert, Janet B. & David Talkin. 1992. Lenition of /h/ and glottal stop. In Gerard J. Docherty & D. Robert Ladd (eds.) Papers in laboratory phonology II: Gesture, segment, prosody, 90 –117. Cambridge: Cambridge University Press. Plomp, Reinier 1964. Rate of decay of auditory sensation. Journal of the Acoustical Society of America 36. 277–282. Poser, William J. 1990. Evidence for foot structure in Japanese. Language 66. 78 –105. Potisuk, Siripong, Jackson Gandour & Mary P. Harper. 1996. Acoustic correlates of stress in Thai. Phonetica 53. 200 –220. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19 –100. Reed, Irene, Osahito Miyaoka, Steven Jacobson, Paschal Afcan & Michael Krauss. 1977. Yup’ik Eskimo grammar. Fairbanks: Alaska Native Language Center. Rehg, Kenneth L. 1993. Proto-Micronesian prosody. In Jerold A. Edmondson & Kenneth J. Gregerson (eds.) Tonality in Austronesian languages, 25 –46. Honolulu: University of Hawaii Press. Revithiadou, Anthi. 2004. The Iambic/Trochaic Law revisited: Lengthening and shortening in trochaic systems. Leiden Papers in Linguistics 1. 37– 62. Sammallahti, Pekka. 1988. Historical phonology of the Uralic languages. In Sinor (1988), 478–554. Schütz, Albert J. 1985. The Fijian language. Honolulu: University of Hawaii Press. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Sinor, Denis (ed.) 1988. The Uralic languages: Description, history and foreign influences. New York: Brill. Sluijter, Agaath M. C. & Vincent J. van Heuven. 1996a. Spectral balance as an acoustic correlate of linguistic stress. Journal of the Acoustical Society of America 100. 2471–2485. Sluijter, Agaath M. C. & Vincent J. van Heuven. 1996b. Acoustic correlates of linguistic stress and accent in Dutch and American English. Proceedings of the International Conference on Spoken Language Processing 4, 630–633. Smith, Jennifer. 2000. Prominence, augmentation, and neutralization in phonology. Proceedings of the Annual Meeting, Berkeley Linguistics Society 26. 247–257. Smith, Jennifer. 2004. Phonological augmentation in prominent positions. New York: Routledge. Smith, Robert L. 1979. Adaptation, saturation and physiological masking in single auditory nerve fibers. Journal of the Acoustical Society of America 65. 166–178. Steriade, Donca. 1997. Licensing laryngeal features. Unpublished ms., University of California, Los Angeles. Taff, Alice, Lorna Rozelle, Taehong Cho, Peter Ladefoged, Moses Dirks & Jacob Wegelin. 2001. Phonetic structures of Aleut. Journal of Phonetics 29. 231–271. Teranishi, R. 1980. Two-moras-cluster as a rhythm unit in spoken Japanese sentence or verse. Paper presented at the 99th Meeting of the Acoustical Society of America, Atlanta. Tereshchenko, N. M. 1979. Nganasankii iazyk. Leningrad: Nauka. Topuria, G. V. 1974. On one regularity in the system of preruptives in the Lezgian language. Annual of Ibero-Caucasian Linguistics 1. 180 –184. Vaysman, Olga. 2009. Segmental alternations and metrical theory. Ph.D. dissertation, MIT. Viemeister, Neal. 1980. Adaptation of masking. In G. van den Brink & F. A. Bilsen (eds.) Psychophysical, physiological and behavioural studies in hearing, 190–198. Delft: Delft University Press. Walker, Rachel. 2004. Vowel feature licensing at a distance: Evidence from Northern Spanish language varieties. Proceedings of the West Coast Conference on Formal Linguistics 23. 787–800. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language and Linguistic Theory 23. 917–989.

25

Matthew Gordon

Wightman, Colin W., Stefanie Shattuck-Hufnagel, Mari Ostendorf & Patti J. Price. 1992. Segmental durations in the vicinity of prosodic phrase boundaries. Journal of the Acoustical Society of America 91. 1707–1717. Williams, Briony. 1985. Pitch and duration in Welsh stress perception: The implications for intonation. Journal of Phonetics 13. 381–406. Wilson, J. P. 1970. An auditory afterimage. In R. Plomp & G. F. Smoorenberg (eds.) Frequency analysis and psychophysics of hearing, 303–315. Leiden: Sijthoff. Woodbury, Anthony. 1981. Study of the Chevak dialect of Central Alaskan Yupik. Ph.D. dissertation, University of California, Berkeley. Yu, Alan C. L. 2004. Explaining final obstruent voicing in Lezgian: Phonetics and history. Language 80. 73–97. Zoll, Cheryl. 1998. Positional asymmetries and licensing. Unpublished ms., MIT (ROA-282).

40 The Foot Michael Hammond

1

Overview

The metrical foot organizes the syllables of words into higher-order units built around stressed syllables. In this chapter, we review the evidence for, and structure of, the foot. Along the way, we treat some of the major issues that have arisen in the development of this notion. The organization of this chapter is as follows. First, we review the background against which the foot was proposed: linear generative phonology and then early footless metrical phonology. We then turn to the earliest foot proposals and the arguments advanced at the time, including arguments from stress theory and prosodic morphology. We then go on to consider how the theory of the foot has changed in Optimality Theory (OT). (For a more general discussion of stress, see chapter 39: stress: phonotactic and phonetic evidence.)

2

Background

In this section, we lay out the necessary background for understanding the earliest proposals about the foot. First and foremost is the background of generative phonology generally and Chomsky and Halle (1968) specifically. We then go on to consider the foundation for the foot laid in early metrical theory.

2.1

Generative phonology and SPE

Here we discuss Chomsky and Halle (1968; henceforth SPE) as the foundation for the foot.1 The main contribution of SPE for our purposes is an explicit treatment of the regularities of stress in English. The analysis is comprised of a number of 1

Some of the issues in this section are developed further in chapter 51: the phonological word, chapter 116: sentential prominence in english and chapter 41: the representation of word stress. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0040

2

Michael Hammond

rules written in the specific formalism proposed there. Their full rule for main stress in English is given in (1) (SPE: 240). (1)

A G−tense J A G avoc J D D G V → [1 stress] / X __ C0 B HcstressK C10 B HaconsK E E I CI V L C I −ant L F F

/ __

5 1 G−stressJ 4 4 〈 +C 〉 H−tense K [+cons] 1 0 1 0 4 J ( fik)At # 4 ! I−cons L 6 L (@ ) 2 [+D]C0 $ 4 4 G −seg J 4 〈1 〉1 C0[bstress]C0〈2V0C0〉2 4 I〈 −FB〉 L 2 2 7 3

Conditions:

〈NSP〈1 VA〉1〉

!2# b=@ $ 1 c≤ 2 X contains no internal #

Setting aside morphological and diacritic variables, and focusing on nouns in particular, the main stress rule assigns main stress and secondary stress to nouns in the following way. First, a [+tense] vowel in a final syllable gets primary or secondary stress, e.g. kangaroo [‘kæIgH’ru] or chickadee [’Œ>kH‘di]. If the final syllable does not have a stress, and the penultimate vowel is followed by appropriate consonants, then it gets stress, e.g. agenda [H’–endH]. Likewise, if the penultimate vowel is [+tense], it gets stress as well, e.g. aroma [H’romH]. Finally, in other cases, the antepenult receives stress, e.g. America [H’merHkH] or remedy [’remHdi]. Abstracting away from the formalism of the time, what we see is a restriction of the primary stress to the last three syllables of the word and a pressure to stress syllables of appropriate weight. The alternating stress rule in (2) (SPE: 240) is responsible for stresses further to the left, e.g. the first stress in kangaroo [‘kæIgH’ru]. Like the main stress rule, it places stress subject to a basically alternating pattern. (2)

V → [1 stress] / __ C0 (=) C0VC0[1 stress]Co]NAV

Both rules alternate in a similar way; both rules assign stress with respect to a following landmark. The fact that both rules exhibit similar patterns and the fact that this kind of alternation is ubiquitous in other languages was a missed generalization in SPE, and one that found an explanation in the development of the metrical foot. A second important aspect of the analysis proposed in SPE is that, unlike other phonological features, the feature [stress] exhibited more than two levels in the phonology. Thus, while a feature like [high] could have the values [+high] or [−high] in the phonology, the feature stress could have an infinite number of values: [0stress], [1stress], [2stress], [3stress], etc. These different numerical values corresponded to degrees of stress that were held to be contrastive. Moreover, the values had a rather odd interpretation, where [0stress] is the least stressed element and [1stress] has the most stress. Since the values can increase without bound, as the integer value increases, the degree of stress gets smaller, but never quite reaches [0stress].

The Foot (3)

Value

Interpretation

[0stress] [1stress] [2stress] [3stress] [nstress]

stressless primary stress secondary stress tertiary stress etc.

3

What made this interpretation work was the Stress Subordination Convention (SSC): “when primary stress is placed in a certain position, then all other stresses in the string under consideration at that point are automatically weakened by one” (Chomsky and Halle 1968: 16–17). The SSC governed application of the stress rules through the cycle (see chapter 85: cyclicity). The cycle held that phonological rules were reapplied as morphological and syntactic operations proceeded. Let us see how this works with the Nuclear Stress Rule (NSR) and Compound Stress Rule (CSR) (Chomsky and Halle 1968: 18): 1

(4)

. . . V . . . ]NAV # G1 stressJ !__ 1 I V L → [1 stress] / @V . . . __ . . . ] $

a.

CSR

b.

NSR

The NSR basically re-assigns primary stress to the rightmost stress of a domain. The CSR re-assigns primary stress to the penultimate primary stress of a domain. This re-assignment of stress is vacuous with respect to the particular vowel that is targeted, but has consequences for all other vowels in the domain via the SSC (Chomsky and Halle 1968: 22). (5)

[NP John’s 1 – – 2

[N [N black 1 1 1 1

board]N eraser ]N ]NP 1 2 3 4

1 – 2 3

NSR CSR CSR NSR

First, stress is assigned to each word in isolation. Then blackboard is assembled, and main stress is assigned to the penultimate/first primary stress, since this is a compound. Next, eraser is appended, and compound stress is re-assigned. Notice how, since there are only two primary stresses present at that point, the penultimate primary is actually the third stress from the right. Finally, we add John’s, and main stress is re-assigned to the rightmost primary, which again is the third stress from the right. Chomsky and Halle thus contributed four key elements to early stress theory: binary alternation, the cycle, the n-ary [stress] feature, and the Stress Subordination Convention (SSC). All of these figured in the development of early metrical theory.

2.2

Early metrical theory

The earliest versions of metrical theory directly addressed the n-ary stress feature, the cycle, and the SSC. It was only later that the foot was introduced.

4

Michael Hammond

Rischel (1972) was the earliest proposal to replace aspects of the SPE stress system with a hierarchical tree. Specifically, Rischel proposes that compound stress in Danish does not require a cycle and that degrees of stress can be easily read off the morphosyntactic tree. Compare the following two compounds in Danish: (6)

fædrelandssang perlehalsbånd

‘patriotic song’ ‘pearl necklace’

[father-land]song pearl[neck-band]

In the first case, the compound is left-branching and has the stress values 132. In the second case, the compound is right-branching and has the stress pattern 123. The SPE rules given for English here would actually accommodate these directly, as shown in the derivations below. (7) [N [N fædre

lands ]N sang ]N

1 1 1 (8)

[N perle

1 2 3 [N hals

1 – 1

1 1 2

1 – 2

NSR CSR CSR

bånd ]N ]N 1 2 3

NSR CSR CSR

Rischel proposes that cyclic effects can be gotten by reading stress levels directly off of trees. He gives trees like the following for the examples above. The pluses and minuses reflect the relative strength of left and right branches and the numbers on nodes reflect the relative effects of those strengths at different levels of the tree. (9)

+ + fædre

2

−

1

− sang

lands

+ perle

1

− +

hals

2

− bånd

Rischel does not propose a specific algorithm for reading stress values off trees like these, but it is easy to see that various interpretations will produce what appear to be reasonable values. The gist is that reapplication of stress rules per se is not required to get the same kind of cyclic effects cited above from SPE.2 Liberman and Prince (1977) made a similar proposal a few years later, proposing a fairly complete analysis of English stress along similar lines. Basically, they propose that [stress] be treated as a binary feature, with the values [+stress] and 2

Ultimately, Kiparsky (1979) argued that cyclicity is still necessary in a metrical theory of stress. The debate resurfaced again a few years later. See Hammond (1989), Halle and Kenstowicz (1991), and Cole and Coleman (1992) for more discussion.

The Foot

5

[−stress]. Degrees of stress would follow from tree structures erected over the stress values. Trees are binary branching, with each pair of nodes labeled either “strong” (s) or “weak” (w). Liberman and Prince posit the following labeling convention (1977: 257). (10)

In a configuration [c A B ]c: a. b.

NSR: If C is a phrasal category, B is strong. CSR: If C is a lexical category, B is strong iff it branches.

In phrases, clause (a) of the convention places strong nodes uniformly on the right. (11) w John

s

s left w Mary

w loves

s John

In compounds, labeling depends on branching. In a left-branching compound, left nodes are uniformly strong. When right nodes branch, however, by clause (b), they are strong. (12) s s black

s w board

w eraser

w union

s w finance committee

In the case on the left, all right nodes are non-branching, so the left node of each pair of nodes is labeled strong. In the case on the right, finance committee has a non-branching right node, so its left node is labeled strong, but in union finance committee the right node is branching, so it is labeled strong. Like Rischel (1972), Liberman and Prince replace degrees of stress with a treebased algorithm. However, Liberman and Prince go one step further and propose a similar word-internal system. There are three parts to their system. First, [+stress] values are assigned to vowels in words. They initially propose two separate rules for this, much as in SPE. There is the English Stress Rule (ESR (preliminary version); 1977: 272). (13)

V → [+stress] / __ C0 (h(C)) (hC0) #

This rule assigns the rightmost [+stress] in a word. There is also a Stress Retraction Rule (SRR; 1977: 278). (14)

V → [+stress] / __ C0 (h(C))a (VC0)b

V [+stress]

Having two separate rules misses a generalization and Liberman and Prince move a major step forward from SPE in recognizing this:

Michael Hammond

6

Both rules measure leftward from a fixed point of reference, the ESR from a word boundary, the SRR from a stressed syllable; and the standard of measure is in both cases virtually the same. This parallelism strongly suggests that we are witnessing a single unified process of stress assignment, repeating itself across the word, feeding on its own output. (1977: 278)

To deal with this, Liberman and Prince propose a unified English Stress Rule (ESR (iterative version); 1977: 278). (15)

V → [+stress] / __ C0 (V(C))a ( V C0)b ( V X)c# d [+stress] Conditions: ¬c ⊃ d; ¬a, ¬b under certain morphological and lexical circumstances.

A variety of conditions must be imposed on the rule, much as on the SPE equivalents. That said, the rule captures the intuition that there is a similar pattern of iteration. The second part of Liberman and Prince’s algorithm is a key step in the development of the metrical foot. Once values of stress have been assigned to a string by the rule(s) above, syllables are gathered into trees. Above the word level, these trees correspond to syntactic and morphological structures. Below the word level, there are two essential components. First: every sequence of syllables + −, + − −, + − − −, etc., forms a metrical tree. Because of the condition limiting [−stress] to weak positions, and because of the bivalent (binary branching) character of metrical trees, the structure and labeling of the sequences is uniquely determined. (1977: 266)

By this algorithm, we get trees like the following: (16) s +

s

w − s +

s s

w −

w −

s +

w −

w −

w −

...

These trees are then gathered into larger right-branching trees: N

(17)

#

D

D

…

D

D

D

#

The Foot

7

As noted above, the labeling of the lower-level trees is unambiguous, because of a general constraint against [+stress] in weak position. The higher-level trees are labeled in accord with the Lexical Category Prominence Rule (LCPR; 1977: 270). (18)

In the configuration [N1N2], N2 is strong iff it branches.

Let us take a look at an example: Winnepesaukee [‘w>nHpH’sDki]. First, [+stress] values are assigned by the ESR, producing: (19)

+ − −+ − Winnepesaukee

Syllables are gathered into feet as below: (20) s s +

w −

w −

s +

w −

Finally, the feet are gathered into a tree, the right node of which is labeled strong, since it is branching. (21) w

s

s s +

w −

w −

s +

w −

What is important about this entire tree-construction and tree-labeling procedure is that it explicitly recognizes two levels: a foot level and a higher word level. This is the first step toward an explicit theory of the foot. Liberman and Prince showed how the foot could be employed in a reanalysis of the basic stress facts of English that SPE introduced.

3

Why we need feet

The next step was the parametric elaboration of the foot. At around the same time as Liberman and Prince (1977), Hyman (1977) offered the first typological treatment of stress. While he was not able to go very far in terms of the technical analysis offered, this paper was an important catalyst in forcing phonologists interested in stress to look at the broader typological implications of their work. The first parametric approaches to the metrical foot showed up in Halle and Vergnaud (1978) and McCarthy (1979), but the most influential early proposal was that of Hayes (1980). Let us look at the Hayes proposal in some depth.

8

Michael Hammond

Hayes offered a theory of the foot based on the trees proposed in Liberman and Prince. In particular, feet were parameterized for the following: (22)

Headedness Is the designated terminal element – the strongest element of the foot – on the left edge or the right? b. Boundedness Are feet binary or unbounded? Do feet contain only two syllables or as many as possible? c. Directionality Are feet built left-to-right or right-to-left? d. Iterativity Are feet constructed iteratively or not? That is, is only a single foot built on some edge or are as many feet built as possible? e. Quantity-sensitivity There are three choices here. First, feet can be quantity-sensitive (QS): weak nodes cannot dominate heavy syllables. Second, feet can be quantity-insensitive (QI): syllable weight is irrelevant. Last, feet can be obligatory-branching (OB): in OB feet, strong nodes must dominate heavy syllables and weak nodes may not. f. Syllable weight If feet are sensitive to syllable weight, are they sensitive to the weight of the syllable nucleus or the syllable rhyme? a.

Let us go through some of the examples Hayes cites in support of this theory. Maranungku (Tryon 1970) is cited as an example of left-headed binary left-to-right QI feet. Here, main stress falls on the first syllable of the word and secondary stresses fall on alternating syllables to the right. (23)

’tiralk ’mere‘pet ’jangar‘mata ’langka‘rate‘ti ’wele‘pene‘manta

‘saliva’ ‘beard’ ‘the Pleiades’ ‘prawn’ ‘kind of duck’

Here are two examples: (24) s w ’me re ‘pet

s w s w ’jan gar ‘ma ta

Notice how the left-to-right construction of feet is apparent from the fact that in words with an odd number of syllables, a monosyllabic, or degenerate, foot is built on the right. The difference between primary and secondary stress is captured by positing a higher level of structure: the word tree. These are left- or right-headed unbounded trees built on the roots of feet (see chapter 41: the representation of word stress). In Maranungku, the word tree is left headed.

The Foot

9

(25) w

s

s w ’me re ‘pet Note that in this and in subsequent diagrams we circle the roots of feet when the word tree is represented.3 Warao (Osborn 1966) provides an example of right-to-left footing with left-headed binary feet. (26)

‘japu‘ruki‘tane’hase ‘naho‘roa‘haku’tai ji‘wara’nae e‘naho‘roa‘haku’tai

‘verily to climb’ ‘the one who ate’ ‘he finished it’ ‘the one who caused him to eat’

Warao differs from Maranungku in that, in words with an odd number of syllables, there is no initial degenerate foot; rather that foot is removed by an additional destressing rule. For example, [ji‘wara’nae] is first footed as follows: (27) s w s w ji ‘wa ra ’na e This intermediate representation is then converted to: (28) s w s w ji ‘wa ra ’na e The word tree in Warao is right headed. Hayes assumes that unfooted syllables are adjoined as weak nodes to the word tree. For example: (29) s w

s

w s w s w ji ‘wa ra ’na e Hayes cites Weri (Boxwell and Boxwell 1966) as an example of binary right-headed feet constructed from right to left with a right-headed word tree. 3

Hayes (1980) uses underlining, rather than circles.

10 (30)

Michael Hammond ‘bee’ ‘hair of arm’ ‘mist’ ‘times’

I>n’t>p ‘kÁl>’pÁ Á‘lÁa’m>t ‘akÁ‘nete’pel

And again, a couple of examples: (31) s

w

w

w s ‘kÁ l/ ’pÁ

s

w s w s Á ‘lÁ a ’m/t

Completing the set of binary QI systems, Hayes cites Southern Paiute (Sapir 1930) as an example of binary right-headed feet built left to right. Main stress is assigned with a left-headed word tree. (32)

mant’çac‘qac ma’roÜ‘qwaj’qq•c

‘to hold out one’s hands’ ‘(I) stretch it’

There are several complications to the Southern Paiute system, both noted by Hayes. First, it appears as if there are elements that might be analyzed as long vowels, but that must be treated as adjacent short vowels. (Hayes cites additional evidence for this claim.) In addition, there is an another mechanism that prevents footing of the final syllable: extrametricality (see chapter 43: extrametricality and non-finality). With these provisos, and marking extrametricality with angled brackets, we get structures like these: (33) s s w

s s

w s w s w mant ’ça ‘qa < >

s

w

w

w s w s ma ’ro Ü ‘qwa j’q

w

Southern Paiute is a rather complex case. A simpler example is Araucanian, as described by Echeverría and Contreras (1965). There are some complications here too, but stress in Araucanian basically falls on even-numbered syllables counting from the left.4

4

Complications involve three-syllable words ending in a consonant: these have a final secondary stress, e.g. [.u’Iu‘lan] ‘I do not speak’. There are also contextual effects on short words.

The Foot (34)

wu’le Íi’panto e’lumu‘ju e’lua‘enew ki’mufa‘luwu‘laj

11

‘tomorrow’ ‘year’ ‘give us’ ‘he will give me’ ‘he pretended not to know’

The analysis here is binary right-headed feet built left to right. (35) s

w

w s w s e ’lu mu ‘ju Monosyllabic feet are generally disallowed (or removed) in odd-syllabled cases: (36) s w s w Íi ’pan to Let us now consider quantity-sensitivity (QS; see chapter 57: quantity-sensitivity). This parameter allows heavy syllables to attract stress. Hayes cites Tübatulabal (Voegelin 1935) as an example of right-to-left right-headed QS bounded feet. The generalization is that stress falls on (a) the final syllable, (b) any long vowel, and (c) any vowel that is two syllables to the left of a stress. Since stresses are unranked, there is no word tree. (37) ’taa’hawi’laap ‘in the summer’ pq’tqpq’tqqdi’nat ‘he is turning it over repeatedly’ ’qq’?qqi’?aani’œa ‘he will meat-fast’ ’pDnih’wqn ‘of his own skunk’ Here are two examples of the footings produced by these parameter settings. (38) w s ’taa ’ha wi ’laap

’qq

’?qq

w i

s w ’?aa ni

s ’œa

Notice that long vowels count as heavy in Tübatulabal; thus QS refers to the nucleus, not the rhyme.5 Hayes’ theory of feet also allows for unbounded feet. When these are quantityinsensitive, they simply position stress on the first or last syllable of the word. No actual examples are cited, but we would expect trees like the following for a language with initial stress and QI left-headed unbounded feet: 5

The data cited by Hayes do not establish unequivocally that codas do not contribute to weight.

Michael Hammond

12

s

(39) s s s w ’q q

w q

w q

Notice how, since the foot expands to fill the domain, no more than a single foot will ever be built. A language like Czech, with regular initial stress, might qualify as such a system. Hayes cites Eastern Cheremis (Sebeok and Ingemann 1961) as an example of unbounded left-headed QS footing: a single such foot is built on the right edge of the word; a word tree is not needed. The generalization is that the rightmost long vowel bears stress. If there is no such vowel, the initial vowel bears stress. (40)

œiin’Œaam œlaa’paaÚHm ’pyygHlmH ’kiidHœtHÚH ’tHlHzHn

‘I sit’ ‘his hat (acc)’ ‘cone’ ‘in his hand’ ‘moon’s’

This footing produces examples like the following. Notice how the foot starts at the right edge, expands as far as possible subject to the QS restriction that its weak nodes cannot dominate a long vowel. Notice that QS here is sensitive to only vowel length. (41) w s œlaa ’paa ,Hm

s

s s ’tH

w lH

w zHn

s s w ’kii dHœ

w tH

w ,H

The OB parameter is required for languages that exhibit a curious parallelism to languages like Eastern Cheremis. Khalkha Mongolian (Street 1963) is an example of this sort. Stress in Khalkha falls on the leftmost long vowel in the word; in the absence of a long vowel, stress falls on the initial vowel.6 (42)

6

bos’guul bari’aad xojHrdu’gaar ga’raasaa ’ali ’xøtHbHrH

‘fugitive’ ‘after holding’ ‘second’ ‘from one’s own hand’ ‘which’ ‘leadership’

See Walker (1997) for a different description and analysis.

The Foot

13

When there are long vowels in the word, the pattern could be treated with a single right-headed unbounded QS foot built on the left edge of the word. The problem with this is that it produces an incorrect result in the case of words with no long vowels: (43) s

w s ga ’raa saa s ba

w s *a ’li

w w ri ’add

To treat such systems, Hayes proposes a new parameter and makes interesting use of word trees. Specifically, he proposes that QS feet can be further restricted so that the strong/dominant node must dominate a heavy syllable: obligatorybranching (OB). If no such syllable is available, no foot is built. In the case of Khalkha, a single right-headed unbounded OB foot is built on the left edge of the word and a left-headed word tree is constructed. If there is at least one long vowel in the word, the OB foot will be built over the leftmost one, assigning stress to it, and the word tree is vacuous. (44) s w ga

s w ’raa saa

If there are no long vowels, then the OB foot fails to be constructed. In that case, the word tree is still built, taking syllables as terminals. Since, the word tree is left headed, this results in initial stress. (45) s s s w w w ’xø tH bH rH Obligatory-branching allows for a treatment of systems where stress is assigned to the first or last of the available heavy syllables and in the absence of heavy syllables, the same end of the word gets primary stress. This is in contrast to systems like Eastern Cheremis where the opposite edge of the word gets the default stress. These latter systems are treated with unbounded QS feet.7 7

The parallel between these systems was first discussed by Kiparsky (1973). An alternative formalization of OB footing was proposed by Hammond (1986).

14

Michael Hammond

We have only exemplified some of the combinations of parameter settings that this theory allows. The claim of the theory is that all settings can be freely combined and that the set of possible stress languages is fully defined by these settings. The argument for feet per se comes from their role in this parametric system. If the set of possible stress languages is best defined in terms of a theory that adopts the foot as a central descriptive device, then the typology of stress is an argument for the foot.

4

Do we need constituency?

One could argue that while the foot is a central computational device in the system Hayes develops, the full predictive power of the foot in that system is not exploited; specifically, while the foot is a constituent in metrical trees, its constituency plays no specific role in the system. Prince (1983) takes this observation to its logical conclusion, proposing an alternative metrical theory without constituency and without feet. To understand this proposal, let us return to Liberman and Prince (1977) and their theory of the metrical grid. Liberman and Prince propose the metrical grid as a mechanism for identifying the environment for the rhythm rule, the phenomenon whereby stress is shifted in certain contexts. Thus, when a word like ‘thir’teen is combined with ’men, we get a shift of stress in the former: ’thir‘teen ’men. The effect of this shift can be diagrammed as follows: (46) w

w

→

s w s ‘thir ’teen ’men

s w s ’thir ‘teen ’men

Interestingly, the shift also happens with phrases like ’achro‘matic ’lens, but not with phrases like ‘Mon’tana ’cow‘boy. Why this should be the case is not apparent from the metrical trees. (47) w w s ‘a

w s chro ’ma

w s

s

→ w s tic ’lens

s ’a

w w s chro ‘ma

w s tic ’lens

The Foot

15

(48) s s

s

w s w s w ‘Mon ’ta na ’cow‘boy To identify cases where rhythm is applicable, Liberman and Prince propose an alternative representation of stress: the metrical grid. The grid represents relative stress as columns of elements where the heights of those columns are projected from metrical trees. Specifically, Liberman and Prince propose the Relative Prominence Projection Rule (1977: 316). (49)

Relative Prominence Projection Rule (RPPR) In any constituent on which the strong–weak relation is defined, the designated terminal element of its strong subconstituent is metrically stronger than the designated terminal element of its weak subconstituent.

The way this is interpreted is that the column for any element must be tall enough so that the RPPR is satisfied for all node pairings defined by its metrical tree. For the phrases we have discussed we then have grids as below. (50) 6 *4 *5 1 2 3 thir teen men

11 *9 *10 6 7 8 1 2 3 4 5 a chro ma tic lens

8 6 7 1 2 3 4 5 Mon ta na cow boy

Each syllable is marked with a grid element. We then go through the tree, making sure that the RPPR is satisfied for each branching node. If it is not, we add a grid element to the relevant column. For example, the second syllable of Montana gets a grid element because of the lowest-level pairing with the syllable na. This same element is sufficient to satisfy the RPPR when we come to the pairing of tana with Mon. On the other hand, in achromatic, the third syllable gets a second-level grid mark because of its pairing with the fourth syllable. It must get an additional grid mark because of the pairing with the first two syllables. We have marked certain grid elements with asterisks. The rhythm rule applies when two columns are too big and too close, i.e. when stresses clash. These properties are formalized in terms of two adjacent elements at two adjacent levels of the grid. For example, in thirteen men, elements 2 and 3, and elements 4 and 5 are adjacent and we therefore mark elements 4 and 5. Likewise, in achromatic lens, elements 7 and 8 and elements 9 and 10 are adjacent, so again we mark this as a clash. The Rhythm Rule applies on the metrical tree to eliminate these clashes.8 The following grids show the results of the relabeling we have already shown for the relevant metrical grids. 8

However, see Hayes (1984) for a treatment of rhythm in English not making use of clash.

16

Michael Hammond

(51) 6 4 5 1 2 3 thir teen men

9 6 1 a

11 10 7 8 2 3 4 5 chro ma tic lens

Notice how there are no longer clashes in these grids. The metrical grid thus correctly distinguishes cases like thirteen men and achromatic lens from cases like Montana cowboy. It is, of course, unfortunate that a single representation for stress was not available. There were two broad responses. One response was a proposal by Hammond (1988) for a blended representation.9 The key insight in this proposal was that the designated element of a constituent should be marked the same way regardless of whether the constituent branches or not. This gives us equivalences as below for degenerate and binary left-headed feet. Notice how the heads of the feet have the same representation in Hammond’s approach, but not in Hayes’s approach. (52)

Feet

Hayes (1980)

Hammond (1988)

q

o q

degenerate

binary s q

w q

o q

q

Here is what a phrase like achromatic lens would look like after application of the rhythm rule. (53)

a chro ma tic lens While this particular formalism did not survive, this general proposal – that heads of constituents should be marked uniformly – did. Halle and Vergnaud (1987) and Hayes (1995) adopt the following equivalent notations that exhibit the same uniformity.10 These are referred to as “bracketed grids.” 9

This notation came to be referred to as “lollipops.” Idsardi (1992) pursues a representation with similar properties. Since Idsardi’s representation allows unpaired parentheses, it entails a somewhat different notion of constituency. 10

The Foot (54)

Feet

Halle and Vergnaud (1987)

Hayes (1995)

x (x) q

(x) q

x (x x) q q

(x .) q q

degenerate

binary

17

The other proposal in response to parallel tree and grid representations of stress was that of Prince (1983). Specifically, Prince proposed a grid-only theory of stress without the foot. The basic idea behind the proposal as far as feet are concerned is that binary patterns of iteration are replaced by appeal to the perfect grid. This device allows for a binary pattern of stress to be assigned in one of four ways, depending on whether the assignment is from left to right or from right to left and on whether one begins with a stressed syllable or a stressless syllable. (55)

Feet

Left-to-right (LR)

Right-to-left (RL)

peak first

x x x x x x x x q q q q q ...

x x x x x x x x ... q q q q q

trough first

x x x x x x x q q q q q ...

x x x x x x x ... q q q q q

Notice how this pattern is achieved with no appeal to binary constituents. To get the effect of word trees and unbounded foot construction, Prince proposes the End Rule. This device assigns a grid mark to the leftmost or rightmost element of the highest level of the grid present. If no stresses have already been assigned, the effect of the End Rule is to assign a stress to the first or last syllable of the word. (56)

End Rule Left x x x x q q q q →

x x x x x q q q q

x x x x q q q q →

x x x x x q q q q

End Rule Right

If stresses are already present, however, then the effect of the End Rule is to promote the leftmost or rightmost stress to primary stress: (57) End Rule Left x x x x x x q q q q →

x x x x x x x q q q q

18

Michael Hammond End Rule Right x x x x x x q q q q →

x x x x x x x q q q q

Again, no constituents per se are required. Finally, the effect of heavy syllables is achieved by allowing heavy syllables to project their own grid marks. Such marks are placed before assignment of the perfect grid. The perfect grid is interrupted by heavy syllable marks or End Rule marks in different ways. One possibility – the default – is that marked heavy syllables are treated as if they were assigned by the perfect grid itself; such syllables may not have a stress assigned to an adjacent syllable. The other possibility – Forward Clash Override (FCO) – is that iteration by the perfect grid continues right up to the marked syllable. The following schematic examples show how FCO works with (left-to-right trough-first) iteration toward a syllable marked by the End Rule Right. (58)

[−FCO] x x x x x q q q q q

→

[+FCO] x x x x x q q q q q →

x x x x x x q q q q q

→

x x x x x x x q q q q q

x x x x x x q q q q q →

x x x x x x x x q q q q q

Similar effects obtain when a syllable has been marked as a heavy syllable. In the following example, heavy syllables are marked with H; light syllables with an L. (59)

[−FCO] x x x x x x x L L L L H L L

→

x x x x x x x x L L L L H L L

→

x x x x x x x x L L L L H L L

[+FCO] x x x x x x x L L L L H L L

→

x x x x x x x x x x L L L L H L L

→

x x x x x x x x x x x L L L L H L L

Forward Clash Override, in conjunction with the End Rule, does the work of degenerate footing and destressing. There is an additional complication involved in whether heavy syllables occupy a single grid position or two positions in sequence. Recall that Hayes (1980) had to make a similar move in the case of Southern Paiute. The central result of Prince (1983) for our purposes is that it established that, on purely stress-based arguments, there is no argument for the foot as a constituent.

5

Other arguments for feet

There were three broad classes of additional arguments for feet: phonological processes, poetic meter, and prosodic morphology. Some of these arguments persevere and some do not.

The Foot

5.1

19

Flapping

One of the earliest arguments for feet outside of stress per se comes from Kiparsky (1979), who argues that flapping in English is best described in terms of feet (see chapter 113: flapping in american english). The basic environment for flapping is as follows. Coronal stops in English ([t d]) are pronounced as flaps before a stressless vowel and after a [−consonantal] element. Thus, word initially we have stops whether the following vowel is stressed or stressless, e.g. toe [’t ho] or doe [’do] and tonight [t h H’nait] or deny [dH’nai]. Medially, before a stressed vowel, we also have stops, e.g. attack [H’t hæk] or adult [H’dZlt]. However, medially before a stressless vowel, we get flaps, e.g. caddy [’khæ7i] or pity [’ph >7i]. Kiparsky describes this with two rules. The first is a cyclic rule of laxing that makes a consonant lax when it follows a [−consonantal] element within a foot (denoted with U here). (60)

Laxing (cyclic) C → [+lax] / U[. . . [−cons] __ . . .]U

The second rule converts some of these lax consonants into voiced ones when they are initial in a syllable. Kiparsky uses strong–weak labeling for syllable structure as well. The tree structure in the environment of this rule encodes the syllable-initial restriction. (61)

Voicing (postcyclic) t +lax

→ [+voiced]/ —

s

A form like pity [’ph >7i] would be syllabified and footed as follows. (62) s w p

w s /

w t

s i

First, the /t/ undergoes Laxing because it is medial in the foot and follows a [−consonantal] element. Then it can undergo Voicing because the /t/ is syllableinitial. The approach nicely accommodates examples like at ease [‘æ ’7iz], where flapping applies across word boundaries, and examples like a tease [H’t h iz], where it does not. For at ease, we get a derivation as follows: (63)

→ s w ’æ

s t

s ’i

w z

w ’æ ’t

s i

w z

Michael Hammond

20

Each word is syllabified separately as a separate syllable and foot. Since the /t/ follows a [−consonantal] element in a foot, it undergoes Laxing. Postcyclically, the words are combined and syllable structure is readjusted so that the /t/ is resyllabified as an onset, reflecting a general preference for onsets over codas. At this stage, Voicing is applicable and we get a flap. For a tease, the /t/ begins in the second syllable. Hence at the cyclic stage of the derivation, there is no opportunity for Laxing to apply. Postcyclically, there is no pressure to resyllabify the onset /t/ as a coda. Even if the /t/ had undergone Laxing, Voicing would still be inapplicable. (64) s

H

w ’t

s i

w z

The main virtue of this approach is that it does not require ambisyllabicity. Kahn (1980) had proposed that flapping occurred when a consonant occurred in two syllables simultaneously: q

(65) p

/

q t

i

This argument has not survived the test of time. Hammond (1982) showed that the ambisyllabicity approach of Kahn (1980) extended to phrasal instances of flapping, while the foot-based approach did not, citing examples like go tomorrow [‘go7H’maro] vs. buy tomatoes [‘bajt h H’me7oz]. Under appropriate phrasal conditions, flapping can apply to examples where the /t/ begins the second word. There is also substantial psycholinguistic evidence for ambisyllabicity (Treiman and Zukowski 1990; Kessler and Treiman 1997; Treiman and Danis 1988), so eliminating it from linguistic representations is not an obvious desideratum.

5.2

Poetic meter

Another initial argument for the foot came from poetic meter. Kiparsky (1977) argued that constraints on meter occasionally require reference to foot constituency. Subsequently, Hayes (1983) showed that this foot constituency was not necessary, arguing for a purely grid-based theory of meter. Let us look at this argument a little more closely. Iambic pentameter in English is characterized by lines with 10 syllables, where even-numbered syllables are “strong” and odd-numbered syllables are “weak.” Traditionally, such a line is seen as a sequence of five iambic feet. For example (Shakespeare, Sonnet 1): (66) w To

s eat

w s the world’s

w due,

s by

w the

s grave

w and

s thee

The Foot

21

The effect of the division of syllables into strong and weak is that poets can restrict the distribution of stresses in these positions. In English, poets restrict stressed syllables in weak positions; strong positions are unrestricted.11 Kiparsky argues that constraints on the distribution of stresses in weak position refer to constituency, including foot constituency. For example, he argues that Milton’s verse is subject to the following constraint, which he labels “Milton I.” (67) *w s

where s is the strongest syllable of its phrase. Here, the tree above is the line structure and the tree below is the actual prosodic structure of the line. A constraint like this rules out a line like the following for Milton (though it is well formed for Shakespeare, Sonnet 7). (68) w Rew

s sembs

w ling w s

s w strong youth w s

s in w

w his w

s mids

s

w dle w

s age s

w

w

s s

Here the word youth is strongest in its phrase, labeled s, and a right branch. On the upper side, this word occurs as the weak left branch of a foot. If the bracketing agrees, however, such a line is acceptable for Milton (Paradise Lost 4.556).

11

See Fabb and Halle (2008) for a recent comprehensive theory.

22

Michael Hammond

(69) w On w

s a w

w s Sun- beam, s w

w swift w

s as w

w a w

s shoots

s

w ing w

s Star s

w

s

s s s

Finally, examples like the following show that the element must be the strongest element of the phrase, and that the bracketing restrictions do not suffice of themselves (Paradise Regained 2.424). (70) w And w

s his w

w Son s

s Hes

w

w rod w

s plac’d ...

w on

s Ju-

w s dah’s Throne

s s

Hayes (1983) argues that references to foot constituency can be done away with if we define stress peaks over metrical grids and refer to higher-level prosodic constituency. His version of Milton I looks like this: (71)

Milton I (grid version) *Peak / [. . . __ ]phrase

A peak is defined in terms of the grid as a grid column that is higher than at least one of its neighbors. Let us now look at how this constraint separates the cases we have considered so far. Hayes represents grids in terms of a single symbol “x,” rather than numbers. In addition, he represents the line template as a simple single-level grid, rather than with nodes labeled “s” and “w.” For the line in (68), we would have this template: (72)

. x . x . x . x . x Resembling strong youth in his middle age

The Foot

23

In subsequent diagrams, we leave the template out. The relevant part of the illegal line in (68) is: (73)

x . x . x x [Resembling strong youth] in his middle age

The syllable youth is a peak, defined with respect to the preceding syllable. It is in a weak position and phrase final. Hence it is illegal by the revised constraint. The two legal cases we considered above involve mismatches that are not phrase final: (74)

a.

x . . x x [On a Sunbeam], swift as a shooting star

b.

x . . x x [And his Son Herod] plac’d on Judah’s Throne

Although Kiparsky (1977) was an important step forward in our understanding of meter, it would be fair to say that his argument for the foot from meter did not survive.

5.3

Reduplication

An additional class of arguments for feet come from prosodic morphology (McCarthy and Prince 1986, 1993). The key claim here is that the size and location of reduplication, infixation, and related morphological operations refer directly to prosodic units, including the metrical foot. In the remainder of this section, we review four arguments of this sort for the foot: reduplication, locus of infixation, minimal word constraints, and language games. Yidiny (Dixon 1977a, 1977b; Hayes 1982; Marantz 1982) offers an example where a foot is reduplicated to mark the plural.12 (75)

Singular

Plural

mulari gindalba

mulamulari gindalgindalba

‘initiated man’ ‘kind of lizard’

The first two syllables of the word are reduplicated; there is extensive evidence that stress is assigned with binary feet in Yidiny and that the first two syllables of words like these would be footed together.13

12 13

For more general discussion of reduplication, see chapter 100: reduplication. The stress system of Yidiny is complex; see Hayes (1982) for further discussion.

Michael Hammond

24

5.4

Locus of infixation

The position of infixation can also be sensitive to feet. One of the most celebrated examples of this is expletive infixation in English (McCarthy 1982; Hammond 1997, 1999). Feet in English are clearly binary and left headed. There are a number of complications involving quantity-sensitivity and various sorts of ternary patterns, but the basic left-headed binary nature of stress feet is clear (Liberman and Prince 1977; Hayes 1981; Halle and Vergnaud 1987).14 The English expletive fuckin’ can be inserted in the middle of a word, e.g. in fantastic, producing fan-fuckin’-tastic, to indicate emphasis. This is relevant in the present context because the expletive must occur between two feet and cannot interrupt a foot. Thus *fantas-fuckin’-tic is illegal. It then follows that if a word has only one stress – and therefore only one foot – it cannot undergo expletive infixation. In the following examples, stresses are marked in the usual way and foot boundaries with square brackets. (76)

Word

Legal

Illegal

a[n’nounce] a[’genda]

– –

A[’meri]ca

–

a-fuckin’-nnounce a-fuckin’-genda agen-fuckin’-da A-fuckin’-merica Ame-fuckin’-rica Ameri-fuckin’-ca

With a two-syllable word with two stresses, expletive infixation is possible between the syllables: (77) Word [‘mun][’dane]

Legal

Illegal

mun-fuckin’-dane

–

With three-syllable words with two stresses, the position of the infix is precisely between the feet. (78)

Word

Legal

Illegal

[‘fan][’tastic] [‘Tenne][s’see] a[‘long][’side]

fan-fuckin’-tastic Tenne-fuckin’-ssee along-fuckin’-side

fantas-fuckin’-tic Te-fuckin’-nnessee a-fuckin’-longside

There are no monomorphemic examples of the third sort – like alongside – so these are confounded with morphological effects. Longer examples behave as expected. Interestingly, if there are more than two feet, many speakers find multiple infixation sites acceptable.

14

Though see Burzio (1994) for another view.

The Foot (79)

Word

Legal

Illegal

[‘Minne][’sota]

Minne-fuckin’-sota

[‘Tim][‘buk][’tu]

Tim-fuckin’-buktu Timbuk-fuckin’-tu Hali-fuckin’-carnassus Halicar-fuckin’-nassus Apa-fuckin’-lachicola Apalachi-fuckin’-cola

Mi-fuckin’-nnesota Minneso-fuckin’-ta –

[‘Hali][‘car][’nassus] [‘Apa][‘lachi][’cola]

25

Ha-fuckin’-licarnassus Halicarna-fuckin’-ssus A-fuckin’-palachicola Apala-fuckin’-chicola Apalachico-fuckin’-la

Strikingly, there are multiple infixation possibilities just in case we find two medial stressless syllables in a row. This follows directly from the claim that feet in English are binary. (80)

Word

Legal

Illegal

[‘Winne]pe[’saukee]

Winne-fuckin’-pesaukee Winnepe-fuckin’-saukee Kala-fuckin’-mazoo Kalama-fuckin’-zoo

Wi-fuckin’-nnepesaukee Winnepesau-fuckin’-kee Ka-fuckin’-lamazoo

[‘Kala]ma[’zoo]

The second stressless syllable is affiliated with neither of the adjacent feet allowing the infix to be positioned to either side of it, still satisfying the requirement that there be feet to each side and that the primary stress follows. There are additional complications to the system (Hammond 1997, 1999). First, the main stress cannot precede the infix. Thus ‘Kalama-fuckin’-’zoo is decidedly better than ’catama-fuckin’-‘ran. In addition, if the syllable preceding the infix is stressed, it must be at least bimoraic. Hence, mun-fuckin’-dane [‘mZn‘fZkHn’den] is better than ra-fuckin’-ccoon [‘ræ‘fZkHn’kun]. Those complications notwithstanding, the locus of infixation provides additional evidence for foot constituency.

5.5

Minimum word size

Lardil (Wilkinson 1988) provides a nice example of a minimum word constraint based on the foot: words in Lardil must have at least two vowels. If they do not, then they are augmented to meet this target with an epenthetic [a]. This provides for alternations in the shape of the stem depending on whether it is suffixed or not; an unsuffixed sub-minimal stem is augmented. Verbs with at least two syllables are inflected as follows: (81) underlying uninflected non-future future

‘tree’

‘dugong’

‘beach’

‘inside’

/}uIal/ }uIal }uIalin }uIaluÈ

/kentapal/ kentapal kentapalin kentapaluÈ

/kela/ kela kelan kelaÈ

/wiÍe/ wiÍe wiÍen wiÍeÈ

Monosyllabic consonant-final roots with long vowels behave in similar fashion.

Michael Hammond

26 (82)

underlying uninflected non-future future

‘ti-tree sp.’

‘spear gen.’

/pee7/ pee7 pee7in pee7uÈ

/maaK / maaK maaKin maaKkuÈ

However, nouns with only a single vowel get augmented when uninflected. (83) underlying uninflected non-future future

‘thigh’

‘shade’

/te7/ te7a te7in te7uÈ

/wik/ wika wikin wikuÈ

The two-vowel target can be seen as foot based if we treat Lardil as Hayes (1980) treated Southern Paiute: each vowel element is a potential terminal element for footing. Alternatively, if we view vowels as the sole bearers of moras in Lardil, we can view this as a bimoraic target, which was later proposed to be a foot.15

5.6

Language games

Hammond (1990) discusses a language game in English that provides further evidence for the foot. The game is played by substituting names into the following rhyme. (84)

Jack, Jack bo back [–æk –æk bo bæk] banana fana fo fack [bHnænH fæ nH fo fæk] me my mo mack [mi maj mo mæk] Ja–ack [–æ–æk]

— , — , bo b — banana fana fo f — me my mo m — —

The onset of the name undergoes various substitutions not relevant here. The relevant point here is that the name must fit a particular prosodic template: from one to three syllables, where the first syllable is stressed and any subsequent syllables are stressless. This corresponds to a single left-headed binary foot plus an optional extrametrical syllable. Marking feet with square brackets and the extrametrical syllable with angled brackets, we get a clear difference between names that are acceptable and those that are not. (85)

15

Permissible

Impermissible

[’Jack] [’To] [’Jenni] [’Gwendo]

A[n’nette] [‘Isa][’do] [’Mira][‘beil] O[’livi]

Garrett (1999) argues, though, that word minimality is not connected to foot structure.

The Foot

27

The game thus provides corroborating evidence for the role of foot constituency in phonology.

6

The typology of feet

Hayes (1987) proposes a radically different foot inventory. Rather than a symmetric parametric system, Hayes develops a non-parametric asymmetric system with only three basic feet: the syllabic trochee, the moraic trochee, and the iamb. This development and the subsequent responses is sufficiently important that it is treated in a chapter of its own: chapter 44: the iambic–trochaic law.

7

Some subsequent proposals

Kager (1993) offers a reformulation of the asymmetric typology that maintains a symmetric foot system, and derives surface quantitative asymmetries from syllable structure. Consider, for example, the asymmetry between an iambic foot and a moraic trochee. The iambic foot can contain a bimoraic right element, but the moraic trochee cannot. Kager argues that this follows from two independently required observations. First, when a heavy/bimoraic syllable bears stress, it is the first mora that does so. Second, languages avoid lapses, two stressless elements in a row. We can see then that lengthening the right node of an iambic foot is well formed, but lengthening the left node of a trochaic foot is not, since the latter results in two stressless elements – moras – in a row. (Syllable boundaries are marked with square brackets here.) (86)

Iamb

Trochee

( . x) ( . x .) [[][[] → [[][[[]

(x .) (x . .) [[][[] → [[[][[]

An interesting advantage of invoking lapse like this is that the same principle can be used to rule out ternary feet. If we assume that the head of a foot must be peripheral, then either sort of ternary foot would result in a lapse. (87) Left-headed (x . . ) q q q

Right-headed ( . . x) q q q

Gordon (2002) offers an OT analysis of quantity-insensitive stress. This includes systems with a single peripheral stress and systems with binary and ternary patterns of iteration. The paper is important for several reasons. First, it brought quantity-insensitive systems back into the theoretical discourse that had focused on quantitative asymmetries for several years. Second, it offered a rigorous application of standard OT constraints to a broad cross-section of languages. The logic of the approach is as follows. First, there are several Align constraints that put stresses on the edges of words. In addition, there is the Non-finality constraint, the OT analog of extrametricality.

28

Michael Hammond

Iterative footing is accomplished with various versions of Clash and Lapse constraints. Let us look at Cayuvava (Key 1961, 1967) to see how the system works. (88)

’eJe ’œakahe ki’hibere ari’uuŒa ‘–ihira’riama ma‘rahaha’eiki ikit‘apare’repeha ‘Œaadi‘roboßu’ruruŒe me‘daruŒe‘Œeiro’hiiJe Œaa‘dairo‘boiro’hiiJe

‘tail’ ‘stomach’ ‘I ran’ ‘he came already’ ‘I must do’ ‘their blankets’ ‘the water is clean’ ‘99’ (1st digit) ‘15 each’ (2nd digit) ‘99’ (2nd digit)

Here are the key constraints Gordon assumes for this case, along with critical rankings: (89)

*ExtendedLapse, *Clash, Non-finality, Align(X2,R) >> Align(X1,L) >> Align(X1,R) >> ...

The *ExtendedLapse constraint rules out stresses more than two syllables apart; this does the work of ternary iteration. The *Clash constraint prevents stresses from being adjacent. Non-finality is final extrametricality. Align(X2 ,R) forces main stress on the right. Finally, the relative ranking of Align(X1,L) and Align(X1,R) forces iteration from the right. Let us now see how this works in the case of [ma‘rahaha’eiki]. (90)

/marahahaeiki/

*EL *Cl NF Al(X1,L) Al(X1,L) Al(X1,R)

☞ a. ma‘rahaha’eiki b. ‘marahaha’eiki

*!

c. ma’rahaha‘eiki d. marahaha’eiki e. ‘maraha’haeiki

*! *!* *!

5

7

4

8

5

7

4 3

2 9

Stresses are positioned at the right distance apart because of the interaction of *ExtendedLapse and *Clash. The former also forces stress to iterate in the first place. In the context of foot theory, the key observation is that the system does not require feet to describe QI stress patterns. As with Prince (1983), however, such an account leaves open how apparent patterns of foot-related prosodic morphology are accommodated. Finally, another interesting proposal is developed by Hyde (2002). The core of the proposal is that there is a metrical grid that is independent of footing. Feet are all binary and have heads, but foot heads need not bear stress. A further point of interest is that feet may overlap. These innovations allow Hyde to maintain exhaustive footing and develop a spare model of directional stress effects.

The Foot

8

29

Summary

There are a number of other important and interesting proposals regarding foot theory, but we have only been able to touch on some very salient ones here. What does the future hold? There are a number of threads that seem like promising avenues of development for feet. First, Optimality Theory has a huge effect on all of phonological theory, but the framework seems to be reaching its limits. There are several alternative versions of the framework that have developed – for example, Stochastic OT (Boersma 1997) and Harmonic OT (Smolensky and Legendre 2006) – but it is unclear where the general framework is going at this time. What OT has left us with is a clear refocusing of attention from structural elements of phonological representations to constraints on those representations. On such a view, feet per se no longer exist. Moreover, there is no one-to-one mapping of constraints and the foot inventory (necessarily). If this thread continues, we would reasonably expect less attention to large-scale foot effects and more attention on the constraints from which those effects derive. For example, we might expect more attention to constraints like *Clash or the Weight-to-Stress Principle. Given the shift in attention from strictly ranked constraints to other ranking algorithms, including probabilistic ranking, we would expect developments along the lines of probabilistic feet. Perhaps some metrical phenomena are best treated in terms of footing which is only probabilistic in nature. A word might not have a fixed structure, but a structure that is only partially fixed, e.g. locus of footing or headedness, might be indeterminate. One might use this to account for variation in stress or exceptions of various sorts, i.e. ternarity. Another direction of current research is increased attention to the extragrammatical features that impinge on the grammar, e.g. phonetics, perception, production, lexical access, and acquisition. It seems quite likely that both our understanding of the basic facts of footing and the theoretical frameworks we use to describe footing phenomena will change as these efforts expand our empirical focus. The role of quantity has already been discussed from this perspective, for example, by Hayes (1987) and Kager (1993). These are extremely sketchy predictions, however.

ACKNOWLEDGMENTS Thanks to Joyce McDonough, Diane Ohala, Adam Ussishkin, several anonymous reviewers, and the editors for useful discussion. All errors are my own.

REFERENCES Boersma, Paul. 1997. How we learn variation, optionality, and probability. Proceedings of the Institute of Phonetic Sciences of the University of Amsterdam 21. 43–58 (ROA-221). Boxwell, Helen & Maurice Boxwell. 1966. Weri phonemes. In S. A. Wurm (ed.) Papers in New Guinea linguistics No. 5, 77–93. (Pacific Linguistics A37.) Canberra: Australian National University.

30

Michael Hammond

Burzio, Luigi. 1994. Principles of English stress. Cambridge: Cambridge University Press. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Cole, Jennifer & John Coleman. 1992. No need for cyclicity in generative phonology. Papers from the Annual Regional Meeting, Chicago Linguistic Society 28. 36 –50. Dixon, R. M. W. 1977a. A grammar of YidiJ. Cambridge: Cambridge University Press. Dixon, R. M. W. 1977b. Some phonological rules in YidiJ. Linguistic Inquiry 8. 1–34. Echeverría, Max S. & Heles Contreras. 1965. Araucanian phonemics. International Journal of American Linguistics 31. 132–135. Fabb, Nigel & Morris Halle. 2008. Meter in poetry: A new theory. Cambridge: Cambridge University Press. Garrett, Edward, 1999. Minimal words aren’t minimal feet. UCLA Working Papers in Linguistics 1, Papers in Phonology 2. 68–105 (ROA-1031). Gordon, Matthew. 2002. A factorial typology of quantity-insensitive stress. Natural Language and Linguistic Theory 20. 491–552. Halle, Morris & Michael Kenstowicz. 1991. The Free Element Condition and cyclic versus noncyclic stress. Linguistic Inquiry 22. 457–501. Halle, Morris & Jean-Roger Vergnaud. 1978. Metrical structures in phonology. Unpublished ms., MIT. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: MIT Press. Hammond, Michael. 1982. Foot-domain rules and metrical locality. Proceedings of the West Coast Conference on Formal Linguistics 1. 207–218. Hammond, Michael. 1986. The obligatory-branching parameter in metrical theory. Natural Language and Linguistic Theory 4. 185 –228. Hammond, Michael. 1988. Constraining metrical theory: A modular theory of rhythm and destressing. New York: Garland. Hammond, Michael. 1989. Cyclic secondary stresses in English. Proceedings of the West Coast Conference on Formal Linguistics 8. 139 –153. Hammond, Michael. 1990. The Name Game and onset simplification. Phonology 7. 159–162. Hammond, Michael. 1997. Vowel quantity and syllabification in English. Language 73. 1–17. Hammond, Michael. 1999. The phonology of English: A prosodic optimality-theoretic approach. Oxford: Oxford University Press. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Published 1985, New York: Garland. Hayes, Bruce. 1982. Metrical structure as the organizing principle of YidiJ phonology. In Harry van der Hulst & Norval Smith (eds.) The structure of phonological representations, part 1, 97–110. Dordrecht: Foris. Hayes, Bruce. 1983. A grid-based theory of English meter. Linguistic Inquiry 14. 357–394. Hayes, Bruce. 1984. The phonology of rhythm in English. Linguistic Inquiry 15. 33–74. Hayes, Bruce. 1987. A revised parametric metrical theory. Papers from the Annual Meeting of the North East Linguistic Society 17. 274 –289. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313–359. Hyman, Larry M. 1977. On the nature of linguistic stress. In Larry M. Hyman (ed.) Studies in stress and accent, 37–82. Los Angeles: Department of Linguistics, University of Southern California. Idsardi, William J. 1992. The computation of prosody. Ph.D. dissertation, MIT. Kager, René. 1993. Alternatives to the iambic-trochaic law. Natural Language and Linguistic Theory 11. 381–432. Kahn, Daniel. 1980. Syllable-based generalizations in English phonology. New York: Garland. Kessler, Brett & Rebecca Treiman. 1997. Syllable structure and the distribution of phonemes in English syllables. Journal of Memory and Language 37. 295 –311.

The Foot

31

Key, Harold H. 1961. Phonotactics of Cayuvava. International Journal of American Linguistics 27. 143–150. Key, Harold H. 1967. Morphology of Cayuvava. The Hague: Mouton. Kiparsky, Paul. 1973. “Elsewhere” in phonology. In Stephen R. Anderson & Paul Kiparsky (eds.) A Festschrift for Morris Halle, 93–106. New York: Holt, Rinehart & Winston. Kiparsky, Paul. 1977. The rhythmic structure of English verse. Linguistic Inquiry 8. 189–247. Kiparsky, Paul. 1979. Metrical structure assignment is cyclic. Linguistic Inquiry 10. 421–441. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. Marantz, Alec. 1982. Re reduplication. Linguistic Inquiry 13. 435 –482. McCarthy, John J. 1979. Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT. McCarthy, John J. 1982. Prosodic structure and expletive infixation. Language 58. 574–590. McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. McCarthy, John J. & Alan Prince. 1993. Prosodic morphology I: Constraint interaction and satisfaction. Unpublished ms., University of Massachusetts, Amherst & Rutgers University. Osborn, Henry A. 1966. Warao I: Phonology and morphophonemics. International Journal of American Linguistics 32. 108–123. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19 –100. Rischel, Jørgen. 1972. Compound stress in Danish without a cycle. Annual Report of the Institute of Phonetics 6. 211–228. Sapir, Edward. 1930. Southern Paiute, a Shoshonean language. Proceedings of the American Academy of Arts and Sciences 65. 1–296. Sebeok, Thomas A. & Frances J. Ingemann. 1961. An Eastern Cheremis manual: Phonology, grammar, texts and glossary. Bloomington: Indiana University. Smolensky, Paul & Géraldine Legendre (eds.) 2006. The harmonic mind: from neural computation to optimality-theoretic grammar. Cambridge, MA: MIT Press. Street, John C. 1963. Khalkha structure. Bloomington: Indiana University & The Hague: Mouton. Treiman, Rebecca & Catalina Danis. 1988. Syllabification of intervocalic consonants. Journal of Memory and Language 27. 87–104. Treiman, Rebecca & Andrea Zukowski. 1990. Toward an understanding of English syllabification. Journal of Memory and Language 29. 66 –85. Tryon, Darrell T. 1970. An introduction to Maranungku (Northern Australia). (Pacific Linguistics B15.) Canberra: Australian National University. Voegelin, Charles F. 1935. Tübatulabal grammar. University of California Publications in American Archaeology and Ethnology 34. 55–189. Walker, Rachel. 1997. Mongolian stress, licensing and factorial typology. Unpublished ms., University of California, Santa Cruz (ROA-171). Wilkinson, Karina. 1988. Prosodic structure and Lardil phonology. Linguistic Inquiry 19. 325–334.

41

The Representation of Word Stress Ben Hermans

1

Introduction

Since the publication of Hayes (1985), the asymmetries between iambs and trochees have been a central theme in the literature on stress. Two types of asymmetries can be distinguished. One has to do with quantity. It has frequently been claimed that iambs do not allow a heavy syllable in the weak position, and require a heavy syllable in the strong position. Kager (1993) deals with asymmetries of this type on the basis of a theory whose central hypothesis is that feet are built over moras, rather than syllables. The recent literature, however, has shown that these “quantitative asymmetries” are not supported empirically. It is simply not true, for instance, that iambs invariably constrain the occurrence of heavy and light syllables in the way just described. A particularly convincing argument to this effect is given in Altshuler (2009), with respect to Osage. (See also chapter 44: the iambic–trochaic law and chapter 57: quantity-sensitivity.) There is a second class of asymmetries, however, which remains valid. These are “parsing asymmetries,” which have to do with the direction of foot construction in a word. Two authors have argued that such parsing asymmetries can only be explained if the representation of word stress is fundamentally changed. Interestingly, however, they disagree as to how it should be changed. While Gordon (2002) proposes to simplify the representation of word stress by eliminating foot structure completely, Hyde (2001, 2002, 2008) recognizes not only foot structure but, in addition, a new type of structure, the “overlapping foot,” thus complicating the representation of stress in order to account for the asymmetries. In this chapter I consider the ongoing debate about the representation of word stress from the perspective of parsing asymmetries. In §2, after presenting some of the most important asymmetries, I briefly sketch Gordon’s account, which is as simple as it is radical. In his view, asymmetries can be accounted for if foot structure is abolished. Word stress representations contain only gridmarks, as in Prince (1983) and Selkirk (1984). In the spirit of Gordon, then, we might say that feet are superfluous if we want to account for the distribution of stress in the words of the world’s languages. This raises the question of whether feet are necessary at all. Interestingly, if we broaden our scope to include other phenomena as well, the evidence for foot The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0041

The Representation of Word Stress

2

structure becomes overwhelming. In §3 I present some cases from the recent literature which support the existence of feet, thus suggesting that representations with only gridmarks are too impoverished. If foot structure does exist, then how do we account for the distribution of stress in the words of the world’s languages? Is it still reasonable to formulate the relevant constraints in terms of the grid only? Or must they be stated in terms of foot construction, with gridmarks playing only a marginal role? In §4 I present an overview of Hyde’s work, in which the claim is made that the distribution of stress can best be explained in terms of constraints regulating foot construction. Gridmarks only read off some of the basic properties of a word’s foot structure. This is a continuation of the tradition initiated by Liberman and Prince (1977). Other authoritative studies, such as Hayes (1984, 1995), have argued for the same idea, which also led to development of the “bracketed grid” notation (Halle and Vergnaud 1987). There are some important differences between the various “tree-cum-grid” theories, however. In §5 I give a brief overview of one issue where theories seem to differ. This concerns the status of headedness in foot structure. Is a foot inherently headed, even if it is not accompanied by a gridmark? Or is it the case that a foot is inherently headless unless there is a gridmark accompanying it, marking one syllable as the head? In the first approach, some or all of the properties of foot structure can simply be read off the grid. In this view foot structure is primary and the grid secondary. In the second approach, in which the grid is imposed on foot structure, the grid is primary and foot structure secondary.

2

Explaining parsing asymmetries without foot structure

Hyde (2001, 2002) shows that some non-existing systems can easily be derived with generally accepted foot inventories (see chapter 40: the foot). Let us consider three of these cases. The first one is the Australian language Garawa, which can be compared with what Hyde (2002: 329) calls “Anti-Garawa,” an unattested system. (1)

Garawa

Anti-Garawa

x x x (1 2)(3 4)(5 6) x x x (1 2) 3 (4 5)(6 7)

x (1 2)(3 x (1 2)(3

x x 4)(5 6) x x 4) 5 (6 7)

In these representations foot structure is indicated by round brackets and headedness is represented by gridmarks. Garawa can be derived with the following rules. One trochee is built at the left edge. Then trochees are built from right to left. Furthermore, degenerate feet are not allowed. In odd-parity words this creates a lapse following the stressed initial syllable. If we change just two ingredients of this system we derive a non-existing pattern. One iamb is built at the right edge, and then a series of iambs from left to right. We then derive a system in which odd-parity words contain a lapse before the final (stressed) syllable.

3

Ben Hermans

The second example is a pair consisting of the Australian language Pintupi, which can easily be derived by the rules of the theory, and Anti-Pintupi, which can be derived just as easily but does not exist. (2)

Pintupi

Anti-Pintupi

x x x (1 2)(3 4)(5 6) x x x (1 2)(3 4)(5 6) 7

x x x (1 2)(3 4)(5 6) x x x 1 (2 3)(4 5)(6 7)

In Pintupi, trochees are built from left to right. Degenerate feet are not allowed. In odd-parity words this creates a lapse at the right edge of a word. Anti-Pintupi can be derived just as easily, by constructing iambs from right to left, so that the lapse is located at the left edge. Finally, consider Piro, spoken in Brazil and Peru, and Anti-Piro, an impossible system. (3)

Piro

Anti-Piro

x x x (1 2)(3 4)(5 6) x x x (1 2)(3 4) 5 (6 7)

x x x (1 2)(3 4)(5 6) x x x (1 2) 3 (4 5)(6 7)

In Piro, one trochee is built at the right edge. Remaining trochees are built from left to right; degenerate feet are not allowed. In odd-parity words this creates a lapse before the stressed penultimate syllable. Anti-Piro can be derived by changing just two ingredients. One iamb is built at the left edge; remaining iambs are built from right to left. In odd-parity words this creates a lapse after the peninitial syllable. The three non-existent cases have in common that iambic structure refers to the right edge. In Anti-Garawa one iamb is constructed at the right edge, while in Anti-Pintupi and Anti-Piro foot construction starts at the right edge. One might suppose, then, that iambs cannot refer to a word’s right edge. This, however, is not true; languages where iambs are constructed from right to left do exist. One example is Suruwaha, spoken in Brazil (Hyde 2002: 320), which is the mirror image of Araucanian, spoken in Chile and Argentina (Hyde 2002: 320). In this language iambs are constructed from left to right. These systems are illustrated in (4): (4)

Araucanian x (1 2)(3 x (1 2)(3

x 4)(5 x 4)(5

Suruwaha x 6) x 6) 7

x x x (1 2)(3 4)(5 6) x x x x (1)(2 3)(4 5)(6 7)

In Araucanian the final syllable is unparsed in odd-parity words. In Suruwaha the first syllable of odd-parity words is assigned a degenerate foot.

The Representation of Word Stress

4

It seems as if these two languages employ two different strategies to achieve the same goal: a rhythmic pattern in which every other syllable is stressed. To achieve this, Araucanian leaves the final syllable unparsed; if this were not the case, a clash would be created at the right edge, disturbing the rhythmic alternation. In Suruwaha, on the other hand, a degenerate foot is constructed at the left edge. Otherwise, there would be a lapse at the left edge, disturbing the binary rhythmic pattern. If it is true that a binary rhythmic pattern is the primary goal, we might just as well express that target directly, without the mediation of feet. A comparison of Araucanian and Suruwaha, then, suggests that feet might be superfluous; all that seems to matter is rhythmic alternation. All the non-existing systems mentioned above share the same property; in an odd-parity word there is a lapse, entailing that the rhythmic alternation is disturbed. This again suggests that binary rhythm is the only relevant factor, which leads Gordon to adopt the strategy of expressing this directly, without foot structure. Essentially, then, the idea is that the “anti-systems” in (1)–(3) cannot exist because they do not realize the ideal of binary rhythm. The ideal of rhythmic alternation is expressed on the grid, using the following two constraints. (5)

a. b.

*Lapse (Gordon 2002: 502) A string of more than one consecutive stressless syllable may not occur. *Clash (Gordon 2002: 506) A stress domain may not contain adjacent stressed syllables.

Having established that the non-existing systems can be explained on the assumption that they lack the ideal rhythm in odd-parity words, the question arises of how we can account for systems which do exist, even though alternating rhythm is disturbed in odd-parity words. Let us turn first to Pintupi. To account for this language, Gordon makes use of Non-finality, a device that can be motivated independently (Prince and Smolensky 1993; see chapter 43: extrametricality and non-finality for a definition and discussion). In this view, Pintupi has alternating rhythm, but the final syllable is excluded from stress assignment. This becomes clear when we compare Pintupi with another Australian language, Maranungku. (6)

Pintupi x x 1234 x x 1234

Maranungku x 5 {6} x 5 6 {7}

x x 1234 x x 1234

x 56 x x 567

I have indicated the Non-finality requirement imposed on the final syllable with curly brackets (braces). The two representations in (6) demonstrate that there is only one difference between Pintupi and Maranungku. In the former, but not the latter, the final syllable cannot be stressed. This explains why Pintupi has a lapse at the end of the word. We can also see why Anti-Pintupi does not exist. There is no equivalent of Non-finality operating at the left edge; there is no Non-initiality. Therefore,

5

Ben Hermans

an initial lapse cannot be created. To see this consider again Anti-Pintupi and Surahawa. (7)

Anti-Pintupi

Suruwaha

x x x 123456 x x x 1234567

x x x 123456 x x x x 1234567

Anti-Pintupi has an initial lapse in odd-parity words. This lapse is automatically eliminated in favor of a binary rhythmic alternation. A gridmark is therefore inserted on the first syllable. This creates a system that does exist: Suruwaha. Before I go on to discuss the other asymmetries, we should note that, although Araucanian and Suruwaha both exhibit binary rhythm, there is still a difference between them. Recall from (4) that they can easily be distinguished by foot structure; Araucanian builds iambs from left to right, and Suruwaha from right to left. From the point of view of the grid-only approach, the difference resides in the number of gridmarks in odd-parity words; there are fewer marks in Araucanian than in Suruwaha. This can be explained with the use of Alignment constraints, according to Gordon (2002: 498). Every gridmark has to be aligned with the right or left edge, depending on the language, as determined by the constraint (family) Align(x1,L/R). Every syllable separating the gridmark from its designated edge adds a violation. Since this is true for every gridmark, the constraint automatically reduces the number of gridmarks. If Alignment is satisfied, one and one only gridmark is created, located at the left or right edge, depending on the language.1 In order to achieve rhythmic alternation, then, *Lapse must be ranked higher than Align(x1,L/R). I illustrate the minimization effect of Align(x1,L/R) in (8): (8)

Align(x1,L/R) 1234567 ☞ a. x x x *,***,***** (9) 1234567 b. x x x x **,****,****!** (12) 1234567

In Araucanian, Align(x1,L/R) has maximal effect, reducing the number of gridmarks. *Lapse, however, must be satisfied. The result is that the minimal number of Such systems do exist. They are analyzed with the ranking Align(x1,L/R) >> *Lapse. These are languages where, in principle, one stress is created at the left/right edge of a word. In older theories, these languages were accounted for with unbounded feet (e.g. Hayes 1980; Hammond 1984; see Kager 1995 for an overview). Since Prince (1985) there has been general agreement that unbounded feet do not exist. The most common force creating additional stresses in languages that tend to prefer just one stress in each word is Weight-to-Stress (Prince 1983; Prince and Smolensky 1993; Kager 1999). If this constraint is highly ranked, a language has one stress in each word, unless a word contains one or more heavy syllables. For a typology of unbounded systems where heaviness does not play a role, see Gordon (2002). For a typology of unbounded systems where weight does play a role, see Hyde (2001, 2006).

1

The Representation of Word Stress

6

gridmarks is inserted, such that no violation of *Lapse is created. To account for the fact that the last syllable is stressed in even-parity words, Align(x1,R) must dominate Align(x1,L) in Araucanian. This is shown in (9). (9)

Align(x1,R) 123456 ☞ a. x x x **,**** (6) 123456 b. x x x *,***,***!** (9) 123456

Align(x1,L) *,***,***** (9) **,**** (6)

To account for the fact that the number of gridmarks is maximized in odd-parity words in Suruwaha, Gordon proposes AlignEdges, which demands that a syllable located at the edge must have a gridmark (Gordon 2002: 497). In Suruwaha, this constraint must dominate Align(x1,R), which explains why four gridmarks are created in a seven-syllable word, rather than just three, as in Araucanian. Notice that AlignEdges also requires a gridmark on the initial syllable in the even-parity words of Suruwaha. However, in this environment, a gridmark on the first syllable either creates a clash, or a lapse, as shown in (10). (10)

xx x x 123456 ↑ clash

x x x 123456 ↑ lapse

Both *Clash and *Lapse dominate AlignEdges in Suruwaha. This explains why the final, rather than the initial, syllable is stressed in even-parity words. This is the basic system proposed by Gordon. I will now demonstrate that it cannot derive Anti-Garawa, which is given again in (11), but without foot structure. (11)

Anti-Garawa x 123 x 123

x x 456 x x 4567

To guarantee that the final syllable is stressed, AlignEdges must be high-ranked. However, in even-parity words the initial syllable should not be stressed by AlignEdges. This constraint must therefore be lower-ranked than *Clash and *Lapse, just as in Suruwaha. To ensure that in even-parity words the final syllable is stressed, we also have to rank Align(x1,R) above Align(x1,L), again just as in Suruwaha. Interestingly, with the two rankings *Clash, *Lapse >> AlignEdges and Align(x1,R) >> Align(x1,L) it is impossible to derive the pattern in (11), which contains a lapse before the last stressed syllable in odd-parity words. This pattern violates AlignEdges and *Lapse. Inserting a gridmark on the initial syllable solves both violations, giving (12):

7 (12)

Ben Hermans Anti-Garawa resolved (Suruwaha) x x x 123456 x x x x 1234567

Insertion of the initial gridmark creates a system with a binary rhythm. This is a possible system, which is exemplified by Suruwaha. The same reasoning applies to Anti-Piro, whose pattern is given again in (13), without foot structure. (13)

Anti-Piro x x x 123456 x x x 1234567

The presence of stress on the final syllable indicates that AlignEdges is high-ranked, as in Suruwaha. To ensure that in even-parity words the final syllable is stressed, Align(x1,R) must dominate Align(x1,L). With these rankings, it is impossible to derive the stress pattern of odd-parity words in (13). Again, this pattern has a lapse, and violates AlignEdges. Both violations can be eliminated by inserting a gridmark over the first syllable, and by moving the gridmark of the second syllable to the third, as shown in (14). (14)

Anti-Piro resolved (Suruwaha) x x x 123456 x x x x 1234567

Again this pattern is actually attested, and is exemplified by Suruwaha. I have shown in this section that grid-only frameworks can easily explain the asymmetries occurring in binary rhythmic patterns. They do so in a way that is as simple as it is radical. Foot structure is eliminated, and only stresses are preserved, represented as gridmarks. As far as the distribution of stress is concerned, grid-only approaches do rather well. Surely this explains why the grid-only approach is still adopted (see for recent applications Karvonen 2005, 2008; also Gordon 2003: 179, note 4, where he confidently states that “a grid-based theory of stress offers a closer fit to the typology of stress than foot-based metrical theories”). It seems, then, that grid-only approaches are sufficient, and that foot structure is therefore superfluous. Interestingly, however, if we go beyond the distribution of stress, and broaden our scope to other phenomena, we find abundant evidence for feet. In the next section I will present a few arguments in favor of foot structure. Then, in §4, I will investigate what this means for the theory of stress.

The Representation of Word Stress

3

8

Evidence for foot structure

In this section I present evidence for the existence of foot structure. First I demonstrate that, in a sequence of two unstressed syllables, not every syllable behaves identically. Then I show that foot structure is necessary to define the domains within which certain phenomena apply. The third type of evidence for foot structure has to do with weight. (For further discussion of foot structure see chapter 39: stress: phonotactic and phonetic evidence and chapter 40: the foot.)

3.1

Unstressed syllables do not behave identically in lapses

De Lacy (2002, 2007a) shows that in Ayutla Mixtec, spoken in Mexico, there is a close relation between the position of stress and tonal quality. One generalization is that the leftmost high tone receives word stress. This is illustrated by the following examples: (15)

ML’H L’HH LM’H

[kenù’rá] [sù’t jáí] [sùt ja’í]

‘his tobacco’ ‘I will swim’ ‘I will not swim’

If all tones in a word are the same, the first syllable becomes stressed: (16)

’HHH ’LLL ’MMM

[’œínírá] [’œàtùì] [’œcJera]

‘he understands’ ‘my trousers’ ‘his pineapple’

Prince (1983) establishes that patterns of this type pose no problems for the grid-only approach, as they are an instance of what he calls the “default-to-sameside” pattern.2 Interestingly, however, Ayutla Mixtec has yet another pattern, which turns out to be quite problematic. If a word has a sequence of a high tone immediately followed by a low tone, then the high tone must be stressed, even if it is preceded by another high tone. If there is more than one such sequence, then the high tone of the first sequence is stressed. This is illustrated in (17). (17)

H’HL LMH’HL ’HLHL

[lú’lúrà] [vìœcí’ráà] [’œáàœíì?]

‘he is small’ ‘he is not cold’ ‘is not eating’

The low tone attracts stress to the high tone immediately to its left. To explain this pattern, de Lacy proposes that HL is parsed as a trochee, (HL). Thus, in a word with the structure HHL, as in the first example in (17), the parse H(HL) is better than (HH)L. The result is that a high tone immediately preceding a low 2

In systems where “default-to-same-side” obtains, a gridmark is assigned to the syllables with a certain property (in this case the syllables with a high tone); one End-Rule applies at a low level and another End-Rule at a higher level. In Ayutla Mixtec, both End-Rules apply at the left.

9

Ben Hermans

tone receives word stress, because it is the head of a trochee. With a grid-only approach it is difficult to explain these facts. It seems as if an unstressed syllable with a low tone behaves differently from an unstressed syllable with a high tone; the former but not the latter seems to avoid a lapse. By its very nature, the grid structure of an unstressed syllable with a low tone is identical to the grid structure of an unstressed syllable with a high tone. This indicates that a grid-only framework cannot account for these facts.3 I now turn to the second type of evidence for foot structure: the proper characterization of domains.

3.2

The characterization of domains

A stressed syllable and a neighboring unstressed syllable often form a domain within which a phenomenon applies. A grid-only approach is notoriously bad at defining domains of this type, as it can only define a primary stressed position, secondary stressed positions, and unstressed positions. It is impossible to express the fact that some unit creates a bond with another unit.4 An example showing this is provided by Guugu Yimidhirr, a language spoken in Australia (Zoll 2004; Elías-Ulloa 2006). In this language long vowels can only occur in the first and/or the second syllable of a word, but nowhere else. Some illustrative examples, from Elías-Ulloa (2006: 231–232), are given in (18): (18)

a.

b.

c.

’gu(gu ’bu(ra‘jaj ’da(ba‘IalIa‘la ma’gi(l ma’ji(Iu ma’gi(lnaj‘gu ’bu(ra(j ’dji(ra(l‘gal ’bu(ra(j‘bigu

‘language-abs’ ‘water-loc’ ‘ask-red-imp’ ‘branch-abs’ ‘food-purp’ ‘branch-pl-emph’ ‘water-abs’ ‘wife-adess’ ‘water-loc-emph’

In a theory that recognizes feet, it is easy to characterize the domain within which the long vowels can occur; it is the initial foot, which is also the head of the word. This foot can either be a trochee or an iamb, depending on the presence and location of a long vowel. On the other hand, in a grid-only account it is difficult to understand why long vowels are restricted to the first two syllables of the word. This is a consequence of the fact that in this approach the first two syllables cannot be characterized as a domain. The facts illustrated in (18) are therefore 3

Other phenomena showing that not all syllables behave in the same way in a lapse are high vowel deletion in Old English (Dresher and Lahiri 1991) and vowel balance effects in Scandinavian (Bye 1996) and Old Frisian (Smith 2004; Smith and van Leyden 2007). Unfortunately, due to lack of space I cannot discuss these phenomena here. 4 The domain-defining character of the foot is the oldest evidence in favor of its existence (Selkirk 1980). It turns out to be very difficult to find cases where it is absolutely impossible to define a domain with grids only. An interesting example is provided by Pearce (2006), who argues that feet in Kera create the domains for tone association. I suspect, however, that an alternative is possible with gridmarks only, such that (certain) tones tend to anchor to strong positions on the grid. The role of the foot as a domain delineator is systematically eliminated in Majors (1998). She argues that feet do not play a role in stress-dependent harmony.

The Representation of Word Stress

10

problematic for a grid-only framework. Let us now turn to the third type of evidence: syllable weight.

3.3

Evidence from syllable weight

Prince (1983) shows that the grid-only framework can explain the relationship between syllable weight and the position of stress. This seems to suggest that the effect which weight has on the position of stress can be described without making use of foot structure. However, once we broaden our view beyond the distribution of stress in the strict sense, it becomes obvious that weight does provide us with convincing evidence for foot structure. Here I present one case: allomorph selection in Shipibo, spoken in Brazil and Peru.5 Shipibo has two allomorphs meaning ‘again’: /ribi(/ and /ri(ba/. The allomorph /ribi(/ has a long vowel in the second syllable, /ri(ba/ in the first syllable. ElíasUlloa (2006) shows that allomorph selection is determined by foot structure. The language has two different feet: moraic trochees and iambs. The former is preferred; iambs are only built if the construction of moraic trochees is not possible. Furthermore, both moraic trochees and iambs must be bisyllabic. Two other high-ranking constraints are relevant here. The constraint Weight-toStress is inviolable, so a heavy syllable cannot be left unparsed. There is also one Alignment constraint that is high-ranked: the left edge of a word must be aligned with the left edge of a foot. Consider the following forms (from ElíasUlloa 2006: 7), where the relevant allomorphs are underlined: (19)

Allomorph selection in Shipibo a. b. c. d.

(pi-’ri()(ba-kq) (’puta)(ri‘bi()-kq (’puta)(-ma-‘ri()(ba-kq) (’puta)(-‘jama)(-ri‘bi()-kq

‘eat-again-past’ ‘throw-again-past’ ‘throw-caus-again-past’ ‘throw-neg-again-past’

In (19a), /ri(ba/ is selected, allowing all syllables of the word to be parsed into feet. If the other allomorph had been selected, the form /pi-ribi(-kq/ would have been created. This form cannot be correctly parsed into foot structure. One possible parse is (’pi-ri)(‘bi(-kq), but this representation contains an uneven trochee, which is not allowed in Shipibo. Another realization might be pi-(ri’bi()-kq. But this structure violates the Alignment constraint, which is not possible either. In (19b), /ribi(/ is selected; this is again explained by foot structure. If the other allomorph had been selected, a form would have been created that could not be properly parsed, viz. /puta-ri(ba-kq/. The parse (’puta)(-‘ri(ba)-kq is unacceptable, because it contains an uneven trochee. The alternative (’puta)(-‘ri()(‘ba-kq) is also bad, because it contains a monosyllabic iamb, a type of foot that is non-existent in this language. Yet another alternative would be pu(ta-’ri()(‘ba-kq), but this representation violates the requirement that a word should begin with the left edge of a foot. This constraint is very highly ranked in Shipibo. The form that is actually realized, (19b), does not suffer from any of these problems. It is parsed 5

Other cases are reduplication in Kosraean (Kennedy 2005) and tonal spread in Capanahua (Hagberg 1993).

11

Ben Hermans

into appropriate feet and only the final syllable is left unparsed. The same can be said for (19c) and (19d). We see, then, that foot structure explains this instance of allomorph selection. In the grid-only framework the distribution of the two allomorphs remains a puzzle. Consider again (19a), or rather its alternative, where the wrong allomorph is selected. This is the form /pi-ribi(-kq/. In a theory that only has gridmarks it is possible to parse this form as in (20): (20)

x x x x x x x pi - ri bi( - kq

There seems to be no particular reason why this form should be rejected. There is a smooth rhythmic alternation, and the heavy syllable is stressed. In a grid-only framework, then, no constraint is able to eliminate this form. The reason, of course, is that there are no feet in this approach. It therefore lacks the essential ingredients with which the distribution of the two allomorphs in Shipibo can be explained. In this section I have presented three types of evidence for foot structure: the phonology of lapses, the characterization of domains, and the interaction between weight and foot parsing. On the basis of this evidence we can conclude that foot structure does exist. Gordon’s approach might be economical, and also explain asymmetries, but its lack of foot structure makes it inadequate. Consequently, we are forced to explain the asymmetries in a theory of stress that does include foot structure. A priori, this looks like an almost impossible task, because, as we saw in §2, it is precisely the presence of foot structure that makes it so difficult to explain why certain systems do not exist. In the next section I will present a brief overview of the most important aspects of Hyde’s (2001, 2002) work, where an account is offered of the asymmetries in a theory that does include foot structure.

4

Explaining parsing asymmetries with foot structure

One of the central ideas in Hyde (2001, 2002) is that iambs and trochees are derived by word-edge Alignment constraints; there are no separate constraints of the type Foot=Iamb and/or Foot=Trochee. In particular, the following two constraints are important: (21)

a.

b.

Hds-L The left edge of every foot-head is aligned with the left edge of some prosodic word. Hds-R The right edge of every foot-head is aligned with the right edge of some prosodic word.

If Hds-L is high-ranked, then a word is parsed with trochees, while if Hds-R is high-ranked, it is parsed with iambs. This is similar to Gordon’s approach.

The Representation of Word Stress

12

The difference is that in Gordon’s theory gridmarks are subject to Alignment, whereas in Hyde’s approach foot-heads undergo Alignment. I demonstrate the effects of Alignment with the tableau in (22). I use Hyde’s notation; vertical lines indicate headedness. (22)

qqqqqq ☞ a. q q q q q q

Hds-L

Hds-R

**,**** (6)

*,***,***** (9)

*,***,***!** (9)

**,**** (6)

b. q q q q q q

In Hyde’s framework all syllables must be dominated by a foot. There is no such thing as weak layering (Itô and Mester 1992). Absence of weak layering triggers foot construction, and Alignment maximally reduces the number of feet. In principle, then, a trochee can only be created by Hds-L. Likewise, iambs can only be created by Hds-R. From this it follows that there is only a two-way contrast. In classical foot theory there is a four-way contrast: in principle, iambs as well as trochees can be right-aligned as well as left-aligned. The reduction to a two-way contrast is the basis of Hyde’s explanation of the asymmetry problem. Hyde’s theory has two other important ingredients. The first one is a constraint, GridmarkMapping, which requires that feet contain a gridmark, which must occupy some head position. Mapping is illustrated in (23). (23)

Trochees

Iambs

qqqqqq x x x

qqqqqq x x x

The second important ingredient is the idea that feet can overlap. This means that the following configuration is allowed: (24)

Overlapping feet

qqq x Because of the presence of the gridmark in the head position of the first foot, both feet satisfy GridmarkMapping. This is because the gridmark is located in the domain of both feet. Hyde is able to explain the minimization patterns by making use of overlapping feet. Recall that these are the patterns where the odd-parity words have relatively few stresses. (25) illustrates this with Araucanian and its mirror image Nengone (Hyde 2002: 320), a language of New Caledonia.

13 (25)

Ben Hermans Nengone

Araucanian

1234 56 x x x

1 2 3456 x x x

1234 567 x x x

1 23456 7 x x x

Nengone has trochees. Its foot-heads are therefore left-aligned. In odd-parity words a sequence of two adjacent foot-heads is created at the left edge. This is the consequence of maximal satisfaction of Hds-L. Inserting one gridmark in the domain of the two feet satisfies GridmarkMapping. Consequently, only three stresses are generated, even though there are four feet. Araucanian is the mirror image of Nengone. Maximal satisfaction of Hds-R creates iambs, and in odd-parity words there is a sequence of two adjacent footheads at the right edge. The two last feet satisfy GridmarkMapping with only one gridmark. The introduction of overlapping feet, then, makes it possible to account for the minimization patterns. The maximization patterns, where odd-parity words receive more stresses than is necessary to create a rhythmic alternation, are explained with two additional constraints, given in (26). (26)

a.

b.

PrWd-L The left edge of every prosodic word is aligned with the left edge of some foot-head. PrWd-R The right edge of every prosodic word is aligned with the right edge of some foot-head.

I illustrate these constraints with Maranungku and Suruwaha. (27)

Maranungku

Suruwaha

1234 56 x x x

1 2 345 6 x x x

12 34567 x x x x

123 4 5 67 x x x x

In Maranungku, the left edge of a prosodic word must be aligned with a foot-head. A trochee is therefore built at the left edge. The heads of all other feet must be aligned with the right edge of a word. This would normally mean that iambs would have to be built, but Hyde stipulates that two adjacent feet can never dominate two adjacent unstressed syllables. This excludes the following structure:

The Representation of Word Stress (28)

14

Excludedbybystipulation stipulation Excluded

1 234 The implication of this stipulation is that if PrWd-L, which requires a trochee at the left edge, is high-ranked, then the other feet cannot be iambic, even though iambs better satisfy right-alignment. The mirror image is also true, of course; if PrWd-R, which requires an iamb at the right edge, is high-ranked, then the other feet cannot be trochaic, even though trochees better satisfy left-alignment. Maranungku has the ranking PrWd-L >> Hds-R, which derives the representation on the left in (27). This is an instance of maximization, because, in an odd-parity word, as many gridmarks are present as there are feet. Suruwaha has the ranking PrWd-R >> Hds-L, deriving the pattern on the right in (27). Again, in the odd-parity words the number of stresses equals the number of feet. Maranungku has traditionally been analyzed as a language with trochees built from left to right, with a degenerate foot at the right. Suruwaha has been described as its mirror image, with iambs built from right to left and a degenerate foot at the left (cf. (4)). In principle, every foot receives a gridmark, although, due to overlapping feet, this does not necessarily mean that there are as many gridmarks as there are feet. At the right edge, Non-finality might exclude the presence of a gridmark over the last syllable. Furthermore, at the left edge, the first syllable can be subject to a constraint called InitialGridmark (Hyde 2002: 320), which requires the presence of a gridmark on the first syllable of a word. If these constraints are high-ranked, this can lead to a situation where a foot is not accompanied by a gridmark. Pintupi has high-ranking Non-finality, while Garawa has high-ranking InitialGridmark, as illustrated in (29). (29)

Garawa

Pintupi

1234 56 x x x

1 2 3 4 56 x x x

1234 567 x x x

123 4567 x x x

Garawa is like Nengone (cf. (25)) in the sense that feet are attracted to the left by high-ranking Hds-L. There are also two differences. In Garawa, InitialGridmark is high-ranked, so the initial syllable must have stress. The constraint against clashing gridmarks is also high-ranked, excluding an immediately following gridmark in the domain of the second foot. With these two constraints dominating GridmarkMapping, the first foot of the two overlapping feet is stressed, whereas the second foot is not. We thus get a stressless trochee. This is a foot with a head, but without a gridmark. The absence of a gridmark in the second foot creates a lapse immediately after the initial stress.

15

Ben Hermans

Pintupi is similar to Maranungku (cf. (27)), but has high-ranking Non-finality. If the final syllable is a foot-head, then it may not receive a gridmark. This creates a lapse at the right edge. With the system described here it is very easy to explain all the asymmetries mentioned in §2, which are listed again in (30). (30)

Anti-Garawa

Anti-Pintupi

Anti-Piro

x (1 2)(3 x (1 2)(3

x x x (1 2)(3 4)(5 6) x x x 1 (2 3)(4 5)(6 7)

x x x (1 2)(3 4)(5 6) x x x (1 2) 3 (4 5)(6 7)

x x 4)(5 6) x x 4) 5 (6 7)

In Anti-Garawa, one iamb is built at the right edge, and the other iambs are built from left to right, so that, in odd-parity words, a lapse is created before the final stress. In Hyde’s theory it is impossible to derive such a pattern. There are two ways to stress the final syllable of a word: either Hds-R or PrWd-R is high-ranked. With these two systems, it is impossible to create a lapse before the final stress. With high-ranking Hds-R, we derive Araucanian, as already shown in (25). This is a language with a minimization pattern, where in odd-parity words stresses are economically placed, so as to avoid a lapse. With high-ranking PrWd-R we get a system like Suruwaha, whose basic structure is given in (27). This is a maximization pattern, where an extra stress is created so as to avoid a lapse. Anti-Garawa, then, is ruled out, because it cannot be derived. There is also no place for Anti-Pintupi. The particular property of this imaginary system is that there is no stress on the initial syllable. There is only one way to block a stress on a peripheral syllable, Non-finality, which, however, can only block stress on a final syllable. Since there is no equivalent Non-initiality preventing stress from the initial syllable, Anti-Pintupi cannot be derived. Anti-Piro is also impossible. In this imaginary system, one iamb is built at the left edge, creating a fixed stress on the peninitial syllable. To place a fixed stress at the left periphery two constraints are available. Either Hds-L or PrWd-L must be high-ranked. In the former case we derive Nengone, whose basic configurations are shown in (25). Nengone is a minimization pattern, where the minimal number of stresses is economically placed so as to avoid a lapse. With highranking PrWd-L Maranungku is derived, as shown in (27). Maranungku displays a maximization pattern, where an additional stress is inserted so as to avoid a lapse. Thus there is no way to generate a system like Anti-Piro. Hyde’s system is similar to Gordon’s in one sense. Neither uses the classical alignment constraints that refer to feet. In particular, AllFeet-L/R and PrWdL/R (McCarthy and Prince 1993) are eliminated. (31)

a.

b.

AllFeet-L/R The left/right edge of every foot is aligned with the left/right edge of some prosodic word. PrWd-L/R The left/right edge of every prosodic word is aligned with the left/ right edge of some foot.

The Representation of Word Stress

16

In these constraints, Hyde replaces the argument foot by foot-head, as shown in (21) and (26), whereas Gordon replaces it by gridmark. These two changes are almost identical, since a foot-head normally has exactly the same distribution as a gridmark. The crucial difference, of course, is the notion of overlapping feet. We might say that, where Gordon eliminates foot structure entirely, Hyde introduces a new type of structure. Of course, Hyde is aware of the fact that this concept must be motivated on independent grounds, a task which he undertakes in Hyde (2008). Optimality Theory struggles with what Hyde calls the “odd-parity problem.” This problem can be divided into two sub-problems: the “even-only problem” and the “odd-heavy problem.” The introduction of overlapping feet provides a solution which is not available in standard approaches. For reasons of space I will only discuss the first instance of the odd-parity problem. The even-only problem is caused by the fact that, in odd-parity words, Faithfulness constraints are in conflict with two other constraints: FootBinarity (the constraint that penalizes degenerate feet) and Parse-q (the constraint requiring that syllables be dominated by feet). Suppose that the Faithfulness constraints are ranked below FootBinarity and Parse-q. Under this ranking it is better to insert or delete a syllable than to create a violation of either FootBinarity or Parse-q, as shown in the following tableau: (32)

1234567 a. (1 2)(3 4)(5 6) 7 b. (1 2)(3 4)(5 6)(7) ☞ c. (1 2)(3 4)(5 6)(7 8) ☞ d. (1 2)(3 4)(5 6)

Parse-q FootBinarity Max Dep *! *! * *

This tableau shows what happens to a word with an uneven number of syllables. The first candidate, (32a), contains an unparsed syllable, violating Parse-q. (32b) has a final, monosyllabic foot, violating FootBinarity. In (32c), a vowel is inserted. In this way an extra syllable is created, so that at the right edge a binary foot is built. FootBinarity and Parse-q are therefore satisfied, although Dep is violated. Finally, in (32d) a vowel is removed, so that both Parse-q and FootBinarity are again satisfied, although Max is violated. We can see, then, that in an odd-parity word Parse-q and FootBinarity can be satisfied if a syllable is inserted or deleted. From this it follows that there should be languages in which all words contain an even number of syllables. These languages would have the ranking Parse-q, FootBinarity >> Max, Dep. However, no language like this has ever been attested. This illustrates the phenomenon referred to as the even-only problem. Hyde (2008) shows that the even-only problem does not arise in a theory with overlapping feet. In such a theory, it is possible to satisfy FootBinarity and Parse-q without violating Faithfulness. In other words, with overlapping feet, no conflict arises between FootBinarity and Parse-q on the one hand and the two Faithfulness constraints on the other. This is shown in (33):

17 (33)

Ben Hermans 1234567

Parse-q FootBinarity Max Dep

☞ a. 1234567 b.

*! 12345678

c.

*! 123456

The candidate with overlapping feet, (33a), does not violate any constraint. Contrary to this, the two candidates in which the uneven number of syllables has been changed into an even number do violate Faithfulness. The candidate with overlapping feet thus harmonically bounds the two candidates where Faithfulness is violated. This means that the two constraints FootBinarity and Parse-q can never have the effect of changing an underlying form with an uneven number of syllables into a surface form with an even number. Overlapping feet, Hyde concludes, are the solution to the even-only problem. In this section I have shown that asymmetries can be explained with a representation of word stress that includes feet. Iambic and trochaic structure is not the result of the constraints Foot=Iamb and Foot=Trochee, but rather of the alignment of foot-heads. The patterns thus established are subject to GridmarkMapping. Syllables containing a gridmark are phonetically stressed. In representations with overlapping feet there is a tendency to minimize the number of gridmarks, because a sequence of two overlapping feet satisfies GridmarkMapping with one gridmark. Under the pressure of certain constraints (*Clash, as in Garawa, or NonFinality, as in Pintupi) it can happen that a foot ends up without a gridmark. Thus, in Hyde’s framework feet are not necessarily stressed, although they are always headed. In the next section I will briefly point out that Hyde is not very explicit about one property of his theory that is quite innovative. His theory is the only one, as far as I know, that makes use of feet that are headed, even though they are not stressed.

5

Stresslessness is not the same as headlessness

In Hyde’s framework feet are not necessarily stressed. In overlapping feet the number of stresses is a proper subset of the number of feet. Let us consider some instances of overlapping feet, and see how they interact with stress. (34)

Nengone

Araucanian

Garawa

qqq x

qqq x

qqq x

The Representation of Word Stress

18

Nengone has overlapping trochees, with a stressed syllable in the middle. Araucanian has overlapping iambs, also with stress in the middle (cf. (25)). Finally, Garawa has overlapping trochees, with a stress in initial position, as shown in (29).6 These representations indicate that headedness is not expressed by gridmarks. In Hyde’s framework a foot can be an iamb, even though it does not have a gridmark in its final syllable, as in Araucanian; similarly, a foot can be a trochee, even without a gridmark in its initial syllable, as in Nengone and Garawa. This is a unique aspect of Hyde’s theory. Most recent theories claim that headedness is expressed on the grid. Some theories claim that feet must have heads, so that for each foot there must be a gridmark accompanying it. Theories of this type invoke a principle like the Faithfulness Condition, originally formulated in Halle and Vergnaud (1987: 15–16).7 Meanwhile, another theory on the relation between foot structure and headedness has been developed. In this theory, headedness is disconnected from foot structure. In these theories there are two foot types; headed feet, which are accompanied by a gridmark, and headless feet, which are not accompanied by a gridmark. The loose connection between headedness and foot structure is expressed by the Separability Hypothesis (Crowhurst and Hewitt 1995b: 39), which states that feet can be headed or headless. Normally, a foot does have a head, but under the pressure of certain constraints, it can happen that a foot is not able to acquire a head. Proponents of the Separability Hypothesis are Hagberg (1993), Crowhurst and Hewitt (1995a, 1995b), Bye (1996), and Crowhurst (1996). Bye occupies a special position among them, because he assumes that feet can sometimes have two heads; when this happens there are two gridmarks in a single foot.8 It is clear that Hyde’s theory differs from both views just mentioned. On the one hand, it (implicitly) uses the Faithfulness Condition, because all feet are inherently headed. (Heads are not expressed on the grid, but by line structure, as we have already seen; a head has a vertical line, whereas a dependent has a slanted line.) Yet the theory also recognizes the Separability Hypothesis, in the sense that there is a separate mode of representation, the grid, where a gridmark may or may not accompany a foot. Hyde’s theory, therefore, does not recognize headless feet, but it does recognize stressed (headed) feet and unstressed (headed) feet. In this sense, Hyde’s theory is certainly representationally richer than all other recent theories. There is some evidence from Vogul, a language of Siberia, that feet can be headed, even if they are not stressed. In Vogul, stress does not seem to be quantitysensitive (Vaysman 2009). The main stress is realized on the initial syllable, and there are secondary stresses on every other syllable thereafter. If, however, the final syllable is a target for secondary stress, this stress is not realized. Syllable quantity is irrelevant for this distribution. Only syllables with long vowels are heavy: words with a heavy syllable have exactly the same stress patterns as words with only light syllables. These regularities are illustrated in (35). 6

The mirror image of Garawa (overlapping iambs with stress on the final syllable) does not exist. This system could only arise with a constraint requiring stress on the final syllable, together with high-ranking *Clash. However, in Hyde’s system there is no constraint requiring stress on the final syllable. Therefore, two overlapping iambs at the right edge will always have stress in the middle (the minimization pattern). 7 Halle and Vergnaud’s Faithfulness Condition should not be confused with the Faithfulness constraints of OT. The former is a condition on the relation between a foot and its head. 8 Level stress in Scandinavian dialects necessitates this representation, according to Bye (1996).

Ben Hermans

19 (35)

Stress in Vogul (Vaysman 2009: 207–242) No stress on the last syllable (short vowel in the first syllable) ’sam-e-nHl ‘his/her/its eye (abl)’ ’at-e-nHl ‘his/her/its smell (abl)’ b. Alternating stress (short vowel in the first syllable) ’sam-a‘:anHl ‘their (dual) eyes’ ’aki-‘:anHl ‘their (dual) uncles’ c. No stress on the last syllable (long vowel in the first syllable) ’joor-e-nHl ‘his/her/its strength (abl)’ ’aat-e-nHl ‘his/her/its hair (abl)’ d. Alternating stress (long vowel in the first syllable) ’saa:rap-‘e-nHl ‘his/her/its axe (abl)’ ’tootap-‘e-nHl ‘his/her/its chest (abl)’ a.

These facts seem to suggest that Vogul has trochees which are assigned without reference to quantity, i.e. quantity-insensitive trochees. A closer look at the language, however, reveals that this hypothesis cannot be maintained. Some affixes have a number of allomorphs. Vaysman convincingly shows that the selection of the allomorphs is determined by the prosody. The moraic trochee plays a decisive role in this process. One example is the morpheme ‘your (sg)’, which has no fewer than four allomorphs. After stems ending in a consonant, either [-an] or [-anHn] is selected, and after stems ending in a vowel, either [-n] or [-nHn]. The choice is narrowed down further by properties of the stem’s prosodic structure. If the initial syllable of the stem has a short vowel (that is, if it is a light syllable), then [-n] or [-an] is selected; if, however, the stem’s initial syllable is heavy, or if the stem is bisyllabic, then [-nHn] or [-anHn] is selected. These facts are illustrated in (36). (36)

Allomorphy in Vogul (Vaysman 2009: 207–242) a.

CVC-stem sam-an ‘your put-an ‘your pos-an ‘your b. CV(C)CVC-stem pasan-anHn ‘your api:-anHn ‘your isnas-anHn ‘your c. CVVC-stem saam-anHn ‘your puut-anHn ‘your oos-anHn ‘your 9 d. CVCV-stem ala-n ‘your pici-n ‘your oma-n ‘your

9

(sg) eyes’ (sg) ice-crusts’ (sg) lights’ (sg) table’ (sg) grandson’ (sg) windows’ (sg) corners’ (sg) cauldrons’ (sg) sheep’ (sg) roofs’ (sg) nests’ (sg) female relatives’

There are no CV-stems in Vogul, i.e. monosyllabic stems ending in a vowel.

The Representation of Word Stress e.

20

CVVCV-stem saali-nHn ‘your (sg) reindeer’ iici-nHn ‘your (sg) evenings’ oopa-nHn ‘your (sg) paternal grandfathers’

Vaysman argues that allomorph selection in Vogul can be explained if it is assumed that the language has moraic trochees. It has a type of foot, in other words, where the head is on the left, and where the number of moras is exactly two. Using moraic trochees we can see that an allomorph is selected in such a way that all syllables of a word are parsed. I demonstrate this in (37) with C-final stems, using Hyde’s notation. (37)

Exhaustive parsing in Vogul CVC-stem F q q sam-an

CV(C)CVC-stem F F q q q q pa san -an Hn

CVVC-stem F F q q q saam -an Hn

If the allomorphs were distributed differently, parsing could not be exhaustive, as shown in (38). (38)

Illicit non-exhaustive parsing in Vogul CVC-stem F q q q sam-an Hn

CV(C)CVC-stem F q q q pa san -an

CVVC-stem F q saam-an

If [-anHn] were added to a CVC-stem, only the first two syllables could be parsed in a moraic trochee; the third syllable would have to remain stray, since no moraic trochee can be built on the monomoraic syllable. Similarly, if [-an] were added to a bisyllabic stem or to a monosyllabic stem with a long vowel, the suffix could not be parsed in a foot, because after a stem with a fully fledged moraic trochee it is impossible to give a light syllable its own moraic trochee. If the language has moraic trochees, then how can we account for the fact that the distribution of stress seems to indicate otherwise? Recall, that, as far as the distribution of stress is concerned, Vogul seems to have syllabic trochees, a type of foot that is assigned without taking quantity into consideration. Vaysman proposes that the moraic trochees of Vogul can only be accompanied by a gridmark (stress) if the immediately preceding syllable is not stressed. Formally, this is an instance of clash resolution: in order to satisfy *Clash, no stress (gridmark) is assigned to the head of a trochee if it is immediately preceded by a stressed syllable. I illustrate this in (39).

Ben Hermans

21 (39)

F

F

q q q q sa ma :a nHl x x

F

F

q q q joo re nHl x

In [’sam-a‘:anHl], both trochees can be assigned a gridmark, since the two stresses are separated by an unstressed syllable. In [’joor-e-nHl], on the other hand, no gridmark can be assigned to the second trochee, because no syllable separates the two foot-heads. The Vogul facts are quite important. On the grounds of allomorph selection, we know that the language must have moraic trochees, i.e. feet in which the syllable on the left is the head. Without a head in this position it would be impossible to characterize the Vogul foot as a trochee, rather than an iamb. Yet not all trochees have stress in this language. This can only mean that feet are always headed, even if they are not accompanied by a gridmark. This is precisely what is assumed by Hyde.

6

Conclusion

The parsing of a word by iambs differs in a number of ways from trochaic parsing. According to Kager (2007), this asymmetry is one of the great puzzles of metrical theory. Two proposals have been put forward to solve the problem. Gordon (2002) proposes to solve the asymmetry by eliminating its cause: foot structure. If there are no iambs or trochees, then we do not expect one type of foot to mirror the other’s behavior. With foot structure eliminated, we are left with gridmarks only. Alignment constraints, interacting with constraints requiring rhythmic alternations, can account for the attested systems, and can also explain why certain systems are impossible. With respect to the distribution of stresses in the words of the world’s languages, grid-only theories seem to be sufficient, and feet therefore seem to be superfluous.10 Nevertheless, feet are necessary. This becomes clear when we take a look at phenomena that go beyond the distribution of stress. There is a wide variety of phenomena leading us to conclude that it is unacceptable to eliminate foot structure from the representation of word stress. Hyde attempts to explain the parsing asymmetries with a theory that does include foot structure. He eliminates the classical constraints on foot form from the theory, proposing instead that iambs and trochees are derived by alignment of foot-heads to the word’s edge. The pattern of feet thus created is subject to interpretation by the grid. Here the notion of overlapping feet plays a major role. With this device it is possible to explain the minimization pattern in odd-parity 10

Van der Hulst (1984) draws a similar conclusion, arguing that grid-only frameworks can do anything a foot-based theory can do. Yet van der Hulst does not eliminate foot structure. He proposes that main stress is generated by foot structure, whereas secondary stresses are created by the grid. In his view, this explains why the phenomena that are related to main stress differ systematically from the type of phenomena that are related to secondary stresses.

The Representation of Word Stress

22

words. Overlapping feet can be motivated on independent grounds: overlapping feet solve the odd-parity problem. In Hyde’s theory, the grid is mostly subservient to foot structure. Feet are always inherently headed, with or without a gridmark.

REFERENCES Altshuler, Daniel. 2009. Quantity-insensitive iambs in Osage. International Journal of American Linguistics 75. 365–398. Bye, Patrik. 1996. Scandinavian “level stress” and the theory of prosodic overlay. Nordlyd 24. 23–62. Crowhurst, Megan J. 1996. An optimal alternative to conflation. Phonology 13. 409–424. Crowhurst, Megan J. & Mark Hewitt. 1995a. Directional footing, degeneracy, and alignment. Papers from the Annual Meeting of the North East Linguistic Society 25. 47–61. Crowhurst, Megan J. & Mark Hewitt. 1995b. Prosodic overlay and headless feet in Yidiny. Phonology 12. 39–84. de Lacy, Paul. 2002. The interaction of tone and stress in Optimality Theory. Phonology 19. 1–32. de Lacy, Paul. 2007a. The interaction of tone, sonority, and prosodic structure. In de Lacy (2007b), 281–307. de Lacy, Paul (ed.) 2007b. The Cambridge handbook of phonology. Cambridge: Cambridge University Press. Dresher, B. Elan & Aditi Lahiri 1991. The Germanic foot: Metrical coherence in Old English. Linguistic Inquiry 22. 251–286. Elías-Ulloa, José A. 2006. Theoretical aspects of Panoan metrical phonology: Disyllabic footing and contextual syllable weight. Ph.D. dissertation, Rutgers University (ROA-804). Gordon, Matthew. 2002. A factorial typology of quantity-insensitive stress. Natural Language and Linguistic Theory 20. 491–552. Gordon, Matthew. 2003. The phonology of pitch accents in Chickasaw. Phonology 20. 173–218. Hagberg, Lawrence R. 1993. An autosegmental theory of stress. Ph.D. dissertation, University of Arizona. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: MIT Press. Hammond, Michael. 1984. Constraining metrical theory: A modular theory of rhythm and destressing. Ph.D. dissertation, University of California, Los Angeles. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Hayes, Bruce. 1984. The phonology of rhythm in English. Linguistic Inquiry 15. 33–74. Hayes, Bruce. 1985. Iambic and trochaic rhythm in stress rules. Proceedings of the Annual Meeting, Berkeley Linguistics Society 11. 429–446. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Hyde, Brett. 2001. Metrical and prosodic structure in Optimality Theory. Ph.D. dissertation, Rutgers University (ROA-476). Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313–359. Hyde, Brett. 2006. Towards a uniform account of prominence-sensitive stress. In Eric Bakovio, Junko Itô & John McCarthy (eds.) Wondering at the natural fecundity of things: Essays in honor of Alan Prince, 139–183. Santa Cruz: Linguistics Research Center, University of California, Santa Cruz. Available at http://repositories.cdlib.org/lrc/prince/8. Hyde, Brett. 2008. The odd-parity parsing problem. Unpublished ms., Washington University (ROA-971).

23

Ben Hermans

Itô, Junko & Armin Mester. 1992. Weak layering and word binarity. Unpublished ms., University of California, Santa Cruz. Kager, René. 1993. Alternatives to the iambic-trochaic law. Natural Language and Linguistic Theory 11. 381–432. Kager, René. 1995. The metrical theory of word stress. In John A. Goldsmith (ed.) The handbook of phonological theory, 367–402. Cambridge, MA & Oxford: Blackwell. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kager, René. 2007. Feet and metrical stress. In de Lacy (2007b), 195–227. Karvonen, Daniel. 2005. Word prosody in Finnish. Ph.D. dissertation, University of California, Santa Cruz. Karvonen, Daniel. 2008. Explaining Nonfinality: Evidence from Finnish. Proceedings of the West Coast Conference on Formal Linguistics 26. 306–314. Kennedy, Robert. 2005. The binarity effect in Kosraean reduplication. Phonology 22. 145–168. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. Majors, Tivoli J. 1998. Stress-dependent harmony: Phonetic origins and phonological analysis. Ph.D. dissertation, University of Texas, Austin. McCarthy, John J. & Alan Prince. 1993. Generalized alignment. Yearbook of Morphology 1993. 79–153. Pearce, Mary. 2006. The interaction between metrical structure and tone in Kera. Phonology 23. 259–286. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19–100. Prince, Alan. 1985. Improving tree theory. Proceedings of the Annual Meeting, Berkeley Linguistics Society 11. 471–490. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Selkirk, Elisabeth. 1980. Prosodic domains in phonology: Sanskrit revisited. In Mark Aronoff & Mary-Louise Kean (eds.) Juncture, 107–129. Saratoga: Anma Libri. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Smith, Laura C. 2004. Cross-level interactions in West Germanic phonology and morphology. Ph.D. dissertation, University of Wisconsin, Madison. Smith, Neil V. & Klaske van Leyden. 2007. The unusual outcome of a level-stress situation: The case of Wursten Frisian. North-Western European Language Evolution 52. 31–66. Vaysman, Olga. 2009. Segmental alternations and metrical theory. Ph.D. dissertation, MIT (ROA-1011). Zoll, Cheryl. 2004. Positional asymmetries and licensing. In John J. McCarthy (ed.) Optimality Theory in phonology: A reader, 365–378. Cambridge, MA & Oxford: Blackwell.

42

Pitch Accent Systems Harry van der Hulst

1

Introduction

This chapter deals with the typology of word prosodic systems and, specifically, discusses the notion of “pitch accent (language),” asking whether there is such a class of pitch accent languages distinct from “stress languages” and “tone languages.” Several issues will turn out to be crucial. Firstly, there is the issue of recognizing (or not) a notion of accent which could be said to underlie both pitch accent and “stress” (or indeed stress accent), and perhaps other phenomena which are not frequently referred to as accentual (such as phonotactic asymmetries). Secondly, there is the question as to whether we wish to distinguish between pitch as a non-distinctive and thus perhaps strictly phonetic property (found in phonetic implementation) and pitch as the exponent of a phonological category (namely tone). Thirdly, there is the possibility of having tone, stress, and accent (in various combinations) “side by side” within the same language, which raises the question of how these notions interact in any given language. The structure of this chapter is as follows. In §2 I will introduce the basic notions and definitions. §3 will briefly discuss examples of languages that have been referred to as pitch accent languages, where accent is apparently realized in terms of non-distinctive pitch. In §4, I examine cases in which tone realization or tone distribution has been said to depend on accent (or stress), a class of languages that is also often included in the pitch accent type. §5 and §6 focus on the different ways in which alleged pitch accent languages have been analyzed, with or without using the notion “accent.” In §7, I define the notions accent and stress as distinct phonological entities and suggest that stress languages may or may not be accentual. In §8 I offer some conclusions.

2 2.1

Accent, tone, and stress: Definitions and usage Accent and stress

For many languages, researchers have reported word-level “prominence,” associated with a specific syllable in the word, which is called “stress” (an English The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0042

Harry van der Hulst

2

term) or “accent” (the term used in, for example, French or German) (see also chapter 41: the representation of word stress). In literature in English on the subject, both “stress” and “accent” have been used for word-level prominence, which has lead to a good deal of confusion, in particular because there are writers who use the terms for different things. Cutler (1984), for example, regards “stress” as a property of words and “accent” as a property of sentences. There is thus a need for clarity on how these two terms are used.

2.2

Accent and its cues

On closer scrutiny, the informal notion of “prominence” can be divided into two distinct phenomena. On the one hand, we have the location of the prominent syllable (e.g. penultimate; final if the final syllable has a long vowel, otherwise penultimate; etc.) and on the other hand, there are phonetic (and phonotactic) cues that signal the location of the prominent syllable (chapter 39: stress: phonotactic and phonetic evidence). In one (fairly old) terminological tradition, the locational aspect of prominence is referred to as accent. The characterization of the accent (location) is essentially sequential (or syntagmatic); only one syllable within the relevant domain can have this property. This is what Martinet (1960) and Garde (1968) refer to as the contrastive or culminative function of “accent,” a term mainly used by Trubetzkoy (1939). The realizational aspect of prominence is, in a sense, paradigmatic (cf. van Coetsem 1996): there are various (not necessarily incompatible) phonetic and phonotactic means for cueing the accent. Some languages may favor one specific cue (e.g. pitch or duration), but several cues may conspire to manifest the accent. This division of “prominence” correlates with traditional terminological systems such as musical accent vs. dynamic accent or (with much the same meaning) pitch accent (systems) vs. stress accent (systems) (see Fox 2000: ch. 3 for an excellent general review of the notion accent; also van Coetsem 1996; van der Hulst 1999b, 2010b). In each case, the modifier of the head noun (“accent”) says something about the way in which the accent is “manifested” or “realized.” In this chapter I will focus on relationships that involve accent and pitch, whether used distinctively (in terms of contrastive tones) or non-distinctively. However, I will also consider the relationship between accent and stress.

2.3

Word prosodic types

While in some languages pitch is a property of words, all languages use pitch features within an intonational system, a system that aligns “sentences” with a melody that can be defined in terms of pitch events that mark boundaries of (syntactic or prosodic) units as well as the informational packaging of the utterance with reference to the notion “focus” (Bolinger 1982; Gussenhoven 2004; chapter 32: the representation of intonation). At the same time there are languages in which pitch is a property of “words.” Within this group of languages we commonly find the labels in (1b) and (1c). The label “stress” in (1a) is reserved for languages that need no specification of pitch at the word level, although, like all other languages, they use pitch for intonation purposes. (1)

a. b. c.

Stress (or stress accent) Pitch accent Tone

Pitch Accent Systems

3

There is, however, a great deal of controversy concerning the use of the terms tone and pitch accent, and, for that matter, the term stress.1 Hyman (2001, 2006, 2009) makes a case for treating systems that we label stress and tone as “prototypes,” meaning that languages that belong to one or the other (or both) type(s) display one or more specific defining properties.2 “Pitch accent,” according to Hyman, is not a prototype, but rather a label for a large class of hybrid systems that mix “tone” and “stress” properties in various ways, or systems that are clearly tonal, although displaying various restrictions on the distribution of tones. In effect, Hyman regards the notion accent as unnecessary, whether as a formal mechanism in analysis or as a prosodic type. Other researchers (such as Gussenhoven, e.g. 2004) who also reject the idea of “pitch accent languages” nonetheless recognize the notion of accent as an analytic device. In this chapter these views will be discussed and compared to views that attribute a fundamental role to the notion accent.

2.4

Definitions and use of tone

A traditional way of defining the notion tone is in terms of “distinctive use of pitch.” Thus, if a language uses pitch to distinguish different otherwise identical morphemes, pitch has a phonological or contrastive (distinctive) status. The following frequently quoted definition captures what is perhaps the canonical use of distinctive pitch: A tone language may be defined as a language having lexically significant, contrastive, but relative pitch on each syllable (Pike 1948: 3).

If tones are distinctive on all syllables (possibly like properties such as frontness, height, or roundness) we can say that the distribution of tones is unrestricted. Most researchers, however, agree that there is no reason to limit the term tonal language to cases in which the distribution of tones is entirely unrestricted (see chapter 45: the representation of tone). Presumably, all tonal systems show restrictions resulting from tonal spreading or assimilation (Schuh 1977; Hyman 2007), from using a limited set of tonal melodies which are properties of morphemes rather than of syllables (Leben 1971; Goldsmith 1976b; Halle and Vergnaud 1982), from the avoidance of sequences of identical tones (dissimilatory or OCP effects), or indeed from relations between tone distribution and accent (or “stress”) (see §4). Also, it is not uncommon to find that the full range of contrasts is not found in affixes (as opposed to roots or stems) (chapter 104: root–affix asymmetries). Finally, initial or final syllables may fail to bear tonal contrast (sometimes to leave room for intonational tones or for other, perhaps “perceptual,” reasons; chapter 98: speech perception and phonology).3 Since it would be unwise 1

Typological studies of word prosodic systems are numerous: e.g. Trubetzkoy (1939); Hockett (1955); Garde (1968); Meeussen (1972); Goldsmith (1976a, 1988); Greenberg and Kaschube (1976); Hyman (1977, 1978, 1981, 2006, 2009); Lockwood (1982); Clements and Goldsmith (1984); Beckman (1986); Clark (1987, 1988); Haraguchi (1988); Hollenbach (1988); van der Hulst and Smith (1988); Mock (1988); Wright (1988); Hayes (1995); van der Hulst (1999a, 2010c); de Lacy (2002); Duanmu (2004). 2 Here Hyman avoids the term “stress accent,” presumably because he no longer (cf. Hyman 1977) recognizes the label “pitch accent” as a useful one and thus essentially wants to eliminate the notion accent altogether. 3 Suárez (1983: 52) observes that in Huichol and Mazahua there is no tone contrast on the last two syllables or the last syllable, respectively. In these languages, inherent lexical tones are removed to free up space for intonational tones.

Harry van der Hulst

4

to maintain the strictness of Pike’s definition (according to which perhaps there is not a single tonal language), van der Hulst and Smith (1988) quote the much more liberal definition of Welmers (1973: 2): A tone language is a language in which both pitch phonemes and segmental phonemes enter into the composition of at least some morphemes.4

Note the use of the term pitch phoneme (chapter 11: the phoneme), which suggests that Welmers requires that pitch is used contrastively, a crucial point to which I return below. This definition includes languages in which there are tonal contrasts in certain positions, or even in only one position, in some morphemes. With this broader definition, tonal languages can be ranked on a scale of tonal density (Gussenhoven 2004), which indicates how many word positions have how much tonal contrast. In a sense, such a scale indicates the relative functional load of tone properties. Stretching Gussenhoven’s notion, we could say that relative density arises not only in the syntagmatic dimension (depending on how many positions display tonal restrictions), but also in the paradigmatic dimensions (depending on the number of contrastive options per position): (2)

Tonal density matrix T1 T2 T3

+ + + x

+ + + x

+ + + x

+ + + x

+ + + x

+ + + x (tone bearing-units)

Each potential minus would indicate a restriction on the distribution of a distinctive tone. However, no matter how dramatic the restrictions, as long as there is tonal contrast (i.e. distinctive use of pitch), phonological tones must be specified in the lexical entries. The smallest tonal system would have two tones, H and L. More extensive systems would add an M tone, or possibly two different M tones (high mid and low mid). In addition, systems can have contour tones (rise, fall, etc.) (chapter 45: the representation of tone).

2.5

Culminativity and obligatoriness

Another frequently cited term in this context is “restricted tone language,” introduced by Schadeberg (1973) and Voorhoeve (1973). This term, too, would seem to indicate a scale of restrictiveness, although Voorhoeve introduced it in the context of Bantu languages whose tonal system is so severely restricted (up to one H per word in a H/L system) that he suggested that an accentual analysis might be considered (chapter 114: bantu tone). Indeed, adding syntagmatic and paradigmatic restrictions on the distribution of tone together, one could see that a language, despite having a H/L contrast, while allowing at most one H tone per word could easily permit an accentual analysis in which the H “tone” is regarded as the predictable pitch cue of an accent, even in a case where there is no indication of any additional independent cues for this accent. 4

Strictly speaking this excludes cases in which a language has tonal affixes without having affixes or other morphemes that combine tone and segmental properties.

Pitch Accent Systems

5

But what is “accent,” and how is it formally represented? Hyman (2009) formulates two necessary properties of what he calls stress, which I will take as a point of departure for establishing what might be seen as characteristics of accent, if these notions are going to be distinguished. One “property” is that each “word” can have stress or accent at most once (only one syllable can be stressed or accented) and, additionally, each word must have it at least once. These two properties, following Hyman (2006, 2009) can be referred to as: (3)

a. b.

Culminativity (at most one) Obligatoriness (at least one)

Let us now ask whether the two properties in (3) must be regarded as necessary properties of accent. An issue that goes to the heart of what is often seen as problematic for the notion “pitch accent” is that languages which allegedly have a pitch accent system, and thus accent, sometimes have (lexical) words that appear to be unaccented (see the discussion of Tokyo Japanese in §6). This, however, is only problematic if obligatoriness is stipulated to be a necessary property of accent. We could investigate a more liberal interpretation of accent, in which unaccented words are permitted in an accentual language. This, of course, has important consequences, because it opens the door to using the presence vs. absence of accent as a contrastive option and thus to analyzing alleged tonal languages that have a H–L contrast as fully accentual languages, with H as the exponent of accent and L as the lack of accent. We might then also question whether culminativity is a necessary requirement for describing accent. If culminativity is not required, even “H/L” languages that allow multiple H “tones” could be analyzed as fully accentual. Allowing words to have multiple accents separates the notions stress and accent even more dramatically than just giving up obligatoriness for accent. Still, if accent is not the same thing as stress, there is no a priori reason for believing that any properties of the latter need to be true of the former. I return to these issues in §5.3.

2.6

Representational issues

Answers to the question of whether or not the properties in (3) are definitional of accent have repercussions for, or are implicit in, the manner in which accents are formally represented. In one type of approach the relevant syllables are marked with an “accent mark,” as is common in dictionaries and in autosegmental theory (e.g. the “star” of Goldsmith 1976a), or in terms of a segmental feature, as in the phonological theory of Chomsky and Halle (1968). In this “lexicographic” approach there is no commitment to culminativity or obligatoriness. A different formal approach is to provide the string of syllables with a headed tree structure, as proposed in various versions of metrical theory (Liberman and Prince 1977) and dependency phonology (Anderson and Ewen 1987) (see also chapter 40: the foot). Metrical structures have one designated terminal unit, the head of the word, which counts as the (primary) “stress.” This notation (assuming that all syllables must be grouped in one structure) implies culminativity, but not necessarily obligatoriness, because it does not follow from the notation that each word must be provided with a metrical tree.

6

Harry van der Hulst

However, rather than seeing asterisks and trees as competing mechanisms, we should entertain the idea that they are complementary, in that the former represent accents, while the latter represent stress. This point is acknowledged by Anderson and Ewen (1987), who, in addition to headed tree structures, also use asterisks to indicate what we might call “potential heads.”5 I will return to this point in §7.

2.7

Problems with the notion “pitch accent”

We have been considering a use of the term accent as an abstract mark of a position that can be cued by various phonetic properties, “stress” being one of them. Beckman (1986) refers to languages that are not stress accent languages, as “non-stress accent” languages (thus avoiding the term “pitch accent language”). This, of course, is compatible with the idea that in many non-stress languages pitch is the most salient property of accent. Van der Hulst (1999a, 2010b) points out that if we maintain the term “pitch accent language” we might also expect to find languages that can be labeled as “duration accent” languages (if duration is the only cue). On this view, pitch accent languages are languages in which accent is (mainly) cued by phonetic pitch. There are, in fact, various factors that make the use of this term problematic. One factor is, obviously, that people may simply define the term differently. For example, as we will see in §4, tonal contrast is often limited to specific syllables in the word; cases of this sort have been analyzed by identifying a notion “accent,” with association of tones being dependent on this accent. While in this case, the presence of tone can be said to function as a cue of accent, the cue is not a phonetic, but rather a phonological, fact (namely the phonotactic distribution of tones). The fact that the possibility of tonal contrast may signal the accent location is part of a much more general pattern, found in many languages, where accented syllables display contrastive or structural options that are exclusive to a particular syllable (see Downing 2010; van der Hulst 2010b).6 Pursuing the terminological path that we started out on above, we might refer to such cases in which tonal contrast is limited to the accented syllable as tone accent (or tonal accent) languages, rather than pitch accent languages. It is apparently the case that accented syllables can be referred to by the phonology as well as by the phonetic implementation system. In fact, accents can be referred to by other grammatical modules as well, e.g. the intonation system. Does this mean that we can refer to English as an “intonation accent” language? Languages cannot be put in a single box as far as cues for accent are concerned. Tonal accent systems, then, differ from pitch accent systems if we agree that in the latter pitch is not used distinctively. However, some writers (e.g. Downing 2010) use the term “pitch accent” for any system in which pitch properties (whether distinctive or not) enter into a relationship with accent or stress. This would include not only what is referred to here as a pitch accent or tone accent language, but also another class of languages which have both tone and accent, in which accent (or “stress”) is assigned with reference to tone. Downing’s use of the term pitch accent is thus much broader than the one considered above. 5

Another formal notation (also proposed in Liberman and Prince 1977) is the metrical grid, which does not imply culminativity. See chapter 41: the representation of word stress for extensive discussion. 6 This relates to the notion of positional faithfulness; cf. Beckman (1998).

Pitch Accent Systems

7

Finally, we return to Hyman’s (2006, 2009) use of the relevant terminology. It would seem that he agrees that there are systems in which pitch could be analyzed as a predictable phonetic cue of a notion accent, but he argues that systems of that sort can always be analyzed as tonal.7 He refers to Gussenhoven’s analysis of Nubi, a language in which each word has precisely one syllable with high pitch. Gussenhoven argues that Nubi presents a case that can be analyzed as a pitch accent language or even as a stress language, but adds that it is also possible to propose a tonal analysis. If a tonal analysis is chosen, it follows that the fact that high pitch in Nubi is culminative and obligatory is “accidental” – Nubi is simply at the far end of a continuum of tonal languages in which the distribution of tones is restricted in various ways. It is important to realize that Hyman (2006, 2009; chapter 45: the representation of tone), like Gussenhoven (2004) (in line with the approach initiated in Pulleyblank 1986; cf. §5), adopts a definition of tone that is even more liberal than that of Welmers (see Hyman 2001 for the introduction of this definition): “A language with tone is one in which an indication of pitch enters into the lexical realization of at least some morphemes.” For these authors, then, the notion tone clearly no longer entails “tonal contrast” (i.e. distinctivity). For this reason, they maintain that a language like Nubi, although it could be analyzed as a pitch accent system, can also be tonal.

2.8

Intonational pitch accents

Before we examine some cases of (alleged) pitch accent systems, let us consider one other use of the term pitch accent. The term is also used in the intonation literature where, following Bolinger (1982), intonational events that associate to phrasal accents (usually called phrasal “stresses”) are called pitch accents. In the autosegmentalmetrical tradition of Liberman (1975), Bruce (1977), Pierrehumbert (1980), Goldsmith (1981), Gussenhoven (2004), and Ladd (2008) (see also chapter 32: the representation of intonation; chapter 50: tonal alignment), intonational pitch accents are phonological tones (H, L or some combination). The reason for this is that in many intonational systems that have been studied within this model, there are tonal contrasts at the intonational level, because different tones or tone combinations have different meanings. However, if in some language each phrasal accent associates with the same pitch event, it would be perfectly possible to analyze that pitch event as a direct phonetic interpretation of the phrasal accent without postulating an intervening phonological tone.8

2.9

The issue of distinctivity

Analyses within the autosegmental-metrical tradition are not, however, much concerned with distinctivity (or indeed with a distinction between “phonological” and “phonetic” phenomena), and all phrase level pitch phenomena are usually analyzed in terms of “tones” (mirroring Hyman’s general use of tones at the word level, which also ignores distinctivity). 7

This is, in fact, how he uses the term in Hyman (1977). It may be the case that languages that have been described or listed as word-level pitch accent systems may be phrasal pitch accent systems. Since the patterns listed for words are often based on elicitation of citation forms, we cannot be sure that the observed word prosodic properties are word level or phrase level. See van Zanten et al. (2010) and Gordon (forthcoming). 8

8

Harry van der Hulst

It could be argued that definitional decisions are, paradoxically, not the crucial issue. Does it really matter whether we “call” Nubi a tone language or a pitch accent language or even a stress language? What is of importance is how specific systems are analyzed and which theoretical tools are used. However, we must also be aware of the bigger issue regarding how we see “phonology” as distinct from and interacting with “phonetic interpretation or implementation.” A traditional stance would be to maintain that using a formal object “H” in the phonology entails that this unit has a contrastive function within the linguistic system (chapter 2: contrast). If pitch is distinctive, we are dealing with phonological entities such as /H/ and /L/, etc. If one sets up the system of phonetic implementation by translating a non-tonal property X (e.g. accent) into a phonetic property “H” which is implemented in terms of relative F0, we seem to be dealing with [H] (rather than with /H/).9

3

Some (alleged) pitch accent systems

In this section I provide references to languages that have been analyzed as pitch accent systems or that have played an important role in the treatment of systems that have pitch or tonal cues correlating with accent.

3.1

A tour around the world

The chapters in van der Hulst et al. (2010b) offer a survey of word accentual systems in the world’s languages. I refer here to the chapters on languages in the Americas (van der Hulst et al. 2010a; Rice 2010; Wetzels and Meira 2010) for many examples of languages that have been described as realizing accent exclusively or mainly in terms of pitch. Several additional examples can be found in the chapters on Papuan languages (van Zanten and Dol 2010), Asian languages (Schiering and van der Hulst 2010), and European languages (van der Hulst 2010a), specifically Caucasian languages (see Kodzasov 1999). Even though these surveys do not prove that the category of pitch accent languages is a genuine prosodic type, it is not without significance that so many systems have been identified with obligatory and culminative (and non-distinctive) high pitch.

3.2

Basque and Japanese

The languages discussed in the following two sections differ from the previous cases in that explicit reference to unaccented words, i.e. lack of obligatoriness, is required. Yet in both cases it would seem that the alleged accents have distributional properties that are very similar to those of stress (accent), which supports a pitch accent analysis.10 9

Cf. Clements (2001, 2009), who defends a broader justification for recognizing phonological features than only distinctivity. If a phonetic property is in some sense “salient,” this would, in his view, justify postulating a phonological feature. 10 Another case which is similar to these two languages is Korean, which, in its many dialects, displays a rich variety that is reminiscent of the Japanese situation in particular; see Fukui (2003) and for a summary Schiering and van der Hulst (2010).

Pitch Accent Systems

9

3.2.1 Basque The dialects of Basque present a great diversity of word prosodic systems (see Hualde 1999). Gussenhoven (2004: ch. 9) presents an analysis of Northern Bizkaian Basque with reference to the Gernika and Lekeito dialects. Both have accented and unaccented roots, the former being in the minority. There are inflectional and derivational suffixes that are accented or pre-accenting. In Lekeito, if a word has an accent, this accent always ends up on the penultimate syllable. In Gernika, the leftmost (non-final) accent prevails; this is the more common case in Basque dialects. In Lekeito, unaccented words are grouped with an accented word to their left or right, whereas sequences of unaccented words together form a single domain. Each such domain either has an accent (if it contains an accented word) or is unaccented. Unaccented domains receive a default final accent in certain syntactic positions, namely at the end of the sentence or before the finite verb. Each accent, whether lexical or default, is associated with a HL pitch accent. The left edge of the accentual domain is marked by a LH boundary sequence; a high plateau is found between the boundary H and the H of the pitch accent. Systems of this sort seem obvious candidates for accentual analyses, which of course begs the question of whether they must be analyzed accentually. One argument that could be made for an accentual approach is that in the various dialects we note a variety of accent locations (ranging from lexical to rule-governed) which is very reminiscent of the distribution of stress in “stress accent languages.” The second argument again involves the fact that pitch is non-distinctive in Basque dialects. Note that in Basque, unaccented words are provided with default accent, at least in some cases.

3.2.2 Japanese We also find a broad array of word prosodic systems among the dialects of Japanese (cf. Haraguchi 1977). An overarching property of all systems is the relevance of pitch at the level of the “word,” or, as some researchers prefer to put it, the “accentual domain.” An interesting overview in the context of autosegmental theory of dialectal differences is offered by Haraguchi (1977, 1988), who divides Japanese dialects into two broad categories: pitch accent systems and unaccented systems. Cross-classifying with this dichotomy, he suggests a “universal” inventory of melodies (H, L, HL, LH, and LHL), from which a system may select at most one or two. In addition to the choice of one or more melodies, the differences among dialects depend on: (4)

a. b.

The location of accent/H: fixed or free.11 The spreading of H: no spreading, rightward or leftward.

Thus, in Tokyo Japanese, the H tone spreads leftwards (leaving an initial mora low, possibly due to a boundary L tone that comes with the left). We will focus on the pitch aspect of Tokyo Japanese in §6. The system of Tokyo Japanese is such that the constituents of words (stems, affixes) can be accented or unaccented (or, in the case of affixes, pre-accented). When more than one accent is present in the accentual domain (which can be larger than the word and therefore needs careful definition; Gussenhoven (2004) calls it the a-domain), the first (or leftmost 11

In §6 we will discuss the way accents are distributed in Tokyo Japanese, which is partly lexical and partly rule-based.

Harry van der Hulst

10

accent) predominates, i.e. will attract the high pitch/tone. If no accent is present, the high pitch occurs on the last (rightmost) syllable (and spreads from there). This “First/Last” pattern constitutes a system that is reminiscent of so-called unbounded stress systems (Hayes 1995). In fact, Haraguchi (1988) notes that three of the possible unbounded patterns occur in Japanese dialects (see also chapter 120: japanese pitch accent). (5)

a.

Systems with unaccented words First/First12 First/Last Last/First Last/Last

Kumi Tokyo, Osaka — Hirosaki

b.

Systems without unaccented words First

Fukuoka

Last

—

Note that systems without unaccented words have no default clause. Haraguchi (1977, 1988) also recognizes unaccented systems, i.e. systems in which no word is accented. He mentions Sendai, Miyakonojo, and Kagoshima. In such systems the tonal melody is associated either from left to right or from right to left in his analysis: (6)

Systems with only unaccented words First Last

For these systems, tones are associated to words in terms of association conventions that make no reference to accents, but rather the word edges. These same conventions are invoked for unaccented words in accentual languages (as in (5a)), which implies that in such systems tones are associated partly to accents and partly directly (i.e. without “intervening” accents). In all dialects that use just one melody, the question can again be raised whether this “melody” is a phonological entity or entirely due to phonetic interpretation. Haraguchi (1988) does not consider this issue, but it could be argued, as before, that only dialects that have more than one word melody are truly tonal.

3.3

Bantu languages

Many Bantu languages are commonly described as having both tone and accentual properties, while a few (such as Swahili) have lost tone, and retain only “stress” (chapter 114: bantu tone). Bantu word prosodic systems have always been of special interest in the debate regarding the appropriate analysis for languages that have both significant word-level pitch movement and indications that accent plays a role as well; see Schadeberg (1973), Voorhoeve (1973), Goldsmith (1976a, 1988), Hyman (1978, 1981, 1982, 1989), Clements and Goldsmith (1984), Odden (1988), and especially Downing (2010). The accentual analysis of Bantu languages was promoted by Goldsmith (1976b, 1988), although the approach has a long history (see the introduction in Clements and Goldsmith 1984 for a historical perspective). 12

This can be glossed as: “Associate a tone with the first accented syllable, or, if no accent is present, with the first syllable.”

Pitch Accent Systems

4

11

Systems with accent and tone

Although the focus of this chapter is on pitch correlates of accent, we must be aware of the fact that in systems that display both tone and accent several relations between these two phenomena are possible (Hyman 1977; van der Hulst and Smith 1988; Fox 2000; de Lacy 2002; Wetzels and Meira 2010; among others): (7)

a. b. c.

Accent and tone are independent. Accent is dependent on tone. Tone is dependent on accent.

De Lacy (2002) proposes a system of constraints and accounts for the different relations in terms of different rankings. In this section I will focus on the systems in which tone is dependent on accent; for a discussion of the other two cases, see van der Hulst (2010c). In §2.7, I used the term tonal accent systems for systems in which the distribution of tone is determined by accent, but we need to be more precise on exactly what kind of relationships may exist. The distribution of distinctive tones can be restricted for a variety of reasons (see §2). While the factors that lead to restrictions in a specific system may be unrelated to the notion “accent” (which may or may not be independently present in the language in question), there comes a point where the tonal system is so restricted that an analysis is possible in which a specific syllable can be identified as “accented” and, as such, function as the domain for the association of the tonal distinctions. If the notion of accent was already present on independent grounds, the common tendency towards reduction of tonal contrast in unaccented syllables may have been a factor in the emergence of a restricted tonal system, in addition to other factors which may have played a role. However, the processes that lead to restrictions may also “accidentally” give rise to an accentual interpretation. Since languages in which accent and tone interact are sometimes included in the class of pitch accent languages, these cases merit attention in this chapter. The effect of accent on tonal contrast can be twofold. It may lead to reduction and eventually neutralization of an underlying contrast (chapter 80: mergers and neutralization). This is what is called here accent-driven reduction. It is commonly claimed that the elimination of tones in certain positions in Mandarin Chinese (chapter 107: chinese tone sandhi) is caused by the fact that tonal contrast can only be maintained in words with accent; see Yip (1980, 2002), Wright (1983), and Duanmu (2000), for analyses and references. A similar case can be found in the Ijo languages (Williamson 1988), where only the first word in a “tone group” retains its underlying tones. In both cases, unaccented words lose their lexical tones (which show up if the words are in an accented position). In these two examples we are dealing with accent at the compound or phrasal level, and thus with neutralization of all tones belonging to words that are not in an accented position. Reduction of tonal contrast within polysyllabic morphemes may lead to restructuring, such that tones formerly associated to unaccented syllables either disappear entirely or are attracted to one particular syllable, the accented syllable. In either case, the end result is that tonal contrast only occurs on the accented syllable. When a restricted tone system is analyzed with reference to a notion of accent, we have accent-driven tonal distribution, and the system can be called a

12

Harry van der Hulst

tonal accent system. A question that arises in these cases is whether the accented syllable is cued merely by its attraction of tonal contrast, or, additionally, by other “stress-like” cues. I will consider this issue in §5. I consider here some examples from Suárez (1983), as well as from Yip (2002), in their surveys of Meso/Middle American languages. Isthmus Zapotec has two tones which associate to the accented syllable and from there spread rightwards. “Pre-stress” syllables are low-toned. Suárez also mentions Northern Pame and Yaitepec Chatino as languages that have a tonal contrast only in the syllable that is said to be “stressed” (in both cases the “final” syllable, presumably of the stem). This can be compared to Huautla Mazatec, where every syllable can have contrastive tone. In between, we find cases where the contrast on certain non-accented syllables is limited. In Palantla Chinantec, for example, there is no tonal contrast on post-stress syllables. Van der Hulst and Smith (1988) cite the case of San Juan Copola Trique, which illustrates how restricted tonal distribution can arise historically (cf. Hollenbach 1988; Yip 2002). In the Otomanguean family in general, we find a continuum of reduction of tonal contrast and, interestingly, an increase in tonal contrasts on the accented syllable. A case where accent has only mildly influenced tonal contrast is found in Cajonos Zapotec (Nellis and Hollenbach 1980). Of the four underlying tones H, L, HL, and M, only M is disallowed in unaccented syllables. In this case, then, we do not have a tone accent system, but simply a tone and accent system, with accent-driven reduction. Among the languages in which the distribution of tone is dependent on accent, there is a subclass of cases in which tonal contrast is only found on or near accented syllables, not because tones have been neutralized in other positions, but simply because a tonal contrast historically developed only in this position. In these cases, the accented syllable, in addition to being an attractor for tonal association, has clear stress-like cues. Hence languages of this kind are both stress accent and tonal accent languages, with the proviso that the tone does not always associate directly to the accented syllable, but sometimes on a syllable near it (although this also depends on the details of the analysis). Two well-known cases of this sort are some of the Scandinavian languages and Serbo-Croatian. For discussions of the Scandinavian type I refer to Bruce (1999) and Gussenhoven (2004) (see also chapter 97: tonogenesis). For Serbo-Croatian, see e.g. Inkelas and Zec (1988).13 We must note that the co-occurrence of stress accent and a lexical pitch contrast enforces a tonal analysis of the latter. If the accent was not manifested in any other way than forming an anchor for lexical pitch, it could be argued that the opposition is one between accented words and unaccented words.

5 5.1

The accent debate Accents or no accents

We have so far discussed two possible interactions between accent and pitch or tone: 13

In his chapter on central Franconian tones, Gussenhoven (2004: ch. 12) discusses the emergence and representation of a tonal distinction that is very similar to the Scandinavian distinction; see also Gussenhoven and Bruce (1999) and Hermans (1994). We also find a similar contrast (due again to different historical factors) in Scottish Gaelic; see MacAulay (1992: 234–236).

Pitch Accent Systems (8)

a. b.

13

Pitch is dependent on accent (pitch accent systems; §3). Tone is dependent on accent (accent-dependent reduction and distribution; §4).

The dividing line between the two types is distinctivity. If pitch is non-distinctive, i.e. if there is no tonal contrast, the system uses pitch to cue accent. But if there is tonal contrast, tones are involved. The Bantu systems mentioned in the preceding section have been analyzed as involving accent and tone. However, the question of whether the occurrence of tone contrast on one specific syllable requires a notion of accent cannot be taken for granted, even when tonal association seems to be limited to an “accent-like” position. Consider the case in which the alleged accented syllable has no independent property apart from being the locus of tonal contrast. One could then say that there really is no accent at all, and instead assume that the tones, being specified as a property of morphemes, associate to their specific locus directly, without first assigning an accent that attracts the tones. In this case we would accept that accent rules and tone association rules fall under the umbrella of a general theory of positional identification, and that the principles for positional identification are similar, if not the same, for both accent placement and tone association. (9)

a.

b.

Indirect (accentual) approach Step one: Accent goes to position X. Step two: Tones go to accent. Direct approach Step one: Tones go to position X.

If the direct approach is taken, the category of tonal accent systems reduces to tonal systems which are then further differentiated in terms of different principles of association (LR, RL, positional). Below we will see that the direct tonal approach can also be applied in systems that have unpredictable (i.e. lexically specified) loci for accents. To what degree should tone placement and accent placement be allowed to overlap? If, for example, a tonal contrast occurs on the final syllable if it is closed, and otherwise on the penultimate syllable, should we say that there is a quantitysensitive accent rule and that tones are attracted to the accent, or should we make the tonal association rules quantity-sensitive? The earlier literature on systems in which tone contrast is limited to specific syllables reflects the view that the theory of accent placement should not be duplicated in a theory of tone placement, so that in these cases accent is usually seen as placing a role in tonal association. On the other hand, Haraguchi (1977, 1988, 1991), as we saw in §3.2, makes a sharp distinction between tones that associate to accents and tones that associate directly to tone-bearing units at edges. The latter case involves only strict directional association in his analysis (from right to left, or from left to right). But, if peripheral tone-bearing units can be “extra-tonal,” we can expand the set of cases in which tonal association can be direct. However, we do not expect direct tonal association to be dependent on syllable weight distinctions. Hence, if tones are attracted to positions that reflect weight criteria one would be inclined to associate tones to accent which are assigned in a weight-sensitive fashion.

14

Harry van der Hulst

Given the inevitable overlap between accent placement and direct tonal association, Pulleyblank (1986) launched an attack on the use of accents and suggested replacing accents by tones. This approach, discussed in the next section, has since become dominant.

5.2

Giving up accents

The direct tone approach was promoted by Pulleyblank (1986) mainly for various African tonal systems and by Poser (1984) for Tokyo Japanese in particular. The most important of Pulleyblank’s arguments against the use of stars (cf. Blevins 1993: 238) are as follows. Firstly, using stars and tone makes the system overly rich, in that we now predict rules referring to stars and to tones, and to both at the same time. Secondly, the inherent culminative nature of stars can also be found in systems that are arguably tonal and non-accentual, i.e. the asymmetry between accent and non-accent finds a counterpart in systems in which H tone contrasts with zero (ending up as default L). Another argument that could be mentioned is that accent (if equated with stress) is a property of syllable, whereas stars sometimes need to be assigned to moras. Finally, as we have already mentioned, the existence of unaccented words in accentual systems, or indeed of words with multiple accents which all surface, can be regarded as problematic. Pulleyblank applied the direct tone approach to a variety of cases, not only cases in which the position of the tone is predictable, but also those where the former accent location is lexically specified; it was subsequently adopted in much other work (Clark 1988; Hyman 1989). We note again that this move entailed the use of phonological features for non-distinctive, i.e. predictable properties. Even though the location of the alleged tone could be a lexical, unpredictable property, the phonetic nature of the entity (high pitch) would nonetheless be predictable.14 The abandonment of stars implies, firstly that the systems discussed in §4, where H tone is restricted (perhaps up to the point of being culminative), but not obligatory, are now analyzed as tonal. However, a further-reaching conclusion is that “straightforward” pitch accent systems (discussed in §3), where high pitch is both obligatory and culminative, are also analyzed as tonal, despite the fact that pitch is not distinctive. This may or may not be considered a (conceptual) problem (cf. Clements 2001, 2009). Another issue is of course that we have rules for tonal association which duplicate the theory of accent which is independently needed for non-tonal accent systems. Abandoning accent cannot make the Scandinavian (and Serbo-Croatian) case purely tonal, since, as mentioned, in these cases we need the notion of stress (accent), independently of the tonal specifications.

5.3

In defense of accents

If accents are rejected for pitch accent and restricted tone languages, the term “accent” can simply be abandoned in favor of the term “stress” (for stress accent languages). Hyman (2007) adopts this position, reducing the typology of word 14

This might suggest a “compromise” position in which “accents” are regarded as unspecified tonal “root nodes.” In an approach which adopts a wider use of accents as possible ingredients of stress accent systems, this idea could not be maintained.

Pitch Accent Systems

15

prosodic systems to tone languages and stress languages. In this section I will focus on the use of accents in “tonal” systems and suggest a different approach, one which maximizes the use of accents at the expense not just of non-contrastive “tones,” but also of (allegedly) contrastive tones. The issue here does not revolve around languages that have obligatory and culminative high pitch such as Nubi. Here, the case for accent could be considered uncontroversial, if we believe that culminativity and obligatoriness are necessary for an analysis using accent (which essentially means that accent and stress are the same thing). Rather, let us focus on languages in which H tones violate one or both of these two constraints. I will argue that languages of this sort can also be analyzed as accentual (and thus non-tonal), if obligatoriness and culminativity, while perhaps being typical or even necessary for stress, are not required for accent. These points were anticipated in §2.5. Let us first consider the type of case in which one syllable per word is either H or L, meaning that H is culminative, but not obligatory. In an accent-cum-tone analysis we would postulate an accent, and from there we have several options, depending on how we characterize the tonal contrast (H/L, H/zero, zero/L). But there is also another option. We can also analyze the contrast as accent vs. no accent (with accent giving rise to phonetic high pitch and low pitch as default). This means that we can analyze these alleged H/L systems as pitch accent systems as long as we “allow” that accentual languages have a class of unaccented words. Secondly, even when a “H/L” system allows multiple (non-adjacent) “H tones,” this does not necessarily enforce a tonal analysis. If neither criterion proposed by Hyman (2007) for stress applies to accent, there is no reason why a word could not have more than one accent. Concluding, if we push the use of accents to its limits (at the expense of using tones), this implies allowing unaccented words (violating obligatoriness) and multiple accents (violating culminativity). In this liberal view on accent, only languages that have more than a binary pitch contrast are necessarily tonal; in addition, we find languages in which culminativity and obligatoriness of accent is independently required (as in the case discussed in §4). One could say that “H/L” systems are the real pivotal cases, where, as linguists (or as language learners), we have a choice between an accentual and a tonal analysis. There may be certain diagnostics that will tip the balance to either an accentual or a tonal analysis, and these need to be made explicit. More work is called for in this area. An accentual approach is favored when the distribution of accent falls squarely within a theory of accent placement that is independently needed for stress accent and other types of accentual languages. This, perhaps, makes an accentual analysis of those languages in which the alleged accents need to be assigned to moras undesirable (cf. the case of Somali; Biber 1981; Hyman 1981; Banti 1988). Another diagnostic pointing to tones is the need to refer to floating tones, on the assumption that the notion “floating accent” is suspect. Thirdly, it could be argued that tonal spreading processes might suggest tone, but implementational mechanisms can also be held responsible for pitch extending over several syllables. A fourth potential way to discriminate between accent and /H/ tone would be to look at the details of phonetic implementation. One could conceivably argue that the phonetic pitch target of phonological categories like /H/ is more specifically defined than the pitch target of accents. Lastly, an accentual analysis could account

Harry van der Hulst

16

for cases in which we need rules that delete apparent accents in clash, or other rules that refer to accents, irrespective of their pitch or “tonal” correlates. McCawley (1978) suggested that in some cases one might want to say that a system is accentual first, and then becomes tonal in the course of the derivation. The question is, however, whether the tonal end of the derivation is still part of the phonology or part of the phonetic implementation. In this section I have suggested that accentual systems should be “allowed” to have unaccented words or multiple accented words, or even both. This seems to imply that obligatoriness and culminativity are not necessary properties of accent and that the case in which accents are both obligatory and culminative is just one of four possibilities; see §7.

6

The case of Tokyo Japanese

Tokyo Japanese is a language that is often mentioned as a prime example of a pitch accent system, but differs from both Nubi and Somali, while apparently sharing properties with each. Every word is said to have high pitch, but, at the same time, some words are accented and others are non-accented. Let us first consider the basic facts; references to various types of analyses are offered below. In Tokyo Japanese, nouns have a specific pitch contour, which in some but not all cases involves LHL. In those words that have the full LHL pattern, the L occurs on the initial mora. This mora is followed by a high plateau, which may drop to low at some point. After the drop, remaining syllables are low. In some words the initial L is missing, and in other words the final L. Thus we find the following patterns, taking trisyllabic nouns to illustrate the possibilities: (10)

a.

HLL qqq inoti ‘life’

b.

LHL qqq kokoro ‘heart’

c.

LHHL qqq(-q) atama ‘head’

d.

LHHH qqq(-q) sakana ‘fish’

This system can be and has been analyzed in many different ways; here we will specifically focus on accounting for the difference between (10c) and (10d). For (10a)–(10c) we have three options; depending on which one is chosen, various approaches can be suggested for class (10d): (11)

(10a)–(10c)

a. Accent → /H/ → [H] b. /H/ c. Accent → [H]

(10d) (i) default accent

(ii) /H/ to last q /H/ to last q

(iii) implementation implementation implementation

In (11a), the accent-cum-tone analysis, the (10d) case would be lexically unaccented. Since such words surface with an apparent H tone throughout (except for the initial mora), one could consider assigning a default final accent (11.i), which then triggers an H tone. This analysis encounters a problem, however. Words that have no lexical accent must be identifiable as such in the phonetic interpretation because there is a phonetic difference between (10c) and (10d). Roughly speaking,

Pitch Accent Systems

17

(10c) is LHH and (10d) is LHM, with the stem-final “H” in the latter not quite as high as the other Hs in both examples. The two types of words also have different effects on following words (or “accentual phrases”) inside the Intermediate Phrase: (10c) causes downstep, (10d) does not; cf. Haraguchi (1988); Pierrehumbert and Beckman (1988); Gussenhoven (2004). Alternative (ii), which would use the H tone assignment rule in (12), resolves this issue, because it can be argued that a H tone on an accented syllable and a H tone on an unaccented syllable are interpreted differently (cf. (5a)): (12)

Assign /H/ to the first accent or, if there is no accent, to the final syllable.15

The difference between (10c) and (10d) could also be described if the pitch properties of the latter class are entirely accounted for in terms of phonetic implementation (11.iii), since this system could respond to the presence vs. absence of an accent. In the second (tone-only) approach, (10d) should be accounted for by method (iii), since method (ii), available in principle, would wrongly conflate (10c) and (10d), as there is no longer an accent to differentiate between them. Finally, in the third method, (11c) (accent only), both classes must be differentiated in the phonetic implementation: accent is interpreted as high pitch, while lack of accent is interpreted differently, although also in terms of elevated pitch. I have briefly discussed three different approaches to a system such as that of Tokyo Japanese nouns, namely those mentioned in (11). All three approaches have been defended in the literature in one form or another. The tone accent approach (although often called “pitch accent approach”), (11a), comes closest to the analysis offered in McCawley (1968). Lexically, the language is accentual, but in the course of the derivation (presumably at the word level) tone is added, and from that point on the language is tonal. This approach was adopted as part of the autosegmental analysis of languages like Japanese and other monomelodic systems (cf. Goldsmith 1976b; Haraguchi 1977, 1988). The tone-only approach, (11b), has been advocated by Meeussen (1972), Poser (1984), Pulleyblank (1986), Clark (1987), and Pierrehumbert and Beckman (1988). Lockwood (1983) is a clear representative of (11c), the pitch accent analysis. To what extent do these linguists recognize the possibilities in (11), other than the one that they propose for Japanese, as valid for other languages? Clark (1988) rejects (11a) as a theoretical option, but claims that (11c) represents an independent possibility, alongside (11b). She makes a distinction between restricted tonal systems, i.e. (11b), and metrical pitch accent systems, i.e. (11c). The difference between the two types is claimed to be that only metrical pitch accent systems have the characteristics that we also find in non-tonal accent languages with respect to accent location (e.g. influence of syllable weight) and other phonetic cues that occur as the manifestation of accent. In her restricted tonal languages, the alleged accent is simply a tone at every level of representation (Clark 1988: 52). An argument for analyzing Tokyo Japanese as tonal would be the fact that we have words like (10d), distinct from (10c). In a tonal analysis, this difference is expected, since words do not have to have a tone. But in an accentual analysis, 15

Here I have added “first” to the rule because, if a word ends up having more than one accent, it is always the leftmost accent that attracts the H tone.

Harry van der Hulst

18

the class of unaccented languages has been seen as unexpected (see also Duanmu 2004). I have shown, however, that accentless words are not an embarrassment if accents need not be obligatory. Let us now ask how the high pitch profile in class (10d) could be analyzed as not resulting from a /H/ tone (supplied by default, i.e. (11.ii)), but rather as emerging in the phonetic implementation (i.e. (11.iii)). In the approach of Pierrehumbert and Beckmann (1988) and Gussenhoven (2004) it is assumed that there are morphemes with lexical accents as well as morphemes that lack accents. Lexical accents are then associated with an H*L “pitch accent.” Up to this point, this essentially follows the accent-cum-tone approach (i.e. (11a)). The high pitch pattern of unaccented words (e.g. 10d) is due to an H “boundary tone.” The claim is that the left edge of “words” is predictably assigned a LH boundary sequence. The L part of this boundary sequence is responsible for the low initial mora of words that do not have initial accent, and the H part is responsible for the high pattern of unaccented words. This H tone associates to the second mora and pitch lowers from there towards the end of the word, explaining why a word with a final accent and an unaccented word have a different high profile. In accented words, the final syllable is realized in terms of a high target for its H*L pitch accent, while an unaccented word’s final syllable does not have a H target, but merely reflects the interpolation of the H boundary tone (which is on the left) toward the end of the word (where we find the boundary L of the next word, or, if the word is utterance-final, an utterance L boundary tone): (13)

a.

x {( ta ta ta ta )} L H _ _ H LL

b. {( ta ta ta ta )} L H _

_

L

Clearly, this analysis does not require a default accent rule for unaccented words (11.i), nor does it appeal to a default pitch accent (11.ii). Before we close this section, let us ask whether this analysis must be regarded as an accent-cum-tone approach or whether it can also be interpreted an accentonly. The fact that there are symbols like “H” and “L” in this approach does not mean that these entities are “lexical” in any sense. I suggest that the pitch accents can be seen as phonetic entities, hence [HL] rather than /HL/. The tonal entities are part of the vocabulary of the implementation system. These entities combine with the other tonal entities that are introduced at the post-lexical level, belonging to the intonational system. Intonational entities themselves may or may not be phonological. Boundary tones that predictably associate with certain types of boundaries without expressing any specific semantics, are, likewise, phonetic entities, e.g. [L] or [H]. Phonetic implementation operates on the representations that the grammar supplies. When it comes to the specification of pitch, the following entities are minimally relevant: tones (lexical or intonational), “accents,” and

Pitch Accent Systems

19

prosodic boundaries. We can, if accents have high pitch, first assign [H] to accent and then do the actual implementation. The same applies to boundaries; we can assign a [H] to the left boundary of a certain type of prosodic phrase. Strictly speaking, we only do this in order to make the implementation rules refer to only one type of entity (namely tonal entities, whether phonetic or phonological) instead of to three different types of entities (tones, accents, and boundaries). In any event, it would seem that the pitch profiles of Tokyo Japanese words do not require reference to word-level tones.

7

Accent and stress

A discussion of pitch accent systems forms part of the broader discussion of word prosodic systems. However, having made reference in the preceding sections to a view that recognizes both accent and stress as independent notions, this section will briefly discuss their properties and interaction. We might entertain the idea that the alleged accents in Tokyo Japanese are simply “marks” which are to be compared to syllable weight. If this comparison holds, we might refer to the accents as “diacritic weight marks” and in that case there is no reason for every word to have one such mark, just as languages that have a contrast between CV (light) and CVX (heavy) syllables typically do not necessarily demand that each word has a “heavy syllable.” Nor, for that matter, do we expect words to have only one “accent,” since words also can have more than one heavy syllable. This interpretation of “accents” explains the occurrence of unaccented words and multiple accented words in specific systems. A problem with this approach is that weight diacritics have characteristics that are more reminiscent of “stress” than of heavy syllables, notably predictability. This can be illustrated by taking a closer look at the accentual system of Tokyo Japanese. See chapter 120: japanese pitch accent, where it is shown that the Tokyo Japanese accent rule is very similar to the Latin-style English accent rule. We now have a new problem. If the Tokyo Japanese accents are like weight, (a) why is their distribution predictable by rule, and (b) why is the rule so similar to the typical “stress” placement rules? And why are there accentual systems in which accent is culminative and/or obligatory? To resolve these issues, van der Hulst (2009, 2010c) proposes to account for accent and “rhythm,” which traditional metrical theories conflate in one representation, in two different modules. The accentual module accounts for the location of so-called primary accent or primary “stress” in systems where this location shows influence of lexical factors (exceptions, morphological classes, etc.), while the rhythmic module associates words with metrical structures. This separation of tasks allows a simpler version of the metrical system, which, as van der Hulst shows, cannot handle all varieties of primary accent locations in bounded systems and is simply not designed to deal with accent locations in unbounded systems. The theory of accent that has been suggested is “liberal,” in that accent is required neither to be culminative nor obligatory. While this allows four different kinds of pitch accentual systems, it might be argued that we now predict four kinds of any sort of accent system, whatever the cues for accent. Focusing on the specific case of stress accent languages, Hyman (2009) argues that in such systems “stress” is always culminative and obligatory.

20

Harry van der Hulst

We can explain the culminativity and obligatoriness of stress by developing a proper understanding of what is meant by “stress.” Instead of saying that the metrical module account, for the rhythmic structure of words, we could simply say that it accounts for stress, thus taking the term to stand for the overall metrical structure of words. In this view, metrical structure is placed on the same level as pitch, i.e. as a word-level property that is assigned to words with reference to accents (if present), which in this capacity are, as previously stated, pre-specified metrical heads. The difference is that, while pitch is an exponent of accent (and thus is absent if there is no accent), metrical structure is a parametric choice that is made for the language as a whole. If a word has an accent, this accent determines the manner in which the metrical structure is associated to the word. If there is no accent, the metrical structure resorts to a default mode of association. This means that languages can have stress without accent (when stress is fully automatic, and often variable),16 and accent without stress (in which case accent has cues such as pitch).

8

Conclusions

In this chapter we have considered the phenomenon of pitch accent, which has necessarily entailed a detailed discussion of the notion “accent.” I have focused on analytical issues, i.e. on how definitions of basic notions such as tone, accent, and stress allow or disallow certain types of analysis. Alongside the idea that lexical relevance or salience of pitch is a sufficient condition for tone, we have considered a more conservative view, which insists on distinctivity. Whereas the former view can essentially do away with pitch accent as a prosodic type, the latter view is compelled to adopt this notion in cases where pitch is not distinctive. I then showed that even systems in which pitch appears to function distinctively can be analyzed in terms of accents, if accents are neither required to be obligatory nor culminative. There is thus a class of systems that is ambiguous between a tonal and an accentual analysis. In summary, the two opposing views in this debate are those that maximize the use of tone (giving up distinctivity as a necessary criterion) and those that maximize the use of accents (which are neither necessarily obligatory nor culminative). By developing a specific notion of accent, we then considered the relationship between accent and non-pitch properties covered by the umbrella term “stress,” making the perhaps obvious connection between stress and rhythmic or metrical structure. This view is further developed in van der Hulst (2010c). Let us finally observe that the status of word-level pitch properties is not entirely unique. All the distinctions that we can establish for relationships between accent and pitch can also be established for accent and properties such as duration and vowel quality. Note that in these domains we do not encounter the claim that any word level occurrence of duration or vowel quality automatically entails the phonological categories “length” and “tense.” This, then, presents an asymmetry in the assessment of what is considered to be phonological: why speak of tone (instead of accent) if pitch is not used distinctively (and is thus a predictable cue 16

A case in point would be Indonesian stress; cf. Odé and van Heuven (2004).

Pitch Accent Systems

21

of accent) if, at the same time, cases in which accent is cued by non-distinctive duration or vowel quality are not analyzed as involving lexical specification of length or of non-distinctive vowel features?

ACKNOWLEDGMENTS I would like to thank Carlos Gussenhoven, Marc van Oostendorp, Keren Rice, and two anonymous reviewers for valuable comments on earlier drafts of this chapter.

REFERENCES Anderson, John M. & Colin J. Ewen. 1987. Principles of dependency phonology. Cambridge: Cambridge University Press. Banti, Giorgio. 1988. Two Cushitic systems: Somali and Oromo nouns. In van der Hulst & Smith (1988), 11–49. Beckman, Jill N. 1998. Positional faithfulness. Ph.D. dissertation, University of Massachusetts, Amherst. Beckman, Mary E. 1986. Stress and non-stress accent. Dordrecht: Foris. Biber, Douglas. 1981. The lexical representation of contour tones. International Journal of American Linguistics 47. 271–282. Blevins, Juliette. 1993. A tonal analysis of Lithuanian nominal accent. Language 69. 237–273. Bolinger, Dwight L. 1982. Intonation and its parts. Language 58. 505–532. Bruce, Gösta. 1977. Swedish word accents in sentence perspective. Lund University Department of Linguistics Working Papers 12. 219–228. Bruce, Gösta. 1999. Word tone in Scandinavian languages. In van der Hulst (1999b), 605–633. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clark, Mary. 1987. Japanese as a tone language. In Takashi Imai & Mamoru Saito (eds.) Issues in Japanese linguistics, 53–106. Dordrecht: Foris. Clark, Mary. 1988. An accentual analysis of the Zulu noun. In van der Hulst & Smith (1988), 51–79. Clements, G. N. 2001. Representational economy in constraint-based phonology. In T. A. Hall (ed.) Distinctive feature theory, 71–146. Berlin & New York: Mouton de Gruyter. Clements, G. N. 2009. The role of features in speech sound inventories. In Raimy & Cairns (2009), 19–68. Clements, G. N. & John A. Goldsmith (eds.) 1984. Autosegmental studies in Bantu tone. Dordrecht: Foris. Coetsem, Frans van. 1996. Towards a typology of lexical accent: Stress accent and pitch accent in a renewed perspective. Heidelberg: Carl Winter. Cutler, Anne. 1984. Stress and accent in language understanding and production. In Dafydd Gibbon & Helmut Richter (eds.) Intonation, accent and rhythm, 77–90. Berlin & New York: Mouton de Gruyter. de Lacy, Paul. 2002. The interaction of tone and stress in Optimality Theory. Phonology 19. 1–32. Downing, Laura J. 2010. Accent in African languages. In van der Hulst et al. (2010b), 381–428. Duanmu, San. 2000. The phonology of Standard Chinese. Oxford: Oxford University Press. Duanmu, San. 2004. Tone and non-tone languages: An alternative to language typology and parameters. Language and Linguistics 5. 891–924.

22

Harry van der Hulst

Fox, Anthony. 2000. Prosodic features and prosodic structure: The phonology of suprasegmentals. Oxford: Oxford University Press. Fromkin, Victoria A. (ed.) 1978. Tone: A linguistic survey. New York: Academic Press. Fukui, Rei. 2003. Pitch accent systems in Korean. In Shigeki Kaji (ed.) Cross-linguistic studies of tonal phenomena: Historical development, phonetics of tone, and descriptive studies, 275–286. Tokyo: ICLAA. Garde, Paul. 1968. L’Accent. Paris: Presses Universitaires de France. Goldsmith, John A. 1976a. Autosegmental phonology. Ph.D. dissertation, MIT. Goldsmith, John A. 1976b. An overview of autosegmental phonology. Linguistic Analysis 2. 23–68. Goldsmith, John A. 1981. English as a tone language. In Didier L. Goyvaerts (ed.) Phonology in the 1980s, 287–308. Ghent: E. Story-Scientia. Goldsmith, John A. 1988. Prosodic trends in the Bantu languages. In van der Hulst & Smith (1988), 81–93. Gordon, Matthew. Forthcoming. Disentangling stress and pitch accent: toward a typology of different prosodic levels. In Harry van der Hulst (ed.) Word accent: Theoretical and typological issues. Berlin & New York: Mouton de Gruyter. Greenberg, Joseph H. & Dorothea Kaschube. 1976. Word prosodic systems: A preliminary report. Working Papers on Language Universals 20. 1–18. Gussenhoven, Carlos. 2004. The phonology of tone and intonation. Cambridge: Cambridge University Press. Gussenhoven, Carlos & Gösta Bruce. 1999. Word prosody and intonation. In van der Hulst (1999b), 233–271. Halle, Morris & Jean-Roger Vergnaud. 1982. On the framework of autosegmental phonology. In Harry van der Hulst & Norval Smith (eds.) The structure of phonological representations, part II, 65–82. Dordrecht: Foris. Haraguchi, Shdsuke. 1977. The tone pattern of Japanese: An autosegmental theory of tonology. Tokyo: Kaitakusha. Haraguchi, Shdsuke. 1988. Pitch accent and intonation in Japanese. In van der Hulst & Smith (1988), 123–150. Haraguchi, Shdsuke. 1991. A theory of stress and accent. Dordrecht: Foris Publications. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hermans, Ben. 1994. The composite nature of accent, with case studies of the Limburgian and Serbo-Croatian pitch accent. Ph.D. dissertation, Free University, Amsterdam. Hockett, Charles F. 1955. A manual of phonology. Baltimore: Waverly Press. Hollenbach, Barbara. 1988. The asymmetrical distribution of tone in Copala Trique. In van der Hulst & Smith (1988), 167–182. Hualde, José Ignacio. 1999. Basque accentuation. In van der Hulst (1999b), 947–994. Hulst, Harry van der. 1999a. Word accent. In van der Hulst (1999b), 3–115. Hulst, Harry van der (ed.) 1999b. Word prosodic systems of the languages of Europe. Berlin & New York: Mouton de Gruyter. Hulst, Harry van der. 2009. Brackets and grid marks or theories of primary accent and rhythm. In Raimy & Cairns (2009), 225–245. Hulst, Harry van der. 2010a. Word accent systems in the languages of Europe. In van der Hulst et al. (2010b), 429–508. Hulst, Harry van der. 2010b. Word accent: Terms, typologies and theories. In van der Hulst et al. (2010b), 3–54. Hulst, Harry van der. 2010c. Word accentual systems. Unpublished ms., University of Connecticut. Hulst, Harry van der & Norval Smith. 1988. Autosegmental studies on pitch accent. Dordrecht: Foris. Hulst, Harry van der, Keren Rice & W. Leo Wetzels. 2010a. Word accent systems in the languages of Middle America. In van der Hulst et al. (2010b), 249–312.

Pitch Accent Systems

23

Hulst, Harry van der, Rob Goedemans & Ellen van Zanten (eds.) 2010b. A survey of word accentual systems in the language of the world. Berlin & New York: Mouton de Gruyter. Hyman, Larry M. 1977. On the nature of linguistic stress. In Larry M. Hyman (ed.) Studies in stress and accent, 37–82. Los Angeles: Department of Linguistics, University of Southern California. Hyman, Larry M. 1978. Tone and/or accent. In Donna Jo Napoli (ed.) Elements of tone, stress and intonation, 1–20. Washington: Georgetown University Press. Hyman, Larry M. 1981. Tonal accent in Somali. Studies in African Linguistics 12. 169–203. Hyman, Larry M. 1982. Globality and the accentual analysis of Luganda tone. Journal of Linguistic Research 2–3. 1–40. Hyman, Larry M. 1989. Accent in Bantu: An appraisal. Studies in the Linguistic Sciences 19. 115–134. Hyman, Larry M. 2001. Tone systems. In Martin Haspelmath, Ekkehard König, Wulf Oesterreicher & Wolfgang Raible (eds.) Language typology and language universals: an international handbook, vol. 2, 1367–1380. Berlin & New York: Mouton de Gruyter. Hyman, Larry M. 2006. Word-prosodic typology. Phonology 23. 225–257. Hyman, Larry M. 2007. Universals of tone rules: 30 years later. In Tomas Riad & Carlos Gussenhoven (eds.) Tones and tunes, vol. 1: Typological studies in word and sentence prosody, 1–34. Berlin & New York: Mouton de Gruyter Hyman, Larry M. 2009. How (not) to do phonological typology: The case of pitch-accent. Language Sciences 31. 213–238. Inkelas, Sharon & Draga Zec. 1988. Serbo-Croatian pitch accent: The interaction of tone, stress, and intonation. Language 64. 227–248. Kodzasov, Sandro. 1999. Caucasian: Daghestanian languages. In van der Hulst (1999b), 995–1020. Ladd, D. Robert. 2008. Intonational phonology. 2nd edn. Cambridge: Cambridge University Press. Leben, William R. 1971. Suprasegmental and segmental representation of tone. Studies in African Linguistics. Supplement 2. 183–200. Liberman, Mark. 1975. The intonational system of English. Ph.D. dissertation, MIT. Published 1979, New York: Garland. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. Lockwood, D. G. 1982. Parameters for a typology of word accent. Lacus Forum 9. 231–240. Lockwood, D. G. 1983. Tone in a non-substantive theory of language. Lacus Forum 10. 131–140. MacAulay, Donald (ed.) 1992. The Celtic languages. Cambridge: Cambridge University Press. Martinet, André. 1960. Eléments de linguistique générale. Paris: Librairie Armand Colin. McCawley, James D. 1968. The phonological component of a grammar of Japanese. The Hague & Paris: Mouton. McCawley, James D. 1978. What is a tone language? In Fromkin (1978), 113–131. Meeussen, A. E. 1972. Japanese accentuation as a restricted tone system. Papers in Japanese Linguistics 1. 267–270. Mock, Carol. 1988. Pitch accent and stress in Isthmus Zapotec In van der Hulst & Smith (1988), 197–223. Nellis, Donald G. & Barbara Hollenbach. 1980. Fortis versus lenis in Cajonos Zapotec phonology. International Journal of American Linguistics 46. 92–105. Odden, David. 1988. Predictable tone systems in Bantu. In van der Hulst & Smith (1988), 225–251. Odé, Cecilia & Vincent van Heuven (eds.) 1994. Phonetic studies of Indonesian prosody. University of Leiden: Vakgroep Talen en Culturen van Zuidoost-Azië en Oceanië. Pierrehumbert, Janet B. 1980. The phonology and phonetics of English intonation. Ph.D. dissertation, MIT. Pierrehumbert, Janet B. & Mary E. Beckman. 1988. Japanese tone structure. Cambridge, MA: MIT Press.

24

Harry van der Hulst

Pike, Kenneth L. 1948. Tone languages. Ann Arbor: University of Michigan Press. Poser, William J. 1984. The phonetics and phonology of tone and intonation in Japanese. Ph.D. dissertation, MIT. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19–100. Pulleyblank, Douglas. 1986. Tone in Lexical Phonology. Dordrecht: Reidel. Raimy, Eric & Charles Cairns (eds.) 2009. Contemporary views on architecture and representations in phonology. Cambridge, MA: MIT Press. Rice, Keren. 2010. Accent in the native languages of North America. In van der Hulst et al. (2010b), 155–248. Schadeberg, Thilo C. 1973. Kinga: A restricted tone system. Studies in African Linguistics 4. 23–47. Schiering, René & Harry van der Hulst. 2010. Word accent systems in the languages of Asia. In van der Hulst et al. (2010b), 509–614. Schuh, Russell G. 1977. Tone rules. In Fromkin (1978), 221–256. Selkirk, Elisabeth. 1996. The prosodic structure of function words. In James L. Morgan & Katherine Demuth (eds.) Signal to syntax: Bootstrapping from speech to grammar in early acquisition, 187–213. Mahwah, NJ: Lawrence Erlbaum. Suárez, Jorge A. 1983. The Mesoamerican Indian languages. Cambridge: Cambridge University Press. Trubetzkoy, Nikolai S. 1939. Grundzüge der Phonologie. Göttingen: van der Hoeck & Ruprecht. Translated 1969 by Christiane A. M. Baltaxe as Principles of phonology. Berkeley & Los Angeles: University of California Press. Voorhoeve, Jan. 1973. Safwa as a restricted tone system. Studies in African Linguistics 4. 1–21. Weidert, Alfons. 1981. Tonologie: Ergebnisse, analysen, vermutungen. Tübingen: Niemeyer. Welmers, William E. 1973. African language structures. Berkeley: University of California Press. Wetzels, W. Leo & Sergio Meira. 2010. A survey of South American stress systems. In van der Hulst et al. (2010b), 313–380. Williamson, Kay. 1988. Tone and accent in :j•. In van der Hulst & Smith (1988), 253–278. Wright, Martha. 1983. A metrical approach to tone sandhi in Chinese dialects. Ph.D. dissertation, University of Massachusetts, Amherst. Wright, Martha. 1988. A level-based model for pitch accent languages. In van der Hulst & Smith (1988), 295–316. Yip, Moira. 1980. The tonal phonology of Chinese. Ph.D. dissertation, MIT. Yip, Moira. 2002. Tone. Cambridge: Cambridge University Press. Zanten, Ellen van & Philomena Dol. 2010. Word stress and pitch accent in Papuan languages. In van der Hulst et al. (2010b), 113–154. Zanten, Ellen van, Ruben Stoel & Bert Remijsen. 2010. Stress types in Austronesian languages. In van der Hulst et al. (2010b), 87–112.

43 Extrametricality and Non-finality Brett Hyde

1

Introduction

Extrametricality and non-finality are often treated as two terms that refer to the same principle of final stress avoidance, the former being implemented in a rule-based framework and the latter being implemented in a constraint-based framework. The labels and native frameworks of the two approaches, however, do not constitute the full extent of their differences. As Prince and Smolensky (1993) explain, extrametricality focuses on the parsability of prosodic constituents, while nonfinality focuses on the position of stress peaks. Extrametricality is concerned primarily with dominance relations, and non-finality is concerned primarily with prominence relations. Liberman and Prince (1977) introduced extrametricality in their foundational work on metrical stress theory to capture the apparent exclusion of certain English suffixes from the domain of stress rules.1 Recognizing the potentially wide range of applications, Hayes (1980) proposed the general formulation for extrametricality rules in (1), where the initial or final constituent of a particular domain is designated as extrametrical. (1)

1___ ]D 5 C → [+extrametrical] / 2 6 3D[___ 7

The result of extrametricality is essentially invisibility to the application of subsequent rules. When a constituent is designated as extrametrical, it is excluded from the domain of rules that might incorporate it into higher levels of prosodic structure. An extrametrical segment cannot be associated with a mora; for example, an extrametrical syllable cannot be footed, and an extrametrical foot cannot be included in a prosodic word. 1

Liberman and Prince introduce the notion of extrametricality to account for the apparent invisibility to stress rules of final -y in English: “From our point of view, -y functions as a kind of ‘extrametrical’ syllable; it simply does not take part in the metrical calculation” (Liberman and Prince 1977: 293). Later in the same paragraph: “-y is effectually hors de combat in the basic determination of metrical structure.” The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0043

2

Brett Hyde

As part of his general approach, Hayes proposed four restrictions on extrametricality. The first, constituency, ensures that only constituents – segments, syllables, feet, affixes, and so on – can be designated as extrametrical. Peripherality restricts extrametrical constituents to the edges of a domain, while edge markedness prefers that they occur at the right edge. Finally, non-exhaustivity ensures that extrametricality cannot exhaust the domain of a rule, preventing it from applying altogether. Prince and Smolensky (1993) incorporated similar restrictions into non-finality when they presented it as a replacement for extrametricality as part of their initial work on Optimality Theory. As (2) indicates, non-finality only applies at the edge of a domain (peripherality), and it only applies at the right edge in particular (edge markedness). The stress peaks that must avoid the right edge are prosodic categories (constituency) that are the heads of larger categories. (2)

Head-based non-finality No head Cat1 of a Cat2 occurs in final position in Cat3 (where Cat1, Cat2, and Cat3 are prosodic categories).

The effect of non-finality constraints is to prevent prominent categories – the heads that represent stress peaks – from occurring at the right edge of a domain. Nonfinality might prevent the head moras of syllables from occurring at the right edges of feet, for example, or head syllables of feet from occurring at the right edge of a prosodic word. Although it is usually a simple matter to distinguish non-finality from extrametricality, some approaches do exhibit characteristics of both. This is especially true of approaches that target relationships between final constituents and entries on the metrical grid, the classical device for representing stress patterns (see chapter 41: the representation of word stress). As (3) indicates, the nonfinality constraints of Hyde (2003) prohibit stress peaks – grid entries – in final position, but they specify a particular final constituent that stress must avoid. (3)

Grid-based non-finality No Cat1-level grid entry occurs over the final Cat2 of Cat3 (where Cat1, Cat2, and Cat3 are prosodic categories).

Under the grid-based approach, a non-finality constraint might prevent foot-level grid entries (secondary stress) from occurring over the final mora of a foot, for example, or prosodic word-level entries (primary stress) from occurring over the final foot of a prosodic word. The grid-based non-finality approach is like the head-based approach, then, in that it focuses on stress peaks, but it is similar to an extrametricality approach in that it excludes a particular final element from associating with some structure (in this case, certain levels of the metrical grid). A similar mixture of characteristics can be found in approaches that are typically considered extrametricality approaches. Since the grid-based account of Prince (1983) lacked feet, for example, the effect of extrametricality was to prevent syllables from mapping to the metrical grid – in other words, from associating with a stress peak – rather than to prevent them from being footed.

Extrametricality and Non-finality

3

As we shall see below, extrametricality and non-finality are among the most well-motivated principles in phonological theory, with support coming from several different lines of evidence. Perhaps the most compelling, however, is that they can be usefully applied in an unusually broad range of contexts. §2 and §3 examine phenomena involving final syllables and final feet, respectively, two types that can be handled equally well by extrametricality or non-finality. §4 examines phenomena involving final moras, a strength of non-finality approaches, and §5 examines effects involving final consonants, a strength of extrametricality approaches. In §6, I review some of the classic arguments marshaled in support of extrametricality and discuss the extent to which they also support nonfinality. Finally, §7 reviews some of the arguments for the asymmetry in edge specifications (edge markedness).

2

Final syllables

One of the most well-known uses for extrametricality and non-finality is avoidance of stress on final syllables. The most compelling examples are languages where a binary pattern is perturbed at a word’s right edge so that an anticipated final stress either arrives early or is absent altogether. An anticipated final stress arrives early, for example, in “iambic reversal” languages such as Southern Paiute (Sapir 1930, 1949), Axininca Campa (Payne et al. 1982), and Aguaruna (Payne 1990; Hung 1994). In the Aguaruna examples below, alternation of unstressed and stressed syllables from left to right would place the final stress on the ultima in even-parity forms, but it actually emerges on the penult.2 (4)

Final stress avoidance in Aguaruna a. b. c. d.

i’Œi‘naka i’Œina‘kana ŒaI’kina‘Iu‘mina ŒaI‘kina‘Iumi‘naki

‘pot (nom)’ ‘pot (acc)’ ‘your basket (acc)’ ‘only your basket (acc)’

An anticipated final stress is absent altogether in the iambic pattern of languages like Hixkaryana (Derbyshire 1985), Carib (Hoff 1968), and Choctaw (Nicklas 1972, 1975). In the Choctaw examples below, alternation of unstressed and stressed syllables would position the final stress on the ultima in even-parity forms, but the ultima and the penult both emerge without a stress. The examples are combinations of /pisa/ ‘to see’, /Œi-/ ‘you (obj)’, /-Œi/ ‘causative’, and /-li/ ‘I (subj)’. (5)

Final stress avoidance in Choctaw a. b. c.

Œi’pisa Œi’pisali Œi’pisa’Œili

An extrametricality approach would produce the Aguaruna and Choctaw patterns by making word-final syllables extrametrical and then constructing iambic feet 2

Hung (1994) infers the position of stress from the absence of vowel reduction processes. Her account is based on Payne’s (1990) description.

4

Brett Hyde

from left to right. With the final syllable extrametrical, the last two syllables in an even-parity form cannot form an iambic foot, so the expected final stress fails to appear. The difference between the two languages would be that Aguaruna tolerates degenerate feet – and can parse the penult as a degenerate foot after iambic footing is no longer possible – but Choctaw does not. Since Aguaruna can parse the penultimate syllable as a degenerate foot, as (6) illustrates, the expected final stress shifts to the penult. (6)

Aguaruna extrametricality analysis i.Œi.na.ka

extrametricality → i.Œi.na〈ka〉

→

parsing (i.’Œi)(‘na)〈ka〉

Since Choctaw cannot parse the penult as a degenerate foot, however, as (7) illustrates, the expected final stress is absent altogether. (7) Choctaw extrametricality analysis Œi.pi.sa.li →

extrametricality Œi.pi.sa〈li〉

→

parsing (Œi.’pi)sa〈li〉

For additional, and more detailed, extrametricality analyses of final stress avoidance, see Halle and Vergnaud (1987) and Hayes (1995). A non-finality approach produces the same patterns, although a bit more directly, simply by prohibiting stress at the right edge of the word. Head-based nonfinality, where heads represent stress, prohibits the head syllable of a foot from occurring in final position. Grid-based non-finality, where grid entries represent stress, prohibits a foot-level gridmark from occurring over the final syllable. In either case, prohibiting final stress effectively prohibits a final iambic foot. The difference between Aguaruna and Choctaw is in the options that they employ to avoid a final iamb. As (8) illustrates, Aguaruna employs a final trochaic foot, shifting the expected final stress to the penult. Notice that the non-finality analysis does not necessarily require underparsing like the extrametricality analysis. (In (8) and examples throughout, the expression “X ›› Y” indicates that X is more harmonic than Y or that some constraint, in this case non-finality, prefers X to Y.) (8)

Aguaruna non-finality analysis non-finality (i.’Œi)(‘na.ka) ›› (i.’Œi)(na.‘ka)

In contrast, as (9) illustrates, Choctaw prefers to leave its final two syllables unparsed in order to avoid a final iamb.3 (9)

Choctaw non-finality analysis non-finality (Œi.’pi)sa.li ›› (Œi.’pi)(sa.’li)

3

An alternative to leaving the final two syllables unparsed is to parse them into a stressless foot: (Œi.’pi)(sa.li). See Hyde (2002) for discussion.

Extrametricality and Non-finality

5

See Hyde (2002, 2003) for more detailed grid-based non-finality analyses of final stress avoidance, and McCarthy and Prince (1993) and Kenstowicz (1995) for more detailed head-based non-finality analyses. In this section, then, we have seen that extrametricality and non-finality provide equally effective analyses for the avoidance of stress on final syllables. Both approaches account for cases where a binary stress pattern is perturbed at the right edge, whether the final stress arrives early or is absent altogether.

3

Final feet

Another important use of extrametricality and non-finality has been to prevent primary stress from occurring over a word-final foot. In the clearest examples of the phenomenon, the primary stress is the penultimate stress, the presence of a secondary stress further to the right being the clearest indication that there is a final foot that primary stress might have occupied. Consider Banawá (Buller et al. 1993; Everett 1996, 1997) and Paumari (Everett 2003). In Banawá, consonant-initial words have a trochaic pattern, and vowelinitial words have an iambic pattern. In both the trochaic pattern and the iambic pattern, however, the primary stress is the penultimate stress. The secondary stress that follows indicates that there is a final foot that primary stress might have occupied if it had been drawn as far to the right as possible. (10)

Primary stress in Banawá a. b. c.

a’bari‘ko ‘metu’wasi‘ma ‘tina’rifa‘bune

‘moon’ ‘find them’ ‘you are going to work’

The primary stress is also the penultimate stress in the consistently iambic Paumari, indicating the presence of a final foot that primary stress has avoided. (11)

Primary stress in Paumari a. b. c. d.

ka’baha‘ki ‘aha’kaba‘ra a‘t hana’rari‘ki bi‘kana‘t hara’ravi‘ni

‘to get rained on’ ‘dew’ ‘sticky consistency’ ‘to cave in, to fall apart quickly’

It is a relatively simple matter to produce the Banawá and Paumari patterns with either extrametricality or non-finality. In the extrametricality approach, a wordfinal foot is designated as extrametrical, excluding it from the prosodic word. As (12) illustrates, when a right-headed prosodic word is constructed, it positions the primary stress over the penultimate foot, rather than the final. (12)

extrametricality

word layer

( x ) (x )( x )( x )→(x )( x )〈( x )〉 → ( x )( x )〈( x )〉 ti na ri fa bu ne ti na ri fa bu ne ti na ri fa bu ne

6

Brett Hyde

The non-finality approach produces a similar result, although it does not require that final feet be excluded from the prosodic word. Head-based non-finality avoids primary stress on final feet by prohibiting head feet from occurring in final position. Grid-based non-finality prohibits a prosodic word-level gridmark from occupying the final foot. non-finality

(13)

( x ) ( x ) (x )( x )( x ) (x )( x )( x ) ti na ri fa bu ne ›› ti na ri fa bu ne In either case, the primary stress and the foot associated with it are pushed back from the right edge. As a result the primary stress is the penultimate stress, and the associated foot the penultimate foot. It should be noted at this point that many of the examples cited in the literature on primary stress avoiding final feet are not as compelling as those discussed above. Hayes (1995) presents several languages as potential examples of foot extrametricality: Bedouin Arabic (Blanc 1970), Cayuga (Chafe 1977; Foster 1982; Michelson 1988), Delaware (Goddard 1979, 1982), Eastern Ojibwa (Piggott 1980, 1983), and Palestinian Arabic (Kenstowicz and Abdul-Karim 1980; Kenstowicz 1983). McCarthy (2003) points out, however, that these are not especially clear cases, because no secondary stress has been reported in a position associated with the supposed extrametrical foot. While McCarthy’s point overreaches a bit – Piggott (1983) reports post-tonic secondary stresses in Ojibwa, and patterns of reduction and non-reduction suggest post-tonic feet in Delaware – it is true of many of the traditional examples. As Banawá and Paumari demonstrate, however, the avoidance of stress on final feet is still one of the important functions performed by extrametricality and non-finality.

4

Final moras

The avoidance of final moras can make stress sensitive to the weight of syllables generally or to the weight of domain-final syllables (see chapter 57: quantitysensitivity). As we shall see below, non-finality offers a relatively straightforward analysis in such cases. §4.1 demonstrates how avoidance of syllable-final moras promotes general weight-sensitivity, §4.2 how avoidance of prosodic word-final moras promotes sensitivity to the weight of prosodic word-final syllables, and §4.3 how avoidance of foot-final moras promotes rhythmic lengthening. §4.4 examines the difficulties confronting extrametricality analyses.

4.1 General weight-sensitivity A fairly common type of weight-sensitivity is the type where stress avoids light syllables. It has been addressed, for example, using the Obligatory Branching Parameter (Hayes 1980) of classical metrical theory and the Peak-Prominence (Prince and Smolensky 1993) and Stress-to-Weight (Hammond and Dupoux 1996; Lorentz 1996) constraints of Optimality Theory. As Hyde (2006, 2007b) demonstrates,

Extrametricality and Non-finality

7

non-finality offers an alternative: when syllable-final moras cannot be stressed, stress cannot occupy a monomoraic syllable. Head-based non-finality prohibits stress on syllable-final moras by prohibiting the head mora of a foot from being final in a syllable. Grid-based non-finality prohibits foot-level gridmarks from occupying syllable-final moras. The clearest examples of stress avoiding light syllables generally are found in quantity-sensitive unbounded stress systems. Murik (Abbott 1985) and Aguacatec (McArthur and McArthur 1956) are typically used to exemplify the default-to-sameside variety.4 As (14) illustrates, Murik avoids stressing a light syllable whenever possible. If a heavy syllable is available, Murik stresses the leftmost. Murik stresses a light syllable, also the leftmost, only in those cases where heavy syllables are absent. Note that in Murik heavy syllables are syllables with long vowels. All others are light. (14)

Murik forms a. b. c. d.

’LL ’LLL LLL’H LL’HL

’damag ’dakhanqmp anHnpha’7e(t h numa’7o(go

‘garden’ ‘post’ ‘lightning’ ‘woman’

As (15) illustrates, Aguacatec stresses the rightmost heavy syllable when one is available. Otherwise, it stresses the rightmost light syllable. As in Murik, heavy syllables in Aguacatec are syllables with long vowels. (15)

Aguacatec forms a. b. c. d.

L’L LL’L L’H ’HL

ka?’pen Œinhoj’lih-ts ?in’ta( ’mi(tu?

‘day after tomorrow’ ‘they search for me’ ‘my father’ ‘cat’

The type of weight-sensitivity found in unbounded stress systems emerges when avoidance of syllable-final moras takes precedence over directional orientation. In Murik, as (16) illustrates, non-finality in the syllable takes precedence over leftward orientation. Stress appears over the leftmost heavy syllable, rather than a light syllable, even if it does not occur exactly at the left edge. (16)

Non-finality preferences in the syllable x x [ [ [[ [ ›› [ [ [[ [ nu . ma . 7oo . go nu . ma . 7oo . go

4

Languages presented as default-to-same-side systems often are not completely convincing in this classification. Since individual forms never contain more than one heavy syllable in Murik, for example, the significance of being the leftmost is less than clear. There is a similar problem with the classification of Aguacatec. McArthur and McArthur do not demonstrate the pattern for forms with more than one heavy syllable. For a more thorough discussion of non-finality’s role in both defaultto-same-side and default-to-opposite-side systems, see Hyde (2006).

Brett Hyde

8

Similarly, the Aguacatec pattern would emerge when non-finality in the syllable takes precedence over rightward orientation. Moraic non-finality constraints applied to the syllable domain, then, have the same effect as Obligatory Branching, Peak-Prominence and Stress-to-Weight. One point that favors the non-finality approach over the others is that nonfinality constraints are motivated by their usefulness in a much wider range of contexts – the avoidance of stress on final syllables (§2) and feet (§3), for example – many of which have nothing to do with syllable weight.

4.2 Weight-sensitivity in word-final syllables In this section, we examine the situation where stress is sensitive to the weight of prosodic word-final syllables only. To make stress sensitive to the weight of prosodic word-final syllables, all that is necessary is to require that prosodic wordfinal moras be stressless. When word-final moras cannot be stressed, stress can occupy a heavy final syllable, but it cannot occupy a light final syllable. One situation where stress is sensitive to the weight of prosodic word-final syllables only arises in syllabic trochee systems. Consider the case of Wergaia (Hercus 1986), where heavy syllables are syllables with long vowels (typically limited to initial position), syllables with diphthongs (limited to initial or final position), and closed syllables. As (17) illustrates, Wergaia stress is largely weight-insensitive. It falls automatically on every odd-numbered syllable counting from the left, except the final syllable. Stress falls on final syllables only if they are odd-numbered and heavy, as in (17f) and (17g). It avoids final syllables if they are light, as in (17d) and (17e). (17) Avoidance of final light syllables in Wergaia a. b. c. d. e. f. g. h. i.

’LL ’HL ’LH ’HLL ’LHL ’LL‘H ’LL‘H ’LL‘LL ’LL‘LH

’wuru ’Ia(ri ’Iarau ’ma(bila ’daguIga ’buna‘Õug ’waJa‘gai ’buna‘mala ’wureg‘wuraI

‘mouth’ ‘oak tree’ ‘wild turkey’ ‘to tell lies’ ‘to punch someone’ ‘broad-leaved mallee’ ‘catfish’ ‘fine-leaved mallee’ ‘speaking together, gabbing’

In Hyde’s (2007b) grid-based non-finality approach, the Wergaia pattern emerges when it is more important that foot-level gridmarks avoid prosodic word-final moras than it is that feet contain a stressed syllable. When the final syllable of an odd-parity form is light, stress cannot occur on the final syllable without occurring on the final mora, so the final foot emerges without a stress, as in (18). (18)

Moraic non-finality preferences in the prosodic word ( x )( ) ( x )( x) [[ [ [ ›› [[ [ [ maa . bi . la maa . bi . la

Extrametricality and Non-finality

9

When the final syllable is heavy, however, stress can occupy the final syllable without occupying the final mora, so the final foot emerges with a stress, as in (19). (19)

( x )( x ) [ [ [[ wa . Ja . gai

The same result can be produced in a more standard structural framework when non-finality in the prosodic word takes precedence over the constraints that require syllables to be parsed into feet. Odd-parity forms with a light final syllable would emerge with the final syllable unparsed and stressless. Odd-parity forms with a heavy final syllable would emerge with the final syllable parsed and stressed. The result can also be produced with head-based non-finality by prohibiting the head mora of a foot from being final in the prosodic word.5

4.3 Rhythmic lengthening Non-finality can be used not only as a simple detector of syllable weight – the use focused on in §4.1 and §4.2 – but also as a trigger to increment syllable weight. When stress would fall on an underlyingly light syllable, non-finality can force the syllable to become heavy on the surface. Rhythmic lengthening is an example of this effect. It results from avoidance of stress on foot-final moras or syllablefinal moras. There are two types of rhythmic lengthening: iambic lengthening and trochaic lengthening. The former adds a mora to the stressed syllable of an iamb; the latter adds a mora to the stressed syllable of a trochee. The iambic type appears to occur more frequently than the trochaic type (Hayes 1985, 1987, 1995; Kager 1993; chapter 44: the iambic–trochaic law), but both are well attested. Iambic lengthening can be found in Carib (Hoff 1968), for example. As (20) illustrates, Carib lengthens even-numbered syllables counting from the left, but not the final syllable, producing a fairly typical iambic pattern. (20)

Iambic lengthening in Carib a. b. c. d.

tonoro kurijara woturoporo woturopotake

→ → → →

tono(ro kuri(jara wotu(ropo(ro wotu(ropo(take

‘large bird’ ‘canoe’ ‘cause to ask’ ‘I shall ask’

Trochaic lengthening can be found in Chimalapa Zoque (Knudson 1975), a dual stress language based on trochaic footing. In Chimalapa Zoque, stress occurs on the initial syllable and the penult, with the stress on the penult being primary. As (21) illustrates, every stressed syllable must be heavy on the surface. When an underlyingly light syllable is stressed, the syllable is made heavy by lengthening its vowel. 5

An alternative approach is to rely on a foot minimality requirement to distinguish between light and heavy final syllables. This is essentially the approach adopted by Hayes (1995). As Hyde (2007b) points out, however, such an approach produces the same type of weight-sensitivity in non-final syllables, as well, where it is, unfortunately, unattested.

Brett Hyde

10 (21)

Trochaic lengthening in Chimalapa Zoque a. b. c. d.

’kosa? ‘hu’kutq ‘wqti hu’kutq ‘witu?paj’nqksq

→ ’ko(sa? → ‘hu(’ku(tq → ‘wq(ti hu’ku(tq → ‘wi(tu?paj’nqksq

‘scold (imp)’ ‘fire’ ‘big fire’ ‘he is coming and going’

Under a non-finality approach, rhythmic lengthening is just a special case of stress avoiding light syllables. To avoid stressing a light syllable, which would mean stressing a domain-final mora, the vowel of the syllable lengthens, making it heavy. Consider first the situation where non-finality prohibits stress on the final moras of feet (Kager 1995; Hyde 2007b), a head-based approach by prohibiting the head mora of the foot from being final in the foot and a grid-based approach by prohibiting foot-level gridmarks from occupying the foot-final mora. In this situation, stress must avoid light foot-final syllables, making it necessary for such syllables to lengthen if they are going to carry a stress. When avoidance of stress on foot-final moras takes precedence over the prohibition against mora insertion, the result is iambic lengthening, as (22) illustrates. (22)

Moraic non-finality preferences in the foot x x [ [[ ›› [ [ ( CV . CVV ) ( CV . CV )

Avoidance of foot-final moras cannot, however, produce lengthening in trochees. Since there is no danger of a stress occupying the final mora in a trochaic foot, lengthening would be gratuitous. Now consider the situation where stress avoids syllable-final moras. In this situation, stress must avoid a light syllable whether it is final in the foot or initial. When it takes precedence over prohibitions against mora insertion, then, as (23) illustrates, avoidance of syllable-final moras produces lengthening in both iambs and trochees. (23)

Non-finality preferences in the syllable a.

Iambic foot x x [ [[ ›› [ [ ( CV . CVV ) ( CV . CV )

b.

Trochaic foot x x [[ [ ›› [ [ ( CVV . CV ) ( CV . CV )

The non-finality approach meets the primary burden for an account of rhythmic lengthening, in that it produces both iambic lengthening and trochaic lengthening, but it has an additional advantage in that it predicts the greater frequency of lengthening among iambic systems. Non-finality in the syllable and non-finality

Extrametricality and Non-finality

11

in the foot both produce iambic lengthening, but only non-finality in the syllable produces trochaic lengthening. Since there are two sources of pressure for lengthening in iambic feet but only one for lengthening in trochaic feet, we would expect lengthening to occur with greater frequency among iambic systems, all else being equal. For further discussion of rhythmic lengthening and related issues, see chapter 44: the iambic–trochaic law.

4.4 The obstacle to an extrametricality approach As Hayes (1995) observes, there are significant obstacles from a structural standpoint to the implementation of mora extrametricality. For an extrametricality approach to produce the types of effects discussed in §4.1–§4.3 it must uniquely exclude the extrametrical mora from some higher prosodic structure. It is not clear, however, how moras can be excluded from higher prosodic structure in a way that produces the desired effects without abandoning syllable integrity (Prince 1976) or preventing extrametrical moras and their associated segments from being syllabified. For example, in the Wergaia case discussed in §4.2, word-final moras are invisible to stress assignment. This prevents stress from falling on light final syllables but still allows it to occupy heavy final syllables. There are only two ways in which this type of effect might be produced under an extrametricality approach. The first is to assume that the final mora is extrametrical and that extrametrical moras cannot be footed. When the final syllable is light, as in (24a), excluding the final mora from the foot level effectively excludes the final syllable, rendering it unstressable. When the final syllable is heavy, however, as in (24b), excluding the final mora does not entirely exclude the final syllable, allowing it to support stress. (24)

a.

Light final syllable (

x) [ [ 〈[〉 [CV] [CV] [CV]

b. Heavy final syllable ( x )( x) [ [ [〈[〉 [CV] [CV] [CVC]

The problem is most obvious in (24b). Uniquely excluding the final mora of a heavy syllable from the final foot means that the foot must split the syllable in violation of syllable integrity. Hayes explicitly rejects this possibility. Abandoning syllable integrity would make it possible for stress to occur on codas (Hayes 1995), and it would make it possible for multiple stresses to occur within a single syllable (Hyde 2007a). The second option is to assume that the final mora is extrametrical and that extrametrical moras cannot be syllabified. When the final mora is unsyllabified in CV-final words, as in (25a), no syllable can be built on the final vowel, so no stress is possible in this position. When the final mora is unsyllabified in CVVor CVC-final words, however, as in (25b), there is still a mora on which a final syllable can be constructed. Though what would otherwise form a heavy syllable only forms a light syllable, the syllable can still be footed, allowing stress to occur in final position.

12 (25)

Brett Hyde a. Final CV i.

Preceding consonant stray ( x ) [ [ 〈[〉 [CV] [CV] CV

b.

Final CVC ( x )( x) [ [ [ 〈[〉 [CV] [CV] [CV] C

ii. Preceding consonant as coda ( x ) [ [ 〈[〉 [CV] [CVC] V Hayes does not actually consider this second option and, therefore, does not reject it explicitly. Its rejection is implied, however, by his assumption that extrametricality never prevents syllabification. Unfortunately, he only justifies the assumption as it relates to extrametrical consonants, but there do seem to be good reasons for applying it to extrametrical moras as well. The analysis presents some fairly obvious problems in CV-final words. When the final mora and its associated vowel remain unsyllabified, there are essentially two options for dealing with any consonants that would otherwise be part of the final vowel’s onset. First, they might be left stray, as in (25a.i), in which case they would be subject to Stray Erasure and deleted (see §6.3). The result would be a language where final CVC and CVV sequences are preserved and their vowels stressed but final CV sequences have their consonants deleted and their vowels left stressless. To my knowledge, such an outcome is unattested. Second, preceding consonants might be incorporated into the preceding syllable as a coda, as in (25a.ii). The result would be a language where final CVV and CVC sequences always have their consonants syllabified as onsets and their vowels stressed but final CV sequences always have their consonants syllabified as codas and their vowels left stressless. To my knowledge, this outcome is also unattested.6 There is an additional, primarily theoretical, reason for rejecting extrasyllabic moras, not only in the particular situation under consideration, but in all situations. Moras are unique in that the primary motivation for including them in the prosodic hierarchy in the first place is to provide an effective representation of syllable weight, a function that cannot be performed outside the syllable. Neither of the options that might be used to achieve the desired results under an extrametricality approach, then, appears to be viable.

5

Final consonants

Evolved from proposals by Mohanan (1979) and Hayes (1980), traditional consonant extrametricality rules prevent word-final consonants from having moraic status and, therefore, from contributing to the weight of final syllables. The result is that final syllables that end in a consonant are lighter than we would otherwise 6

Similar arguments can be made against proposals that involve extrasyllabic moras acting as a sort of prosodic licenser for otherwise stray segments in order to protect them from deletion through Stray Erasure (Downing 1993; Everett 1996). Although such licensing has only been employed at the left edge, there would seem to be nothing to prevent it applying at the right edge, as well, leading to the results illustrated in (25).

Extrametricality and Non-finality

13

expect. As (26) illustrates, final CV and CVV are unaffected. Final CVC, CVVC, and CVCC, however, are all lighter than they would be otherwise. Final CVC, normally bimoraic, emerges as monomoraic and counts as light. Final CVVC and CVCC, normally trimoraic, emerge as bimoraic and count as heavy. (26)

Weight contrasts under consonant extrametricality a.

b.

Light syllables i. [ CV

ii.

[ CV〈C〉

Heavy syllables i. [[ CVV

ii.

[[ CVV〈C〉

iii.

[[ CVC〈C〉

Among the languages that have been argued to exhibit consonant extrametricality are English (Hayes 1982), various dialects of Arabic (McCarthy 1979; Hayes 1995), Ancient Greek (Steriade 1988), Spanish (Harris 1983), and Estonian (Hint 1973; Prince 1980). Examples from Estonian are provided in (27). (27) Final syllables in Estonian a. b. c. d.

’kava‘latt ’pahe‘mait ’pimestav ’pimes‘tavale

‘cunning’ ‘worse (part pl)’ ‘blinding’ ‘blinding (ill sg)’

Like Wergaia, Estonian automatically stresses every odd-numbered syllable except the final syllable. Final syllables are stressed only if they are heavy, as in (27a) and (27b). When a final syllable is light, as in (27c) and (27d), it is unstressed. Since final CVV, CVVC, and CVCC are always stressed, they must pattern together in counting as heavy. Since final CV and CVC are always stressless, they must pattern together in counting as light. This is exactly the division predicted by consonant extrametricality. Since moras are not stress peaks, non-finality cannot directly prohibit moras from associating with final consonants. Non-finality can only affect a final consonant’s moraic status by referring to a stress peak that coincides with moras. The success of a non-finality approach, however, depends crucially on the representation of stress peaks. Under head-based non-finality, no stress peak coincides with moras generally. A mora coincides with a stress peak only if it is a head mora, and banning head moras from final position does not ban all moras.7 In contrast, 7

As an anonymous reviewer points out, the claim that head-based non-finality cannot prevent final consonants from being moraic depends on the assumption that moras – unlike the higher prosodic categories – do not have heads. If moras have head segments, as argued by de Lacy (1997), then final consonants might be prevented from being moraic by prohibiting head segments from being final in the prosodic word. There are several arguments against this approach, however, one of which is that segments are not constituents of moras in the usual sense. It is often the case that multiple moras are associated with single segments. In such cases, not only would each mora have exactly the same single constituent, but it would also have exactly the same head. Neither situation is tolerated at higher prosodic levels, even in fairly permissive theories, like Hyde (2002), that allow prosodic categories to share constituents.

Brett Hyde

14

assuming that moras map to the base-level of the grid, there are stress peaks that coincide with moras generally under grid-based non-finality. By prohibiting base-level gridmarks from occurring over prosodic-word final consonants, nonfinality can prevent final consonants from associating with moras. To illustrate, when it is more important for final consonants to avoid associating with base-level gridmarks than it is for coda consonants to be moraic, final consonants will give up their moraic status to avoid associating with base-level gridmarks. Final CVC syllables emerge as monomoraic and light, and final CVVC and CVCC syllables emerge as bimoraic and heavy, resulting in the same weight distinctions among final syllables as those created by consonant extrametricality. (28)

Consonantal non-finality preferences a.

x [ CVC

››

xx [[ CVC

b.

xx [[ ›› CVVC

xx x [[[ CVVC

c.

xx [[ CVCC

xx x [[[ CVCC

››

Given its parsability focus, then, the extrametricality analysis is the most straightforward for cases like Estonian. Since it makes final consonants invisible to the process of mora assignment, consonant extrametricality produces the desired weight distinctions in a fairly direct fashion. While it is also possible to provide a nonfinality analysis, it is only possible to do so with a grid-based approach. For additional discussion of this and other issues concerning final consonants, see chapter 36: final consonants.

6

The classic arguments

We turn now to some of the classic arguments marshaled in support of extrametricality and briefly consider whether or not they also provide support for non-finality. Below we consider three of extrametricality’s traditional uses: establishing trisyllabic stress windows, helping to capture generalizations about the stress patterns of different lexical classes, and helping to provide a general account of the deletion of unsyllabifiable segments.

6.1 Eliminating ternary foot templates In many languages, a word’s final three syllables form a domain that is crucial in creating the appropriate stress pattern. The most direct option for creating such a domain – establishing it with a trisyllabic foot – has the disadvantage of making

Extrametricality and Non-finality

15

it necessary to expand the foot inventory beyond the well-motivated binary templates to include less well-motivated ternary templates. As Hayes (1980) demonstrates, a less direct extrametricality approach allows us to maintain the smaller inventory. It allows the theory to create trisyllabic windows using a binary foot followed by an unparsed syllable. Consider the stress pattern of Latin. In Latin words of at least three syllables, stress falls on either the antepenult or penult, depending on the weight of the latter. If the penult is heavy, it is stressed; otherwise, the antepenult is stressed. (29)

Trisyllabic stress window in Latin a. b. c. d.

L’HH LH’HH L’LLH L’HLH

a’mi(kus mone(’ba(mus ko’mitium do’mestikus

‘friend (nom sg masc)’ ‘warn (1pl imperf indic act)’ ‘the election site in the forum (nom sg)’ ‘domestic (nom sg masc)’

Without extrametricality, the Latin pattern requires the quantity-sensitive ternary template (Ä L q) to establish the appropriate trisyllabic domain at the right edge. When the penult is light, the template is used to construct a ternary foot at the right edge of the word, resulting in antepenultimate stress. When the penult is heavy, however, the template allows only a binary foot, resulting in penultimate stress. (30)

Ternary foot analysis a. b.

do.mes.ti.kus mo.ne(.ba(.mus

→ →

parsing do(’mes.ti.kus) mo.ne((’ba(.mus)

Extrametricality makes the ternary template unnecessary, allowing the trisyllabic domain to be formed with an unparsed syllable and a maximally disyllabic foot. The unparsed syllable is the result of syllable extrametricality. The maximally disyllabic foot is produced with the quantity-sensitive template (Ä L). If the penult is light, as in (31a), the template allows for a disyllabic foot at the right edge. In combination with the extrametrical syllable, the result is stress on the antepenult. If the penult is heavy, however, as in (31b), the template only allows for a monosyllabic foot, resulting in stress on the penult. (31)

Extrametricality analysis a. b.

do.mes.ti.kus mo.ne(.ba(.mus

→ →

extrametricality do.mes.ti〈kus〉 mo.ne(.ba(〈mus〉

→ →

parsing do(’mes.ti)〈kus〉 mo.ne((’ba()〈mus〉

As Prince and Smolensky (1993) demonstrate, a head-based non-finality approach can also construct trisyllabic domains from a binary foot and an unparsed syllable. When it is more important for the head foot to avoid final position than it is for the head foot to occur as far to the right as possible, the desired pattern emerges.

Brett Hyde

16 (32)

Head-based non-finality analysis a. b.

non-finality do(’mes.ti)kus ›› do.mes(’ti.kus) mo.ne((’ba()mus ›› mo.ne(.ba((’mus)

With the head foot pushed back from the right edge by non-finality, a disyllabic foot positions stress on the antepenult when the penult is light, and a monosyllabic foot positions it on the penult when the penult is heavy. In the case of Latin, extrametricality and head-based non-finality have a very similar effect. They both result in the final syllable being left unparsed. The similarity arises because the stress peak that must avoid final position in the non-finality analysis happens to be a foot, the head foot of the prosodic word. If the head foot must be the rightmost foot but cannot be final, then the final syllable must remain unfooted, the very situation demanded when a final syllable is made extrametrical. Two points should be kept in mind, however. The first is that grid-based nonfinality is unable to produce this same result. Since stress peaks do not double as prosodic constituents, grid-based non-finality cannot require that final syllables remain unfooted.8 Second, as Hyde (2008) points out, even head-based nonfinality does not offer a general approach to trisyllabic stress windows. It is unable to produce the stress window of Macedonian (Comrie 1976), for example. An alignment-based analysis actually provides a more successful general approach. For a discussion of ternary stress intervals more generally, not just those limited to word edges, see chapter 52: ternary rhythm.

6.2 Similarities between lexical classes In many languages, one class of lexical items exhibits one stress pattern, while a different class exhibits a slightly different pattern. In many cases, the difference can be reduced to an extrametricality effect that one class exhibits and the other does not. Once the extrametricality effect is recognized, the similarities between the patterns become apparent, and it is possible to address both with a more unified approach. English (Hayes 1982), Spanish (Harris 1983), and Yawelmani (Archangeli 1984) are among the languages where extrametricality has played an important role in this context. English is used to illustrate below. At first glance, English verbs and nouns seem to have very different stress patterns. In verbs, the position of stress depends on the shape of the final syllable. If the ultima is CVV, CVVC, or CVCC, the ultima is stressed. If the ultima is CV or CVC, the penult is stressed.

8

As an anonymous reviewer points out, whether or not grid-based non-finality can prevent the final syllable from being footed depends on the particular structures that are assumed to be the constituents of feet. If feet are actually built on base-level gridmarks, rather than syllables, preventing the final syllable from mapping to a base-level gridmark would also prevent it from being footed. The gridbased non-finality approach presented here, however, assumes that metrical structure and prosodic structure are independent, so that feet are built on syllables. Under this approach, the failure of a final syllable to map to the grid would not prevent it from being footed.

Extrametricality and Non-finality (33)

17

English verbs o’bey a’tone tor’ment

de’velop as’tonish

In nouns, the position of stress depends on the weight of the penult. If the penult is heavy, stress appears on the penult. Otherwise, it appears on the antepenult.9 (34)

English nouns a’genda e’litist Ari’zona

A’merica ’discipline ’labyrinth

As Hayes (1982) demonstrates, the difference between verbs and nouns is that they show the effects of two different types of extrametricality. The verb pattern is influenced by consonant extrametricality, the evidence being the characteristic weight distinctions among final syllables (see §5). The noun pattern is influenced by syllable extrametricality, the evidence being the presence of a trisyllabic stress window (see §6.1). Once we allow for the two different types of extrametricality, the correct patterns emerge for both verbs and nouns when we use the quantity-sensitive binary template (Ä L) to construct a foot at the right edge. In verbs, the (Ä L) template positions stress on the penult when the ultima emerges as light, once the effects of consonant extrametricality have been taken into account. It positions stress on the ultima when the ultima emerges as heavy. (35)

English verbs and consonant extrametricality a. [ [ [ de . ve . lo〈p〉

→

(x ) [ [ [ de . ve . lo〈p〉

→

( x ) [[ [[ tor . men〈t〉

b. [[ [[ tor . men〈t〉

Once syllable extrametricality excludes final syllables from the foot layer in nouns, the same (Ä L) template positions stress on the antepenult when the penult is light. It positions stress on the penult when the penult is heavy. (36)

English nouns and syllable extrametricality a.

→

( x ) A . me . ri 〈ca〉

→

(x) a . gen 〈da〉

A . me . ri 〈ca〉 b. a . gen 〈da〉 9

This generalization applies to English nouns with a stressless final syllable. Nouns with final stress must be treated differently.

Brett Hyde

18

Extrametricality, then, allows us to extract the aspects of the English verb and noun patterns that differ, in order to capture the similarities in a single general stress rule. The analysis consists of two independently motivated extrametricality rules, the source of the differences, and a single, general stress rule, the source of the similarities. If extrametricality were unavailable, we would be forced to incorporate its effects directly into separate stress rules for verbs and nouns, making both that much more complicated and obscuring the similarities between the patterns. It is not a straightforward matter to reproduce the extrametricality analysis in this case with a non-finality analysis. As mentioned in §6.1, head-based nonfinality can produce the type of stress window found in Latin and in English nouns, but grid-based non-finality cannot. As mentioned in §5, however, grid-based nonfinality can reproduce the consonant extrametricality effect seen in English verbs, but head-based non-finality cannot. Although non-finality could, in principle, help to capture similarities between the stress patterns of different lexical classes, then, its success depends very much on the facts of the particular case.

6.3 Licensing segments Itô (1986) puts extrametricality to a use that is quite different from those discussed thus far. In the types of effects discussed above, extrametricality makes a domain-final constituent invisible to rules that create prosodic structure. Itô, however, uses extrametricality to make domain-final segments invisible to Stray Erasure (Harris 1983), a rule that deletes unsyllabified segments. The result is a theory of syllabification that relies on general, rather than idiosyncratic, deletion rules. As a simple illustration, consider deletions that occur as part of the syllabification process in Diola Fogny (Sapir 1965). Diola prefers not to syllabify obstruents as codas. As seen in (37a)–(37c), a medial obstruent that would otherwise be syllabified as a coda ends up being deleted instead. The preference to avoid obstruent codas seems to be thwarted at the right edge of the word, however, as seen in (37d). Final obstruents are not deleted, even though they cannot be syllabified as anything other than a coda. (37) Obstruent deletion in Diola Fogny a. b. c. d.

letkuÁaw uÁukÁa kobkoben kuJilak

→ → → →

lekuÁaw uÁuÁa kokoben kuJilak

‘they won’t go’ ‘if you see’ ‘yearn, long for’ ‘the children’

Extrametricality accounts for the different treatment of final and medial obstruents. In the lexical phonology, Diola’s coda condition prevents obstruents from being syllabified if they would syllabify as codas. Stray Erasure then deletes any segment that remains unsyllabified and has not been designated as extrametrical. Since medial consonants cannot be designated as extrametrical – due to the Peripherality restriction – medial obstruents that fail to syllabify are always deleted, as in (38a). Since final consonants can be designated as extrametrical, however, final obstruents are invisible to Stray Erasure and escape deletion, as in (38b), even

Extrametricality and Non-finality

19

though they are not attached to a syllable. The extrametrical consonant is syllabified later in the post-lexical phonology where the coda condition does not apply. syllabification (lexical)

(38) a. b.

extrametricality

stray erasure

syllabification (post-lexical)

[le]t[ku][Áaw] → [le]t[ku][Áaw] → [le][ku][Áaw] → [le][ku][Áaw] [ku][Ji][la]k → [ku][Ji][la]〈k〉 → [ku][Ji][la]〈k〉 → [ku][Ji][lak]

In this case, then, extrametricality accounts for an asymmetry in the deletion of medial and final obstruents, making it possible to avoid an idiosyncratic deletion rule that targets medial consonants specifically. Since it does not seem to be connected to stress peaks, at least not in any direct way, it is not immediately clear how a non-finality approach could replicate this type of segmental licensing effect.

7

The edge asymmetry

Shortly after extrametricality’s introduction, it became clear that the vast majority of phenomena that might be analyzed in terms of extrametricality occur at the right edge of the relevant domain. It is for this reason that Hayes (1980) proposed the Edge Markedness restriction on extrametricality. As its name implies, the asymmetry is more absolute under the non-finality approach. In addition to the distributional evidence, two arguments have emerged to support this more absolute view. First, initial extrametricality and “non-initiality” do not have the same strong phonetic and rhythmic motivations as their right edge counterparts. Second, the inclusion of initial extrametricality or “non-initiality” in the grammar results in a significant decline in the accuracy of typological predictions.

7.1 Rhythmic and phonetic evidence In searching for potential phonetic motivations, Lunden (2007) connects final stress avoidance to phonetic final lengthening. Since phonetic lengthening also occurs in initial syllables, we might expect stress to avoid initial syllables as well. Upon closer inspection, however, this expectation rests on very shaky ground. While the characteristics of final lengthening are more compatible with the characteristics of stresslessness, the characteristics of initial lengthening are in fact more compatible with the characteristics of stress. First, consider the typical characteristics of final and initial lengthening. Oller (1973) and Wightman et al. (1992), amongst numerous others, report that final lengthening typically affects all rhyme segments to some degree, is often associated with decline in amplitude and devoicing, and is often cumulative when multiple prosodic boundaries coincide. In contrast, Oller (1973) and Keating et al. (2003), amongst others, report that initial lengthening is typically limited to the initial segment, is often associated with longer voice onset time and aspiration, and is less typically cumulative when multiple boundaries coincide. Now consider the typical characteristics of stressed syllables. Lieberman (1960), Beckman (1986), and Gordon (2002), amongst others, report that stressed syllables often exhibit increased duration in the rhyme, increased intensity in the rhyme, and fortition, lengthening, or aspiration of the onset. The fact that stressed

Brett Hyde

20

syllables often have a longer rhyme might make them seem more compatible with final lengthening. The fact that intensity declines in the rhyme under phonetic final lengthening but increases under stress, however, suggests that this is really not the case. The increased intensity in the rhyme and the strengthening of the onset makes stress more compatible with initial lengthening. Based on a parallel phenomenon in music (Gabrielsson 1987, 1993), Hyde (2009) suggests that different types of tempo changes at prosodic boundaries might account for the different characteristics of initial and final lengthening. Initial lengthening is the result of a strong attack and acceleration to medial tempo, while final lengthening is the result of a deceleration from medial tempo. An initial acceleration results in strengthening of initial segments and increased intensity in initial syllables, characteristics consistent with stress. A final deceleration results in declining intensity in final rhymes, a characteristic consistent with stresslessness.

7.2 Stress typologies The second line of evidence against initial extrametricality and “non-initiality” is that they result in a decline in the accuracy of typological predictions (Hyde 2002; Altshuler 2009). Consider, for example, the iambic patterns of Aguaruna and Choctaw, discussed in §2. They emerge when rightward binary alternation of unstressed and stressed syllables is perturbed at the right edge of even-parity forms in order to avoid final stress. Aguaruna avoids final stress by shifting it one syllable to the left, and Choctaw avoids it simply by not assigning it. (39)

a.

Unattested qÄÄqÄq qÄqÄqÄq

b. Aguaruna qÄqÄÄq qÄqÄqÄq

c.

Unattested qqÄqÄq qÄqÄqÄq

d. Choctaw qÄqÄqq qÄqÄqÄq

Although the trochaic mirror images of these patterns are both unattested, they would be predicted to occur if leftward binary alternation of unstressed and stressed syllables could be perturbed at the left edge in order to avoid initial stress. Among the attested binary patterns in general, final stress avoidance is often a reason to perturb binary alternation, but initial stress avoidance is not. Including a principle of initial stress avoidance in the grammar, then, would only result in the prediction of unattested patterns. The only requirement for initial syllables that produces attested patterns is a requirement that initial syllables be stressed. For example, in the trochaic Passamaquoddy (LeSourd 1993) and Garawa (Furby 1974) patterns, an initial stress requirement perturbs leftward binary alternation at the left edge. (40)

a. Passamaquoddy ÄqÄqÄq ÄÄqÄqÄq

b. Unattested qÄqÄqÄ qÄqÄqÄÄ

c. Garawa ÄqÄqÄq ÄqqÄqÄq

d. Unattested qÄqÄqÄ qÄqÄqqÄ

Extrametricality and Non-finality

21

Not coincidentally, given the repulsion of stress by final syllables, the iambic mirror images of these patterns are both unattested. Both final stresslessness and initial stress, then – the two aspects of the asymmetry suggested by the phonetic and rhythmic considerations discussed above – are confirmed by the typology of binary stress patterns.

7.3 Potential counterexamples While the vast majority of extrametricality and non-finality effects have been found at the right edges of prosodic domains, a few languages have been argued to exhibit extrametricality effects at the left edge. In most such cases, however, alternative analyses are readily available. Halle and Vergnaud (1987), for example, attribute the unstressability of initial vowels in Western Aranda to initial segment extrametricality. Subsequent research, however, has resulted in a number of alternative analyses of Western Aranda and similar languages, analyses that do not require initial extrametricality or noninitiality. Typically, they require the left edge of an appropriate prosodic structure to align with a consonant, preventing initial vowels from being included in that structure and, therefore, from being stressed. In Goedemans (1996), for example, the left edges of feet must align with a consonant. This prevents the initial vowel from being footed and, therefore, from being stressed. In Hyde (2007a), it is the left edges of head syllables that must align with a consonant; in Downing (1998), it is the left edges of prosodic words. Smith’s (2002) approach simply requires stressed syllables to have onsets. As a second example, in the stress patterns of Winnebago (Miner 1979; Hale and White Eagle 1980) and Kashaya (Oswalt 1961, 1988; Buckley 1992) the primary stress in a form is the leftmost stress, but it typically does not appear until the third syllable. Since this ternary interval is characteristic of both even- and oddparity forms, the most straightforward analysis is to establish a trisyllabic stress window at the left edge of the word. An initial extrametricality approach could establish the stress window by making the initial syllable extrametrical and then constructing a maximally disyllabic foot just to the right of the initial syllable. This is not necessarily strong evidence for initial extrametricality, however. As mentioned in §6.1, there are alternative approaches to trisyllabic stress windows in the literature, some of them addressing a greater variety of windows than is possible with extrametricality.

8

Summary

Extrametricality and non-finality have much in common. Both deal with peripheral positions in a domain, both deal primarily with the right edge of the domain, and both often result in final stresslessness. An important difference, however, is that extrametricality focuses on constituent parsability, while non-finality focuses on the position of stress peaks. Extrametricality rules typically prevent some domain-final constituent from being parsed into higher prosodic structure; non-finality constraints typically prevent a stress peak from occurring in some domain-final position. While they have been used to address many of the same phenomena, the difference in focus ensures that they do not address all types with equal success.

22

Brett Hyde

In §2 and §3, we saw that extrametricality and non-finality provide equally effective analyses of situations where stress avoids larger final constituents like syllables and feet. In situations where stress is avoided on final syllables, an expected final stress either arrives early or is absent altogether. An extrametricality analysis achieves the desired effect by excluding the final syllable from the foot layer, a non-finality analysis by prohibiting head syllables in final position or by prohibiting foot-level gridmarks over final syllables. In situations where primary stress avoids final feet, the primary stress emerges as the penultimate stress. An extrametricality analysis excludes the final foot from the prosodic word; nonfinality either prohibits head feet in final position or prohibits prosodic-word level gridmarks over final feet. In contrast, extrametricality and non-finality do not perform equally well in accounting for phenomena involving smaller final constituents. In §4, we saw how the avoidance of stress on word-final moras makes stress sensitive to the weight of word-final syllables, how the avoidance of stress on foot-final moras results in iambic lengthening, and how the avoidance of stress on syllable-final moras results in general weight-sensitivity, iambic lengthening, and trochaic lengthening. In these cases, a non-finality analysis is much more straightforward than an extrametricality analysis. With its stress peaks focus, non-finality can prohibit stress on domainfinal moras directly. With its parsability focus, however, extrametricality can only prohibit stress on domain-final moras by excluding them from some higher prosodic structure, a requirement that seems impossible to implement without either violating syllable integrity or requiring moras to remain unsyllabified. In §5, we saw how the failure of final consonants to contribute to syllable weight affects the stressability of final syllables. Extrametricality achieves the desired result directly by making final consonants invisible to mora assignment. A grid-based non-finality approach achieves the desired result indirectly by prohibiting moralevel gridmarks – and, thus, the moras associated with them – from occurring over final consonants. A head-based non-finality approach, however, appears to be unable to capture the effect at all. In §6, we examined some of the classic arguments for extrametricality, focusing on trisyllabic stress windows and segmental licensing, and we considered the possibility of non-finality approaches. While head-based non-finality offers analyses for some types of trisyllabic windows, grid-based non-finality does not. Recent alternative proposals for a general approach to stress windows, however, make non-finality’s limitations in this area less problematic. With respect to segmental licensing, it is not clear that a non-finality approach is even possible. Finally, §7 outlined the evidence for the edge asymmetry in extrametricality and non-finality formulations. First, the types of effects analyzable in terms of extrametricality or non-finality occur almost exclusively at right edges. Second, phonetic and rhythmic considerations motivate stresslessness in final positions, but they actually motivate stress in initial position. Third, the inclusion of initial extrametricality or non-initiality in the grammar negatively impacts the accuracy of typological predictions.

REFERENCES Abbott, Stan. 1985. A tentative multilevel multiunit phonological analysis of the Murik language. Papers in New Guinea Linguistics 22. 339–373.

Extrametricality and Non-finality

23

Altshuler, Daniel. 2009. Quantity-insensitive iambs in Osage. International Journal of American Linguistics 75. 365 –398. Archangeli, Diana. 1984. Extrametricality in Yawelmani. The Linguistic Review 4. 101–120. Beckman, Mary E. 1986. Stress and non-stress accent. Dordrecht: Foris. Blanc, Haim. 1970. The Arabic dialect of the Negev Bedouins. Proceedings of the Israeli Academy of Science and Humanities 4. 112–150. Buckley, Eugene. 1992. Theoretical aspects of Kashaya phonology and morphology. Ph.D. dissertation, University of California, Berkeley. Buller, Barbara, Ernest Buller & Daniel L. Everett. 1993. Stress placement, syllable structure, and minimality in Banawá. International Journal of American Linguistics 59. 280 –293. Chafe, Wallace L. 1977. Accent and related phenomena in the Five Nations Iroquois languages. In Larry M. Hyman (ed.) Studies in stress and accent, 169–181. Los Angeles: Department of Linguistics, University of Southern California. Comrie, Bernard. 1976. Irregular stress in Polish and Macedonian. International Review of Slavic Linguistics 1. 227–240. de Lacy, Paul. 1997. Prosodic categorisation. M.A. thesis, University of Auckland. (ROA-236.) Derbyshire, Desmond C. 1985. Hixkaryana and linguistic typology. Dallas: Summer Institute of Linguistics & University of Texas at Arlington. Downing, Laura J. 1993. Unsyllabified vowels in Aranda. Papers from the Annual Regional Meeting, Chicago Linguistic Society 29. 171–185. Downing, Laura J. 1998. On the prosodic misalignment of onsetless syllables. Natural Language and Linguistic Theory 16. 1–52. Everett, Daniel L. 1996. Prosodic levels and constraints in Banawá and Suruwaha. Unpublished ms., University of Pittsburgh (ROA-121). Everett, Daniel L. 1997. Syllable integrity. Proceedings of the West Coast Conference on Formal Linguistics 16. 177–190. Everett, Daniel L. 2003. Iambic feet in Paumari and the theory of foot structure. Linguistic Discovery 2(1). 22–44. Foster, Michael. 1982. Alternating weak and strong syllables in Cayuga words. International Journal of American Linguistics 48. 59–72. Furby, Christine. 1974. Garawa phonology. (Pacific Linguistics A37.) Canberra: Australian National University. Gabrielsson, Alf. 1987. Once again: The theme from Mozart’s “Piano sonata in A major” (K. 331). A comparison of five performances. In Alf Gabrielsson (ed.) Action and perception in rhythm and music, 81–103. Stockholm: Royal Swedish Academy of Music. Gabrielsson, Alf. 1993. The complexities of rhythm. In Thomas J. Tighe & W. Jay Dowling (eds.) Psychology and music: Understanding of melody and rhythm, 94–120. Hillsdale, NJ: Lawrence Erlbaum. Goddard, Ives. 1979. Delaware verbal morphology: A descriptive and comparative study. New York: Garland. Goddard, Ives. 1982. The historical phonology of Munsee. International Journal of American Linguistics 48. 16–48. Goedemans, Rob. 1996. An optimality account of onset-sensitive stress in quantityinsensitive languages. The Linguistic Review 13. 33 –47. Gordon, Matthew. 2002. A phonetically driven account of syllable weight. Language 78. 51–80. Hale, Kenneth & Josie White Eagle. 1980. A preliminary metrical account of Winnebago accent. International Journal of American Linguistics 46. 117–132. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: MIT Press. Hammond, Michael & Emmanuel Dupoux. 1996. Psychophonology. In Jacques Durand & Bernard Laks (eds.) Current trends in phonology: Models and methods, vol. 1, 274 –297. Salford: ESRI. Harris, James W. 1983. Syllable structure and stress in Spanish: A nonlinear analysis. Cambridge, MA: MIT Press.

24

Brett Hyde

Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Published 1985. New York: Garland. Hayes, Bruce. 1982. Extrametricality and English stress. Linguistic Inquiry 13. 227–276. Hayes, Bruce. 1985. Iambic and trochaic rhythm in stress rules. Proceedings of the Annual Meeting, Berkeley Linguistics Society 11. 429 –446. Hayes, Bruce. 1987. A revised parametric metrical theory. Papers from the Annual Meeting, North East Linguistics Society 17. 274 –289. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hercus, Luise A. 1986. Victorian languages: A late survey. (Pacific Linguistics B77.) Canberra: Australian National University. Hint, Mati. 1973. Eesti Keele Sõnafonoloogia I. Talinn, Estonia: Eeste NSV Teaduste Akadeemia. Hoff, Berend J. 1968. The Carib language: Phonology, morphonology, morphology, texts and word index. The Hague: Martinus Nijhoff. Hung, Henrietta. 1994. The rhythmic and prosodic organization of edge constituents. Ph.D. dissertation, Brandeis University (ROA-24). Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313 –359. Hyde, Brett. 2003. Nonfinality. Unpublished ms., Washington University in St Louis (ROA-633). Hyde, Brett. 2006. Towards a uniform account of prominence-sensitive stress. In Eric Bakovio, Junko Itô & John McCarthy (eds.) Wondering at the natural fecundity of things: Essays in honor of Alan Prince, 139–183. Santa Cruz: Linguistics Research Center, University of California, Santa Cruz. http://repositories.cdlib.org/lrc/prince/8. Hyde, Brett. 2007a. Issues in Banawá prosody: Onset sensitivity, minimal words, and syllable integrity. Linguistic Inquiry 38. 239 –285. Hyde, Brett. 2007b. Non-finality and weight-sensitivity. Phonology 24. 287–334. Hyde, Brett. 2008. Alignment continued: distance-sensitivity, order-sensitivity, and the Midpoint Pathology. Unpublished ms., Washington University in St Louis (ROA-998). Hyde, Brett. 2009. The rhythmic foundations of Initial Gridmark and Nonfinality. Papers from the Annual Meeting, North East Linguistics Society 38(1). 397–410. Itô, Junko. 1986. Syllable theory in prosodic phonology. Ph.D. dissertation, University of Massachusetts, Amherst. Kager, René. 1993. Alternatives to the iambic–trochaic law. Natural Language and Linguistic Theory 11. 381–432. Kager, René. 1995. Review article of Hayes (1995). Phonology 12. 437–464. Keating, Patricia, Taehong Cho, Cécile Fougeron & Chai-Shune Hsu. 2003. Domain-initial articulatory strengthening in four languages. In John Local, Richard Ogden & Rosalind Temple (eds.) Phonetic interpretation: Papers in laboratory phonology VI, 145–163. Cambridge: Cambridge University Press. Kenstowicz, Michael. 1983. Parametric variation and accent in the Arabic dialects. Papers from the Annual Regional Meeting, Chicago Linguistic Society 19. 205 –213. Kenstowicz, Michael. 1995. Cyclic vs. non-cyclic constraint evaluation. Phonology 12. 397–436. Kenstowicz, Michael & Kamal Abdul-Karim. 1980. Cyclic stress in Levantine Arabic. Studies in the Linguistic Sciences 10(2). 55–76. Knudson, Lyle M. 1975. A natural phonology and morphophonemics of Chimalapa Zoque. Papers in Linguistics 8. 283 –346. LeSourd, Philip S. 1993. Accent and syllable structure in Passamaquoddy. New York: Garland. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. Lieberman, Philip. 1960. Some acoustic correlates of word stress in American English. Journal of the Acoustical Society of America 32. 451–454. Lorentz, Ove. 1996. Length and correspondence in Scandinavian. Nordlyd 24. 111–128.

Extrametricality and Non-finality

25

Lunden, Anya. 2007. Weight, final lengthening and stress: A phonetic and phonological case study of Norwegian. Ph.D. dissertation, University of California, Santa Cruz (ROA-833). McArthur, Henry & Lucille McArthur. 1956. Aguacatec (Mayan) phonemes within the stress group. International Journal of American Linguistics 22. 72–76. McCarthy, John J. 1979. On stress and syllabification. Linguistic Inquiry 10. 443 –465. McCarthy, John J. 2003. OT constraints are categorical. Phonology 20. 75 –138. McCarthy, John J. & Alan Prince. 1993. Prosodic morphology I: Constraint interaction and satisfaction. Unpublished ms., University of Massachusetts, Amherst & Rutgers University. Michelson, Karin. 1988. A comparative study of Lake-Iroquoian accent. Dordrecht: Kluwer. Miner, Kenneth L. 1979. Dorsey’s Law in Winnebago-Chiwere and Winnebago accent. International Journal of American Linguistics 45. 25 –33. Mohanan, K. P. 1979. Word stress in Hindi, Malayalam, and Sindhi. Paper presented at MIT, as reported by Hayes (1980). Nicklas, Thurston Dale. 1972. The elements of Choctaw. Ph.D. dissertation, University of Michigan, Ann Arbor. Nicklas, Thurston Dale. 1975. Choctaw morphophonemics. In James M. Crawford (ed.) Studies in Southeastern Indian Languages, 237–250. Athens: University of Georgia Press. Oller, D. K. 1973. The effect of position in utterance on speech segment duration in English. Journal of the Acoustical Society of America 54. 1235–1247. Oswalt, Robert L. 1961. A Kashaya grammar (Southwestern Pomo). Ph.D. dissertation, University of California, Berkeley. Oswalt, Robert L. 1988. The floating accent of Kashaya. In William Shipley (ed.) In honor of Mary Haas, 611– 622. Berlin: Mouton de Gruyter. Payne, David. 1990. Accent in Aguaruna. In Doris L. Payne (ed.) Amazonian linguistics: Studies in lowland South American languages, 161–184. Austin: University of Texas Press. Payne, David, Judith Payne & Jorge Sanchez Santos. 1982. Morfología, fonología, y fonética del asheninca del Apurucayali. Pucallpa, Peru: Instituto Lingüístico de Verano. Piggott, Glyne L. 1980. Aspects of Odawa morphophonemics. New York: Garland. Piggott, Glyne L. 1983. Extrametricality and Ojibwa stress. McGill Working Papers in Linguistics 1. 80 –117. Prince, Alan. 1976. “Applying” stress. Unpublished ms., University of Massachusetts, Amherst. Prince, Alan. 1980. A metrical theory for Estonian quantity. Linguistic Inquiry 11. 511–562. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19 –100. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Sapir, Edward. 1930. Southern Paiute, a Shoshonean language. Proceedings of the American Academy of Arts and Sciences 65. 1–296. Sapir, Edward. 1949. The psychological reality of phonemes. In David G. Mandelbaum (ed.) Selected writings of Edward Sapir in language, culture, and personality, 46–60. Berkeley: University of California Press. Sapir, J. David. 1965. A grammar of Diola-Fogny. Cambridge: Cambridge University Press. Smith, Jennifer. 2002. Phonological augmentation in prominent positions. Ph.D. dissertation, University of Massachusetts, Amherst. Steriade, Donca. 1988. Greek accent: A case for preserving structure. Linguistic Inquiry 19. 271–314. Wightman, Colin W., Stefanie Shattuck-Hufnagel, Mari Ostendorf & Patti J. Price. 1992. Segmental durations in the vicinity of prosodic phrase boundaries. Journal of the Acoustical Society of America 91. 1707–1717.

44

The Iambic–Trochaic Law Brett Hyde

1

Introduction

In the development of metrical stress theory, several influential approaches (Hayes 1985, 1987, 1995; McCarthy and Prince 1986; Prince 1990) have employed the Iambic– Trochaic Law (ITL) to provide extralinguistic grounding for an account of the differences between iambic and trochaic stress systems (see also chapter 39: stress: phonotactic and phonetic evidence; chapter 40: the foot; chapter 41: the representation of word stress). The ITL, given in (1), is a statement about the naturalness of two types of rhythmic groupings in two different contexts. According to the ITL, sequences of elements that contrast in intensity most naturally divide into groups with trochaic prominence, and sequences of elements that contrast in duration most naturally divide into groups with iambic prominence. (1)

The Iambic–Trochaic Law (Hayes 1985, 1987) a. b.

Elements contrasting in intensity naturally form groupings with initial prominence. Elements contrasting in duration naturally form groupings with final prominence.

For approaches to metrical stress based on the ITL, this difference in naturalness is responsible for the duration-related differences found in iambic and trochaic stress patterns. The ITL is based on a long tradition of experimental investigation into the perception of rhythmic grouping (Bolton 1894; Woodrow 1909). In the typical experiment, participants are asked to group a sequence of artificially created alternating sounds. The sounds alternate either in intensity, as in (2a), or in duration, as in (2b). (2)

a. b.

... o O o O o O o O o O o O o O o O o O o ... ... – — – — – — – — – — – — – — – — – — – ...

The outcome, under certain circumstances, is that participants tend to divide intensity alternations into groups where the more intense element appears first, The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0044

The Iambic–Trochaic Law

2

as in (3a), and they tend to divide duration alternations into groups where the longer element appears second, as in (3b).1 The ITL is essentially a statement of these results. (3)

a.

Intensity contrasts: Left-prominent groupings . . . [O o][O o][O o][O o][O o][O o][O o][O o][O o] . . .

b.

Duration contrasts: Right-prominent groupings . . . [– —][– —][– —][– —][– —][– —][– —][– —] . . .

Though the ITL is an extralinguistic principle, it seems to be reflected in the stress patterns of numerous languages, suggesting, at least initially, that it plays an important role in shaping them. For example, many trochaic languages are like Cahuilla (Seiler 1965, 1967, 1977; Seiler and Hioki 1979). They exclude heavy syllables from disyllabic feet, ensuring that durational contrasts never arise in a foot with trochaic prominence. In Cahuilla, heavy syllables are CVV and CV?. (4)

Exclusion of H from disyllabic feet in Cahuilla (’LL)(‘L) (’LL)(‘LL) (’H)(‘L) (’H)(‘LL) (’L)(‘H)(‘L) (’L)(‘H)(‘LL)

(’taxmu)(‘?at) (’taka)(‘liŒem) (’pa?)(‘li) (’qa()(‘niŒem) (’su)(‘ka?)(‘ti) (‘nesun) (’ka)(‘vi()(‘Œi-wen)

‘song’ ‘one-eyed ones’ ‘the water (obj)’ ‘palo verde (pl)’ ‘the deer (obj)’ ‘I was surprised’

Many iambic languages are like Hixkaryana (Derbyshire 1985). They lengthen the vowel of stressed syllables, if necessary, to ensure that feet with iambic prominence always contain durational contrasts. In Hixkaryana, heavy syllables are CVV and CVC. (5)

Iambic lengthening in Hixkaryana a.

(L’L)(’H)L (khæ’næ)(’nQh)nD b. (L’L)(L’H)L (mQ’hæ)(næ’nQh)nD c. (’H)(L’L)L (’Dw)(tD’hD)næ d. (’H)(L’L)LL (’tDh)(ku’rji)hDnæ

→ → → → → → → →

(L’H)(’H)L (khæ’næ()(’nQh)nD (L’H)(L’H)L (mQ’hæ()(næ’nQh)nD (’H)(L’H)L (’Dw)(tD’hD()næ (’H)(L’H)LL (’tDh)(ku’rji()hDnæ

‘I taught you’ ‘you taught him’ ‘to the village’ ‘to Tohkurye’

This chapter reviews the strengths and weaknesses of ITL approaches to metrical stress, and examines some of the most promising alternatives. We shall see that the ITL does not actually offer an adequate foundation for an account 1

The effect emerges in the range of one half to five beats per second. (The syllable rate of “ordinary conversational speech” is typically toward the upper limits of this range; Bell 1977.) Hayes (1995) states that the right-prominent effect illustrated in (3b) requires a durational contrast where the longer elements are 1.5 to 2 times as long as the shorter elements, noting that Woodrow (1909) found that smaller durational contrasts actually result in left-prominent groupings.

Brett Hyde

3

of stress systems in general, but it may provide an adequate foundation for an account of quantity-sensitive stress systems in particular (see chapter 57: quantity-sensitivity). This is not to say that it provides the best foundation. There is a clear sense in which the superficial and descriptive ITL is itself an observation in need of an explanation, much like the stress patterns found in natural language. Part of the appeal of the most promising alternatives is that they have the potential to account not only for the stress patterns of natural language, but also for the ITL itself. Before reviewing the various proposals, we should note the results of more recent investigations into the perception of rhythmic grouping. In some cases, more recent studies have confirmed the grouping preferences found in the earlier studies on which the ITL was based. In other cases, they have challenged their universality. The studies of Rice (1992), Vos (1977), and Hay and Diehl (2007), for example, found grouping preferences among English, French, and Dutch speakers similar to those found in the earlier studies of Bolton (1894) and Woodrow (1909).2 The studies of Kusumoto and Moreton (1997) and Iversen et al. (2008), however, found significant differences between speakers of English and Japanese. Iversen et al., for example, found that English speakers had a fairly strong preference (68 percent) for dividing sequences of amplitude contrasts into trochaic (loud–soft) groups, but Japanese speakers had a much stronger preference (91 percent) for trochaic grouping. English speakers showed a very strong preference (89 percent) to divide duration contrasts into iambic (short–long) groups, but Japanese speakers showed no preference. While the challenge to universality may be troubling to those particularly concerned with extralinguistic grounding, and it certainly presents an interesting problem in this connection, it does not necessarily tell us anything about the ITL’s ability to predict differences between iambic and trochaic stress patterns in language. Having noted the problem with respect to extralinguistic grounding, then, I will not address the issue further.

2

Interpretations of the ITL

The most recent ITL accounts (McCarthy and Prince 1986; Hayes 1987, 1995; Prince 1990) reflect two distinct interpretations. The stronger of the two, given in (6), takes the ITL to be concerned with the actual presence or absence of durational contrasts within rhythmic groupings. (6)

Strong interpretation of the ITL a. b.

If a foot contains a durational contrast, it is iambic. If a foot lacks a durational contrast, it is trochaic.

The weaker interpretation, given in (7), takes the ITL to be concerned with sensitivity to the positions of the heavy syllables that might help to create durational contrasts.

2

Rice’s study also found a preference for iambic grouping when elements contrasted in pitch, a result not found in previous studies.

The Iambic–Trochaic Law (7)

4

Weak interpretation of the ITL (Hayes 1985) a. b.

If parsing is sensitive to the position of heavy syllables, it is iambic. If parsing is insensitive to the position of heavy syllables, it is trochaic.

Even at the point at which the ITL was introduced to metrical stress theory, it was clear that the strong interpretation in (6) was unsustainable, at least when applied to stress systems generally. Under the strong interpretation, iambic footing and the presence of durational contrasts are intimately connected: only iambs contain durational contrasts; durational contrasts arise only in iambic feet; and only iambic systems employ rules that create durational contrasts. Similarly, trochaic footing and the absence of durational contrasts are intimately connected: only trochees lack durational contrasts; durational contrasts are absent only in trochaic feet; and only trochaic systems employ rules that destroy durational contrasts. Even a cursory look at the general typology of attested stress patterns reveals that the strong interpretation misses the mark by a wide margin. As mentioned above, many iambic languages are like Hixkaryana, lengthening the vowel of stressed syllables, if necessary, to ensure that surface iambs contain durational contrasts. Many other iambic languages are like Araucanian (Echeverría and Contreras 1965), however. They tolerate surface iambs that have no durational contrasts. (8)

Even iambs in Araucanian (L’L) (L’L)L (L’L)(L‘L) (L’L)(L‘L)L

(wu’le) (ti’pan)to (e’lu)(mu‘ju) (e’lu)(a‘e)new

‘tomorrow’ ‘year’ ‘give us’ ‘he will give me’

A similar situation obtains with trochaic languages. As mentioned above, several are like Cahuilla in prohibiting foot-internal durational contrasts. Several others, however, are like Chimalapa Zoque (Knudson 1975). They tolerate foot-internal durational contrasts and even have rules that create them. In Chimalapa Zoque, heavy syllables are CVV and CVC. (9)

Trochaic lengthening in Chimalapa Zoque a.

(’LH) (’kosa?) b. (‘L)(’LL) (‘hu)(’kutq) c. (‘LL)L(’LL) (‘wqti) hu(’kutq) d. (‘LH)H(’HL) (‘witu?)paj(’nqksq)

→ → → → → → → →

(’HH) (’ko(sa?) (‘H)(’HL) (‘hu()(’ku(tq) (‘HL)L(’HL) (‘wq(ti) hu(’ku(tq) (‘HH)H(’HL) (‘wi(tu?)paj(’nqksq)

‘scold (imp)’ ‘fire’ ‘big fire’ ‘he is coming and going’

There appears to be no close connection between iambs and the presence of durational contrasts, then, or between trochees and the absence of durational contrasts, at least in the general case. Given the shortcomings of the strong interpretation, Hayes (1985) introduced the ITL to metrical stress theory under the weak interpretation in (7). Under the weak interpretation, the crucial connections are between iambic footing and

5

Brett Hyde

quantity-sensitivity and trochaic footing and quantity-insensitivity. While iambic feet and trochaic feet might both contain durational contrasts, parsing is iambic if and only if it is sensitive to the positions of heavy syllables. Parsing is trochaic if and only if it is insensitive to the positions of heavy syllables. There are three problems with the weak interpretation. The first is conceptual. The ITL is plainly a generalization about the appropriateness of durational contrasts within two different types of feet. Since its requirements concerning durational contrasts affect both types, the ITL does not countenance situations where either type is quantity-insensitive (where either type simply ignores the differences in syllable weight that help to create durational contrasts). In viewing the primary concern of the ITL to be the appropriateness of quantity-sensitivity for different types of feet, the weak interpretation seems really to be a misinterpretation. The second problem is a loss of empirical coverage. Since it only addresses quantity-sensitivity, the weak interpretation tells us nothing about the status of lengthening and shortening rules addressed by the strong interpretation. The final problem is that the weak interpretation is false. A significant number of trochaic systems are quantity-sensitive, falsifying (7a), and a significant number of iambic systems are quantity-insensitive, falsifying (7b). In (4), for example, we saw that heavy syllables consistently perturb the basic stress pattern of the trochaic Cahuilla, indicating that it is quantity-sensitive. In (10), we see that heavy syllables consistently fail to perturb the basic pattern of the iambic Paumari (Everett 2003), indicating that it is quantity-insensitive. In the basic pattern, stress appears on every odd-numbered syllable from the right. CVV syllables are heavy. (10)

Quantity-insensitive iambs in Paumari (‘L)(L’L) (L‘L)(L’L) (‘H)(H’L) (H‘L)(L’L)

(‘ma)(si’ko) (ka‘–o)(wi’7i) (‘kai)(hai’hi) (wai‘Œa)(na’wa)

‘moon’ ‘island’ ‘type of medicine’ ‘little ones’

Additional quantity-sensitive trochaic languages include Palestinian Arabic (Brame 1973, 1974; Kenstowicz and Abdul-Karim 1980; Kenstowicz 1983) and Fijian (Schütz 1978, 1985; Dixon 1988). Additional quantity-insensitive iambic languages include Araucanian, Osage (Altshuler 2009), Suruwaha (Everett 1996), and Weri (Boxwell and Boxwell 1966). As we shall see in §3, parts of both interpretations, (6a) of the strong interpretation and (7b) of the weak interpretation, are brought together to form the basis for two subsequent ITL accounts, those of Hayes (1987, 1995) and of McCarthy and Prince (1986). This marriage between the halves of two very different interpretations often makes the connection between the ITL and the phenomena that these approaches attempt to account for less than clear. This is part of the reason, perhaps, that some have concluded that there is actually little of the ITL left in ITL-based approaches (see van der Hulst 1999, for example). A third ITL approach, that of Prince (1990), employs only the strong interpretation, but seeks to avoid the problems discussed above by employing it only in the context of quantity-sensitive systems and only as a relative “preference” rather than an absolute “law.” Though I will point out the aspects of the more recent ITL accounts that derive from the weak interpretation, it should be clear at this point that the weak

The Iambic–Trochaic Law

6

interpretation does not give us an accurate picture of the potential for quantitysensitivity with different types of feet, so I will not address the issue in any detail. I will address in some detail, however, the support that aspects deriving from the strong interpretation find among quantity-sensitive systems. While both iambic and trochaic languages can be quantity-sensitive, differences in the way that they resume basic stress alternations after a heavy syllable is encountered indicate that they are quantity-sensitive in different ways. Iambic systems require that heavy syllables occupy the prominent position in a disyllabic foot, but trochaic systems exclude heavy syllables from disyllabic feet entirely. Since the strong interpretation predicts this difference, it might provide the foundation for an account, not of stress systems generally, but of quantity-sensitive systems in particular.

3

Quantity-sensitivity

In quantity-sensitive languages, heavy syllables are always stressed, and this often has the effect of perturbing basic stress alternations. The feature of quantitysensitivity that is most significant to the discussion here is that there is sometimes a difference between trochaic systems and iambic systems in how they resume their basic alternations after encountering a heavy syllable. Under certain circumstances, the particular way in which a system resumes its basic alternation can indicate whether it prefers to parse heavy syllables into disyllabic feet or monosyllabic feet. Whether or not a difference in resumption actually emerges, however, depends on the combination of foot type and parsing directionality the system employs. No difference emerges when the headedness of the foot and parsing directionality match. As illustrated in (5), for example, in left-to-right iambic languages like Hixkaryana, a heavy syllable is always followed by a stressless syllable. Similarly, in right-to-left trochaic languages like Fijian, as (11) illustrates, heavy syllables are always preceded by a stressless syllable. In Fijian, heavy syllables are CVV. (11)

Fijian loanwords (Schütz 1978)3 ‘LL’LL L‘LL’LL ‘LL‘H’LL L‘LL’H ‘LL‘LL’H L‘HL’H ‘H‘LL’H

‘ndiko’nesi pe‘resi’tendi ‘mbele‘mbo(’tomu pa‘lasi’ta( ‘mini‘siti’ri( pa‘raima’ri( ‘ndai‘reki’ta(

‘deaconess’ ‘president’ ‘bell-bottoms’ ‘plaster’ ‘ministry’ ‘primary’ ‘director’

There is no difference, then, between left-to-right iambic systems and right-to-left trochaic systems in this context – both resume their basic alternations with a stressless syllable – so there is no evidence for a difference in their treatment of heavy syllables. 3

Loanwords are typically employed to illustrate the Fijian stress pattern, since long native stems are uncommon and morphology can influence the position of stress.

7

Brett Hyde

The reason is simply that the foot that is constructed immediately after the heavy syllable is parsed, rather than the foot that is constructed to parse the heavy syllable itself, determines how the basic alternation resumes. Whether heavy syllables are included in disyllabic feet or parsed as monosyllabic feet, the basic alternations of both iambic and trochaic systems would resume with an unstressed syllable (the underlined syllable in the examples below). (12)

Parsing directionality matches headedness a. Left-to-right iambic i. Iamb ( x)( x) ...L H L L...

ii. Monosyllable (x)( x) ...H L L...

b. Right-to-left trochaic i. Trochee (x )( x ) ...L L H L...

ii. Monosyllable (x )(x) ...L L H...

In left-to-right iambic systems, the heavy syllable must occur at the right edge of a foot whether the foot is an iamb, as in (12a.i), or a monosyllable, as in (12a.ii). Since the next foot constructed would be iambic in either case, the alternation resumes with an unstressed syllable. In right-to-left trochaic systems, the heavy syllable would be parsed at the left edge of a foot whether the foot is a trochee, as in (12b.i), or a monosyllable, as in (12b.ii). Since the next foot constructed would be trochaic in either case, the alternation again resumes with an unstressed syllable. A difference in the resumption of basic alternations emerges only in situations where parsing directionality and the headedness of the foot do not match. In right-to-left iambic languages like Tübatulabal (Voegelin 1935), as (13) illustrates, heavy syllables are always preceded by stressless syllables. In Tübatulabal, heavy syllables are CVV(C). (13)

Resumption with a stressless syllable in Tübatulabal ’LL’L L’LL’H L’LL’L ’LL’LL’HL’L ’H’LL’L ’H’LL’H

’ŒiIi’jal ti’Iija’laap wi’taIha’tal ’witaI’hata’laaba’tsu ’taa’hawi’la ’taa’hawi’laap

‘the red thistle’ ‘on the red thistle’ ‘the Tejon Indians’ ‘away from the Tejon Indians’ ‘the summer’ ‘in the summer’

In left-to-right trochaic languages like Cahuilla, however, as illustrated in (4), heavy syllables are always followed by stressed syllables. The difference between right-to-left iambic systems and left-to-right trochaic systems, then, is that the former resume their basic alternations with stressless syllables while the latter resume them with stressed syllables. The reason that a difference emerges when headedness and parsing directionality do not match is that the resumption of basic alternations depends directly on how the heavy syllable itself is footed.

The Iambic–Trochaic Law (14)

8

Mismatch between parsing directionality and headedness a. Right-to-left iambic i. Iamb ( x)( x) ...L L L H...

ii. Monosyllable *( x) (x) ...L L H...

b. Left-to-right trochaic i. Trochee * (x )(x ) ...H L L L...

ii. Monosyllable (x)(x ) ...H L L...

In right-to-left iambic systems, parsing the heavy syllable into an iamb, as in (14a.i), would position an unstressed syllable between the heavy syllable and the next stress to the left. Parsing it into a monosyllabic foot, however, as in (14a.ii), would make the next stress adjacent. The fact that right-to-left iambic languages resume their basic alternation with a stressless syllable indicates that they prefer to accommodate heavy syllables with disyllabic feet. In left-to-right trochaic systems, parsing the heavy syllable into a trochee, as in (14b.i), would position a stressless syllable between the heavy syllable and the next stress to the right. Parsing it into a monosyllabic foot, as in (14b.ii), would not. The fact that left-to-right trochaic languages resume their basic alternation with a stressed syllable indicates that they prefer to accommodate heavy syllables with monosyllabic feet. Though iambic and trochaic languages display no difference, then, in the resumption of basic stress alternations when parsing directionality and headedness match, they do show a difference when parsing directionality and headedness do not match. The difference indicates that iambic systems prefer to parse heavy syllables into disyllabic feet and trochaic systems prefer to parse heavy syllables as monosyllabic feet. As we shall see next, the foot inventories of ITL accounts capture these divergent preferences as directly as possible.

3.1

The asymmetric foot inventory

As mentioned above, some of the most recent ITL accounts fuse together parts of the strong interpretation and the weak interpretation. Hayes (1987, 1995), for example, relies on the combination to motivate two disparities in the inventory of parsing feet. (15)

a. b.

Quantity-insensitive Syllabic trochee

(q q)

Quantity-sensitive i. Moraic trochee ii. Standard iamb

(L L) or (H) (L q) or (H)

Clause (7b) of the weak interpretation, “if parsing is insensitive to the position of heavy syllables, it is trochaic,” motivates a disparity in the types of feet that can be quantity-insensitive. As (15a) indicates, Hayes’s account allows only trochaic feet to be quantity-insensitive. As discussed above, the approach is undermined by the

Brett Hyde

9

existence of several quantity-insensitive iambic languages. While Hayes argues that the quantity-insensitivity of such systems is only apparent, as they do not actually contain heavy syllables to perturb the basic pattern, the argument is plausible only in the cases of Araucanian and Weri. It is not plausible in the cases of Osage, Paumari, and Suruwaha, each of which has long vowels, diphthongs, or both. Given that both iambic and trochaic systems can be quantity-sensitive, clause (6a) of the strong interpretation, “if a foot contains a durational contrast, it is iambic,” motivates a disparity in precisely how the two types can be quantity-sensitive. As (15b) indicates, Hayes’s account requires trochaic systems to deal with heavy syllables differently than iambic systems. Iambs allow heavy syllables in strong position in disyllabic feet, where they can create durational contrasts. They exclude them only from weak position. Trochees, however, exclude heavy syllables from disyllabic feet entirely. Trochaic systems must parse heavy syllables into monosyllabic feet, where no durational contrast is possible. The disparity in how the two types of feet can be quantity-sensitive predicts the difference, discussed above, in how right-to-left iambic languages and left-to-right trochaic languages resume basic stress alternations after encountering a heavy syllable. The fact that iambic systems parse heavy syllables into disyllabic feet in Hayes’s account correctly predicts that right-to-left iambic languages will resume their basic alternation with a stressless syllable, as in (14a.i). The fact that trochaic systems must parse heavy syllables into monosyllabic feet correctly predicts that left-to-right trochaic languages will resume their basic alternation with a stressed syllable, as in (14b.ii). McCarthy and Prince (1986) arrive at a foot inventory similar to Hayes’s, but they arrive at it through a slightly different route and in service of a different purpose. They posit one type of quantity-insensitive foot: the balanced [q q] template, and two types of quantity-sensitive feet: the balanced [[ [] template and the unbalanced [q[ q[[] template. (16)

a.

Quantity-insensitive Balanced [q q]

b.

Quantity-sensitive i. Balanced [[ [] ii. Unbalanced [q[ q[[]

The ITL contributes to McCarthy and Prince’s account in two ways. First, clause (6a) of the strong interpretation, “if a foot contains a durational contrast, it is iambic,” motivates the iambic configuration of the quantitatively unbalanced foot. To guarantee that quantitatively iambic feet are also iambic with respect to stress, they posit the Quantity/Prominence Homology principle. It ensures that the heavier syllable in feet with a quantity contrast – in effect, the heavy syllable in a [q[ q[[] foot, given the limited possibilities in (16) – bears the stress. (17)

Quantity/Prominence Homology For a, b § F, if a > b quantitatively, then a > b stresswise.

Prominence in unbalanced feet is determined by the Trochaic Default principle, which ensures that [q q] and [[ [] feet both stress their initial syllable.

The Iambic–Trochaic Law (18)

10

Trochaic Default For a, b § F, if a = b quantitatively, then F = [s w].

That the single quantity-insensitive foot template, [q q], always emerges as a trochee is a second contribution of the ITL. It derives from clause (7b) of the weak interpretation, “if parsing is insensitive to the position of heavy syllables, it is trochaic.” Although they provide some minimal discussion of the asymmetric foot inventory’s role in creating stress patterns, McCarthy and Prince’s primary concern is to derive the types of feet encountered in morphological templates. The fact that the types of feet that seem to be required for creating the appropriate stress patterns in Hayes’s account are the same types that seem to be involved in morphological templates in McCarthy and Prince’s account significantly strengthens the case for the asymmetric foot inventory and the ITL. Prince’s (1990) Harmonic Parsing account also involves crucial asymmetries but, in this case, the asymmetries emerge in the preference hierarchies of iambic and trochaic systems rather than in the foot inventory itself. For Prince, quantitysensitive systems are those that obey the Weight-to-Stress principle and quantityinsensitive systems are those that do not. (19)

Weight-to-Stress If heavy, then stressed.

Focusing on the former, Prince provides three principles that can be used to determine the relative well-formedness of different types of iambic and trochaic feet in quantity-sensitive systems. The first principle, Binarity, is given in (20). It requires that feet be either disyllabic or bimoraic. (20)

Binarity Feet should be analyzable as binary.

The second and third principles, given in (21), are equivalent to the strong interpretation of the ITL in (6). |X| means “the size of X.” (21)

a. b.

Iambic Quantity In a rhythmic unit [W S], |S| > |W|, preferably. Trochaic Quantity In a rhythmic unit [S W], |S| = |W|, preferably.

Iambic Quantity expresses the preference that iambic feet contain a durational contrast, and Trochaic Quantity expresses the preference that trochaic feet not contain a durational contrast. As (22) indicates, the best-formed feet are those that respect both the relevant quantity principle and Binarity. The next best are those that respect Binarity only. The worst are those that respect neither.

Brett Hyde

11

satisfy:

(22) a. b.

Iambic Trochaic

IQ/TQ, Binarity [L H] [L L], [H]

Binarity only › ›

[L L], [H] [H L]

neither › ›

[L] [L]

The asymmetry in this case arises in how the iambic and trochaic hierarchies order balanced and unbalanced feet. Iambs prefer unbalanced [L H] to balanced [L L] and [H], but trochees prefer balanced [L L] and [H] to unbalanced [H L]. The main difference between the inventory of quantity-sensitive feet in Prince’s and Hayes’s accounts is the possibility of unbalanced [H L] trochees. As Prince notes, however, with their lesser status, there are limited situations in which unbalanced trochees might arise. First, consider the result of Harmonic Parsing in a left-to-right trochaic system. Since feet are constructed serially, and parsing the next syllable in line is the overriding concern, Harmonic Parsing would parse the H of an HL sequence as monosyllabic foot, just like Hayes’s moraic trochees. If the following L could not combine with another L to form a disyllabic foot, giving [H][LL], it might be parsed as a monosyllable, giving [H][L], or left unparsed, giving [H]L, depending on whether or not degenerate feet are tolerated. These are the same options available under Hayes’s moraic trochees. The results are rather different in right-toleft systems. Harmonic Parsing would always parse an HL sequence into an [HL] foot, but Hayes’s moraic trochees would yield either [H][L] or [H]L, depending on whether or not degenerate feet are tolerated. The latter option results in the same stress pattern, but the former does not. I am not aware, however, of a right-to-left trochaic language that would allow us to distinguish between the two approaches.

3.2

A symmetric foot inventory

As discussed above, the evidence for a difference between iambic and trochaic quantity-sensitivity comes from systems where parsing directionality is opposite the headedness of the foot. After left-to-right trochaic parsing encounters a heavy syllable, binary alternation resumes with a stressed syllable. In contrast, after rightto-left iambic parsing encounters a heavy syllable, binary alternation resumes with a stressless syllable. Where ITL approaches posit an asymmetric foot inventory to account for this difference, Kager (1993) proposes a symmetric foot inventory, arguing that the difference is best explained in terms of the metrical principles of clash and lapse avoidance. Kager distinguishes between parsing feet and the surface feet that can be formed later through adjunction of unparsed syllables. The inventory of parsing feet is symmetric. The quantity-insensitive syllabic trochee corresponds to a mirror-image syllabic iamb. The quantity-sensitive moraic trochee corresponds to a mirrorimage moraic iamb. Iambic quantity-sensitivity and trochaic quantity-sensitivity are identical, then, in that both exclude heavy syllables from disyllabic feet. (23)

Parsing feet

Syllabic (quantity-insensitive)

trochaic iambic (x ) ( x) q q q q

Moraic (quantity-sensitive)

(x ) [ [

( x) [ [

The Iambic–Trochaic Law

12

Crucial to Kager’s account is the claim that heavy syllables contain an internal prominence contrast corresponding to a decline in sonority between the first and second mora. According to Kager, the internal contrast is the characteristic of heavy syllables responsible for their attraction of stress (Prince’s 1990 Weight-to-Stress principle). The decline in sonority ensures that stress occurs over the first mora and that the second mora is stressless, as in (24). (24)

(x

)

H A heavy syllable’s strong–weak contour translates into different results with respect to clash at the mora level. When a stressed heavy syllable immediately follows another stressed syllable, as in (25a), the result is a clash. When the order is reversed, as in (25b), there is no clash. (25)

a. Clash

b. No clash

x

x

x

x

L

H

H

L

Assuming that clash is never tolerated at the point the basic alternation is resumed, the internal prominence contrast accounts for the different modes of resumption after a heavy syllable.4 In left-to-right trochaic systems, a trochaic foot can immediately follow the heavy syllable without creating clash, so the pattern resumes with a stressed syllable. (26)

Left-to-right trochees: No clash (x

L

) (x

L

H

) (x

)

L

L

In right-to-left iambic systems, however, an iambic foot cannot immediately precede the heavy syllable without creating clash. This being the case, the parsing algorithm must skip a syllable before constructing an iambic foot, and the pattern resumes with an unstressed syllable.

4

More precisely, Kager assumes that the construction of a foot cannot introduce clash within the parsing window. The parsing window consists of the syllables being parsed in the current iteration plus the string of syllables encountered by the parsing algorithm in previous iterations. It does not include syllables that the algorithm has not yet encountered.

Brett Hyde

13 (27)

Right-to-left iambs a. Clash * (

L

x) (x

L

L

) (

x)

L

H

b. Clash avoided (x ( x)

L

L

L

L

H

) (x

)

L

L

An asymmetric inventory of parsing feet is not actually necessary, then, to account for the different ways in which alternating patterns resume after a heavy syllable. Kager shifts the asymmetry to the prominence contrast within heavy syllables, and the difference in pattern resumption falls out from the principle of clash avoidance. Although the inventory of surface feet is still asymmetric in Kager’s account, the assymetry falls out from the principle of foot-internal lapse avoidance. The unparsed syllable in L(H) sequences can adjoin to the following foot to form an (LH) iamb, as in (28a), because it does not create a foot-internal lapse. In contrast, the unparsed syllable in (H)L sequences cannot adjoin to the preceding foot to form an (HL) trochee, as in (28b), because it would create a foot-internal lapse. (28)

Adjunction of stray syllables (x

a.

)

(

x

H

)

→ L

H

L

b. (x

)

* (x

)

H

L

→ H

L

Van de Vijver’s (1998) “Incidental Iamb” approach is similar to Kager’s Symmetrical Foot Inventory approach, in that it rejects a difference between iambic and trochaic quantity-sensitivity. Rather than accounting for the different behavior of iambic and trochaic systems in terms of rhythmic principles, however, van de Vijver claims that examples of the crucial iambic case, right-to-left iambs, simply do not exist. Following an earlier idea from Kager (1989), van de Vijver re-analyzes rightto-left iambic languages like Tübatulabal as right-to-left trochaic languages. He argues that diachronic processes have resulted in a lexical stress on word-final syllables in Tübatulabal and that trochaic feet are constructed from right to left away from the lexical stress, as in (29). (29)

qqqq’q qqqqq’q

→ →

(’qq)(’qq)(’q) q(’qq)(’qq)(’q)

While such an analysis does have the virtue of producing the correct stress pattern, one undesirable feature is the necessity of representing an entirely predictable aspect of the pattern – the word-final stress – in the lexicon (chapter 1: underlying representations).

The Iambic–Trochaic Law

14

To actually rule out the possibility of right-to-left iambic systems, and the necessity of accounting for the type of quantity-sensitivity that such systems would have to exhibit, the Incidental Iamb approach posits a constraint specifically promoting trochaic feet but not a constraint specifically promoting iambic feet. Iambic systems can only arise from the combined demands of two constraints. The first, Align-L(PrWd, Ft), requires that every word begin with a foot. The second, *Edgemost, requires that peripheral syllables be stressless. (30)

a.

b.

Align-L(PrWd, Ft) The left edge of the prosodic word should be aligned with the left edge of a foot. *Edgemost Edge-adjacent elements may not be prominent.

Combined, the constraints essentially make two demands. They demand that final syllables be stressless, due to *Edgemost, and they demand that every word begin with a disyllabic iamb, due to both *Edgemost and Align-L(PrWd, Ft). (Each word must begin with a foot, but it must be a foot that leaves the initial syllable stressless.) While trochaic systems often meet the demand that final syllable be stressless, they do meet the demand that words begin with an iambic foot, creating an opening for an iambic system to emerge. The type of iambic system that emerges, of course, must meet the combined demands of *Edgemost and Align-L(PrWd, Ft). While left-to-right iambic patterns like (31a.i) do begin with an iambic foot and leave final syllables stressless, leftto-right patterns like (31a.ii) ignore the latter requirement, and right-to-left patterns like those in (31b) ignore both. (31)

a.

b.

Left-to-right iambic systems i. Predicted (attested) (q’q)(q’q)qq (q’q)(q’q)(q’q)q Right-to-left iambic systems i. Not predicted (attested, but re-analyzed as trochaic) (q’q)(q’q)(q’q) (’q)(q’q)(q’q)(q’q)

ii.

Not predicted (attested) (q’q)(q’q)(q’q) (q’q)(q’q)(q’q)q

ii.

Not predicted (unattested) (q’q)(q’q)(q’q) q(q’q)(q’q)(q’q)

The results are mixed. The Incidental Iamb approach predicts the (31a.i) pattern, a pattern found in Carib (Hoff 1968), Hixkaryana, and Choctaw (Nicklas 1972, 1975; Lombardi and McCarthy 1991). It does not predict the (31a.ii) pattern, however, a pattern found in Araucanian. It also does not predict (31b.i), an attested pattern re-analyzed as trochaic, as discussed above, and it does not predict (31b.ii), an unattested pattern.

4

Rhythmic lengthening and rhythmic shortening

Rhythmic lengthening and rhythmic shortening are two processes where a syllable’s duration is altered because it occupies a particular position in an alternating

Brett Hyde

15

pattern. Rhythmic lengthening appears to be based solely on the alternation of strong and weak positions, affecting only the former. In contrast, rhythmic shortening can affect both strong and weak positions and seems in many cases to be motivated, at least partially, by a preference for exhaustive parsing. Rhythmic lengthening increases the duration of stressed syllables either through vowel lengthening (chapter 20: the representation of vowel length) or gemination of an adjacent consonant (chapter 37: geminates), the former method being more common than the latter. It can be found in both iambic and trochaic languages. In (5), for example, we saw that the stressed vowels of underlyingly light syllables lengthen in the iambic Hixkaryana, making them heavy on the surface. Other iambic lengthening languages include Carib, Choctaw, and several varieties of Yupik (Woodbury 1981, 1987; Jacobson 1984, 1985; Krauss 1985a; Leer 1985; among others). In (9), we saw that the stressed vowels of underlyingly light syllables lengthen in the trochaic Chimalapa Zoque. Other trochaic lengthening languages include Chamorro (Topping and Dungca 1973; Chung 1983), Icelandic (Árnason 1980, 1985), Mohawk (Michelson 1988), and Selayarese (Mithun and Basri 1986). An interesting difference between iambic and trochaic lengthening is that lengthening occurs in iambic systems only when they are quantity-sensitive, and in trochaic systems only when they are quantity-insensitive.5 When it is seen as shaping the possibilities of stress patterns generally, then, the existence of trochaic lengthening in quantity-insensitive systems clearly undermines the ITL. If we restrict the ITL’s scope to quantity-sensitive systems, however, the distribution of lengthening gives it considerable support. The presence of lengthening in iambic languages, where durational contrasts are encouraged, is consistent with the ITL, as is the absence of lengthening in trochaic languages, where durational contrasts are prohibited. Another important generalization concerning rhythmic lengthening is the correlation between what I will refer to as regular lengthening and certain types of minimal words. Regular lengthening is the exceptionless lengthening in nonminimal forms characteristic of many lengthening languages: vowels lengthen in underlyingly light syllables whenever they receive the appropriate degree of stress.6 As (32) indicates, languages with regular lengthening allow only three types of minimal word: H, LL, and HL. (32)

Minimal words associated with regular lengthening a.

Monosyllabic L unattested H Chimalapa Zoque (trochaic) Choctaw nouns (iambic) Icelandic (trochaic) Yupik varieties (iambic)

b.

Disyllabic LL Choctaw verbs (iambic) HL Carib (iambic) Hixkaryana (iambic) Selayarese (trochaic) LH

5

unattested

The clearest cases of quantity-insensitive iambs – Osage, Paumari, and Suruwaha – do not exhibit lengthening. The less clear cases – Weri and Araucanian – also do not exhibit lengthening. 6 Lengthening is not “regular” when it is prohibited in various positions in non-minimal forms, especially in final position. Syllables with primary stress in Italian, for example, lengthen if they are penultimate but not if they are antepenultimate or final. Languages like Unami and Munsee Delaware (Goddard 1979), which make stressed syllables heavy through consonant gemination, also fall outside the generalization.

The Iambic–Trochaic Law

16

Iambic lengthening languages and trochaic lengthening languages can both insist on H or HL minimal words, but only iambic lengthening languages can insist on an LL minimal word. There appear to be no regular lengthening languages with L minimal words, and none with LH minimal words. If we exclude alternations better described as vowel reduction (see below), rhythmic shortening is a marginal phenomenon. It occurs in only a few trochaic systems, and, as Mellander (2003) points out, each of these few is quantity-sensitive. Trochaic shortening can affect either a stressed syllable or an unstressed syllable. In Fijian, for example, stressed syllables shorten, converting HL sequences into LL sequences. (33)

Trochaic shortening in Fijian ,bu(-nIu ha(-j-a nre(-ta

→ → →

’mbu-nIu ’ta-j-a ’nre-ta

‘my grandmother’ ‘chop-trans-3 sg obj’ ‘pull-trans’

In Pre-Classical Latin (Allen 1973; Mester 1994), stressless syllables shorten, converting LH sequences into LL sequences.7 (34)

Trochaic shortening in Latin ego( → male( → ami(kitiam →

’ego ’male ‘ami’kitiam

‘I’ ‘bad’ ‘friendship’

Rhythmic shortening, though marginally attested, is consistent with the ITL. Among quantity-sensitive languages, it occurs only in trochaic systems, destroying the durational contrasts that the ITL prohibits. It does not occur in iambic systems, where it would destroy durational contrasts that the ITL requires. Quantity-insensitive languages, of either type, apparently do not exhibit rhythmic shortening. Before we proceed, it should be noted at this point that there are at least two languages with shortening phenomena that are potential counterexamples to the generalizations presented in the preceding paragraph: Central Slovak (DvonL 1955; Bethin 1998; Mellander 2003) and Gidabal (Geytenbeek and Geytenbeek 1971; Rice 1992; Mellander 2003). The stress patterns of both languages are fairly complex, however, and their analyses are not at all straightforward. It is not clear whether they are examples of shortening in trochaic feet (resulting in unbalanced HL trochees), shortening in iambic feet, or both, or whether they are simply examples of shortening in non-head syllables generally. Since it is unclear exactly how such examples are relevant, I have set them aside here. I have also set aside phenomena involving vowel strengthening in stressed syllables and vowel reduction and deletion in unstressed syllables. Some of the alternatives to an ITL approach draw to a significant extent on such phenomena as evidence that the difference between iambs and trochees with respect to rhythmic lengthening and shortening is not as great as previously thought (Revithiadou and van de Vijver 1997; van de Vijver 1998; Revithiadou 2004). Strengthening, reduction, and deletion phenomena are fairly common in both iambic and trochaic systems. 7

The Latin-type shortening is often referred to as iambic shortening, because it affects the second syllable in a two-syllable sequence rather than the first.

Brett Hyde

17

While Hayes (1985) introduced strengthening, reduction, and deletion into the discussion to support the ITL, claiming that the phenomena arose primarily in iambic systems and helped to create durational contrasts, he later observed that there were significant difficulties with this line of evidence (Hayes 1995). First, while they have some impact on phonetic duration, they are not primarily duration phenomena – as opposed to sonority or parsing phenomena. They do not involve phonological duration in the same way as the canonical examples of lengthening and shortening. The second reason, Hayes admits, is that they are not specifically iambic phenomena, which essentially grants the position of Revithiadou and van de Vijver. Although the introduction of strengthening, reduction, and deletion into the debate was probably a misstep in terms of defending the ITL, the resulting discussion has helped to identify its most plausible areas of influence. Just as the case for the ITL is much stronger in quantity-sensitive systems than in quantity-insensitive systems, the case is much stronger in the contexts of genuine lengthening and shortening than in the contexts of strengthening, reduction, and deletion. This being the case, I will not address the latter contexts further.

4.1

Lengthening and shortening rules

Just as Hayes’s (1987, 1995) asymmetric foot inventory effectively captures the different types of quantity-sensitivity that arise in iambic and trochaic systems, his approach to rhythmic lengthening and shortening effectively captures their distribution in quantity-sensitive systems. In Hayes’s account, rhythmic lengthening is only possible when it would create a durational contrast in iambs, a restriction motivated by the strong interpretation of the ITL. (35)

Iambic lengthening ( q Ø→ [/[

x) q [ __

As (35) indicates, iambic lengthening only occurs in iambic feet where both the first and second syllable are light. The second syllable becomes heavy, creating a durational contrast. Rhythmic shortening is only possible when it would avoid a durational contrast in trochees. The effect of the Fijian-type shortening rule in (36a) is to convert an (H)L sequence to an (LL) sequence. It helps to minimize underparsing without resorting to an (HL) foot. Since the appropriate context can also arise in limited circumstances in iambic systems, Hayes stipulates that the rule can only apply in trochaic systems. (36)

Trochaic shortening a.

Fijian-type →

where

is metrically stray

The Iambic–Trochaic Law b.

Latin-type (x )

(x

18

)

→

The effect of the Latin-type shortening rule in (36b) is to convert an ill-formed (LH) trochee to a well-formed (LL) trochee. An ill-formed (LH) trochee might be created inadvertently, for example, when an extrametrical H syllable is adjoined to a degenerate (L) foot, and (36b) repairs the defect. Though the ITL account captures the distribution of rhythmic lengthening and shortening in quantity-sensitive systems, it falls short in two ways. First, it fails to allow for trochaic lengthening in quantity-insensitive systems (a phenomenon whose existence Hayes denies). Second, it does not account for the correlation between regular lengthening and the group of minimal words in (32). Based on the ITL, lengthening rules might be employed to create durational contrasts in iambic feet (or, possibly, to destroy them in trochaic feet), but there is nothing in the law entailing that lengthening languages should prefer H, HL, and LL minimal words above L and LH minimal words. If the ITL is actually the motivation for lengthening, then the correlation of regular lengthening with these particular minimal words is a mystery.

4.2

Lapse avoidance and non-finality

Kager’s (1993) approach to the asymmetries in rhythmic lengthening and shortening is based on the same principles that governed his approach to quantitysensitivity (see §3.2). In conjunction with a prohibition against foot-internal lapse, the internal prominence contrast in heavy syllables restricts the occurrence of lengthening. Kager views lengthening of stressed syllables in general as phonetically motivated, but the restriction against foot-internal lapse ensures that such lengthening is more common in iambs than in trochees. As (37) illustrates, the grammar tolerates lengthening that creates (LH) iambs, because they contain no foot-internal lapse, but it does not tolerate lengthening that creates (HL) trochees, because they do contain a foot-internal lapse. (37)

Lengthening asymmetry through lapse avoidance a.

No lapse after iambic lengthening ( x) ( x ) → L

b.

L

L

H

Lapse after trochaic lengthening (x ) * (x ) → L

L

H

L

Brett Hyde

19

The prohibition against foot-internal lapse also accounts for the shortening asymmetry, but in this case it acts as a trigger. In Kager’s view, the purpose of trochaic shortening is to eliminate foot-internal lapses like those found in (HL) trochees. Since there is no foot-internal lapse in (LH) iambs, the motivation for shortening never arises in iambic systems. (38)

Lapse avoidance predicts shortening asymmetry a.

No lapse in iambs to trigger shortening ( x ) *( x)

L b.

H

L

L

Lapse in trochees triggers shortening (x ) (x )

H

L

L

L

Although foot-internal lapse avoidance effectively addresses the lesser frequency of lengthening in trochaic systems, it does not address the actual phonological triggers for lengthening. Hyde’s (2007) non-finality approach addresses the lesser frequency of lengthening in trochaic systems, but it also provides phonological triggers for rhythmic lengthening and addresses its correlation with certain types of minimal words (chapter 43: extrametricality and non-finality). Under the non-finality approach, rhythmic lengthening is a special case of the type of weight-sensitivity where stress avoids light syllables. To avoid stressing a light syllable, the syllable is lengthened to make it heavy. Non-finality produces this type of weight-sensitivity by prohibiting stress on domain-final moras. Following Kager (1995), Hyde applies non-finality to the foot domain to promote iambic lengthening. Going a step further, he also applies non-finality to the syllable domain. This gives the approach a second mechanism for promoting iambic lengthening but it also gives it a mechanism for promoting trochaic lengthening. (39)

a. b.

Non-finality(Ft) No stress occurs over the final mora of a foot. Non-finality(q) No stress occurs over the final mora of a syllable.

Non-finality(Ft) effectively prohibits stress on light foot-final syllables. Since it bans foot-level gridmarks from foot-final moras, foot-final syllables must be at least bimoraic to support stress. Non-finality(q) effectively prohibits stress on light syllables generally. Since it bans foot-level gridmarks from syllable-final moras, syllables generally must be at least bimoraic to support stress. To produce lengthening, one of the non-finality constraints must dominate Dep-[, the faithfulness constraint that prevents mora insertion. Under such rankings,

The Iambic–Trochaic Law

20

when stress would otherwise occupy a light syllable, a mora can be added to make the syllable heavy on the surface. The two non-finality constraints do not, however, have equal ability to promote lengthening in every type of foot. Since Non-finality(Ft) prohibits stress over light foot-final syllables in particular, it can lengthen the stressed syllable of an iamb but not the stressed syllable of a trochee. In contrast, since Non-finality(q) prohibits stress over light syllables in general, it can lengthen the stressed syllables of both. Consider first the situation where the stressed syllable occurs in an iamb. When Non-finality(Ft) dominates Dep-[, a second mora is added to underlyingly light syllables to avoid stress on foot-final moras. Non-fin(Ft) Dep-[

LLLL

(40)

x x x x x x x x [ [ [ [ [ [

☞ a.

** … q q

q q

x x x [ [

x x x [ [

b.

…

*!* … q q

q q

…

The result is similar when Non-finality(q) dominates Dep-[: a second mora is added to the underlyingly light syllables to avoid stress on syllable-final moras. Non-fin( ) Dep-

LLLL

(41)

x x x x x x x x

☞ a.

** …

…

x x x

b.

x x x *!*

…

…

21

Brett Hyde

Now consider the situation where the stressed syllable occurs in a trochee. When Non-finality(Ft) dominates Dep-[, as in (42), there is no lengthening. Because stress does not occupy the foot-final syllables in either candidate, there is no danger that it will occupy the foot-final moras, and Non-finality(Ft) cannot distinguish between them. The lower ranked Dep-[ settles on the faithful (42b) candidate. (42) a.

Non-fin(Ft) Dep-[

LLLL x x x x x x x x [ [ [ [ [ [

*!* … q

q q

q…

x x [

x x x [ [

x [

… q

q q

q…

☞ b.

When Non-finality(q) dominates Dep-[, however, as in (43), the lengthening candidate emerges as the winner. The stressed syllables become heavy, to allow stress to avoid syllable-final moras. (43) ☞ a.

Non-fin( ) Dep-

LLLL x x x x x x x x

** …

…

x x

b.

x x x

x *!*

…

…

One advantage of the non-finality approach is that it has a built-in explanation for the lesser frequency of lengthening among trochaic systems. Non-finality in the syllable and non-finality in the foot both produce iambic lengthening, but only

The Iambic–Trochaic Law

22

non-finality in the syllable produces trochaic lengthening. Every ranking that produces trochaic lengthening, then, also produces iambic lengthening, but some rankings that produce iambic lengthening do not produce trochaic lengthening. Since the percentage of possible rankings that produce iambic lengthening is greater than the percentage of possible rankings that produce trochaic lengthening, we would expect lengthening to occur with greater frequency in iambic systems than it does in trochaic systems, all else being equal. A second advantage of the non-finality approach is that it helps to account for the particular group of minimal words associated with regular lengthening. As discussed above, languages that automatically lengthen appropriately stressed vowels only allow three types of minimal word: H, LL, and HL. They never allow L or LH minimal words. Using the same non-finality constraints to produce rhythmic lengthening and the minimal word restrictions predicts this situation. L minimal words are absent, because the lengthening constraints themselves both establish H minimal words. Non-finality(q) has the same effect in monosyllabic feet that it has in disyllabic feet, and Non-finality(Ft) has the same effect that it has in iambs. They both force the stressed syllable to lengthen. As (44) indicates, if either of the lengthening constraints ranks highly enough to produce lengthening in the disyllabic feet of longer forms, then it also ranks highly enough to produce lengthening in the monosyllabic feet of monosyllabic forms. (44)

a. b.

Non-finality(q) >> Dep-[ Iambic or trochaic lengthening + H minimal word Non-finality(Ft) >> Dep-[ Iambic lengthening + H minimal word

Two desirable predictions result from this situation: regular lengthening is always accompanied by a minimal word that is at least bimoraic, and iambic lengthening languages and trochaic lengthening languages can both have H minimal words. Although the lengthening constraints cannot produce disyllabic minimal words on their own, they do help to determine which type of disyllable emerges. Assuming that disyllabic minimal words have a trochaic strong–weak stress contour, we can explain the two-syllable requirement with an additional nonfinality constraint, Non-finality(w), which bans stress from the final syllable of a prosodic word.8 Once the strong–weak contour is established, lengthening constraints determine the weight of the initial syllable. Non-finality(q), which produces lengthening in both iambic feet and trochaic feet, requires that the initial syllable be heavy. Non-finality(Ft), which produces lengthening only in iambic feet, tolerates a light initial syllable. (45)

a. b.

8

Non-finality(w), Non-finality(q) >> Dep-[ Iambic or trochaic lengthening + HL minimal word Non-finality(w), Non-finality(Ft) >> Dep-[ Iambic lengthening + LL minimal word

Plausible cases of iambic minimal words appear to be extremely rare.

Brett Hyde

23

This correctly predicts that either iambic or trochaic lengthening can be accompanied by an HL minimal word, but only iambic lengthening can be accompanied by an LL minimal word. Van de Vijver (1998) and Revithiadou (2004) propose an approach to rhythmic lengthening that is similar in some respects to the non-finality approach. Although it does not rely on the non-finality formulation, it does posit two lengthening mechanisms. One lengthens stressed syllables generally, which produces both iambic lengthening and trochaic lengthening, and the other lengthens foot-final syllables in particular, which produces only iambic lengthening. (46)

Lengthening constraints (van de Vijver 1998) a. b.

Stressed Syllable Length A stressed syllable is long and an unstressed syllable is short. FootFinal Foot-final elements are lengthened.

Since there are two sources for iambic lengthening and only one for trochaic lengthening, Revithiadou’s and van de Vijver’s proposals, like the non-finality approach, provide an account of the different frequencies with which the two types of lengthening occur. The advantage of the non-finality approach is that it incorporates the lengthening mechanisms into the much more general non-finality formulation, a formulation independently motivated by its ability to account for a surprisingly broad range of phenomena at different prosodic levels. (See chapter 43: extrametricality and non-finality.)

5

Summary

The most interesting interpretation of the Iambic–Trochaic Law is a strong interpretation that focuses on the presence or absence of durational contrasts in disyllabic feet. Since the general typology of attested stress systems offers very little support for the strong interpretation, Hayes (1985) introduced the ITL to metrical theory under a weaker interpretation that focused on quantity-sensitivity. This also turned out to be inadequate, however, as it was soon recognized that both iambic languages and trochaic languages could be either quantity-sensitive or quantity-insensitive. Two subsequent accounts – McCarthy and Prince (1986) and Hayes (1987, 1995) – pursued a hybrid approach, combining aspects of the weak interpretation and the strong interpretation. Another, Prince (1990), returned to a strong interpretation of the ITL, but applied it, in effect, only to quantity-sensitive systems and as a preference rather than an absolute requirement. Since the ITL is inherently quantity-sensitive, it seems more natural to employ it as a foundation for an account of quantity-sensitive systems in particular than as the foundation for an account of stress systems generally. There is, in fact, considerable support for the ITL among quantity-sensitive systems. Iambic quantity-sensitivity differs from trochaic quantity-sensitivity, as indicated by the different ways in which alternating patterns resume after encountering a heavy syllable, and the asymmetric foot inventory of ITL accounts very effectively for this difference. Standard iambs exclude heavy syllables from weak position

The Iambic–Trochaic Law

24

in disyllabic feet; moraic trochees exclude them from disyllabic feet entirely. The result, that (disyllabic) iambic feet can contain durational contrasts but (disyllabic) trochaic feet cannot, is consistent with a strong interpretation of the ITL. Lengthening and shortening asymmetries also support a strong interpretation. Among quantity-sensitive systems, lengthening only occurs in iambic systems, and shortening only occurs in trochaic systems. When we restrict our attention to quantity-sensitive systems, then, the ITL does capture important differences between iambs and trochees with respect to the particular type of quantity-sensitivity exhibited and the employment of lengthening and shortening rules. This does not necessarily mean, of course, that the ITL is the best explanation for these differences. Rather than being an explanation, the descriptive and superficial ITL actually seems to be an observation in need of an explanation, much like the attested stress patterns themselves. Particularly important to alternative accounts is an assumed prominence asymmetry that arises within heavy syllables (Kager 1993). When a heavy syllable is stressed, the stress occupies its first mora and its second mora is stressless. This allows Kager (1993) to account for differences in quantity-sensitivity in terms of the rhythmic principle of clash avoidance and to account for lengthening and shortening asymmetries in terms of the rhythmic principle of lapse avoidance. Hyde (2007) exploits the same syllable-internal prominence asymmetry to provide a non-finality-based account of lengthening asymmetries and minimal words. An important advantage of these alternatives is that they offer the potential to account not only for many of the asymmetries found in the typologies of attested stress patterns, but also for the Iambic–Trochaic Law itself.

REFERENCES Allen, W. Sidney. 1973. Accent and rhythm. Cambridge: Cambridge. University Press. Altshuler, Daniel. 2009. Quantity-insensitive iambs in Osage. International Journal of American Linguistics 75. 365–398. Árnason, Kristján. 1980. Quantity in historical phonology: Icelandic and related cases. Cambridge: Cambridge University Press. Árnason, Kristján. 1985. Icelandic word stress and metrical phonology. Studia Linguistica 39. 93–129. Bell, Alan. 1977. Accent placement and perception of prominence in rhythmic structures. In Larry Hyman (ed.) Studies in stress and accent, 1–13. Los Angeles: Department of Linguistics, University of Southern California. Bethin, Christina Y. 1998. Slavic prosody: Language change and phonological theory. Cambridge: Cambridge University Press. Bolton, Thaddeus L. 1894. Rhythm. American Journal of Psychology 6. 145–238. Boxwell, Helen & Maurice Boxwell. 1966. Weri phonemes. In S. A. Wurm (ed.) Papers in New Guinea Linguistics No. 5, 77–93. (Pacific Linguistics A37.) Canberra: Australian National University. Brame, Michael K. 1973. On stress assignment in two Arabic dialects. In Stephen R. Anderson & Paul Kiparsky (eds.) A Festschrift for Morris Halle, 14–25. New York: Holt, Rinehart & Winston. Brame, Michael K. 1974. The cycle in phonology: Stress in Palestinian, Maltese and Spanish. Linguistic Inquiry 5. 39–60. Chung, Sandra. 1983. Transderivational relationships in Chamorro phonology. Language 59. 35–66.

25

Brett Hyde

Derbyshire, Desmond C. 1985. Hixkaryana and linguistic typology. Dallas: Summer Institute of Linguistics & University of Texas at Arlington. Dixon, R. M. W. 1988. A grammar of Boumaa Fijian. Chicago: University of Chicago Press. DvonL, Ladislav. 1955. Rytmicky zákon v spisovnej slovenLine. Bratislava: Vydavatel’stvo Slovenskej Akadémie Vied. Echeverría, Max S. & Heles Contreras. 1965. Araucanian phonemics. International Journal of American Linguistics 31. 132–135. Everett, Daniel L. 1996. Prosodic levels and constraints in Banawá and Suruwaha. Unpublished ms., University of Pittsburgh (ROA-121). Everett, Daniel L. 2003. Iambic feet in Paumari and the theory of foot structure. Linguistic Discovery 2(1). 22–44. Geytenbeek, Brian B. & Helen Geytenbeek. 1971. Gidabal grammar and dictionary. Canberra: Australian Institute of Aboriginal Studies. Goddard, Ives. 1979. Delaware verbal morphology: A descriptive and comparative study. New York: Garland. Hay, Jessica S. F. & Randy L. Diehl. 2007. Perception of rhythmic grouping: Testing the iambic/trochaic law. Perception and Psychophysics 69. 113–122. Hayes, Bruce. 1985. Iambic and trochaic rhythm in stress rules. Proceedings of the Annual Meeting, Berkeley Linguistics Society 11. 429–446. Hayes, Bruce. 1987. A revised parametric metrical theory. Papers from the Annual Meeting of the North East Linguistic Society 17. 274–289. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hoff, Berend J. 1968. The Carib language: Phonology, morphonology, morphology, texts and word index. The Hague: Martinus Nijhoff. Hulst, Harry van der. 1999. Issues in foot typology. In S. J. Hannahs & Mike Davenport (eds.) Issues in phonological structure: Papers from an international workshop, 95–127. Amsterdam & Philadelphia: John Benjamins. Hyde, Brett. 2007. Non-finality and weight-sensitivity. Phonology 24. 287–334. Iversen, John R., Aniruddh D. Patel & Kengo Ohgushi. 2008. Perception of rhythmic grouping depends on auditory experience. Journal of the Acoustical Society of America 124. 2263–2271. Jacobson, Steven A. 1984. The stress conspiracy and stress-repelling bases in the Central Yup’ik and Siberian Yupik Eskimo languages. International Journal of American Linguistics 50. 312–324. Jacobson, Steven A. 1985. Siberian Yupik and Central Yupik prosody. In Krauss (1985b), 25–45. Kager, René. 1989. A metrical theory of stress and destressing in English and Dutch. Dordrecht: Foris. Kager, René. 1993. Alternatives to the iambic–trochaic law. Natural Language and Linguistic Theory 11. 381–432. Kager, René. 1995. Review article of Hayes (1995). Phonology 12. 437–464. Kenstowicz, Michael. 1983. Parametric variation and accent in the Arabic dialects. Papers from the Annual Regional Meeting, Chicago Linguistic Society 19. 205–213. Kenstowicz, Michael & Kamal Abdul-Karim. 1980. Cyclic stress in Levantine Arabic. Studies in the Linguistic Sciences 10(2). 55–76. Knudson, Lyle M. 1975. A natural phonology and morphophonemics of Chimalapa Zoque. Papers in Linguistics 8. 283–346. Krauss, Michael. 1985a. Supplementary notes on Central Siberian Yupik prosody. In Krauss (1985b), 47–50. Krauss, Michael (ed.) 1985b. Yupik Eskimo prosodic systems: Descriptive and comparative studies. Fairbanks: Alaska Native Language Center. Kusumoto, Kiyomi & Elliott Moreton. 1997. Native language determines parsing of nonlinguistic rhythmic stimuli. Journal of the Acoustical Society of America 102. 3204.

The Iambic–Trochaic Law

26

Leer, Jeff. 1985. Toward a metrical interpretation of Yupik prosody. In Krauss (1985b), 159–172. Lombardi, Linda & John J. McCarthy. 1991. Prosodic circumscription in Choctaw morphology. Phonology 8. 37–71. McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. Mellander, Evan W. 2003. (HL)-creating processes in a theory of foot structure. The Linguistic Review 20. 243–280. Mester, Armin. 1994. The quantitative trochee in Latin. Natural Language and Linguistic Theory 12. 1–61. Michelson, Karin. 1988. A comparative study of Lake-Iroquoian accent. Dordrecht: Kluwer. Mithun, Marianne & Hasan Basri. 1986. The phonology of Selayarese. Oceanic Linguistics 25. 210–254. Nicklas, Thurston Dale. 1972. The elements of Choctaw. Ph.D. dissertation, University of Michigan, Ann Arbor. Nicklas, Thurston Dale. 1975. Choctaw morphophonemics. In James M. Crawford (ed.) Studies in Southeastern Indian languages, 237–250. Athens: University of Georgia Press. Prince, Alan. 1990. Quantitative consequences of rhythmic organization. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26(2). 355–398. Revithiadou, Anthi. 2004. The Iambic/Trochaic Law revisited: Lengthening and shortening in trochaic systems. Leiden Papers in Linguistics 1. 37–62. Revithiadou, Anthi & Ruben van de Vijver. 1997. Durational contrasts and the iambic/ trochaic law. Proceedings of Western Conference on Linguistics 9. 229–242. Rice, Curt. 1992. Binarity and ternarity in metrical theory: Parametric extensions. Ph.D. dissertation, University of Texas, Austin. Schütz, Albert J. 1978. English loanwords in Fijian. In Albert J. Schütz (ed.) Fijian language studies: Borrowing and pidginization, 1–50. Suva: Fiji Museum. Schütz, Albert J. 1985. The Fijian language. Honolulu: University of Hawaii Press. Seiler, Hansjakob. 1965. Accent and morphophonemics in Cahuilla and in Uto-Aztecan. International Journal of American Linguistics 31. 50–59. Seiler, Hansjakob. 1967. Structure and reconstruction in some Uto-Aztecan languages. International Journal of American Linguistics 33. 135–147. Seiler, Hansjakob. 1977. Cahuilla grammar. Banning, CA: Malki Museum Press. Seiler, Hansjakob & Kojiro Hioki. 1979. Cahuilla dictionary. Banning, CA: Malki Museum Press. Topping, Donald M. & Bernadita C. Dungca. 1973. Chamorro reference grammar. Honolulu: University Press of Hawaii. Vijver, Ruben van de. 1998. The iambic issue: Iambs as a result of constraint interaction. Ph.D. dissertation, University of Leiden. Voegelin, Charles F. 1935. Tübatulabal grammar. University of California Publications in American Archaeology and Ethnology 34. 55–189. Vos, P. 1977. Temporal duration factors in the perception of auditory rhythmic patterns. Scientific Aesthetics 1. 183–199. Woodbury, Anthony. 1981. Study of the Chevak dialect of Central Alaskan Yupik. Ph.D. dissertation, University of California, Berkeley. Woodbury, Anthony. 1987. Meaningful phonological processes: A consideration of Central Alaskan Yupik Eskimo prosody. Language 63. 685–740. Woodrow, Herbert. 1909. A quantitative study of rhythm. Archives of Psychology (New York) 14. 1–66.

45

The Representation of Tone Larry M. Hyman

1

Introduction

No issue has had a greater impact on phonological representations than the study of tone. Although receiving only passing attention in both pre- and early generative phonology, tone quickly moved away from its marginal status to occupy center stage in the development of non-linear phonology. While both level and contour tones had been traditionally transcribed with either accents or numerals, as in (1) and (2), the assumption in early generative phonology, e.g. Wang (1967), was that tones consisted of features that could be added at the bottom of a segmental feature matrix, as in (3). (1) High (H) Low (L) HL (falling) LH (rising)

(2)

Falam

Obokuitai

ná ‘breast’ nà ‘house’ nâ ‘taro’ pa ‘fish’ (Loving 1966: 25)

páa ‘mushroom’ kèe ‘leg’ sâa ‘animal’ z"u ‘bear’ (personal notes)

kuik1 ‘rock’ kuik2 ‘insect (sp.)’ kuik12 ‘lizard (sp.)’

Jingpho H Mid (M) L HL (Qingxia

(3)

Awa

a.

(Jenison and Jenison 1991: 85)

Ayutla Mixtec mO55 ‘word’ mu33 ‘delicious’ mu31 ‘to see’ nu51 ‘mother’ and Diehl 2003: 401)

H tone /á/ G+syll J H−cons K H+back K H+low K I+highL

H–H œi1Ju?1 H–L œi1ni?3 M–L œi2ni?3 L–L ti3ku?3 (Pankratz and

‘pineapple’ ‘hat’ ‘head’ ‘louse’ Pike 1967: 291)

b. HL falling tone /â/ G+syll J H−cons K H+back K H+low K I+fallingL

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0045

The Representation of Tone

2

A major representational problem was how to account for the properties of contour tones. Although falling and rising tones could be expressed less formally by combining accents (â, a) or numerals (31, 13, etc.), features such as [±falling] and [±rising] fail to capture what are known as “edge effects”: a high to low falling tone acts like a H tone with respect to what precedes, but as a L tone with respect to what follows. Similarly, a low to high rising tone acts like a L tone with respect to what precedes, but as a H tone with respect to what follows. Representations such as in (4), which were occasionally entertained, would simply be incoherent in a framework in which a segment consists of a single vertical matrix of features: (4)

a.

Falling tone /â/ G+syll H−cons H+back H+low I[+high][+low]

b. Rising tone /a/ J K K K L

G+syll J H−cons K H+back K H+low K I[+low][+high]L

In order to solve this and other representational problems, Goldsmith (1976a, 1976b) proposed a theory of “autosegmental” phonology in which segments and tones appear on separate “tiers,” as in (5). (5)

Autosegmental representations of H, L, HL, LH a. level tones a

a

H

L

b. contour tones a H

a L

L

H

By so doing, Goldsmith was able to capture the traditional intuitions implicit in the accent and numeral notations and make predictions about what should vs. should not be found in tone systems. Armed with the autosegmental framework, enormous strides were made in the analysis of tone as well as in other applications of the framework, e.g. segmental harmonies (Clements 1977, 1981; Hoberman 1988), feature geometry (Clements 1985; Clements and Hume 1995) and prosodic morphology (McCarthy 1981; McCarthy and Prince 1986). The main question to be addressed in this chapter is the extent to which the key insight of autosegmental phonology, expressed in (6), is still valid: (6)

Tones are semi-autonomous from their tone-bearing units a. b.

tones are on a separate tier, but they are linked to their tone-bearing units by association lines.

The chapter is organized as follows: in §2 we consider some of the predictions of autosegmental tonology in order to see how they have fared since the 1970s. In §3 we consider the issue of underspecification, while §4 addresses the issues of tone features, tonal geometry, and tone-bearing units. §5 evaluates potential limitations of autosegmental representations, while §6 concludes the chapter with a brief consideration of tone in constraint-based phonology. We will see that there is much

Larry M. Hyman

3

reason to hold on to the autosegmental insight even as phonological frameworks have evolved.

2

Autosegmental tonology

One can distinguish two consequences of autosegmental phonology (see also chapter 14: autosegments) as applied to tone: those that follow directly from the architecture vs. those that involve additional principles or conventions. We take up the first of these here and postpone discussion of the second until §5. There are at least three direct consequences of the two-tier autosegmental architecture proposed by Goldsmith, as in (7): (7)

2.1

a. b. c.

non-isomorphism between the tiers zero representation on one vs. the other tier stability effects

Non-isomorphism

Using tbu to represent the tone-bearing unit to which tones link (see §4), the first of these is schematized in (8). (8)

a.

tbu H

b. tbu tbu

tbu L

L

H

H

tbu tbu L

We have already seen that more than one tone can link to a single tbu, resulting in falling and rising contour tones, as in (8a). Complex contours are also attested, as exemplified in (9). (9)

a.

falling-rising:

b.

rising-falling:

be243 ba243 Nzadi mwaìn dz≤`

Iau

‘tree fern’ ‘sticking to’ ‘child’ ‘eye’

(Bateman 1990) (personal notes)

In addition, Lomongo is said to have LHLH on one syllable derived from elision (Hulstaert 1961: 164): /èmí là w( basàngì/ → [èm â w a!sàngì] ‘it’s you and I who are related’, where marks the two places where elision occurs. The second type of non-isomorphism in (8b) shows that the same tone can link to more than one tbu. What this means is that there is a potential contrast between the representations in (10a) vs. those in (10b). (10)

a. tbu

tbu

b. tbu tbu

H

H

H

tbu

tbu

tbu tbu

L

L

L

The Representation of Tone

4

The representations in (10a) were originally thought to be prohibited by the obligatory contour principle (OCP), which disallows successive identical features on the same tier (Leben 1973; Goldsmith 1976a). However, the contrasts in (10) are clearly needed. Typically, the one-to-one representations in (10a) occur when the tones belong to different morphemes (ultimately, words), while those in (10b) are expected when the H–H or L–L sequence occurs tautomorphemically. This makes sense if one considers that morphemes are spelled out separately, such that they may be concatenated with other morphemes carrying identical tones. Examples follow. As reported by Paulian (1975), Kukuya noun stems are limited to five shapes (CV, CVV, CVCV, CVVCV, CVCVCV) and four tonal “melodies”: /H, L, HL, LH/. (A fifth melody, /LHL/, occurs only on polymorphemic verb stems.) When a /H/ noun stem consists of more than one tbu, the H is multiply linked, producing representations such as in (11). (11)

a. (mà-) bágá

‘show knives’

b. (lì-) bálágá

H

‘fence’

H

Now consider the two words [má-bá] ‘they are oil palms’ and [wátá] ‘bell’. The first consists of a /H/ copular tone assigned to the toneless prefix /ma-/ followed by a /H/ stem, while the latter lacks a prefix and instead has a single /H/ linked to both tbus of the stem. The two words are pronounced identically as H–H in medial position, as in (12a), but not in prepausal position in (12b), where there is a lowering to mid tone: (12)

a. Medial

b. Prepausal

má-bá

wátá

má-ba

wata

H H

H

H M

M

If we assume that the prepausal lowering rule refers only to the tonal tier, targeting the last H “autosegment,” as in (13), the right results are obtained (// = pause boundary): (13)

H → M / __ //

One H is lowered in [má-ba], while both Hs are lowered in [wata] (Hyman 1987). While the two representations in (12a) are expected to correlate with heteromorphemic vs. tautomorphemic H tones, Odden (1982) shows that Shambala requires both representations within noun stems. A noun such as [njóká] ‘snake’, pronounced H–H, has a single /H/ linked to both syllables of the stem. The question, then, is how to analyze stems such as [ngó↓tó] ‘sheep’, pronounced H–↓H (H followed by downstepped H tone; tonal downstep is indicated by ↓). Since successive heteromorphemic /H/ tones produce downsteps, i.e. /H/ + /H/ → H–↓H, the logical move is to represent [ngó↓tó] with two successive /H/s on the monomorphemic stem. (Odden analyzes Shambala with a privative /H/ tone, which

Larry M. Hyman

5

contrasts with /Ø/ rather than /L/.) Although a rare occurrence, the tautomorphemic version of the OCP is thereby violated. Similar contrasts and OCP problems are found with respect to /L/ tones in Dioula d’Odienné (Braconnier 1982). In this language there is a rule of the form in (14), where pause (//) may be interpreted with a %L boundary tone: (14)

L → H / {//, L} __ H

As seen in (15), the rule is not iterative, as only one L autosegment can be raised: (15)

a.

/ì mà dÜ tá/

→

b.

/ì mà tùrù tá/

→

ì *ì ì *ì

mà má mà má

dÅ tá dÅ tá túrú tá túrú tá

‘you didn’t take any child’ ‘you didn’t take any oil’

However, the following examples show that not all monomorphemic L tone nouns have the same properties before H: (16)

a.

before pause sèbè tùrù kàràkà sùmàrà

b. before H sèbé túrú kàràká sùmárá

‘paper’ ‘oil’ ‘bed’ ‘soumbala (a spice)’

One solution is to give the above nouns the representations in (17a), where the OCP is violated similarly to Odden’s (1982) tautomorphemic Hs in Shambala. (17)

a. sebe LL b. sebe L

turu L turu

karaka L L karaka

L

L

sumara L

L

sumara L

Another solution in (17b) might be to posit a distinction between /L/ and toneless tbus, which later become L by default. While the OCP is violable in the ways just seen, it is important to note that the autosegmental representations provide a straightforward way of encoding the contrastive tonal properties of Kukuya, Kishambaa, and Dioula d’Odienné. Other frameworks would have to resort to ad hoc junctures or diacritics.

2.2

Zero representation

The second consequence that directly flows from the autosegmental architecture is the possibility that a tone can exist without a tbu, and vice versa. When a morpheme consists solely of a tone, it is referred to as a tonal morpheme (see also chapter 82: featural affixes). An oft-cited example is the associative (genitive)

The Representation of Tone

6

marker in Igbo (see Williamson 1986 and references cited therein), which is most transparently observed when preceded and followed by a L–L noun. As seen in (18), the H tonal morpheme is assigned to the preceding tbu in Central Igbo and to the following tbu in the Aboh dialect: (18)

a. b.

Central Igbo àgbà + ´ + èIwè → àgbá èIwè ‘jaw of monkey’ Aboh Igbo Qgbà + ´ + èIwè → Qgbà éIwè ‘jaw of monkey’

Floating tones can also be part of a lexical morpheme. In Aghem, the nouns [kRfú] ‘rat’ and [kRwó] ‘hand’ are both pronounced with H–H tones in isolation (Hyman 1979). However, as seen in (19), they have different effects on the following word (the [kR-] prefix drops out when the noun is modified): (19)

a.

fú

kîa

H

L

wó

kSa

‘your (sg) rat’

‘your (sg) hand’

HL L

b.

fú

kRn

H

H

wó

↓

kRn

‘this rat’

‘this hand’

HL H

In (19a), the H of the stem /-fú/ spreads onto the /L/ of /kSa/ ‘your sg (class 7)’ creating a HL contour. As seen, the floating L of /-wó`/ ‘hand’ blocks the spreading. When followed by the /H/ tone demonstrative /kRn/ ‘this (class 7)’ in (19b), ‘this rat’ is realized H–H, while ‘this hand’ has a downstep conditioned by the same lexical floating L of /-wó`/, which was originally due to a historically lost syllable (cf. Proto-Bantu *-bókò ‘hand’). Corresponding to floating tones, which lack a tbu, are tbus that lack tones. Such toneless morphemes may receive their tonal specification by context or by default tone assignment (see §3). An oft-cited example of the former comes from Mende (Leben 1973, 1978): base noun

(20) a. b. c. d. e.

/H/ /L/ /HL/ /LH/ /LHL/

kF bèlè mbû mba njàhâ

‘war’ ‘trousers’ ‘owl’ ‘rice’ ‘woman’

+hu ‘in’

+ma ‘on’

kF-hú bèlè-hù mbú-hù mbà-hú njàhá-hù

kF-má bèlè-mà mbú-mà mbà-má nyàhá-mà

As seen, the tone of the two locative postpositions is the same as the last underlying tone of the nouns to which they attach. The /HL/, /LH/, and /LHL/ “melodies” are linked one-to-one to the noun + postposition constituent.

2.3

Stability effects

The third consequence of autosegmental representations, stability, is related to the second: when a tbu is deleted, its tone may still remain, and vice versa. An example of this comes from Tangale, which Kenstowicz and Kidda (1987: 230)

7

Larry M. Hyman

interpret as having an underlying /H/ vs. /Ø/ contrast. As seen in the following examples, the final vowel of a word is deleted “when it is followed by another word in a close syntactic configuration”: (21)

a. /tuuÚe/ + /lawo/ → tuuÚ H

H

b. /jaara/ + /lawo/ H

lawo → tùuÚ

→ jaar

láwò

‘child’s horse’

làwò

‘child’s arm’

H lawo → jáar

H

H

In (21a), we observe that the deletion process targets only the tbu: When the final /e/ of /tuuÚé/ is deleted, its H tone thus floats and is relinked to the first tbu of the second noun. In (21b), on the other hand, when the final /á/ of /jáará/ is deleted, its link to the H tone is also lost. Since a floating H tone does not result from vowel deletion, /lawo/ ‘child’ is realized L–L by default. This is exactly what is expected from the doubly linked H enforced by the OCP in the autosegmental representation in the input in (21b). If /jáará/ had been represented with two H autosegments, the second would have floated, and could potentially have been relinked to the following noun. While one might preclude this possibility by introducing a floating-H deletion process, the autosegmental representation makes exactly the right prediction without stipulation.

3

Tonal underspecification

Another representational issue in the study of tone concerns underspecification: the possibility that some tbus do not have a tone at all (see chapter 7: feature specification and underspecification). The example in (21a) above demonstrates how a tone can shift onto another tbu that was claimed to be underlyingly toneless. Such shifts can be quite long-distance, as seen in following examples from Giryama, which contrasts /H/ vs. /Ø/ (Philippson 1998: 321): (22)

a. ku-tsol-a b. ku-on-a

ki-revu ‘to choose a beard’ ki-révu ‘to see a beard’

/-tsol-/ ‘choose’ /-ón-/ ‘see’

H In (22a) both the infinitive /ku-tsol-a/ ‘to choose’ and the object noun /ki-revu/ ‘beard’ are underlyingly toneless (and pronounced all L by default). In (22b), on the other hand, the infinitive /ku-ón-a/ ‘to see’ consists of an underlying H-tone root, which however shifts to the penultimate syllable of the following word. Such long-distance processes, which make little sense in a segmental interpretation of tone, are best described as displacing a tone, here H, across toneless tbus. If one instead proposed that these tbus have L tones that are permeable, one would have to establish a system of, say, markedness conventions to predict when tones can

The Representation of Tone

8

cross each other. The representation of L as /Ø/, on the other hand, allows for such long-distance processes without requiring further complications. Just as in the case of single vs. multiply linked tones, underspecification potentially allows for more distinctions than fully specified representations. Limiting ourselves first to languages that have only two tone heights, (23) summarizes the possible analyses: (23)

a. b. c. d.

/H/ /H/ /L/ /H/

vs. /L/ vs. /Ø/ vs. /Ø/ vs. /L/

vs. /Ø/

As seen, the tonal contrast may be binary, with /H/ contrasting with /L/; privative, with either /H/ or /L/ contrasting with /Ø/; or both, in the case of where a ternary system of /H, L, Ø/ is required. Thus, the central question concerning any such system is to determine which of the above representations best accounts for the properties of the surface [H] vs. [L] contrast. A /H, L/ system is required when both features are phonologically active (see also chapter 4: markedness for more discussion of markedness). This is seen in Kuki-Thaadow, which has both H- and L-tone spreading (Hyman 2010): (24)

a. /kà + zóoI + lìen + thúm/ L

H

L

[kà zòoI líen th"m] ‘my three big monkeys’

H

b. /kà + kèel + góoI + gùup/ [kà kèel gòoI gûup] ‘my six thin goats’ L

L

H

L

Not only do both /H/ and /L/ spread onto a following /L/ and /H/ syllable, respectively, but the result in final position is a rising (LH) or falling (HL) contour tone. Both spreading and contour tones would be difficult to represent if one of the tones were underspecified. The same would be true of a language which has both floating H and L. While /H, L/ systems are those in which the phonology refers to both tone values, in a privative tone system only one of the two tones is phonologically active. Many Bantu languages have a /H, Ø/ system, as was seen in the Giryama examples in (22). A few have a /L, Ø/ system, such as Ruund, for which Nash (1992–94) gives the following arguments: (25)

a. b. c. d.

Hs are by far more numerous than Ls, hence “unmarked.” Floating L exists, while floating H does not. Morphological rules assign L tones, not Hs. Phonological rules manipulate L tones, not Hs.

Athabaskan languages are also known for having H- vs. L-marked tone systems (see the various studies in Hargus and Rice 2005). The closely related South American dialects, Bora (Weber and Thiesen 2000) and Miraña (Seifart 2005), have

Larry M. Hyman

9

been reported to have /L, Ø/ systems based on properties similar to (25), but with the additional OCP argument that two word-level /L/ tones cannot occur in sequence. A similar argument is made for Munduruku (Picanço 2002), but with the need to distinguish two kinds of L tone: a /L/, which both triggers and undergoes a change to H after L, vs. a /Ø/ tone, which does not trigger, but does undergo, the change to H. Other /H, L, Ø/ systems in which toneless /Ø/ receives its tone from context or by default include Margi (Pulleyblank 1986) and Kinande (Mutaka 1994). A variation on this is Ganda, which Hyman et al. (1987) analyze with underlying /H, Ø/ and surface [H, L], but with a three-way H vs. L vs. Ø contrast at an intermediate level of representation. While Ganda introduces the L at the word level (where the dissimilatory process known as Meeussen’s Rule converts /H–Hn/ to H–Ln, e.g. /a-bá-láb-á/ → [a-bá-làb-à] ‘they who see’), other /H, Ø/ systems introduce L tones at the phonological phrase level or as part of the intonational system. Still others may introduce L targets in the phonetic interpretation or perhaps not at all (Myers 1998). In Tangale, an utterance-final /Ø/ tbu is pronounced extra-low. However, when the final syllable of a /Ø–H/ sequence undergoes lowering before pause, the final syllable does not become extra-low. While Kenstowicz and Kidda (1987) account for this by a rule delinking a singleton H before pause, an analysis more in line with what happens in other tone systems, e.g. in Grassfields Bantu (Hyman and Tadadjeu 1976), would involve L-tone spreading into a prepausal H, as in (26a). (26)

a. tùuNè˚ ‘horse’

b. laIóró

L H //

‘donkey’

L H

As in Grassfields Bantu, the symbol L° means that a L tone doesn’t fall (or lower) before pause (//). Since Kenstowicz and Kidda propose an underlying /H, Ø/ system, default Ls would first have to be introduced to trigger the delinking of the prepausal H. As seen in (26b), the L-spreading rule only targets an immediately following H tbu that occurs directly before pause. Corresponding to the four different analyses of two-height tone systems in (23) are those characterizing systems with three tone heights in (27). (27)

a. b. c. d. e.

/H/ /H/ /H/ /Ø/ /H/

vs. vs. vs. vs. vs.

/M/ /Ø/ /M/ /M/ /M/

vs. vs. vs. vs. vs.

/L/ /L/ /Ø/ /L/ /L/

vs. /Ø/

As in the case of /H, L/, if the three tone heights are phonologically active, the system will most likely require underlying /H, M, L/. Examples are MDnD (Kamanda-Kola 2003; Olson 2005), where all three heights occur as floating tones and combine to form all six tonal contours (HM, HL, MH, ML, LM, LH), and Gwari (Hyman and Magaji 1970), which has H-, M-, and L-tone spreading. It is generally assumed that M is the unmarked tone of a three-height tone system (Maddieson 1978; Pulleyblank 1988). Both Akinlabi (1985) and Pulleyblank

The Representation of Tone

10

(1988) propose that Yoruba has the system in (27b), where M tone is not only unmarked, but underspecified. Among the arguments is the fact that Yoruba allows only HL and LH contours, but none involving M tone (*MH, *LM, *HM, *ML). If M = /Ø/, it follows that Ø cannot form a contour with either H or L. Another argument derives from alternations in which H or L overrides M tone. While all of these properties naturally follow from the underspecification of M, the latter can be set up as /M/, as long as other mechanisms are put in place to capture the recessiveness of M tones. Pulleyblank (2004), for instance, proposes an account within Optimality Theory in which, essentially, M tones need not be preserved in outputs to the same degree as H or L. An argument for a /H, Ø, L/ system comes from Peñoles Mixtec (Daly and Hyman 2007), which has an OCP constraint disallowing successive Ls, or Ls which are separated by any number of /Ø/ tbus. The result is the L-tone deletion rule in (28a). (28)

a. b.

L → Ø / L __ (N = nasalization) qqN dìi-ni-kwe-œi kada-kwe-œi qqN qqN ŒiuN l l ↓ Ø ‘only one of them will do each of the jobs’

As seen, this rule is responsible for the loss of the second L on /ŒìuN/ ‘work’ in (28b), which is separated from the first L by twelve toneless tbus. That the two L tones can ‘see’ each other over long distances is strong evidence for the underspecification account. Returning to the systems in (27), (27c) and (27d) seem to be rare or nonoccurring. Paster (2003) proposes /H, M, Ø/ for Leggbó, with /Ø/ alternating between H and L in the verb morphology (cf. §4). If correct, the relatively few short-vowel LH, HL, and LM contours would have to be exceptionally marked: [ègg"] ‘catfish (sp.)’, [gbppjôn] ‘afternoon’, [lèssò¯l] ‘last year’. While I am unaware of any three-height tone system being analyzed as /Ø, M, L/, the system in (27e) seems at first appropriate for Yatzachi Zapotec (Pike 1948). Although this language has only the three surface tones H, M, L (and HM and MH contours on monosyllabic words), there are two kinds of L tones: those that alternate with M vs. those that do not. Pike identifies these, respectively, as class A vs. class B. For example, a class B L will become M when followed by a M or H within or across words, while a class A L tone will not be affected. This is seen in the minimal pair in (29). (29)

a. b.

La Lb

[bìa] ‘cactus’ [bìa] ‘animal’

bìa gdlc ‘old cactus’ bca gdlc ‘old animal’

A natural interpretation would be to recognize La as /L/ and Lb as /Ø/. Where the /Ø/ tone is not realized M, i.e. when not occurring before a /M/ or /H/, it will be realized with a default L tone, hence merging with class A /L/. However, it is hard to evaluate this proposal without considering what the featural representation is of all of the tones in the language (see Hyman 2009). It is likely that Yatzachi Zapotec has two kinds of L tone because it originally contrasted four tone heights. The Lb /Ø/ could therefore have been a tone level between M and

11

Larry M. Hyman

L, which Pulleyblank (1986), interestingly, considers the underspecified tone of a four-height tone system. In all of the above discussion, we have been considering only systems that have the same number of underlying and surface tone levels (perhaps with downstep). It is also possible for languages to have fewer underlying tone levels than on the surface. For example, Kom has underlying /H/ and /L/, which spread and float, and form LH and HL contours. However, as discussed in §5 below, there is a rule H → M / L __ which results in a single H tbu being realized M on the surface. Although Welmers (1962) analyzes Kpelle with three tone heights, the surface Ms are clearly the realization of the /LH/ melody, whether realized on one tbu or more. This raises the possibility of a /H, M/ or /L, M/ system, where the two highest or two lowest tones are underlying and the third tone derived. Smith (1968) and Stahlke (1971) posit /H, M/ for Ewe dialects with H, M, and L tone, while Clements (1977) does the same for a dialect which has four tone heights: H, M, L, and ↑H (raised H). While the second-to-lowest height has been hypothesized to be the unmarked, if not underspecified, tone in a four-height tone system, it is not clear what to propose for systems with five contrastive tone heights, the maximum number that has been attested (Maddieson 1978; Edmondson and Gregersen 1992). The value of underspecification seems greatest in two-height tone systems, the ultimate case being culminative tone systems such as Somali, which allow at most one H tone per word. The status of underspecification becomes less clear in systems that have multiple tone levels and tonal contours. As will be seen in the next section, multilevel systems raise the important question of whether tones should be analyzed in terms of (binary) features.

4

Tone features, tonal geometry, and tone-bearing units

In the preceding sections we have followed the common practice of symbolizing tones as H, M, and L. It is generally assumed, however, that tones should be analyzed in terms of features and feature geometry (chapter 27: the organization of features). Take, for example, the question of tonal contours. We have already seen several pieces of evidence that these should be decomposed into level tones (or tone features), but there have been recurrent claims that at least some contour tones should be analyzed as units, i.e. they are “true” contour tones rather than being sequences of tones which happen to be linked to the same tbu (Biber 1981; Newman 1986). Although most or all of the putative cases are reinterpretable, nowhere is the intuition “contour = unit” stronger than in the study of Chinesetype tone systems. Yip (1989) tackles this issue and proposes two different tonal geometries for true contour tones vs. tone clusters: (30)

a. tonal contours tbu

H

b. tone clusters

tbu

L

L

tbu

H

H

tbu

L

L

H

The Representation of Tone

12

In (30) the tones link to a tonal node, which in turn links to the tbu. As seen, a tonal contour is one where the H and L tones link to a single tonal node, while a cluster is one where each tone has its own tonal node. This, then, nicely captures the intuition that there is something different about a Chinese 51, 35, or 214 as opposed to a LH rising or HL falling tone found in other tone systems. Thus, in Chinese it is not uncommon for a sandhi rule to replace one tonal contour by another, possibly unrelated, contour. Such is unheard of in African tone systems, where the tone clusters rarely, if ever, behave as units. Since Yip (1980) it has generally been assumed that two tone features are needed: one to make a basic tonal distinction between high and low, and the other to make a further split into higher vs. lower registers of high and low. Adopting Yip’s upper and Pulleyblank’s (1986) raised for these functions, this produces the distinctions in (31), where M represents a lower mid tone: (31) upper raised

H + +

M + −

M − +

L − −

As seen, the natural classes are {H, M} vs. {M, L}, which differ in the feature upper, and {H, M} vs. {M, L}, which differ in the feature raised. The motivation for a grouping of non-contiguous levels, {4, 2} vs. {3, 1}, is seen from cases of “tonal bifurcations” in East and Southeast Asia (Matisoff 1973): if [±upper] represents the original tonal opposition, often attributable to a laryngeal distinction in syllable finals, [±raised], typically attributable to a laryngeal distinction in syllable initials, can potentially modify the original contrast and provide the four-way opposition in (31). Yip (1980), Pulleyblank (1986), and others have implemented this, or a slightly revised, interpretation of the two binary contrasts, including the assignment of the second feature to a register node in an elaboration of the tonal geometry in (30). One issue that immediately arises is how to represent the fifth tone level attested in, for example, Kam (Shidong) (Edmondson and Gregerson 1992): (32)

Ía11 Ía22 Ía33 Ía44 Ía55

‘thorn’ ‘eggplant’ ‘father’ ‘step over’ ‘cut down’

With this many contrasts, numbers seem to be more useful than letters to represent the tone levels (Chao 1930). Returning to the distinctions that upper and raised do make in (31), how would one express {M, M} as a natural class? Although they are contiguous on the pitch scale, they do not share a feature. Work on this topic has been inconclusive. Rules such as M → M / L __ typically lower a contiguous sequence of Ms after L and are therefore equally interpretable as M → ↓ M (Hyman 1986). While the phenomenon of tonal downstep (↓) is best known for establishing H vs. ↓H contrasts in underlying two tone-height systems, M vs. ↓ M and L vs. ↓L contrasts are also attested. The outstanding question here is how to represent downstepped tones. One possibility in (33a) is that the downstep in a H–↓H sequence is represented only structurally, e.g. as a floating L tone

Larry M. Hyman

13

wedged between H tones, as Clements and Ford (1979) originally proposed for Kikuyu: (33)

a. tbu H

tbu L

b. tbu tbu

H

H

H

However, it was mentioned with respect to Shambala in §2.1 that the downstep might instead be left to the phonetic interpretation of two or more H features on successive tbus, as in (33b). Such an analysis is not possible when a contrastively downstepped tone follows a non-identical tone. A H–↓M sequence would thus require an intervening floating L, as has also been proposed for H–↓L, L–↓L, and L–↓H sequences which contrast with H–L, L–L, and L–H in Bamileke-Dschang (Hyman and Tadadjeu 1976). The question still remains whether M and M ever function as a natural class. The feature specifications in (31) predict that they should not. Actually, they predict that in a three-height tone system, M tone might have either of two specifications: [+upper, −raised] or [−upper, +raised]. In fact, in a three-level tone system, there could be two phonetically equivalent M tones that realize both of these feature combinations. In a number of three-height systems where there is a lexical contrast of /H/, /M/, and /L/ on nouns, verbs instead fall into two tone classes, whose realizations vary according to inflectional features and clause type. In (34) I distinguish two such systems: Type I

(34) a. b.

Higher tone class Lower tone class

H M

M L

Type II M M

H L

In Type I languages, e.g. Day (Nougayrol 1979), the contrasting tones have a higher and lower variant in the two environments: H ~ M vs. M ~ L. The problem here is double. First, one has to decide what the underlying (input) tones are: /H, M/, /M, L/, or maybe even /H, L/? It all depends on whether one views the alternations as raising or lowering (or both). The second problem is determining the feature representations. One possibility would be to start with one tone as [+upper] and the other as [−upper]. The assigned grammatical feature would then be [+raised] in (34a) and [−raised] in (34b). In this analysis M would be [+upper, −raised] in (34a) vs. [−upper, +raised] in (34b). In the Type II languages, e.g. Leggbó (Hyman et al. 2002), one verb tone is /M/, which does not change, while the other tone varies between H and L. Since H and L do not share either feature value of upper and raised (and certainly not to the exclusion of M), we can either introduce another feature that they could share, such as [+extreme] (Maddieson 1971), or seek another solution. Paster (2003) posited /M/ vs. /Ø/ for Leggbó, with the morphology assigning a H or L prefix, which takes the place of /Ø/ but cannot override /M/. A third possibility is simply to represent the contrast as /H/ vs. /M/, with H → L being a morphologically conditioned rule. In the above it has been assumed without discussion that tone heights should be characterized by features that are binary rather than multivalued (Stahlke 1977)

The Representation of Tone

14

or binary and hierarchized (Clements 1983). Anderson (1978) provides a comprehensive appraisal of the different feature proposals up to that date. If assimilations are assumed to spread individual features such as [upper] and [raised], these latter must occur on independent tiers. This then raises the question of where these tiers link up, e.g. to a laryngeal node or directly to the tbu? There have been numerous proposals in the literature (see the surveys and discussion in Yip 1995, 2002; Bao 1999; Snider 1999; and Lee 2008). The advantage of the first proposal is that tones frequently interact with other laryngeal properties. Halle and Stevens’ (1971) system in (35) – an early attempt to capture the relation between tone and obstruent voicing – fails, however, to characterize more than three tone heights: tones

(35) stiff slack

H M + − − −

L − +

voiceless obstruents

sonorants

voiced obstruents

ptkfs + −

mnlwj − −

bdgvz − +

Clearly, one wants to account for the relation between tone and (non-modal) phonations, or the interference of obstruent voicing with tonal assimilations, but not at the expense of losing the generalization that tones are distributed by tbus. While tones have an autosegmental independence, they ultimately must be realized on something, e.g. a vowel or syllabic consonant. A language may consistently assign tones by mora, such that a CVV syllable receives two tones, or it may assign tones by syllable. Sometimes it is difficult to distinguish between the two. The complexities and corresponding representational possibilities are many. Both Clements et al. (2009) and Hyman (2009) have expressed doubts that tones should be analyzed in terms of features at all.

5

Possible limits of autosegmental representation

Two general conclusions can be drawn from the preceding sections. First, there is still much merit in the autosegmental insights on tone. Second, there is much more work to do. In this section we first address real or apparent problems for autosegmental tonology. The arguments for autosegmental tonology were enumerated in §2. Some of the evidence concerned the non-isomorphism between the tones and their tbus: more than one tone can link to a tbu, in which case we get a tonal contour or cluster; conversely, one tone can link to more than one tbu. Contrasting representations such as those in (12a) and (17a) were said to be needed. The question here is whether they ever get in the way: are there cases where it is disadvantageous to represent a tautomorphemic H–H or L–L as a single tone linked to two tbus? One such awkward case would seem to arise in Kom, which has an underlying contrast between /H/ and /L/, but a surface contrast between H, M, and L. All M tones are derived by a rule that lowers a single H tbu to M when preceded by a L or initial phrase boundary (which can be represented by a %L boundary tone). This produces outputs such as the following:

Larry M. Hyman

15 (36)

a.

/fe-:am/ H L b. /fe-Jwin/ H LH c. /fe-bu?/ H HL d. /fe-tam/ HH

→ fb-:âm

‘mat’

[M–HL]

→

fb-JwíT

‘bird’

[M–HM]

→

fb-bú?

‘gorilla’

[M–H]

→

fb-tám

‘fruit’

[M–H]

As in these examples, most nouns in Kom have a /H/-tone prefix followed by a monosyllabic stem that can have any of the four tone patterns exemplified in (36). As schematized in (37), the H of the prefix spreads onto a following L or LH stem: (37)

fb-:âm

a. %L

H L

fb-Jwí¶

b. %L

H

L

H

As seen in the transcriptions, the output is M–HL in (37a) and M–HM in (37b). The M tones in question are conditioned by the rule that lowers a H to M after a (linked, floating, or boundary) L. In both forms the prefix is thus lowered to M; in addition, in (37b), the (delinked) stem L lowers the following H to produce the HM contour. Since the prefix lowers to M without affecting the stem, the lowering rule cannot be written as a single-tier rule, as in (38), or we would get the wrong outputs *M–ML and *M–M: (38)

H → M / L __ (cf. the Kukuya rule in (13))

It is clear that the doubly linked H representation is not useful, but it is not fatally contradictory to the autosegmental approach. At least four responses come to mind. First, assuming that the tbu is the mora and that stems are bimoraic, the M can be derived by delinking a H from the first tbu (mora) that immediately follows a L. Second, one can complicate the tonal geometry to include a tonal root node, as in (39), to which the preceding L can link as a register feature, perhaps [−raised] (cf. the surveys of similar proposals in Bao 1999 and Snider 1999): (39)

[

[ (tonal root node) (tonal node)

L

H

Third, assuming that floating tones persist into the output, one might argue that the M outputs are derived by phonetic implementation, where the M may be interpolated as part of the aligning of output tones with their segmental supports. A fourth possibility is to spread the L to form a LH tbu, which is phonetically interpreted as M.

The Representation of Tone

16

In other cases the autosegmental representations may have to be seriously “fixed up.” In the Mijikenda languages Chikauma and Chirihe, the H tone of a prefix shifts to penultimate position (Cassimjee and Kisseberth 1992: 29). (40)

a. /ni-a-sukum-a/ → n-a-sukum-a b. /ú-a-sukum-a/ → w-a-sukúm-a H

‘I am pushing’ ‘s/he is pushing’

H

All of the morphemes in (40a) are underlyingly toneless, including the 1st person singular subject prefix /ni-/. In (40b), the class 1 (human singular) subject prefix /ú-/ has an underlying tone, which, however, is realized on the penult. Cassimjee and Kisseberth argue that this should be accounted for by spreading and delinking of all but the last link of the H, as shown. Evidence for this analysis comes from the interaction of so-called depressor consonants (voiced obstruents) with H-tone spreading. Thus consider the forms in (41), where the verb stem /-galuk-/ ‘change’ begins with a depressor consonant: (41)

a. b.

→ →

/ni-a-galuk-a/ /ú-a-galuk-a/

n-a-galuk-a w-á-galúk-a

‘I am changing’ ‘s/he is changing’

Again the input morphemes are all toneless in (41a). In (41b) we see that the one underlying H is realized both on the penult, where expected, but also on the marker -á-. In the final set of data, we see the toneless form in (42a) contrasting with three H tones in (42b). (42)

a. b.

/ni-a-zi-fugul-a/ /ú-a-zi-fugul-a/

→ →

n-a-zi-fugul-a ‘I am untying them’ w-á-zi-fú↓gúl-a ‘s/he is untying them’

Cassimjee and Kisseberth’s analysis is schematized in (43), where the only underlying H tone is on the subject prefix /ú-/: (43)

a.

b.

L

L

d.

c.

u-a-zi-fugul-a → w-a-zi-fugul-a → w-a-zi-fugul-a → w-á-zi-fú↓gúl-a H

H

H L H LH

H L HLH

As indicated in (43a), the H of /ú-/ spreads to the penult. In (43b) a L-tone feature is assigned to each depressor consonant, in this case to the /z/ of the object prefix /zi-/ and to the stem-internal /g/ of /-fugul-/ ‘untie’. In (43c) these Ls are folded in with the multiply linked H tone. In a process which Cassimjee and Kisseberth term “fission,” these Ls divide up the one H into separate H tone domains. After this, delinking occurs in (43d), where the downstepped ↓H is conditioned by the depressor L, which is wedged between two H tbus. As in the Kom case, something needs to be added to fix the multiply linked structures derived from tone spreading. Despite the complications as shown by Kom and Chikauma/Chirihe, there is no reason to abandon the autosegmental insights; rather, we can extend and build on them as languages require.

Larry M. Hyman

17

On the other hand, there were some proposals of early autosegmental tonology that had to be abandoned. Most of these, listed in (44), concerned conventions that turned out to be too strong: (44)

a. b. c.

Every tbu must have a tone, hence automatic spreading. Every tone must have a tbu, hence automatic contouring. Tonal melodies should automatically map left to right.

These conventions were felt to be needed to account for the dashed associations in (44). (45)

a. tbu tbu

b. tbu

H

H

c.

tbu tbu tbu L

L

H

In (45a) a toneless tbu follows a H tone tbu. As we saw in the case of the toneless locative postpositions in Mende in (20), it was proposed that the last tone on the left would automatically spread to any toneless tbus to its right. Pulleyblank (1986) argued, however, against “automatic spreading” by showing that another option was for the toneless tbu(s) to receive a default tone, e.g. L. In (45b), the opposite is the case: there are more tones than tbus. Early autosegmental tonology assumed that the leftover tones would automatically link to the last tbu, as shown. However, numerous studies argued that certain tones should be left floating, e.g. a floating L to condition downstep, while Hyman and Ngunga (1994) present data from Ciyao to argue that even a floating H does not undergo “automatic contouring.” Finally, in (45c), we see the “automatic” left-to-right mapping of a LH melody to three tbus. The proposal in early autosegmental tonology was that free tones link one-to-one from left to right. As seen, this produces a L–H–H sequence. However, an alternative of “edge-in” association is proposed by Yip (1988), by which the first tone would map to the first tbu and the last tone to the last tbu, and then intervening available tbus would receive their tones from the left, thereby producing a L–L–H sequence. Edge-in association works well for Kukuya (Paulian 1975; Hyman 1987), where /LH/ maps as L–L–H and /HL/ as H–L–L on trimoraic stems (with the internal tbu being L). On the other hand, Zoll (2003) argues that apparent directional effects may result from whether a language prefers multiple Hs or multiple Ls. Kukuya definitely prefers L sequences, violating Zoll’s Lapse constraint, over multiple Hs, which would violate her Clash constraint. As a dramatic confirmation of Zoll’s tone-specific approach, consider the realization of tonal schemas in Fore in (46), where H, L = floating tones (Scott 1990): (46)

schema

1q

2q

3q

4q

/L/ /H/ /LH/ /HL/ /LHL/ /HLH/ /LHLH/ /HLHL/

L H LH HL

L–L H–H L–H H–L L–HL H–LH

L–L–L H–H–H L–L–H H–L–L L–H–L H–L–H L–H–LH H–L–HL

L–L–L–L H–H–H–H L–L–L–H H–L–L–L L–H–L–L H–L–L–H L–H–L–H H–L–H–L

The Representation of Tone

18

As seen, there are no sequences of H tones if a L is present in the input. Scott (1990) summarizes the system as follows: The simplest system which may be hypothesized for Fore is one in which only changes between high and low tone are recognised as being contrastive. (1990: 141) There are no contrastive contours. (1990: 147) . . . tones appear to spread by increasing the domain of the L tones in preference to the spreading of H tones . . . From this it appears that H tones are to be considered as peaks of prominence or pitch targets. (1990: 147)

In other words, underlying /H/ and /L/ in Fore are constrained as follows. First, sequences of L are to be preferred to sequences of H and, second, there is no default L tone: if a word has a /H/ melody, the H will link to all of the tbus. In the next and final section we consider the question of whether non-derivational approaches to phonology need change our view of tonal representations.

6

Constraint-based tonology

The above characterization of Fore is of course a way of describing the tonal distributions from a constraint-based perspective. Within Optimality Theory (Prince and Smolensky 1993; McCarthy 2002), the constraints might be ranked as follows: (47)

Max(Tone) >> Dep(Tone) >> *H >> *L

Max(Tone) says that any input tone has a corresponding tone in the output, while Dep(Tone) says that any output tone has a corresponding tone in the input. In other words, tones are neither deleted nor inserted. The constraint ranking *H >> *L is designed to account for the limited number of H tones in (46): each tbu linked to a H tone is counted as a violation. The ranked constraints in (47) are but one way to conceptualize the Fore tonal distributions in such a framework. This therefore raises the question of what Optimality Theory (OT) has to offer tone and vice versa. High among the contributions of tone to our understanding of phonological systems is the family of OCP constraints that find a natural home in OT (Myers 1997). Building on his own and others’ previous work, Myers shows that the OCP functions as a conspiracy in a number of languages. In Chibemba, which contrasts /H/ and /Ø/, there is a process of bounded H-tone spreading illustrated in (48). (48)

a. b. c.

tu-la-kak-a tu-la-mu-kak-a bá-lá-kak-a bá-lá-mu-kak-a tu-la-bá-kák-a bá-la-bá-kák-a

‘we tie up’ ‘we tie him up’ ‘they tie up’ ‘they tie him up’ ‘we tie them up’ ‘they tie them up’

tu-la-súm-á tu-la-mu-súm-á bá-la-súm-á bá-lá-mu-súm-á tu-la-bá-súm-á bá-la-bá-súm-á

‘we bite’ ‘we bite him’ ‘they bite’ ‘they bite him’ ‘we bite them’ ‘they bite them’

In the above examples, all of the morphemes except /-súm-/ ‘bite’ and the plural class 2 prefixes /bá-/ ‘they’ and /-bá-/ ‘them’ are underlyingly toneless. As seen in the examples in the right column, the /H/ of /-súm-/ spreads onto the final

19

Larry M. Hyman

inflectional vowel /-a/. In (48b) the H of the subject prefix /bá-/ spreads onto the toneless present tense marker /-la-/, except in [bá-la-súm-á] ‘they bite’. This is because the spreading of H would have derived two H autosegments on successive tbus, thereby producing an OCP violation. In (48c), the /H/ of the object prefix /-bá-/ spreads onto the toneless verb root /-kak-/, as expected. Once again, the H of the subject prefix /bá-/ fails to spread, because *[-lá-bá-] would have constituted an OCP violation. However, note in the examples on the right in (48c) that, when concatenated, the object prefix /-bá-/ and the verb root /-súm-/ constitute an OCP violation, which is allowed to surface. In other words, the OCP blocks H-tone spreading when the result would be a violation, but does not condition any change on input violations. Such a situation, known as The Emergence of the Unmarked (McCarthy and Prince 1994; chapter 58: the emergence of the unmarked), is predicted by and can be elegantly modeled within OT: (49)

Max(H) >> OCP(H) >> Spread(H)

Max(H) guarantees that input Hs will be preserved in the output. OCP(H) says that there should not be Hs on successive tbus. The constraint Spread(H) is designed to say that every input H should spread onto the following tbu. The above ranking establishes that the OCP will be violated rather than deleting offending input Hs, while spreading will not occur if the result is an OCP violation. As we have seen, tone systems allow for a multiplicity of interpretations and solutions, producing an indeterminacy that can be referred to as “the too many analyses problem.” The problem is perhaps more pronounced in OT, where it is not clear how to interpret certain potentially complex tonal processes. For example, in a rule-based framework, bounded tone shift, where each input tone is realized on the following tbu, is straightforwardly described in terms of bounded spreading and delinking. While each of the two processes is independently motivated, it is not clear what is optimized in a system where the processes have been “telescoped” into a synchronic input–output relation of local tone shift. With few exceptions, the constraints that have been proposed for tone have not been universally adopted (see Akinlabi and Mutaka 2001 and Yip 2002 for partial inventories of tonal constraints). While OT tonology is perhaps in an unsettled state, a lot depends on the tonal representations that one assumes. At least one approach, Optimal Domains Theory (Cassimjee and Kisseberth 1998), groups tones together into tonal domains rather than exploiting the autosegmental manyto-one linkings between tones and tbus (see also McCarthy’s 2004 notion of “headed spans”). One area that seems crucial, but has not received the attention it deserves, is floating tones. Do floating tones violate Max(Tone) if they are allowed to survive as such in surface representations? To answer this one must first demonstrate that they are needed to survive in surface representations. We saw in (33a) that a floating L wedged between two linked Hs is one of the proposed representations of downstep. Since there would be no problem converting this structure into one where the floating L feature is assigned as a register feature on the following tone (and tonal sequence), the example is not conclusive. Let us then return to the tonal contrasts on Kom nouns in (36). In (50) their input and output tones are shown as they occur between the /L/ tone preposition /nè/ ‘with’ and the /H/-tone postposition /-fé/ (Hyman 2005: 317):

The Representation of Tone (50)

a.

-fe → nè fè-:aå -fè° ‘with a mat’

ne fe-:am L H L

b.

H

L H L

c.

L H

L H HL

H

d. ne fe-tam L H H

-fb

L H HL

H

-fe → nè fè-tám

-fé

H

L HH

‘with a bird’

[L–L–M–H]

‘with a gorilla’

[L–L–H–M]

‘with a fruit’

[L–L–H–H]

LH H

-fe → nè fè-bú?

ne fe-bu?

[L–L–ML–L°]

H

ne fe-Jwin -fe → nè fè-Jwcn -fé L H L H H

20

H

As seen, Kom has both H- and L-tone spreading. In (50a), L-tone spreading causes the following H to delink (producing a level L° tone on the postposition), while H-tone spreading does not cause the following L to delink. With a single H tbu lowering to M after L (cf. (37)), the noun stem /-:àm/ ‘mat’ surfaces with a ML contour tone. Similarly, the stem /-Jwì5/ ‘bird’ surfaces as M, each of the two linked H tones being lowered to M by a preceding L. The important examples are (50c) and (50d). As seen, when the L of /nè/ spreads and delinks the H of the noun prefix /fé-/, the latter’s H tone floats. As a result, the floating H shields the roots /-bú?`/ ‘gorilla’ and /-tám/ ‘fruit’ from lowering to M. (The floating L of /-bú?`/ is responsible for the lowering of the postposition /-fé/ to M in (50c).) Since the H to M lowering process is a late one for which we have even entertained the possibility that it applies in phonetic implementation, it should be clear that the floating Hs must persist into the surface representations. The alternative, that L-tone spreading does not delink a following H, but rather creates LH rising tones that are simplified in phonetic implementation, is perhaps suspect, if not ad hoc. If, on the other hand, the floating tones are allowed to occur in surface representations, the solution is straightforward. Such a conclusion would also seem to have consequences for other analyses. Particularly within OT, the question arises whether a surviving, delinked floating tone should be able to satisfy Max(Tone). Data from Kuki-Thaadow suggest maybe not. In this largely monosyllabic Tibeto-Burman language, words are underlyingly /H/, /L/, or /HL/ (Hyman 2010). As seen in (51a), however, a rising tone results from L-tone spreading: (51)

a.

L b.

→ hùon z oI

/hùon + zóo /

‘garden monkey’

H

/hùon + zóo + gùup/ → hùon zòoI gûup ‘six garden monkeys’ L

H

L

Larry M. Hyman

21

In (51b) we see that Kuki-Thaadow also has H-tone spreading, and the H of /zóoI/ is delinked. This is because rising (LH) and falling (HL) tones are permitted only phrase-finally. Thus, the L of /hùon/ ‘garden’ is realized on /zóoI/ ‘monkey’, and the H of /zóoI/ is realized on /gùup/ ‘six’. Now consider the forms in (52). (52)

a.

/hùon + zóoI + thúm/ → hùon zóoI thúm ‘three garden monkeys’ L

b.

H

H

/hùon + zóoI + gîet/ → hùon zóoI gîet L

H

‘eight garden monkeys’

HL

As seen, L-tone spreading does not occur when the targeted H tone is followed in turn by a H or HL tone. The question is: Why not? To answer this we have but to consider what the output would have been if L-tone spreading had applied in (52). The H of /zóoI/ would necessarily have had to delink, since a LH rising tone is well formed only in phrase-final position. As a result, the underlying /H/ would, at best, have to float. The question, then, is why this would not be well formed. It cannot be that an input link must remain as such in the output, since the H does delink in (51b). Instead, what seems to allow L-tone spreading to apply in (51a) and (51b) is that the input /H/ of /zóoI/ is preserved in the output: It is realized within the LH contour on its own syllable in (51a) and as part of the HL contour on the following syllable in (51b). In other words, Max(H) is satisfied. We must suppose, therefore, that if L-tone spreading applied in (52) and delinked the /H/ of /zóoI/, the resulting floating H would not satisfy Max(H). By contrast, H-tone spreading applies whether or not the targeted L remains linked on the surface (see Hyman 2010). We therefore can establish the ranking in (53). (53)

Max(H) >> Spread(H,L) >> Max(L)

In Kuki-Thaadow all input H tones make it to the surface, vs. closely related Hakha Lai, where the opposite ranking Max(L) >> Max(H) results in all input L tones being realized on the surface (Hyman and VanBik 2004). It should be noted that the non-application of L-tone spreading in (52) results more straightforwardly from the ranked constraints in (53), rather than from a requirement on recoverability: had L-tone spreading applied in (52a), the output L–L–H sequence would have unambiguously pointed to underlying /L–H–H/, since /L–L–H/ would have been realized as L–L–LH. (The same is not true of (52b), since L-tone spreading does not apply to a following HL tone.) If (53) is correct, OT will have made a unique contribution in providing a constraint, Max(H), which is responsible for the nonapplication of L-tone spreading in (52). By contrast, a rule-based approach would have to stipulate that L-tone spreading occurs to a L–H sequence when the H is followed either by pause or by a L tone, and would not provide any motivation for why the rule does not apply when the following tone is H or HL. Whether this is an indication that OT is on the right track – and can hence offer more improvements over the pre-OT conceptions of tonology – remains to be seen. In fairness,

The Representation of Tone

22

it must be said that the blocking of an otherwise general L-tone spreading process in (52) is unique to Kuki-Thaadow. As pointed out by Hyman (1973b: 157), we expect L–H–H to be a better target for L-tone spreading than L–H–L. Except for Kuki-Thaadow, this is true whether the H–H sequence is from two /H/ tones or from one /H/ that is doubly linked. Where does this leave the representation of tone? To summarize the foregoing sections, although the well-formedness and mapping conventions in (44) have been superseded in subsequent work, most of the essential representational insights of autosegmental tonology are still intact. The above discussion has only touched on a small part of the vast world of tone and of the growing constraint-based literature treating tone. Whatever the outcome of ongoing OT interpretations of tone, it is likely that questions of representation will remain central.

REFERENCES Akinlabi, Akinbiyi. 1985. Tonal underspecification and Yorùbá tone. Ph.D. dissertation, University of Ibadan. Akinlabi, Akinbiyi & Ngessimo M. Mutaka. 2001. Tone in the infinitive in Kinande: An OT analysis. In Ngessimo M. Mutaka & Sammy B. Chumbow (eds.) Research mate in African linguistics: Focus on Cameroon, 333–356. Cologne: Rüdiger Köppe Verlag. Anderson, Stephen R. 1978. Tone features. In Fromkin (1978), 133–175. Bao, Zhiming. 1999. The structure of tone. New York & Oxford: Oxford University Press. Bateman, Janet. 1990. Iau segmental and tonal phonology. Miscellaneous Studies of Indonesian and Other Languages in Indonesia 10. 29–42. Biber, Douglas. 1981. The lexical representation of contour tones. International Journal of American Linguistics 47. 271–282. Bogers, Koen, Harry van der Hulst & Maarten Mous (eds.) 1986. The phonological representation of suprasegmentals. Dodrecht: Foris. Braconnier, Cassian. 1982. Le système du dioula d’Odienné, vol. 1. Abidjan: University of Abidjan. Cassimjee, Farida & Charles W. Kisseberth. 1992. The tonology of depressor consonants: Evidence from Mijikenda and Nguni. Proceedings of the Annual Meeting, Berkeley Linguistics Society 18(2). 26–40. Cassimjee, Farida & Charles W. Kisseberth. 1998. Optimal Domains Theory and Bantu tonology: A case study from Isixhosa and Shingazidja. In Hyman & Kisseberth (1998), 33–132. Chao, Yuen-Ren. 1930. A system of tone-letters. Le maître phonétique 45. 24–27. Clements, G. N. 1977. The autosegmental treatment of vowel harmony. In Wolfgang U. Dressler & Oskar E. Pfeiffer (eds.) Phonologica 1976, 111–119. Innsbruck: Innsbrucker Beiträge zur Sprachwissenschaft. Clements, G. N. 1981. Akan vowel harmony: A non-linear analysis. Harvard Studies in Phonology 2. 108–177. Clements, G. N. 1983. The hierarchical representation of tone features. In Ivan R. Dihoff (ed.) Current approaches to African linguistics, vol. 1. 145–176. Dordrecht: Foris. Clements, G. N. 1985. The geometry of phonological features. Phonology Yearbook 2. 225–252. Clements, G. N. & Kevin C. Ford. 1979. Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry 10. 179–210. Clements, G. N. & Elizabeth Hume. 1995. The internal organization of speech sounds. In Goldsmith (1995), 245–306.

23

Larry M. Hyman

Clements, G. N., Alexis Michaud & Cédric Patin. 2009. Do we need tone features? Paper presented at the Symposium on Tones and Features, University of Chicago. Daly, John P. & Larry M. Hyman. 2007. On the representation of tone in Peñoles Mixtec. International Journal of American Linguistics 73. 165–208. Edmondson, Jerold A. & Kenneth J. Gregerson. 1992. On five-level tone systems. In Shina Ja J. Huang & William R. Merrifield (eds.) Language in context: Essays for Robert E. Longacre, 555–576. Arlington: Summer Institute of Linguistics & University of Texas at Arlington. Fromkin, Victoria A. (ed.) 1978. Tone: A linguistic survey. New York: Academic Press. Goldsmith, John A. 1976a. Autosegmental phonology. Ph.D. dissertation, MIT. Goldsmith, John A. 1976b. An overview of autosegmental phonology. Linguistic Analysis 2. 23–68. Goldsmith, John A. (ed.) 1995. The handbook of phonological theory. Cambridge, MA & Oxford: Blackwell. Halle, Morris & Kenneth N. Stevens. 1971. A note on laryngeal features. MIT Research Laboratory of Electronics Quarterly Progress Report 101. 198–213. Hargus, Sharon & Keren Rice (eds.) 2005. Athabaskan prosody. Amsterdam & Philadelphia: John Benjamins. Hoberman, Robert D. 1988. Emphasis harmony in modern Aramaic. Language 64. 1–26. Hulstaert, Gustaaf. 1961. Grammaire du LDmDngD, vol. 1: Phonologie. Tervuren: Musée Royal de l’Afrique Centrale. Hyman, Larry M. 1973a. The role of consonant types in natural tonal assimilations. In Hyman (1973b), 151–179. Hyman, Larry M. (ed.) 1973b. Consonant types and tone. Los Angeles: University of Southern California. Hyman, Larry M. 1979. Phonology and noun structure. In Larry M. Hyman (ed.) Aghem grammatical structure, 1–72. Los Angeles: University of Southern California. Hyman, Larry M. 1986. The representation of multiple tone heights. In Bogers et al. (1986), 109–152. Hyman, Larry M. 1987. Prosodic domains in Kukuya. Natural Language and Linguistic Theory 5. 311–333. Hyman, Larry M. 2005. Initial vowel and prefix tone in Kom: Related to the Bantu Augment? In Koen Bostoen & Jacky Maniacky (eds.) Studies in African comparative linguistics with special focus on Bantu and Mande: Essays in honour of Y. Bastin and C. Grégoire, 313–341. Cologne: Rüdiger Köppe Verlag. Hyman, Larry M. 2009. Do tones have features? Paper presented at the Symposium on Tones and Features, University of Chicago. Hyman, Larry M. 2010. Kuki-Thaadow: An African tone system in Southeast Asia. In Franck Floricic (ed.) Essais de typologie et de linguistique générale, 31–51. Lyon: Presses de l’École Normale Supérieure. Hyman, Larry M. & Charles W. Kisseberth (eds.) 1998. Theoretical aspects of Bantu tone. Stanford: CSLI. Hyman, Larry M. & Daniel J. Magaji. 1970. Essentials of Gwari grammar. Ibadan: Institute of African Studies, University of Ibadan. Hyman, Larry M. & Armindo Ngunga. 1994. On the non-universality of tonal association “conventions”: Evidence from Ciyao. Phonology 11. 25–68. Hyman, Larry M. & Maurice Tadadjeu. 1976. Floating tones in Mbam-Nkam. In Larry M. Hyman (ed.) Studies in Bantu tonology, 57–111. Los Angeles: Department of Linguistics, University of Southern California. Hyman, Larry M. & Kenneth VanBik. 2004. Directional rule application and output problems in Hakha Lai tone. Phonetics and Phonology, Special Issue, Language and Linguistics 5. 821–861. Taipei: Academia Sinica. Hyman, Larry M., Francis Katamba & Livingstone Walusimbi. 1987. Luganda and the strict layer hypothesis. Phonology Yearbook 4. 87–108.

The Representation of Tone

24

Hyman, Larry M., Heiko Narrog, Mary Paster & Imelda Udoh. 2002. Leggbó verb inflection: A semantic and phonological particle analysis. Proceedings of the Annual Meeting, Berkeley Linguistics Society 28, 399–410. Jenison, Scott D. & Priscilla B. Jenison. 1991. Obokuitai phonology. Workpapers in Indonesian Languages and Culture 9. 69–90. Kamanda-Kola, Roger. 2003. Phonologie et morphosyntaxe du mDnD: Langue oubanguienne du Congo R. D. Munich: Lincom Europa. Kenstowicz, Michael & Mairo Kidda. 1987. The obligatory contour principle and Tangale phonology. In David Odden (ed.) Current approaches to African linguistics, vol. 4, 223–238. Dordrecht: Foris. Leben, William R. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT. Leben, William R. 1978. The representation of tone. In Fromkin (1978), 177–219. Lee, Seunghun Julio. 2008. Consonant–tone interactions in Optimality Theory. Ph.D. dissertation, Rutgers University. Loving, Richard E. 1966. Awa phonemes, tonemes, and tonally differentiated allomorphs. Papers in New Guinea Linguistics 5. 23–32. Maddieson, Ian. 1971. The inventory of features. In Ian Maddieson (ed.) Tone in generative phonology, 3–18. Ibadan: Institute of African Studies, University of Ibadan. Maddieson, Ian. 1978. Universals of tone. In Joseph H. Greenberg, Charles A. Ferguson & Edith A. Moravcsik (eds.) Universals of human language, vol. 2: Phonology, 335–365. Stanford: Stanford University Press. Matisoff, James A. 1973. Tonogenesis in Southeast Asia. In Hyman (1973b), 71–95. McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12. 373–418. McCarthy, John J. 2002. A thematic guide to Optimality Theory. Cambridge: Cambridge University Press. McCarthy, John J. 2004. Headed spans and autosegmental spreading. Unpublished ms., University of Massachusetts, Amherst (ROA-685). McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. McCarthy, John J. & Alan Prince. 1994. The emergence of the unmarked: Optimality in prosodic morphology. Papers from the Annual Meeting of the North East Linguistic Society 24. 333–379. Mutaka, Ngessimo. 1994. The lexical tonology of Kinande. Munich: Lincom Europa. Myers, Scott. 1997. OCP effects in Optimality Theory. Natural Language and Linguistic Theory 15. 847–892. Myers, Scott. 1998. Surface underspecification of tone in Chichewa. Phonology 15. 367–391. Nash, Jay A. 1992–94. Underlying low tones in Ruwund. Studies in African Linguistics 23. 223–278. Newman, Paul. 1986. Contour tones as phonemic primes in Grebo. In Bogers et al. (1986), 175–193. Nougayrol, Pierre. 1979. Le day de bouna (Tschad), vol. 1: Eléments de description linguistique. Paris: SELAF. Odden, David. 1982. Tonal phenomena in KiShambaa. Studies in African Linguistics 13. 177–208. Olson, Kenneth S. 2005. The phonology of Mono. Dallas: SIL International & University of Texas at Arlington. Pankratz, Leo & Eunice V. Pike. 1967. Phonology and morphophonemics of Ayutla Mixtec. International Journal of American Linguistics 33. 287–299. Paster, Mary. 2003. Tone specification in Leggbo. In John M. Mugane (ed.) Linguistic typology and representation of African languages, 139–150. Trenton, NJ: Africa World Press. Paulian, Christiane. 1975. Le kukuya: Langue teke du Congo. Paris: Societé d’Études Linguistiques et Anthropologiques de France.

25

Larry M. Hyman

Philippson, Gérard. 1998. Tone reduction vs. metrical attraction in the evolution of Eastern Bantu tone systems. In Hyman & Kisseberth (1998), 315–329. Picanço, Gessiane. 2002. Tonal polarity as phonologically conditioned allomorphy in Mundurukú. Proceedings of the Annual Meeting, Berkeley Linguistics Society 28. 237–248. Pike, Eunice V. 1948. Problems in Zapotec tone analysis. International Journal of American Linguistics 14. 161–170. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Pulleyblank, Douglas. 1986. Tone in Lexical Phonology. Dordrecht: Reidel. Pulleyblank, Douglas. 1988. Vocalic underspecification in Yoruba. Linguistic Inquiry 19. 233–270. Pulleyblank, Douglas. 2004. A note on tonal markedness in Yoruba. Phonology 21. 409–425. Qingxia, Dai & Lon Diehl. 2003. Jingpho. In Graham Thurgood & Randy J. LaPolla (eds.) The Sino-Tibetan languages, 401–408. London & New York: Routledge. Scott, Graham. 1990. A reanalysis of Fore accent. La Trobe University Working Papers in Linguistics 3. 139–150. Seifart, Frank. 2005. The structure and use of shape-based noun classes in Miraña (North West Amazon). Ph.D. dissertation, Radboud University Nijmegen. Smith, Neil V. 1968. Tone in Ewe. MIT Research Laboratory of Electronics Quarterly Progress Report 88. 290–304. Snider, Keith L. 1999. The geometry and features of tone. Dallas: Summer Institute of Linguistics & University of Texas at Arlington. Stahlke, Herbert F. W. 1971. The noun prefix in Ewe. Studies in African Linguistics. 141–159. Stahlke, Herbert F. W. 1977. Some problems with binary features for tone. International Journal of American Linguistics 43. 1–10. Wang, William S.-Y. 1967. Phonological features for tone. International Journal of American Linguistics 33. 93–105. Weber, David & Wesley Thiesen. 2000. A synopsis of Bora tone. Work Papers of the Summer Institute of Linguistics 45. Available (May 2010) at http://www.und.edu/dept/ linguistics/wp/2001.htm. Welmers, William E. 1962. The phonology of Kpelle. Journal of African Languages 1. 69–93. Williamson, Kay. 1986. The Igbo associative and specific constructions. In Bogers et al. (1986), 195–208. Yip, Moira. 1980. The tonal phonology of Chinese. Ph.D. dissertation, MIT. Yip, Moira. 1988. Template morphology and the direction of association. Natural Language and Linguistic Theory 6. 551–577. Yip, Moira. 1989. Contour tones. Phonology 6. 149–174. Yip, Moira. 1995. Tone in East Asian languages. In Goldsmith (1995), 476–494. Yip, Moira. 2002. Tone. Cambridge: Cambridge University Press. Zoll, Cheryl. 2003. Optimal tone mapping. Linguistic Inquiry 34. 225–268.

46

Positional Effects in Consonant Clusters Jongho Jun

1

Introduction

It is commonly observed across languages that phonological processes may apply only in certain (“non-prominent”) positions. In contrast, elements in other (“prominent”) positions typically resist or trigger these processes. Such prominent vs. non-prominent positional distinctions are further applicable to more general patterns of licensing and neutralization of phonological contrasts; namely, featural/segmental contrasts are likely to be licensed in prominent positions, whereas these contrasts are likely to be neutralized in non-prominent positions. Pairs of prominent and non-prominent positions include word-initial vs. non-initial, stressed vs. unstressed, root vs. affix, and prevocalic vs. preconsonantal positions. Among the positional effects involving these pairs, this chapter is mainly concerned with those involving the two members of intervocalic consonant clusters. See Beckman (1998) and Barnes (2006) for recent extensive investigations of other positional effects; see also chapter 104: root–affix asymmetries and chapter 102: category-specific effects. In intervocalic C1C2 clusters, the preconsonantal C1 is more likely to undergo phonological processes such as voicing and place assimilation, in contrast with the prevocalic C2, which is rarely subject to such processes. I will refer to this asymmetric positional effect as the C2 dominance effect. This chapter discusses empirical data patterns which display positional effects. Its focus will be on how to explain the C2 dominance effect. I will begin with a discussion of typical data patterns of the C2 dominance effect and proceed to less common, somewhat exceptional, patterns which are nonetheless crucial in comparing the previous approaches. Specifically, I will concentrate on the comparison between prosody-based approaches (Itô 1986, 1989; Cho 1990; Goldsmith 1990; Rubach 1990; Lombardi 1995, 1999, 2001b; Beckman 1998; Kabak and Idsardi 2007) and cue-based approaches (Steriade 1993, 1995, 1999, 2001, 2009; Flemming 1995; Jun 1995, 2004; Padgett 1995; Boersma 1998; Hume 1999; Côté 2000; Wilson 2001; Blevins 2003; Seo 2003). It will be shown that current evidence is mixed. Much of the commonly observed data, to be discussed in the following two sections, can be equally well accommodated in the two approaches. However, there exist less common patterns which can be understood under only one of the two The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0046

Jongho Jun

2

approaches. Evidence exclusively supporting the cue-based approach will be discussed in §4, and evidence for the prosody-based approach in §5.

2

C2 dominance effect

Assimilation occurs in consonant clusters when one of two neighboring consonants takes on some property of the other. I will call the former (i.e. the undergoer of assimilation) the target, and the latter (i.e. the source of the assimilating property) the trigger. With respect to the assimilation in C1C2 clusters, it is crosslinguistically true that C1 and C2 are the target and the trigger, respectively, and thus the direction of assimilation is regressive. To illustrate this C2 dominance effect, I will first consider patterns of voicing assimilation and then patterns of place assimilation. Finally, patterns of consonant deletion will be discussed. As can be seen in (1), in Catalan, Polish and Russian, voiced and voiceless obstruents are separate phonemes, and they may occur unhindered in prevocalic position. But in clusters composed of obstruents, the first constituent of the cluster must agree in voicing with the following constituent. As shown in (1.i), underlyingly voiced obstruents in C1 become voiceless before a voiceless obstruent in C2 whereas underlyingly voiceless obstruents in C1 become voiced before a voiced obstruent in C2, as in (1.ii). Thus, voicing assimilation occurs in clusters, targeting C1. This C2 dominance effect in voicing assimilation can be seen in other languages, including Dutch, Yiddish, Sanskrit, Romanian, Serbo-Croatian, Ukrainian, Hungarian, Egyptian Arabic, and Lithuanian. Steriade (1999) and Beckman (1998) provide in-depth discussion of voicing assimilation, i.e. a type of laryngeal neutralization, in these languages, emphasizing that it is normally regressive, and thus its C2 dominance effect is quite robust. (1)

Regressive voicing assimilation a.

Catalan (from Beckman 1998, citing Hualde 1992) __ voiceless i. /b/ ’OoßH ‘wolf (fem)’ ’OoppH’tit /g/ H’m/:H ‘friend (fem)’ H’m/kpH’t/t __ voiced ii. /t/ ’gatH ‘cat (fem)’ ’gaddu’len /k/ ’pDkH ‘little (fem)’ ’pDg’du

‘small wolf’ ‘little friend’ ‘bad cat’ ‘a little hard’

b.

Polish (from Rubach 1996, Beckman 1998) __ voiceless ‘small frog’ i. /b/ za[b]a ‘frog’ za[pk]a ‘vodka’ /d/ wo[d]a ‘water’ wo[tk]a __ voiced ii. /Œ/ li[Œ]yo ‘count’ li[dzb]a ‘numeral’ /k/ szla[k]-u ‘route (gen sg)’ szla[gb]ojowy ‘war route’

c.

Russian (from Kenstowicz 1994, Kiparsky 1985, Padgett 2002) __ voiceless ‘bastbax (dim)’ i. /b/ korob-a ‘bastbax (gen sg)’ kirop-ka /d/ pod-nesti ‘to bring (to)’ pot-pisat j ‘to sign’

Positional Effects in Consonant Clusters

ii. /s’/ pros’-it’ /t/ ot-jexat j

‘to ask’ ‘to ride off’

__ voiced proz’-ba od-brosit j

3

‘request’ ‘to throw aside’

Consonant place assimilation occurs in clusters when one of two adjacent consonants, i.e. the target, takes on the place of articulation of the other, i.e. the trigger. Consonants differ in the likelihood of being targeted in place assimilation depending on the manner and place of articulation. Nasals (as opposed to stops and continuants) and coronals (as opposed to labials and velars) are the most likely targets of place assimilation (Mohanan 1993; Jun 1995, 2004). For instance, as shown in (2.i), in Diola Fogny only nasals can be targeted in place assimilation, and in Yakut only coronals can be targeted. However, assimilation applies only when such potential target consonants occupy C1, not C2, position. Notice in (2a.ii) that nasals such as /m/ in C2 only trigger, not undergo, assimilation. In (2b.ii), coronals in C2 resist place assimilation. (2)

Regressive place assimilation a.

b.

Diola Fogny (from Sapir 1965) i. /ni+gam+gam/ [nigaIgam] ii. /na+mi:n+mi(n/ [nami(mmi(n] Yakut (from Krueger 1962) i. Coronals in C1 are the target /at+ka/ [akka] /yn+kyr/ [yIkyr] ii. Coronals in C2 are not the target /tobuk+ta/ [tobukta] /silim+te/ [silimne]

‘I judge’ ‘he cut (with a knife)’

‘to a horse’ ‘sloping, aslant’ ‘knee (part)’ ‘glue (part)’

As shown above, place assimilation is typically regressive, just as voicing assimilation is; and thus it targets C1, and its trigger is C2. This C2 dominance effect in place assimilation is very robust across languages, as shown in typological studies by Webb (1982), Mohanan (1993), and Jun (1995, 2004), and there are only a small number of exceptions. (See Jun 1995, 2004 and McCarthy 2008 for discussion of such exceptional progressive assimilation patterns.) Optional patterns of place assimilation are not different from obligatory and categorical patterns in the preponderance of regressive assimilation over progressive assimilation. In Korean, the occurrence of place assimilation is subject to speech style and rate, and thus optional (Kim-Renaud 1974; Jun 1996; Son et al. 2007; Kochetov and Pouplier 2008; Son 2008). This optional assimilation is regressive. Specifically, only coronal and labial stops and nasals in C1 are the target, and non-coronals in C2 are the trigger, as can be seen in (3a.i). Notice in (3a.ii) that coronals in C2 do not undergo the assimilation, and non-coronals in C1 cannot trigger the assimilation (see chapter 12: coronals). English casual speech assimilation is very much like Korean assimilation in the direction of assimilation and segmental characteristics of the target. The only difference is that only coronals, not labials, can be the target in English assimilation. Notice that coronals undergo assimilation only when they are in C1, not C2, as can be seen in (3b).

Jongho Jun

4 (3)

Casual speech place assimilation a.

b.

Korean1 i. /mit+ko/ /ip+ko/ /cinan pam/ ii. /ik+ta/ /ip+ta/ /paI+pota/

[mitk’o] ~ [mikk’o] [ipk’o] ~ [ikk’o] [cinanbam] ~ [cinambam] [ikt’a], *[itt’a], *[ikk’a] [ipt’a], *[itt’a], *[ipp’a] [paIbota], *[pambota], *[paIgota]

‘believe (conn)’ ‘wear (inf)’ ‘last night’ ‘ripe (inf)’ ‘wear (inf)’ ‘(more) than room’

English (based on Bailey 1970) i. right poor righ[p] poor good-bye goo[b]bye ii. keep track *kee[t] track *keep [p]rack back track *ba[t] track *back [k]rack

This C2 dominance effect in casual speech assimilation can also be seen in other languages, such as German (Kohler 1990, 1991a, 1991b, 1992), Malay, Thai (Lodge 1986, 1992), Toba Batak (Hayes 1986), Spanish (Harris 1969), and Ponapean (Rehg and Sohl 1981); see also chapter 79: reduction on assimilation as a casual speech phenomenon. The C2 dominance effect is not limited to assimilation in consonant clusters, but extends to consonant deletion in clusters. Consonant deletion occurs in clusters when one of two adjacent consonants, i.e. the target, deletes. It has been observed and emphasized in the literature (Côté 2000; Wilson 2001; Jun 2002; McCarthy 2008) that C1, as opposed to C2, is always the target in such deletions. For instance, as shown in (4), stops in C1, not C2, delete in Diola Fogny, West Greenlandic, and Basque. (4)

Consonant deletion: C1 is the target a.

1

Diola Fogny (Sapir 1965) /let+ku+jaw/ [le kujaw] /kuteb sinaIas/ [kute sinaIas] [eke bo] /eket bo/

b.

West Greenlandic /qanik+lerpoq/ /ukijuq+tuqaq/ /anguti+kulak/

c.

Basque (Hualde 1987) /bat+paratu/ [ba-paratu] /bat+kurri/ [ba-kurri] /guk+piztu/ [gu-piztu] /guk+kendu/ [gu-kendu]

‘they won’t go’ ‘they carried the food’ ‘death there’

(Rischel 1974; Fortescue 1980) [qani lerpoq] ‘begins to approach’ [ukiju tuqaq] ‘old year’ [angu kulak] ‘he goat’ ‘put one’ ‘run one’ ‘we light’ ‘we take away’

In Korean, which has a three-way laryngeal contrast among obstruents, i.e. lenis, aspirated, and tense, lenis obstruents become tense after an obstruent (Post-obstruent tensing), and voiced between sonorants (Inter-sonorant voicing). See Kim-Renaud (1986) and Ahn (1998) for details of these automatic processes.

Positional Effects in Consonant Clusters

5

In addition to those shown above, consonant deletion with a C1 target can be found in Akan (Lombardi 2001b, citing Schachter and Fromkin 1968), Axininca person prefixes (Lombardi 2001b, citing Payne 1981), Carib (Gildea 1995), and Tunica (Wilson 2001, citing Haas 1946). Consequently, as summarized below, the cross-linguistic generalization which is common to the three phonological processes (voicing assimilation, place assimilation, and consonant deletion) is that C1 in intervocalic C1C2 clusters is the target, whereas C2 is the trigger. (5)

The C2 dominance effect in voicing assimilation, place assimilation, and consonant deletion In intervocalic C1C2 clusters, C1 is a typical target, and C2 is a typical trigger.

Let us consider how to capture this C2 dominance effect. It is not difficult to derive patterns with the C2 dominance effect within the framework of previous theories such as classical generative theory, autosegmental phonology, and underspecification theory. For instance, regressive assimilation can easily be characterized by a rule of the type shown in (6a). However, progressive assimilation can also be formulated with equal complexity, as shown in (6b), and its absence, or at best rarity, in the typology would be a surprise. Representational theories such as autosegmental phonology, feature geometry, and underspecification theory would be no better in this respect than classical generative theory, as there is no plausible reason to differentiate in the complexity of the representation between the two members of a consonant cluster. (See Jun 1995 for the relevant discussion.) (6)

Rules for consonant place assimilation a. b.

C → [aplace] / __ [C, aplace] C → [aplace] / [C, aplace] __

(regressive assimilation) (progressive assimilation)

Compared to the rule-based theories with a focus on the correct formulation of the language-specific phonological processes, Optimality Theory (McCarthy and Prince 1995; Prince and Smolensky 2004) is more concerned with universal patterns, thus being in a better position to explain positional effects such as the C2 dominance effect and understand the motivation behind processes showing the effect. Along with the development of Optimality Theory, there have been two major lines of approach to the analysis of the positional effects, Licensing-by-cue and Licensing-by-prosody (in Steriade’s 1999 terminology). The prosody-based approach explains positional asymmetries by reference to prosodic structure. It attributes the C2 dominance effect to the coda–onset asymmetry since C1 and C2 are usually syllabified as a coda and an onset, respectively. The C1 in the coda is likely to be targeted in the processes since the coda is phonologically non-prominent and marked. In contrast, the C2 in the onset resists these processes since the onset is phonologically prominent and unmarked. For an analysis of the data of the C2 dominance effect, either greater faithfulness to the onset and/or dispreference for the coda or (marked) properties in the coda have been called on. Specifically, within the framework of Optimality Theory, positional faithfulness constraints for the onset or/and positional markedness constraints for the coda have been adopted in the literature. (See Casali 1996, 1997 and Beckman 1998 for positional

Jongho Jun

6

faithfulness analyses, and Zoll 1998 for the comparison between positional faithfulness and positional markedness.) In contrast, the cue-based approach (Flemming 1995; Steriade 1995, 1999, 2001, 2009; Boersma 1998; Côté 2000; Wilson 2001; Blevins 2003; Jun 2004) explains the C2 dominance effect by relying on the perceptual factors involved. The C1 has low perceptibility since it is preconsonantal and thus may lack important perception cues, such as release bursts and C-to-V formant transitions, to laryngeal/place features and segmenthood under overlap with C2 (Lamontagne 1993; Wright 1996). In contrast, the C2 is perceptually prominent since it is prevocalic, being able to maintain such perception cues. (See Wright 2004 for a detailed discussion of perception cues.) From the assumption that change in perceptually prominent positions would cause drastic input–output difference, and thus be greatly dispreferred, whereas the comparable change in non-prominent positions would cause less difference, and thus be less dispreferred, it is derived that non-prominent C1 is more likely to be modified, i.e. targeted in phonological processes, than prominent C2. Thus, the cue-based approach attributes the C2 dominance effect to higher perceptibility of C2 over C1. The two approaches under consideration differ in whether the constraints (and rules) adopted to explain positional effects should be expressed as prosody-based or string-based (more precisely, cue-based) statements. However, the empirical data presented thus far will not distinguish the two approaches, since the preconsonantal C1 which is expected to be the target of the phonological processes in the cue-based approach is normally syllabified as a coda, which the prosody-based approach also expects to be the more likely target. In §4 and §5, I will present the data patterns for which the two approaches make distinct predictions.

3

Non-assimilatory neutralization

Assimilation can be considered as a case of contrast neutralization. As shown in the previous section, assimilation is primarily regressive (i.e. C1 is the target), and thus potential contrasts of the assimilating feature are neutralized in C1 position. For instance, in regressive voicing assimilation, consonants in C1 with distinct voice feature values in their underlying form would have identical phonetic realization with respect to voicing, i.e. they are voiced before a voiced segment and voiceless before a voiceless one. Non-assimilatory neutralization of voicing, as well as other laryngeal features and place of articulation features, targets the C1 and word-final position. In languages which have voicing assimilation in consonant clusters, the word-final position is the only available target of non-assimilatory neutralization, and in fact languages with voicing assimilation mentioned in the previous section show final devoicing, as in (7) (see also chapter 69: final devoicing and final laryngeal neutralization). (7) Final devoicing in languages with voicing assimilation a.

Catalan (Beckman 1998, citing Hualde 1992) __V __# i. /t/ [’gatH] ‘cat (fem)’ [’gat] /k/ [’pDkH] ‘little (fem)’ [’pDk]

‘bad (masc)’ ‘little (masc)’

Positional Effects in Consonant Clusters ii. b.

c.

/b/ /g/

[’OoßH] ‘wolf (fem)’ [H’m/:H] ‘friend (fem)’

Dutch (Kager 1999) __V i. /t/ [petHn] [betHn] ii. /d/ [bedHn] [hudHn]

‘caps’ ‘(we) dab’ ‘beds’ ‘hats’

Polish (Rubach 1996) __V /d/ sa[d]-y ‘orchards’ /z/ ko[z]-a ‘goat’ /v/ pra[v]-o ‘law’

[’Oop] [’Hm/k]

‘wolf (masc)’ ‘friend (masc)’

__# [pet] [bet] [bet] [hut]

‘cap’ ‘(I) dab’ ‘bed’ ‘hat’

__# sa[t] kó[s] pra[f]

‘orchards (nom sg)’ ‘goat (gen pl)’ ‘law (gen pl)’

7

d. Russian (Hayes 1984; Kiparsky 1985; Padgett 2002) __V __# /b/ klub-a ‘club (gen sg)’ klup ‘club’ /d/ sa[d]-a ‘garden (gen sg)’ sa[t] ‘garden (nom sg)’ /g/ knig-a ‘book (nom sg)’ knik ‘book (gen pl)’ Notice that voiced obstruents may appear before a vowel, more precisely a sonorant, but only neutralized, normally voiceless, obstruents are allowed to occur word-finally in these languages.2 In many other languages, such as Korean, Maidu, Greek, German, Thai, and Sanskrit, the non-assimilatory neutralization of the laryngeal features such as voicing, aspiration, and glottalization occurs not only at the end of the word, but also in preconsonantal C1 position. As mentioned above, Korean has a three-way laryngeal contrast between lenis, aspirated, and tense (i.e. glottalized) obstruents. As shown in (8a), the three-way contrast can be maintained before a vowel. However, Korean obstruents are neutralized to their homorganic lenis stop counterparts in word-final and preconsonantal C1 position (see chapter 111: laryngeal contrast in korean). As discussed in Lombardi (1995), Maidu also has a three-way contrast between voiceless, implosive, and glottalized obstruents. Just as in Korean, the three-way contrast in Maidu is maintained only syllable-initially, i.e. prevocalically, whereas laryngeally marked consonants, i.e. implosive and glottalized, occur neither in C1 nor at the end of the word. Some examples of alternations displaying neutralization of glottalized stops are shown in (8b). As discussed in Steriade (1999), Ancient Greek has voiceless unaspirated, voiced, and voiceless aspirated stops. The laryngeal distinction among stops can be made before vowels (and sonorant consonants) whereas it is neutralized before obstruents. Some examples of relevant alternations are shown in (8c). Notice that stops in Greek are not allowed to occur word-finally, and thus there is no way to observe active word-final neutralization.

2

See Steriade (1999) for a discussion of the phonetic realization of the neutralized obstruents and Blevins (2006) for an in-depth discussion of final devoicing and the related survey.

Jongho Jun

8 (8)

Laryngeal neutralization: target = word-final and preconsonantal C1 a.

Korean underlying form

i. ii. iii. b.

c.

lenis /kuk/ aspirated /puHkh/ glottalized /pak’/

‘soup’ ‘kitchen’ ‘outside’

Maidu (Lombardi 1995) __V /pit’/ pit’i k’atanoky ‘manure-rolling beetle’ /jep’/ jep’im symi ‘male’ ‘buck deer’

__V (-i (nom)) kugi puHkh i pak’i

__# isolation form kuk puHk pak

__C (-to ‘too’) kukt’o puHkt’o pakt’o

__# __C pit pitk’ololo ‘defecate, feces’ ‘intestines’ jepsy ‘men’

Greek (Steriade 1999) __V i. voiceless t horak-os ‘thorax (gen sg)’ ii. voiced laryng-os ‘larynx (gen sg)’ iii. aspirated trikh-os ‘hair (gen sg)’

__C t horak-si ‘thorax (dat pl)’ larynk-si ‘larynx (dat pl)’ t h rik-si ‘hair (dat pl)’

In summary, the preconsonantal C1 position is the cross-linguistically common target position of both assimilatory and non-assimilatory laryngeal neutralizations, and the word-final position is additionally the common target position of the non-assimilatory neutralization. Let us now consider non-assimilatory place neutralization. Just like laryngeal neutralization, place neutralization targets C1 and the word-final position. In languages like Spanish, Ancient Greek, and Japanese, certain place distinctions are removed through regressive assimilation in clusters, and in a non-assimilatory way at the end of the word. As mentioned in Steriade (2001), in Ancient Greek [n] and [m] in C1 assimilate in place to the following consonant, and only [n] is allowed to occur at the end of the word. Similar neutralizations can be seen in Spanish and Japanese. In Spanish, there are three nasal phonemes, bilabial, alveolar, and palatal, which are contrastive only in prevocalic position: ca[m]a, ca[n]a, and ca[J]a (Harris 1984). In intervocalic C1C2 clusters, place distinctions of nasals in C1 are neutralized since they always agree in place with the following consonant in C2, as shown in (9a.i). Moreover, only a single nasal may occur at the end of the word, although dialects differ in the exact place of articulation of the default nasal, i.e. alveolar in standard varieties and velar in non-standard varieties, as shown in (9a.ii, iii). In Japanese, in intervocalic C1C2 clusters, consonants in C1 must agree in place with those in C2, as shown in (9b.i, ii). Thus, within a word, only homorganic nasal and geminate clusters can occur. In addition, the word-final position can be occupied only by a single nasal, which is called the mora nasal and is usually transcribed as [N] or [I]. This nasal is produced with no fixed oral constriction (Vance 1987: 35), and thus it is sometimes argued to be a placeless segment (for instance, McCarthy 2008: 278). (See chapter 22: consonantal place of articulation.)

Positional Effects in Consonant Clusters (9)

9

Place assimilation and final place neutralization a.

Spanish (from Harris 1984) i. Homorganic NC clusters: ca[mp]o, ma[ns]o, ma[Ik]o, á[,f]ora, ma[|t]o ii. Final [n] in standard varieties: e[n] Chile, ta[n] frío, u[n] elefante iii. Final [I] in non-standard varieties: e[I] Chile, ta[I] frío, u[I] elefante

b.

Japanese (Vance 1987; Yip 1991; Itô et al. 1995; Kager 1999) i. Geminates kap.pa ‘a legendary being’ kit.te ‘stamp’ gak.koo ‘school’ ii. Homorganic nasal + obstruent tom.bo ‘dragonfly’ non.do ‘tranquil’ kaI.gae ‘thought’ /jom + te/ → jonde ‘reading’ /œin + te/ → œinde ‘dying’ iii. Final mora nasal hoN ‘book’ zeN ‘goodness’

There are also languages in which place neutralization occurs only in a nonassimilatory fashion. As discussed in Lombardi (2001b, citing Rice 1989), nonsonorant consonants in Slave (Athabaskan) are realized as /h/ syllable-finally, as shown in (10a). Sonorants are like obstruents in having no place distinctions, although the exact final neutralization patterns are not the same. Syllable-final nasals delete, nasalizing the preceding vowel, and /j/ is the only possible coda among non-nasal sonorants. Another example of non-assimilatory place neutralization is from the Kelantan dialect of Malay (Teoh 1988). Final stops /k t p/ are realized as [?], and final fricatives like /s/ as [h], as shown in (10b). (10)

Final place neutralization (from Lombardi 2001b) a.

b.

Slave i. /ts’ad/ ii. /xaz/ iii. /see:/ /tl’ux/

ts’ah xah seeh tl’uh

Kelantan Malay i. /ikat/ ii. /sHsak/ iii. /dakap/ iv. /tapis/

ika? ‘tie’ sHsa? ‘crowded’ daka? ‘embrace’ tap/h ‘to filter’

-ts’ade ‘hat’ -:aze ‘scar’ -zee:e ‘saliva’ -tl’uxe ‘rope’

In these languages, contrasts of some features other than place can be maintained finally, for instance obstruents vs. sonorants in Slave and stops vs. fricatives in Kelantan Malay. But there are also many languages like Burmese in which all

10

Jongho Jun

consonants are neutralized to [?]. (See Lombardi 2001b and references therein for more details.) In summary, as stated below, the preconsonantal C1 is not only the typical target of assimilation, but also the typical target of the non-assimilatory neutralization. The word-final position is an additional typical target of the neutralization. (11)

The C2 dominance effect in (assimilatory and non-assimilatory) laryngeal and place neutralization Preconsonantal C1 and word-final positions are common target positions.

It is usually the case that preconsonantal C1 and word-final positions form a natural class, i.e. coda. Thus it is obvious that the prosody-based approach can provide a unified account of positional neutralizations of C1 and word-final positions, attributing both cases to the coda–onset asymmetry. The cue-based approach also provides a somewhat unified account for the two common target positions based on the fact that the word-final position lacks C-to-V transition cues, just as preconsonantal C1 does, and thus it has lower perceptibility compared to the prevocalic C2 position. Consequently, the relatively common positional neutralization patterns presented thus far do not significantly distinguish the two approaches. In the remainder of this chapter, I will consider the patterns which crucially distinguish them.

4 4.1

Evidence for the cue-based approach Neutralization sites ≠ syllable positions

According to the prosody-based approach, neutralization contexts should be described in prosodic terms: for instance, “codas are the target of laryngeal neutralization.” But, as discussed by Steriade (1999), there are cases in which there is no consistent connection between neutralization sites and syllable structure. First, there are languages in which neutralization targets only C1, not word-final, positions. Languages in which laryngeal neutralization occurs only in C1, not at the end of the word, include Yiddish, Romanian, Serbo-Croatian (Lombardi 2001b: 269), French, Hungarian, and Kolami (Steriade 1999). In addition, place neutralization of nasals occurs only in C1, not at the end of the word, in Diola Fogny (Sapir 1965) and the Souletin dialect of Basque (Hualde 1993). Under the prosody-based approach, it is not clear why word-medial and final codas behave differently, and even less clear why medial codas are more likely to be targeted in the neutralization than word-final codas (see chapter 36: final consonants for more discussion). In contrast, in the cue-based approach, the asymmetry between preconsonantal C1 and final positions may be derived from their relative perceptibility difference. C1 may be considered less perceptible than word-final position, because stops in C1, which overlap with consonants in C2, are more likely to be unreleased, thus lacking the release burst and closure duration cues, than those in word-final position.3 3

Blevins (2006: 143) discusses data from Dhaasana, Chadic Araic, and Maltese in which devoicing occurs exclusively at the end of the word, not in C1 position. This word-final, but not syllable-final, devoicing can be a problem not only for the prosody-based approach but also for the cue-based approach.

Positional Effects in Consonant Clusters

11

Second, laryngeal neutralization targets C1 only before obstruents, not before sonorants, regardless of its syllabic assignment in many languages, including Lithuanian, German, Russian, Greek, Sanskrit, Polish, Hungarian, and Kolami. Notice that it is possible that C1 in intervocalic C1C2 clusters is syllabified as a coda when sonorant consonants occupy C2, and its voicing contrast is then expected to be neutralized in the prosody-based approach. Also, it is usually the case that an obstruent is syllabified as an onset when it occurs as the first constituent of the word-initial clusters composed of obstruents, and then its voicing contrast is expected to be licensed in the prosody-based approach. These two expectations of the prosody-based approach are not satisfied in the languages mentioned above. For instance, in Lithuanian, where consonant clusters are heterosyllabic regardless of composition (e.g. /’auk.le/), voicing of obstruents may be contrastive in the coda when they occur before sonorants, as in (12c). The voicing of an obstruent is neutralized in the onset when it precedes another obstruent, as in (12b). Consequently, it is difficult to provide an adequate description of neutralization contexts in syllabic terms. (12)

Lithuanian obstruents in clusters (Steriade 1999) a. b. c. d.

licensed onsets neutralized onsets licensed codas neutralized codas

voiceless sam’gus ‘cheerful’ spalva ‘color’ aug.muo ‘growth’ daj[k] ‘much’

voiced Nmo’gus ‘man’ ’lizdas ‘nest’ ak.muõ ‘stone’

In contrast, to explain the difference in the likelihood of the neutralization between pre-obstruent and pre-sonorant positions, the cue-based approach may still rely on the perceptibility difference of the two positions. Specifically, the preobstruent position lacks the main contextual cues (VOT and other release-related cues), and thus is less perceptible than the pre-sonorant position, where the main cues can be maintained. Finally, there are languages in which neutralization patterns are fixed despite the variable syllabification. As discussed by Steriade (1999), for both Sanskrit and Ancient Greek syllable divisions in obstruent–sonorant clusters were variable, depending on “the dialect, the period, the literary style and the juncture separating the consonants.” In contrast, there was no variation in the pattern of laryngeal neutralization: in styles or dialects where VC1.C2V divisions were the norm for all clusters, laryngeal neutralization did not take place before heterosyllabic sonorants. This indicates that laryngeal features in these languages are neutralized irrespective of the syllabic affiliation of clusters, and thus the neutralization patterns cannot be adequately described in syllable terms. The above patterns indicate that syllable positions like codas are neither a sufficient nor a necessary condition for the occurrence of neutralization. Codas are not a sufficient condition in the patterns in which only word-medial, as opposed to final, codas and pre-obstruent, as opposed to pre-sonorant, codas are neutralized. Codas are not a necessary condition in the patterns in which an obstruent onset in word-initial clusters is neutralized. Codas would be totally useless in describing the Sanskrit and Greek patterns with variable syllabification but fixed neutralization patterns. Consequently, all these patterns can be taken as evidence against the prosody-based approach and in favor of the cue-based approach.

Jongho Jun

12

4.2

Apical neutralization

As mentioned above, consonant place assimilation is predominantly regressive across languages. This strong cross-linguistic tendency toward regressive assimilation has been considered a subset case of the C2 dominance effect in place neutralization, as stated in (11) and repeated below as (13). (13)

The C2 dominance effect in (assimilatory and non-assimilatory) laryngeal and place neutralization Preconsonantal C1 and word-final positions are common target positions.

The place neutralization typology displaying this C2 dominance effect provides an important basis for the prosody-based approach’s analysis of the coda–onset asymmetry. However, Steriade (2001) notes that this C2 dominance effect in place neutralization is true only when contrasts between labials, alveolars, velars, and palato-alveolars (referred to by Steriade as major C-Place contrasts) are neutralized. In the case of the neutralization of contrasts between apico-alveolars and retroflexes (referred to by Steriade as apical contrasts), a completely opposite tendency is observed: postconsonantal C2 and word-initial positions are typically targeted. Let us consider first assimilatory neutralization of apical contrasts and then nonassimilatory neutralization. First, apical assimilation is typically progressive, as can be seen in (14). Notice that in both Sanskrit and Urali, postconsonantal C2 alveolars in the underlying form are realized as retroflexes after post-vocalic C1 retroflexes. Thus, apical assimilation targets C2, not C1, which is the opposite of major C-Place assimilation. (14) (Patterns of word-internal) apical assimilation (Steriade 2001) a.

b.

Sanskrit /av-iÕ-dh i/ /ÏaK-na(m/ /jyotiÏ-su/

[aviÕÕh i] [ÏaKKa(m] [jyotiÏÏu]

Urali /eK-nuuru/ /keÕ-t-a-/

[eKKuuru] ‘hundred-8 = 800’ [keÍÍa-] ‘spoil-intrans’

‘favor’ ‘of six’ ‘in planets’

This C1 dominance effect in apical assimilation seems as robust as the C2 dominance effect in major C-Place assimilation, although the total number of cases of apical assimilation is relatively small. Based on a typological survey of apical assimilation, Steriade (2001) reached the conclusion that apical assimilation in clusters is predominantly progressive, and there is no exception to it when clusters belong to the same word and the two constituents of the clusters are identical in stricture (e.g. both stops). Non-assimilatory neutralization of the apical contrast also targets the postconsonantal C2, possibly along with word-initial positions. For instance, in Muri|ba}a, alveolars and retroflexes contrast in both C1 and word-final position. In contrast, the apical contrast is neutralized in postconsonantal C2. Apicals in C2 are always realized as alveolars after non-apicals, and as homorganic with an apical C1. Thus, the non-assimilatory neutralization of the apical contrast targets

Positional Effects in Consonant Clusters

13

C2 in Muri|ba}a. Miriwung is just like Muri|ba}a, in that the apical contrast is maintained in C1, but neutralized in C2. Apical neutralization additionally occurs at the beginning of the word in which only alveolars, not retroflexes, are allowed to occur. Consequently, the typical targets of apical neutralization may be summarized as below: (15)

The C1 dominance effect in (assimilatory and non-assimilatory) apical neutralization Postconsonantal C2 and word-initial positions are common target positions.

This is therefore the complete opposite of the C2 dominance effect in the neutralization patterns of laryngeality and major C-Places of articulation. Given that C2 is usually an onset, the prosody-based approach cannot explain the C1 dominance effect in apical neutralization in the same way as the C2 dominance effect summarized in (13). In contrast, in the cue-based approach, the C1 dominance effect may be derived naturally from the perception fact that cues to the apical distinction lie primarily in the V-to-C, not C-to-V, transitions, and thus C1 is more prominent in the perception of the apical contrast than C2. As discussed by Steriade (2001), citing Ladefoged and Maddieson (1986), Dave (1976), Stevens and Blumstein (1975), and Bhat (1973), the formant transitions into retroflexes in C1 show distinctively low F3 and F4 values, compared to those of denti-alveolars, whereas the transitions out of retroflexes in C2 are not distinct from those of dentialveolars. This acoustic asymmetry originates from the characteristic articulation of retroflexes, in which the tongue tip moves forward during the closure and releases from the same constriction location as apico-alveolars. Consequently, both the C2 dominance effect in major C-Place assimilation and the C1 dominance effect in apical assimilation may be derived from the main argument of the cue-based approach, i.e. the neutralization targets positions which lack prominent perceptual cues to the contrasts in question. Thus, the apical neutralization typology may form very strong evidence for the cue-based approach by showing a case of contrast-specific neutralization. (See Zhang’s 2004 discussion of contour tone typology for an additional case of contrast-specific licensing/neutralization.)

5 5.1

Evidence for the prosody-based approach Obstruent–sonorant clusters in Catalan

The cue-based approach provides a string-based, not prosody-based, account for positional neutralizations. If two sequences are segmentally identical, and thus not significantly different in the perceptibility involved, the cue-based approach expects the two to behave similarly with respect to neutralization even when they have different prosodic structures. Suppose that in obstruent–sonorant C1C2 clusters, the C1 obstruent may be syllabified either as an onset or as a coda, depending on the environment. The cue-based approach expects that the C1 obstruent will behave invariably with respect to positional neutralization, regardless of whether it is an onset or a coda. If, as predicted by the prosody-based approach, the C1 is licensed when syllabified as an onset, but neutralized when syllabified

Jongho Jun

14

as a coda, it will raise a serious problem for the cue-based approach. Wheeler (2005) shows that such a pattern exists in Catalan. As shown in (16a, b), in Catalan, stops other than dentals are contrastive in voicing before liquids as well as glides, while sibilants are contrastive only before glides when the obstruent–sonorant sequences are within a word. However, sibilants are always voiced before liquids, as in (17a), and dentals are always voiced before laterals, as in (17b). There are no word-internal sibilant–liquid and dental–lateral sequences with distinct voicing values. In summary, in Catalan, although the voicing contrast of the C1 obstruents in obstruent–sonorant sequences is in general licensed, sibilants and dentals are neutralized in voicing before liquids and laterals, respectively. In the cue-based approach, it is difficult to explain why some obstruent–sonorant sequences behave differently in voicing neutralization from other obstruent–sonorant sequences. In contrast, Wheeler (2005) argues that all obstruents with contrastive voicing in (16) occur in the onset whereas all with neutralized voicing in (17) occur in the coda.4 Thus, the prosody-based approach can easily describe the voicing neutralization of obstruents in obstruent–sonorant sequences in Catalan. (16)

Contrastive obstruent voicing in Catalan obstruent–sonorant clusters a.

non-sibilant / __ liquid, glide voiceless p vs. b prou [’p7Dw] ‘enough’

voiced brou

[’b7Dw] ∼ [’ß7Dw] sempre [’sem.p7H] ‘always’ sembra [’sem.b7H] t vs. d truita [’t7uj.tH] ‘trout’ druida [’d7uj.ÏH] ∼ [’Ï7uj.ÏH] k vs. g creu [’k7ew] ‘cross’ greu [’g7ew] ∼ [’:7ew] classe [’klasH] ‘class’ glaça [’glasH] ∼ [’:lasH] qualla [’kwaOH] ‘congeals’ guatlla [’gwaO.OH] ∼ [’:waO.OH] b.

sibilant / __ glide voiceless gràcia [’g7a.sjH] ‘humor’

‘broth’ ‘sows’ ‘druid’ ‘serious’ ‘freezes’ ‘quail’

voiced afàsia [H’fa.zjH] ‘aphasia’

(17) Neutralized obstruent voicing in Catalan obstruent–sonorant clusters a.

4

sibilant / __ liquid voiced legislar [z.l] ‘to legislate’ Israel [z.r] ~ [r] ‘Israel’

voiceless none

Although Wheeler does not provide details of syllabification in Catalan, Blevins (2003: 399), who also argues that voicing neutralization in Catalan occurs syllable-finally, suggests that Catalan syllabification is quite predictable by stating that “Catalan syllabification judgments were entirely consistent across speakers.”

Positional Effects in Consonant Clusters b.

dental / __ lateral voiced atleta [d.l] ‘athlete’

15

voiceless none

Further, comparable active neutralization, which is even more difficult to explain within the cue-based approach, can be observed in the obstruent–sonorant sequences occurring across word boundaries. In Catalan, word-final obstruents assimilate in voicing to the following consonants including liquids. Word-final stops, which are voiceless before an initial vowel of the following word (18.ii), become voiced before an initial sonorant (18.i). Thus the C1 obstruents here are neutralized with respect to voicing. This final obstruent neutralization would not be expected within the cue-based approach, since the same obstruent-sonorant sequences within a word are not subject to the voicing neutralization, as shown in (16a). So this is a case in which identical sequences behave differently with respect to neutralization, depending on where they occur. According to Wheeler (2005), the only difference between the obstruent–sonorant sequences of (16a) and (18.i) is in syllabic affiliation: the C1 obstruents in (16a) are onsets, whereas those in (18.i) are codas. Thus, only the prosody-based, not cue-based, approaches can explain the voicing patterns of obstruents in Catalan. (18)

5.2

Neutralized obstruent voicing in Catalan obstruent sonorant clusters a.

non-sibilant / __ #liquid i. poc logic /’pDk#H’lDÚik/ [’pDg.’lDÚik] ‘not very logical’ ii. poc amable [’pD.kH’mab.blH] ‘not very friendly’

b.

non-sibilant / __ #glide i. poc whisky /’pDk#’wiski/ ii. poc usual

[’pDg.’wiski] [’pD.ku’zwal]

‘not much whisky’ ‘not very usual’

Obstruent–sonorant clusters in Eastern Andalusian Spanish

Eastern Andalusian Spanish shows an additional case in which sequences with similar perception cues behave differently in contrast distribution and neutralizing processes, thus posing a problem to the cue-based approach. The discussion of this section is mostly based on Gerfen (2001). As shown in (19), in Standard Peninsular Spanish, /s/ is allowed in preconsonantal and word-final positions. In contrast, in Eastern Andalusian Spanish, /s/ is not allowed to occur in those positions. As shown in (20a), word-final /s/ deletes and aspirates the preceding vowel. Preconsonantal C1 /s/ also deletes but the deletion is accompanied by the gemination of the following C2 consonant. (19)

/s/ in Standard Peninsular Spanish (Gerfen 2001) a. b.

word-final coda: pre-C coda:

[ga.fas] [gas.ko]

‘eyeglasses’ ‘helmet’

Jongho Jun

16 (20)

/s/ in Eastern Andalusian Spanish (Gerfen 2001) a.

word-final coda /ganas/ [ga.nah] ‘desire’

b.

pre-C coda SPS i. [bos.ke] ii. [es.la.ßo]

EAS [bohk.ke] ‘forest’ [ehl.la.ßo] ‘Slavic’

The prosody-based approach can easily attribute the distributional restrictions of /s/ in Eastern Andalusian Spanish to a syllable coda restriction. In contrast, the Licensing-by-cue approach can adopt a string-based statement that /s/ is allowed to occur only before a vowel in Eastern Andalusian Spanish. But, as argued by Gerfen (2001), this is hard to justify, since sibilants such as /s/ have very salient internal cues in the frication noise, and thus C-to-V transition cues would not be important in the perception of the presence or absence of /s/, unlike in the perception of laryngeal and place contrasts. Thus, there is no plausible phonetic motivation for such a string-based requirement, i.e. that a vowel must be present immediately after /s/, within the cue-based approach. Gerfen discusses two more related patterns which are even more problematic for the cue-based approach. In Eastern Andalusian Spanish, the C1 coda deletion (followed by the gemination of C2) is not limited to /s/, but to all obstruents, as shown in (21) (see also chapter 38: the representation of sc clusters). (21)

Aspiration of all obstruent codas in Eastern Andalusian Spanish (Gerfen 2001) SPS [ap.to] [pi..ka] [ak..jon] [ob.tu.so]

EAS [ah t.to] [pih k.ka] [ah ...jon] ([ahs.sjon]) [oh t.tu.so]

‘apt’ ‘pinch, small amount’ ‘action’ ‘obtuse’

Notice that the C1 obstruents in obstruent–liquid sequences, word-initially or medially, are not subject to this deletion, as shown in (22). In the prosody-based approach, this difference can be attributed to the coda–onset asymmetry: stop + liquid clusters are syllabified as onsets, thus resisting coda deletion, whereas obstruent clusters shown in (21) are syllabified as coda–onset sequences, and thus C1 obstruents in the coda are subject to the deletion. In contrast, to explain the above difference between obstruent–obstruent and stop–liquid sequences, the cue-based approach would rely on the asymmetry in perceptibility between pre-obstruent and pre-sonorant positions. Specifically, C2 liquids may have richer perceptual cues to the preceding obstruents than C2 obstruents. (22)

Stop + liquid clusters in Eastern Andalusian Spanish (Gerfen 2001) a.

Initial stop + liquid clusters i. [klaro] ‘clear’ ii. [grado] ‘grade’ iii. [plano] ‘flat’ iv. [trapo] ‘rag’

Positional Effects in Consonant Clusters b.

17

Word-internal obstruent + liquid clusters i. [a.klara] ‘s/he/it clears up’ ii. [a.grada] ‘s/he/it pleases’ iii. [a.plana] ‘s/he/it applies’ iv. [a.trapa] ‘s/he/it traps’

But it is still unclear why /sl/, an obstruent–liquid sequence, does not behave like stop–liquid sequences. As shown in (20b.ii), /s/ is subject to coda deletion before /l/. Also, as shown below, /tl/, a stop–liquid sequence, behaves like /sl/, not like /kl/ and /gl/. If rich perceptual cues to C1 before a liquid can guarantee the surface realization of stop–liquid sequences like /kl/, /gr/, /pl/, and /tr/, it is difficult to understand why /tl/ and /sl/ cannot surface as such. (23)

/tl/ clusters in Eastern Andalusian Spanish (Gerfen 2001) /atleta/

[ah lleta]

‘athlete’

To summarize, both Catalan and Eastern Andalusian Spanish show asymmetric patterns in which only a subset of obstruent–sonorant sequences is subject to the distributional restrictions and related alternations targeting an obstruent in C1. For the analysis of these patterns, the prosody-based approach can still attribute the difference among the obstruent–sonorant sequences to the coda-onset asymmetry, but an equally plausible, string-based, solution seems to be unavailable in the cue-based approach. See Kabak and Idsardi (2007) for an additional support for the prosody-based, as opposed to the cue-based, approach. They investigated Korean listeners’ perception of non-native sequences, and argue that only syllablebased, not string-based, phonotactic constraints can explain their experimental results.

6

Conclusion

From the literature on phonological typology, we know that it is common for phonological processes not to apply in all positions, and more generally for phonological contrasts not to be licensed in all positions. Such positional effects are characterized by reference to certain pairs of prominent and non-prominent positions such as word-initial vs. non-initial, stressed vs. unstressed, root vs. affix, and prevocalic C2 vs. preconsonantal C1 positions. Among these, this chapter has been mainly concerned with the C2 dominance effect in which preconsonantal C1 in intervocalic C1C2 clusters is likely to be targeted for neutralization, whereas prevocalic C2 is likely to trigger or resist such neutralization. This C2 dominance effect is quite robust in laryngeal and place neutralization and consonant deletion. I have looked at the relevant data patterns, ranging from well known and common to less known and relatively exceptional, while comparing the cue-based and prosody-based approaches. Common data patterns can be explained equally well by both approaches. In contrast, less common or somewhat exceptional patterns may distinguish the two approaches. However, the evidence so far is mixed. Not only neutralization patterns in which there is no connection between neutralization sites and syllable positions, but also apical neutralization patterns,

18

Jongho Jun

may form evidence against the prosody-based approach. In contrast, obstruent– sonorant sequences in which the voicing of C1 obstruent may be licensed or neutralized, arguably depending on syllable structure, may form evidence against the cue-based approach. Any adequate theory of positional effects, including the C2 dominance effect, should be able to account for all these patterns, common or exceptional.

REFERENCES Ahn, Sang-Cheol. 1998. An introduction to Korean phonology. Seoul: Hanshin. Aronoff, Mark & Richard T. Oehrle (eds.) 1984. Language sound structure. Cambridge, MA: MIT Press. Bailey, Charles-James N. 1970. Toward specifying constraints on phonological metathesis. Linguistic Inquiry 1. 347 –349. Barnes, Jonathan. 2006. Strength and weakness at the interface: Positional neutralization in phonetics and phonology. Berlin & New York: Mouton de Gruyter. Beckman, Jill N. 1998. Positional faithfulness. Ph.D. dissertation, University of Massachusetts, Amherst. Bhat, D. N. S. 1973. Retroflexion: An areal feature. Working Papers on Language Universals 13. 27 –58. Blevins, Juliette. 2003. The independent nature of phonotactic constraints: An alternative to syllable-based approaches. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 375 –403. Cambridge: Cambridge University Press. Blevins, Juliette. 2006. A theoretical synopsis of Evolutionary Phonology. Theoretical Linguistics 32. 117 –166. Boersma, Paul. 1998. Functional phonology: Formalizing the interactions between articulatory and perceptual drives. The Hague: Holland Academic Graphics. Casali, Roderic F. 1996. Resolving hiatus. Ph.D. dissertation, University of California, Los Angeles. Casali, Roderic F. 1997. Vowel elision in hiatus context: Which vowel goes? Language 73. 493–533. Cho, Young-Mee Yu. 1990. Parameters of consonantal assimilation. Ph.D. dissertation, Stanford University. Côté, Marie-Hélène. 2000. Consonant cluster phonotactics: A perceptual approach. Ph.D. dissertation, MIT. Dave, R. 1976. Retroflex and dental consonants in Gujarati: A palatographic and acoustic study. Annual Report of the Institute of Phonetics, University of Copenhagen (ARIPUC) 11. 27–155. Flemming, Edward. 1995. Auditory representations in phonology. Ph.D. dissertation, University of California, Los Angeles. Fortescue, Michael. 1980. Affix ordering in West Greenlandic derivational processes. International Journal of American Linguistics 46. 259 –278. Gerfen, Chip. 2001. A critical view of licensing by cue: Codas and obstruents in Eastern Andalusian Spanish. In Lombardi (2001a), 183–205. Gildea, Spike. 1995. A comparative description of syllable reduction in the Cariban language family. International Journal of American Linguistics 61. 62 –102. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Haas, Mary. 1946. A grammatical sketch of Tunica. In Cornelius Osgood (ed.) Linguistic structures of native America, 337–366. New York: Viking Fund. Harris, James W. 1969. Spanish phonology. Cambridge, MA: MIT Press.

Positional Effects in Consonant Clusters

19

Harris, James W. 1984. Autosegmental Phonology, Lexical Phonology, and Spanish nasals. In Aronoff & Oehrle (1984), 67 –82. Hayes, Bruce. 1984. The phonetics and phonology of Russian voicing assimilation. In Aronoff & Oehrle (1984), 318–328. Hayes, Bruce. 1986. Assimilation as spreading in Toba Batak. Linguistic Inquiry 17. 467–499. Hayes, Bruce, Robert Kirchner & Donca Steriade (eds.) 2004. Phonetically based phonology. Cambridge: Cambridge University Press. Hualde, José Ignacio. 1987. On Basque affricates. Proceedings of the West Coast Conference on Formal Linguistics 6. 77 –89. Hualde, José Ignacio. 1992. Catalan. London & New York: Routledge. Hualde, José Ignacio. 1993. Topics in Souletin Phonology. In José Ignacio Hualde & Jon Ortiz de Urbina (eds.) Generative studies in Basque linguistics, 289–327. Amsterdam & Philadelphia: John Benjamins. Hume, Elizabeth. 1999. The role of perceptibility in consonant/consonant metathesis. Proceedings of the West Coast Conference on Formal Linguistics 17. 293 –307. Itô, Junko. 1986. Syllable theory in prosodic phonology. Ph.D. dissertation, University of Massachusetts, Amherst. Itô, Junko. 1989. A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7. 217–259. Itô, Junko, Armin Mester & Jaye Padgett. 1995. Licensing and underspecification in Optimality Theory. Linguistic Inquiry 26. 571–613. Jun, Jongho. 1995. Perceptual and articulatory factors in place assimilation: An Optimality Theoretic approach. Ph.D. dissertation, University of California, Los Angeles. Jun, Jongho. 1996. Place assimilation is not the result of gestural overlap: Evidence from Korean and English. Phonology 13. 377 –407. Jun, Jongho. 2002. Positional faithfulness, sympathy and inferred input. Unpublished ms., Yeungnam University, Daegu, Korea. http://ling.snu.ac.kr/jun. Jun, Jongho. 2004. Place assimilation. In Hayes et al. (2004), 58–86. Kabak, Baris & William J. Idsardi. 2007. Perceptual distortions in the adaptation of English consonant clusters: Syllable structure or consonantal contact constraints. Language and Speech 50. 23 –52. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Kim-Renaud, Young-Key. 1974. Korean consonantal phonology. Ph.D. dissertation, University of Hawaii. Kim-Renaud, Young-Key. 1986. Studies in Korean linguistics. Seoul: Hanshin. Kiparsky, Paul. 1985. Some consequences of Lexical Phonology. Phonology Yearbook 2. 85–138. Kochetov, Alexei & Marianne Pouplier. 2008. Phonetic variability and grammatical knowledge: An articulatory study of Korean place assimilation. Phonology 25. 399 –431. Kohler, Klaus J. 1990. Segmental reduction in connected speech in German: Phonological facts and phonetic explanations. In W. J. Hardcastle & A. Marchal (eds.) Speech production and speech modelling, 69 –92. Dordrecht: Kluwer. Kohler, Klaus J. 1991a. The phonetics/phonology issue in the study of articulatory reduction. Phonetica 48. 180 –192. Kohler, Klaus J. 1991b. The organization of speech production: Clues from the study of reduction processes. Proceedings of the 12th International Congress of Phonetic Sciences, vol. 1, 102 –106. Aix-en-Provence: Université de Provence. Kohler, Klaus J. 1992. Gestural reorganization in connected speech: A functional viewpoint on “articulatory phonology.” Phonetica 49. 205 –211. Krueger, John R. 1962. Yakut manual. Bloomington: Indiana University.

20

Jongho Jun

Ladefoged, Peter & Ian Maddieson. 1986. Some of the sounds of the world’s languages. UCLA Working Papers in Phonetics 64. 1–137. Lamontagne, Greg. 1993. Syllabification and consonant cooccurrence conditions. Ph.D. dissertation, University of Massachusetts, Amherst. Lodge, Ken. 1986. Allegro rules in colloquial Thai: Some thoughts on process phonology. Journal of Linguistics 22. 331–354. Lodge, Ken. 1992. Assimilation, deletion paths and underspecification. Journal of Linguistics 28. 13 –52. Lombardi, Linda. 1995. Laryngeal neutralization and syllable wellformedness. Natural Language and Linguistic Theory 13. 39 –74. Lombardi, Linda. 1999. Positional faithfulness and voicing assimilation in Optimality Theory. Natural Language and Linguistic Theory 17. 267 –302. Lombardi, Linda (ed.) 2001a. Segmental phonology in Optimality Theory: Constraints and representations. Cambridge: Cambridge University Press. Lombardi, Linda. 2001b. Why Place and Voice are different: Constraint-specific alternations in Optimality Theory. In Lombardi (2001a), 13 –45. McCarthy, John J. 2008. The gradual path to cluster simplification. Phonology 25. 271–319. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 249 –384. Amherst: GLSA. Mohanan, K. P. 1993. Fields of attraction in phonology. In John A. Goldsmith (ed.) The last phonological rule: Reflections on constraints and derivations, 61–116. Chicago & London: University of Chicago Press. Padgett, Jaye. 1995. Partial class behavior and nasal place assimilation. In Keiichiro Suzuki & Dirk Elzinga (eds.) Proceedings of the 1995 Southwestern Workshop on Optimality Theory (SWOT), 145 –183. Tucson: Department of Linguistics, University of Arizona (ROA-113). Padgett, Jaye. 2002. Russian voicing assimilation, final devoicing, and the problem of [v]. Unpublished ms., University of California, Santa Cruz (ROA-528). Payne, David L. 1981. The phonology and morphology of Axininca Campa. Austin, TX: Summer Institute of Linguistics. Prince, Alan & Paul Smolensky 2004. Optimality Theory: Constraint interaction in generative grammar. Malden, MA & Oxford: Blackwell. Rehg, Kenneth L. & Damien G. Sohl. 1981. Ponapean reference grammar. Honolulu: University of Hawai’i Press. Rice, Keren. 1989. A grammar of Slave. Berlin & New York: Mouton de Gruyter. Rischel, J. 1974. Topics in Greenlandic phonology: Regularities underlying the phonetic appearance of wordforms in a polysynthetic language. Copenhagen: Akademisk Forlag. Rubach, Jerzy. 1990. Final devoicing and cyclic syllabification in German. Linguistic Inquiry 21. 79 –94. Rubach, Jerzy. 1996. Nonsyllabic analysis of voice assimilation in Polish. Linguistic Inquiry 27. 69 –110. Sapir, J. David. 1965. A grammar of Diola-Fogny. Cambridge: Cambridge University Press. Schachter, Paul & Victoria P. Fromkin. 1968. A phonology of Akan: Akuapem, Asante, Fante. UCLA Working Papers in Phonetics 9. Seo, Misun. 2003. A segment contact account of the patterning of sonorants in consonant clusters. Ph.D. dissertation, Ohio State University. Son, Minjung. 2008. The nature of Korean place assimilation: Gestural overlap and gestural reduction. Ph.D. dissertation, Yale University. Son, Minjung, Alexei Kochetov & Marianne Pouplier. 2007. The role of gestural overlap in perceptual place assimilation in Korean. In Jennifer Cole & José Ignacio Hualde (eds.) Laboratory phonology 9, 507 –534. Berlin & New York: Mouton de Gruyter.

Positional Effects in Consonant Clusters

21

Steriade, Donca. 1993. Neutralization and the expression of contrast. Paper presented at the 24th Annual Meeting of the North East Linguistic Society, University of Massachusetts Amherst. Steriade, Donca. 1995. Positional neutralization. Unpublished ms., University of California, Los Angeles. Steriade, Donca. 1999. Phonetics in phonology: The case of laryngeal neutralization. UCLA Working Papers in Linguistics 2: Papers in Phonology 3. 25 –146. Steriade, Donca. 2001. Directional asymmetries in place assimilation: A perceptual account. In Elizabeth Hume & Keith Johnson (eds.) The role of speech perception in phonology, 219–250. San Diego: Academic Press. Steriade, Donca. 2009. The phonology of perceptibility effects: The P-map and its consequences for constraint organization. In Kristin Hanson & Sharon Inkelas (eds.) The nature of the word: Studies in honor of Paul Kiparsky, 151–179. Cambridge, MA: MIT Press. Stevens, Kenneth N. & Sheila Blumstein. 1975. Quantal aspects of consonant production and perception: A study of retroflex consonants. Journal of Phonetics 3. 215 –233. Teoh, Boon Seong. 1988. Aspects of Malay phonology revisited: A non-linear approach. Ph.D. dissertation, University of Illinois. Vance, Timothy J. 1987. An introduction to Japanese phonology. Albany: State University of New York Press. Webb, Charlotte. 1982. A constraint on progressive consonantal assimilation. Linguistics 20. 309–321. Wheeler, Max W. 2005. Voicing contrast: Licensed by prosody or licensed by cue? Paper presented at the 13th Manchester Phonology Meeting (ROA-769). Wilson, Colin. 2001. Consonant cluster neutralisation and targeted constraints. Phonology 18. 147 –197. Wright, Richard. 1996. Consonant clusters and cue preservation in Tsou. Ph.D. dissertation, University of California, Los Angeles. Wright, Richard. 2004. A review of perceptual cues and cue robustness. In Hayes et al. (2004), 34–57. Yip, Moira. 1991. Coronals, consonant clusters, and the coda condition. In Carole Paradis & Jean-François Prunet (eds.) The special status of coronals: Internal and external evidence, 61–78. San Diego: Academic Press. Zhang, Jie. 2004. The role of contrast-specific and language-specific phonetics in contour tone distribution. In Hayes et al. (2004), 157–190. Zoll, Cheryl. 1998. Positional asymmetries and licensing. Unpublished ms., MIT (ROA-282).

47

Initial Geminates Astrid Kraehenmann

1

Introduction

There are a number of different ways in which languages make phonemic differences between consonant segments. One, for example, is on the basis of phonological features. The groups of sounds in (1a) differ in terms of voicing (chapter 69: final devoicing and final laryngeal neutralization), the ones in (1b) in terms of continuancy (chapter 13: the stricture features), and the ones in (1c) in terms of place of articulation (labial, coronal, dorsal) (chapter 22: consonantal place of articulation), etc. (1)

a. b. c.

/p t k f s x/ /p t k b d g/ /p b f v/

vs. vs. vs.

/b d g v z :/ /f s x v z :/ /t d s z/

vs. /k g x :/

Another way in which consonants may contrast is on the basis of inherent prosodic structure such as sound length and/or weight.1 This distinction is also called a difference in quantity (see also chapter 37: geminates; chapter 57: quantitysensitivity). Thus the groups of sounds in (2) differ in terms of quantity, which is generally reflected (a) phonetically as long vs. short articulatory duration, (b) phonologically as heavy vs. light, and (c) structurally in the way they are associated to syllabic and higher-level prosodic structure. (2)

/pp tt kk ff ss xx/ vs. /p t k f s x/

Cross-linguistically, there are some constraints and implicational relationships on the existence of such quantity contrasts. As will be shown in the next section, initial geminates – i.e. geminates that contrast with singletons at the beginning of (lexical) words – are a special case, because they are even rarer than medial geminates. In the subsequent sections we will discuss in turn some issues regarding the phonological representation, prosodic behavior, phonetic properties, perception, and word-edge effects of initial geminates. These issues will provide some interesting insight on why initial geminates are so special. 1

In this chapter, length/quantity is transcribed by a sequence of identical phonetic symbols.

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0047

Initial Geminates

2

2

Typological issues

Not many languages are known to have a lexical quantity contrast at word edges as well as word-medially. As an illustration we give examples from Thurgovian Swiss German (Indo-European)2 in (3), which are representative of the phonological quantity distinction common to the majority of Swiss German dialects. (3)

Swiss German quantity contrast a.

initial

b.

medial

c.

final

/ppaaö/ /ttaIkx/ /kkaaö/ /vappH/ /mattH/ /makkH/ /alpp/ /veött/ /maökk/

‘pair’ ‘tank’ ‘coach’ ‘crest’ ‘mat’ ‘fault’ ‘alp’ ‘value’ ‘marrow’

vs. vs. vs. vs. vs. vs. vs. vs. vs.

/paaö/ /taIkx/ /kaaö/ /vapH/ /matH/ /makH/ /xalp/ /heöt/ /aök/

‘bar’ ‘thank’ ‘cooked’ ‘honeycomb’ ‘maggot’ ‘stomach’ ‘calf’ ‘hearth’ ‘bad’

In order to eventually get a better feel for the typological status of initial geminates such as the ones in (3a), it is worth taking a look at the established typological facts of geminates in general. Surprisingly little comprehensive work has been done in this area to date. If we consult the largest extant phonological database – the UCLA Phonological Segment Inventory Database (UPSID), which is an extended and revised online version of Maddieson (1984), listing the phonemes of 451 languages – only 12 languages are coded for having at least one geminate consonant segment. But not all of them have a quantity contrast to speak of. There are only seven (less than 2 percent of the whole sample) if we discount those languages which – based on the inventories given – have either only a single geminate–singleton pair (!Xu (Khoisan), Iraqw (Afro-Asiatic), and Telugu (Dravidian)) or only geminates that come without a singleton counterpart (Inuit (Eskimo-Aleut) and Trumai (South American isolate)). Those seven are: the North Caucasian Archi, Avar, and Lak; Ocaina (Witotoan), Waray (Austronesian), Wichita (Caddoan), and Wolof (Niger-Congo). While the general picture that emerges seems to be that quantity contrasts are rather uncommon, it is doubtful that they are quite as uncommon as that. For example, the database also contains languages such as the Indo-European Bengali, Breton, and Norwegian, as well as Finnish (Uralic), for which there is general consensus among linguists that all have quantity contrasts. However, they are not coded as such in UPSID. Thus the typological interpretation of the UPSID data – not only with respect to quantity – must be taken with the necessary pinch of salt (see also Simpson 1999, among others). One particular piece of information missing in UPSID but of primary interest for our purposes is where within a word the geminates may occur. This information is given, albeit rather implicitly, in the survey of 63 languages with geminates by 2

Language family classification in all cases follow Lewis (2009).

3

Astrid Kraehenmann

Thurgood (1993). In his sample, Thurgood identifies a number of frequency facts and prerequisite conditions for geminates and proposes some implicational tendencies. For example, he finds preferences for certain places of articulation (alveolar > labial > velar > glottal)3 and certain prosodic environments (post-tonic > pre-tonic/unstressed; after short vowel > after long vowel). But most importantly, he establishes that geminates are most favored in intervocalic position or, more specifically, if preceded by a short (and stressed) vowel and followed by another short vowel (Thurgood 1993: 129). Although not explicitly stated, these flanking vowels preferably also belong to the same word. This means that if a language has any quantity contrast at all, it will be word-medially. Thus, the existence of initial and/or final geminates implies the existence of medial geminates (see also Muller 1999). While Thurgood (1993) lists the three Austronesian counterexamples Sa’ban, Kelantan Malay, and Pattani Malay, no further explanation is given as to how it could be that they exclusively have initial geminates, which is an interesting issue that we will come back to shortly. Thurgood does not mention word-final geminates separately, let alone whether or not the languages under investigation allow word-final (i.e. syllable-final) consonants in the first place. But, based on the preference factors he identifies, we can conclude that a quantity contrast is cross-linguistically most common medially, somewhat less common finally, and least common initially. Is there also a specific consonant class that stands out as the most preferred to display a quantity contrast? Based purely on the sonority of consonants (chapter 49: sonority), Morén (1999: 110) answers this question negatively. He finds languages with quantity distinctions in stops, continuants, and sonorants alike (Hungarian (Uralic), Brahui (Northern Dravidian), Italian, Baloch, Gajarati (Indo-European)), only in continuants (Tartar (Altaic)), only in sonorants (Hausa (Afro-Asiatic)), and only in obstruents (Chechen, Lak (North Caucasian)), but also languages that allow geminate stops and sonorants to the exclusion of continuants (Kurdish (Indo-European)). He does not list any languages either allowing only stop geminates or excluding only stop geminates, both of which he considers to be accidental gaps. In comparison, Thurgood’s (1993) generalizations are a bit more fine-grained for place of articulation and voicing within the major classes of obstruents and sonorants. He observes that geminate stops seem to be a prerequisite for geminate affricates to occur, and presents data that suggest a voiceless alveolar stop or fricative /tt ss/ or an alveolar nasal or liquid /nn ll/ to be good candidates for the prototypical geminate. To conclude our look at geminates in general, it is very surprising to notice that 46 of the 63 languages (= 73 percent) in Thurgood’s (1993) sample are also listed in UPSID, yet obviously with different phoneme analyses from the ones consulted by Thurgood. As regards language diversity, six language families figure very prominently, accounting for more than 50 percent of the whole set: namely, in descending frequency, Afro-Asiatic (10 languages), Indo-European (7), Austronesian (6), Altaic (3), Dravidian (3), and Uralic (3). 3

This means that Thurgood (1993) finds, on the one hand, that alveolar geminates are the most frequent and glottal geminates the least frequent and, on the other hand, that if a language has, for example, velar geminates it also tends to have labial and alveolar ones. He presents his data as an overview of the phonological systems and does not use them to make any claims about preferences for the emergence of geminates.

Initial Geminates

4

As can be seen in (4), the first three are also very prominently represented in languages that exhibit initial geminates. This set of languages is given in the appendix to Muller (2001), which lists the consonant inventories and some distributional facts of 29 different languages and represents the most comprehensive collection of languages with initial geminates to date.4 (4)

Language families and languages with initial geminates (Muller 2001: 204–233) Austronesian (13)

Indo-European (4) Afro-Asiatic (3) North Caucasian (2) Arawakan Austro-Asiatic Japanese Niger-Congo Nilo-Saharan Oto-Manguean West Papuan

Trukese, Dobel, Kiribati, Leti, Ngada, Pattani Malay, Ponapean, Puluwat, Roma, Sa’ban, Taba, Woleaian, Yapese Breton, Cypriot Greek, Swiss German (Bernese and Thurgovian) Cypriot Maronite Arabic, Moroccan Arabic, Tamazight Berber Circassian, Lak Piro Nyaheun Hatoma Luganda Lugbara Atepec Zapotec Hatam

Within this set of languages, five seem to have only initial geminates (see Figure 47.1): Leti, Ngada, Pattani Malay, Yapese, and Nyaheun. While Leti is listed as having medial geminates, all of them are derived rather than lexical (see also Hume et al. 1997). For Ngada, no information on medial geminates is given. Pattani Malay and Yapese are known to lack medial geminates (cf. Abramson 1987, 1991, 1999; Jensen 1977). However, it is important to note that in Yapese only two of the 27 possible consonant phonemes show a length distinction, i.e. /gg ll/,

Number of languages

25 20

20 15 10 5

5

4

0 only initial

initial & medial

initial, medial, & final

Figure 47.1 Geminates by word position in the 29 languages listed in Muller (2001: 204–233) 4

There is a substantial number of languages with initial geminates – for example, Maltese and various southern Italian dialects – that do not appear in Muller’s (2001) list. For the purpose of this brief overview, however, we restrict ourselves to Muller’s sample.

5

Astrid Kraehenmann

which – as we will see – are not prototypical geminates. For Nyaheun, Muller (2001: 234) gives a structural reason for the lack of medial geminates: words in this language are mostly monosyllabic. One could even include Sa’ban as the sixth language of the “initial-only” set, because it has but very few medial geminates, while initial ones are abundant. The fact that we do find languages with initial quantity contrasts to the exclusion of medial and final ones weakens Thurgood’s (1993) generalization – and, indeed, goes against the most basic definition of what a geminate is (see §3) – namely that medial intervocalic geminates are the default case. One explanation may be morpheme and/or syllable structure constraints (chapter 33: syllable-internal structure; chapter 86: morpheme structure constraints), as mentioned for Nyaheun. Alternatively, for the other five languages – all of which are Austronesian – the existence of initial geminates may simply have arisen historically through morpheme concatenation or reduction processes (chapter 79: reduction).5 However, as far as historical comparative evidence goes, Austronesian languages, especially the Western Micronesian branch, are reconstructed with initial geminates in place (cf. Jackson 1984; Bender et al. 2003), and the modern-day systems show a strong dislike for and various avoidance strategies against initial geminates (e.g. Kennedy 2005). Thus what we see in the Micronesian language of Yapese may just be the last remains of what used to be a stable quantity distinction. Incidentally, the other Micronesian languages in Muller’s (2001) sample (Trukese, Kiribati, Ponapean, Puluwat, Woleaian) still have a healthy medial quantity contrast, in addition to the initial distinction. Regarding geminates at the end of words, Moroccan Arabic, Tamaziht Berber, and the two Swiss German dialects allow all the geminates in the system to occur in this position. No information is given for Breton, Circassian, and Cypriot Maronite Arabic, while this category is listed as “not applicable” for Hatam (Muller 2001: 215). The remaining 21 languages do not seem to have any final geminates, but it is not indicated whether this is a systemic gap, as it presumably is for Hatam. On the face of it, this paucity of final geminates appears to conflict with Thurgood’s (1993) finding that final geminates should be more frequent than initial ones. However, it is important to keep in mind that the criterion for inclusion in Muller’s (2001) language set was the existence of initial geminates. Therefore, languages with medial and/or final geminates but without initial ones do not figure in her sample. It might still be the case that Thurgood’s (1993) generalization holds true overall, but that the frequency distribution within the languages with initial geminates is skewed.6 In sum, within Muller’s 29 languages with initial geminates, final geminates imply medial geminates. There are no languages with initial and final geminates that lack medial geminates (see Figure 47.1 for attested combinations). Also, the set of geminate phonemes allowed in initial position is always a proper subset of the whole, either equal to or smaller than the one allowed in medial (and final) position. 5

See Yupho (1989) for this finding in Pattani Malay. Dmitrieva (2009), quoting Thurgood (1993), claims that word-initial geminates are cross-linguistically preferred over word-final ones. However, on the interpretation here, the facts presented by Thurgood seem to point in the opposite direction. Therefore, it is not quite clear what data Dmitrieva bases her basic assumption on. She essentially presents a functional explanation, namely a mismatch between production and perception of the quantity distinction in word-initial vs. word-final position. The data of her production and perception experiments involved (Russian) non-words rather than real words.

6

Initial Geminates

6

Number of languages

25 20 15 10 5 0

nn

tt

mm

kk

ss

pp

bb

ll

dd

ff

rr

Figure 47.2 The 10 most frequent geminate phonemes in the 29 languages listed in Muller (2001: 204–233)

Looking at the frequency of initial geminate phonemes, there are some interesting observations to be made with respect to major consonant class, manner, place of articulation, and voicing.7 But before looking at these factors, let us first consider the 10 most frequent geminates in these languages, as illustrated in Figure 47.2. The nasals /nn mm/ and the voiceless stops /tt kk pp/ are among the most commonly occurring geminates. The voiceless fricatives /ss ff/, the voiced stops /bb dd/, and the liquids /ll rr/ are next on the top-10 frequency list. Thus, within the sonorants /nn/ is most universally present, and within the obstruents it is /tt/ for the stops and /ss/ for the fricatives. Very significantly, all these sounds follow the universal preference patterns (cf. Ladefoged and Maddieson 1996; Ladefoged 2001): they are coronal as well as voiced if sonorant and voiceless if obstruent, which is the less marked voicing characteristic within each respective class. Considering the entire set of initial geminates, stops and fricatives together make up almost two-thirds, with nasals taking the lion’s share in the last third (see Figure 47.3). Geminate fricatives imply geminate stops, except for Hatoma, which has /ff/ and /ss/, but does not allow any geminate stops or affricates. Overall, geminate affricates are least frequent and only occur in conjunction with 120

113

Frequency

100 74

80

61

60 40

25

20

9

5

glides

affricates

0 stops Figure 47.3 7

fricatives

nasals

liquids

Initial geminates by manner (data from Muller 2001: 204–233)

In this chapter, I refer only to type frequency of phonemes and members of natural classes. I have no information about token frequency of the sounds in the respective languages.

Astrid Kraehenmann

7

both stops and fricatives. As for the sonorants, nasals are a prerequisite for any other type of sonorant geminate (chapter 8: sonorants). The only exception is Yapese, which has only a coronal lateral /ll/ and no nasal or other sonorant geminate, although it has labial, coronal, and dorsal nasal singletons. In the complete set of languages, there are 22 that have both obstruent and sonorant geminates. However, Circassian, Lak, Bernese, and Thurgovian Swiss German allow only obstruents, while Kiribati, Piro, and Ponapean allow only sonorants. In the latter group, all three languages have geminate nasals, most often a coronal /nn/. Figure 47.4 illustrates the frequencies of the different places of articulation8 in the sample. As expected, the predominance of coronals over labials and dorsals mirrors the general attested cross-linguistic preferences. Finally, regarding voicing where it is contrastive, namely within the obstruents, voiceless geminates outnumber voiced ones by a little over 100 percent (see Figure 47.5). There are seven languages without voicing distinction, for which only “voiceless” geminates are listed: Atapec Zapotec, Trukese, Hatoma, Lak, Leti, Puluwat, and Thurgovian Swiss German. In three languages, there seem to be “voiced” geminates without any “voiceless” counterpart: Lugbara has /bb dd/ but no */pp tt/, Ngada has /bb/ but no */pp/, and Yapese has /gg/ but no */kk/. However, in all three cases, these phonemes are the only obstruent geminates in the respective systems. Dobel is a slightly different case, in that it lacks */pp/, with /bb/ being part of a bigger set, which includes voiceless /tt kk ?? ss/.

Frequency

160

148

120 78

80

53 40 4

3

glottal

pharyngeal

0 coronal Figure 47.4

Frequency

80

labial

Initial geminates by place of articulation (data from Muller 2001: 204–233) 73

voiceless voiced

54

60 40

dorsal

40 20

20

4 0 stops Figure 47.5 204–233) 8

fricatives

1

affricates

Initial obstruent geminates by manner and voicing (data from Muller 2001:

In Muller’s (2001) terminology, labials comprise bilabials and labio-dentals, coronals refer to (inter)dentals, alveolars, and palatals, and dorsals are velars.

Initial Geminates

8

Within the subset of languages for which voicing is distinctive, 11 display more than one pair of voiceless and voiced geminate obstruents. These are Breton, Circassian, Cypriot Maronite Arabic, Luganda, Nyaheun, Sa’ban, Hatam, Pattani Malay, Moroccan Arabic, Taba, and Tamazight Berber. Interestingly, there are two types of case in which there are more voiced geminates than voiceless ones. The first type involves an apparent dislike of voiceless bilabial stops. In addition to Lugbara, Ngada, and Dobel, mentioned above, Moroccan Arabic, Taba, and Tamaziht Berber allow /bb/ but not */pp/. The other type is less striking and shows a dislike of voiceless palatal stops. For example, both Hatam and Pattani Malay have /ÁÁ/ but not */cc/. However, we find the reverse preference as well: although /ÁÁ/ is part of the phoneme system of Nyaheun, it is the only obstruent that is not allowed word-initially. Voicing seems to be a more systematic restricting factor in Cypriot Greek and Bernese Swiss German. The former distinguishes voiced and voiceless fricatives word-medially, but only allows the voiceless ones initially. The latter apparently has the reverse preference: it distinguishes voiced and voiceless stops in all word positions except initial, where only voiced stops are allowed. It must be noted, though, that the phonemic analysis for the stops in the source reference (Ham 1998) is disputed (e.g. Marti 1985) and it is not firmly established that stops differ phonologically in terms of both voicing and quantity. Finally, voicing is only marginally distinctive in Roma and Woleaian. Both have but one voicing pair – /tt dd/ and /pp bb/, respectively – alongside other voiceless obstruents: /pp kk/ and /pp kk tt ff ss/, respectively. To summarize the main points, languages do not often make a phonological distinction between geminates and singletons, particularly not in word-initial position. Initial quantity contrasts do occur concentrated in a handful of language families, i.e. in Austronesian, Indo-European, Afro-Asiatic, and North Caucasian. Initial geminates, by and large, only occur if there are also medial geminates in the language system. Asymmetries and preferences for certain place of articulation, manner, major class, and voice characteristics follow the preferences of sounds that are not in length opposition: e.g. coronals and voiceless obstruents are most frequent. The rarity of quantity contrasts, which has now been established, raises the question of why this should be the case. On purely physiological grounds, there are no substantial hindrances to articulating speech sounds in long and short versions, although with some sounds (e.g. continuants such as sonorants and fricatives) it might be easier than with others (e.g. non-continuants such as stops). In the following sections it will become clear that the relative nature of the distinction is the main reason. It poses challenges on many levels, namely in the way it must be stored in mental representations, in the way it is realized by the speakers in different linguistic (and social) contexts, and in the way it is perceived by the listeners in those different contexts.

3

Underlying representation, syllabification, and metrical issues

A classical definition of geminates requires these sounds to be single sounds that are long and ambisyllabic when intervocalic (e.g. Swadesh 1937: 10; Trubetzkoy 1939: 156; see also chapter 37: geminates). This means that a geminate both closes

Astrid Kraehenmann

9

a preceding syllable and is the onset of a following syllable; see the structures in (5). As a rule, “intervocalic” is interpreted to mean “within one and the same word.” (5)

Syllabification of geminates and singletons9 a.

ambisyllabic link: geminate q q a

p a [appa]

b.

tautosyllabic link: singleton q q a

p a [apa]

While there is considerable agreement as to the ambisyllabic nature of geminates (but see Topintzi 2008 for a different view), theories vary in what unit they posit to link up with these syllable constituents. These assumptions have far-reaching consequences for how geminates are phonologically represented, especially so if the goal is to establish a single representation that is to capture not only languagespecific but also cross-linguistic patterns of geminates (e.g. Selkirk 1990; Tranel 1991; Broselow 1995; Broselow et al. 1997; Ham 1998; Davis 1999a, 1999b, 2003; Muller 2001; Curtis 2003; Kraehenmann 2003; Topintzi 2008; Ringen and Vago 2010; among others). For convenience, we briefly sketch the main tenets of the two competing theories at this point.10 In moraic theory, a geminate is associated with a unit of weight, the mora – see (6a) – and should therefore participate in weight-related phenomena such as stress patterns, compensatory lengthening, and minimal word conditions. In skeletal theory (chapter 54: the skeleton), though, a geminate is doubly linked to the timing tier – see (6b) – and is prosodically long without pre-association to any weight tier. As a consequence, weightless geminates pose a challenge for the former theory, moraic geminates for the latter. (6)

Underlying representation and syllabification a. moraic theory (e.g. Hyman 1985; Hayes 1989) q q q q

9

[

[

/pp/

a

[

[

p a [appa]

/p/

[

[

a

p a [apa]

syllable tier weight tier melodic tier

The segmental labels linking up to syllable nodes are shorthand for the root nodes representing these segments. 10 See Curtis (2003) for a concise survey of the different models of geminate representation. chapter 37: geminates also discusses in detail the basic controversy between the weight account and the timing account, as well as the Composite Model (e.g. Hume et al. 1997; Muller 2001; Curtis 2003; Kraehenmann 2003; Baker 2008). A full exploration of this theory goes beyond the scope of this chapter. Suffice it to say, however, that it is a combination of the other two models: geminates are associated to two timing positions and weight, if it plays a role at all, is a consequence of syllabification, mainly one of Weight-by-Position (Hayes 1989). One worry with this view is that it runs the risk of overgeneration, since it allows for all types of combinations between length and weight, possibly even unattested ones.

Initial Geminates b. skeletal theory11 q

q

q

x x

x

x

x

x

x

x

x

/pp/

a

p [appa]

a

/p/

a

p

q

syllable tier

x

skeletal/ timing tier

10

a melodic tier [apa]

The major twist for initial geminates is that they are primarily seen as elements belonging to the syllable onset, the prototypical non-weight position. In the majority of languages, onsets play no role in weight-sensitive processes (Hayes 1989; Curtis 2003; Topintzi 2008; Davis, forthcoming; among many others; see also chapter 55: onsets). This holds true for singletons as well as geminates. For example, in Leti (Hume et al. 1997; Muller 2001: 117–142), CV syllables consisting of initial geminates and short vowels behave like light syllables: (a) they do not attract main stress, in contrast to syllables with long vowels, and (b) they also do not satisfy the bimoraic minimality condition on words (e.g. *[ppa] is not a possible Leti word). On the assumption that geminates always are moraic, these facts are difficult to explain. The authors therefore propose the non-moraic skeletal representation in (7a).12 In moraic theory, the bisegmental structure in (7b) has been suggested for the Leti-type geminates (e.g. Hayes 1989; Davis 1999a). (7)

Representations of non-moraic initial geminates a.

single root doubly linked q

b.

bisegmental

[ x

x

root

x

root root

c.

two-root nodes

q

q

[

[ root root [features]

For Topintzi (2008: 176), weightless geminates “cannot be real geminates, since geminates are by definition underlyingly moraic.” She claims that they are fake geminates, i.e. doubled consonants with two separate root nodes. Assuming it is important to differentiate between underlying and surface form, it is somewhat surprising that she does not opt for the bisegmental representation in (7b), but 11

In the interest of simplicity, the skeletal theory is illustrated with x-slot notation (following Levin 1985). There is no claim being made about it being superior to other existing timing models, such as Selkirk’s (1990) two-root theory (see also Curtis 2003: 310) or CV representation (e.g. McCarthy 1979; Clements and Keyser 1983). The point here is simply to illustrate the different prosodic structures in a weight account, as opposed to a length/timing account. 12 See also Kraehenmann (2001, 2003) for initial geminates in Swiss German not participating in a bimoraic minimality constraint. The proposed structures link a single root node to two timing positions, the first of which is associated to the prosodic word level rather than the syllable level.

Astrid Kraehenmann

11

rather Selkirk’s (1990) two-root node model (7c), which posits real geminates with a branching structure underlyingly (chapter 1: underlying representations). The bisegmental analysis is in line with the basic tenet of moraic theory requiring onset segments not to contribute to syllable weight. This is achieved by denying the onset consonants geminate status. Yet, while this analysis can properly account for the fact that Leti geminates pattern like (other) consonant clusters, it cannot explain why the geminates show integrity behavior by patterning like single consonants for processes such as reduplication, in which the onset must not be split up (Hume et al. 1997; Muller 2001). Turning now to the rarer instances in which onsets do seem to add weight to syllables, Trukese (also called Chuukese by some authors) is one of the favorite and most convincing examples cited (e.g. Davis 1999a, forthcoming; Muller 1999, 2001; Curtis 2003; Topintzi 2008; Ringen and Vago 2010; see chapter 37: geminates for more data). Similarly to Leti, Trukese has a bimoraic minimum condition on nouns, but in contrast to Leti, Trukese monosyllables consisting of an onset geminate and a short vowel satisfy this condition, i.e. [tto] ‘clam (sp.)’ is fine but *[tH] ‘islet’ is not (it retains its underlying long vowel: [tHH]) (Davis 1999a: 95). Different representations have been proposed, all of which have in common that the geminates come pre-associated with a mora. Within the skeletal framework, Muller (2001: 174, 179) assumes the semisyllabic13 representation illustrated in (8a) both for Trukese and Luganda initial geminates, the latter of which participate in tone processes restricted to moraic units. Kiparsky (2003: 165) proposes a similar structure in the moraic framework for initial geminates in VC- and C-dialects of Arabic; see (9a). (8)

Representations of moraic initial geminates a.

semisyllabic w

[ x

x

b. syllabic σ σ

q

µ

[

root

µ

c.

stray mora σ µ

µ root

d. tautosyllabic σ µ

µ

root

x

root An important difference, though, is that the Arabic geminates are morphologically derived and, moreover, they are invisible to stress, i.e. weightless (chapter 124: word stress in arabic). In contrast to Topintzi’s (2008) treatment of weightless initial geminates, Kiparsky grants them geminate status as a matter of course, with all the consequences: they are associated with a mora in spite of the fact that they are in a non-weight position. The adjunction of the geminate mora directly 13

The label “semisyllabic” is adopted from Kiparsky (2003: 154), who bases his terminology on Sievers’s (1901) concept of “Nebensilbe.”

Initial Geminates

12

to the prosodic word level is intended to account for the fact that it does not count for stress purposes. Although designed for the Arabic geminates, according to Kiparsky the structure in (9a) is representative for weightless initial geminates in general. (9)

Representations of Arabic initial geminates (Kiparsky 2003: 165) a. light semisyllabic w

[ root

b.

heavy semisyllabic

t

U

q

q

[

[

[

root

The question thus arises of how a mora adjoined to the word level can be weightcontributing in one instance (8a), and not in another (9a). The answer seems to be in the nature of the phenomenon that the structures need to capture. In the Arabic-type case the weight of the syllable (and/or foot) is crucial since stress – evaluated at the syllable/foot level – is at issue; in the Trukese-type case, the weight of the whole word is at issue when minimality conditions are imposed.14 Similarly, Muller (2001: 174) proposes adjunction to the word level, and that only in the absence of evidence for any independent syllable-based prosodic processes. But it must be pointed out that mora adjunction such as in (8a) and (9) introduces different levels of weight projection, namely one in which moras contribute to syllable weight, which is the commonly held view, and others in which moras contribute to foot or word weight. And these levels need not coincide. More importantly, the principle that weight is predictable from syllable structure (e.g. Gordon 2005; see also chapter 39: stress: phonotactic and phonetic evidence) is no longer valid: the structures in (8a) and (9) by themselves might not unequivocally tell us whether or not the mora is indeed weight-contributing. To return to the other proposed representations for moraic initial geminates, the debate is primarily about where and/or how to associate the underlying mora to make it count. For example, Kiparsky (2003: 165) puts forward the structure in (9b), where the mora is adjoined to the foot level and is intended to participate in weight-related processes. Hayes (1989: 302) suggests the projection of a separate syllable as in (8b). This representation has not found much support in the literature on geminates, presumably because there is no evidence that initial geminates can or indeed must form their own syllable. For Trukese initial geminates and Italian initial palatals, Davis (1999a) opts for the representation in (8c) with a stray mora, first proposed in Hayes (1989: 303). However, as pointed out by Hume 14

See also Curtis (2003: 298), who proposes this structure for Trukese, arguing that weight should count at the word level rather than the syllable level.

13

Astrid Kraehenmann

et al. (1997: 393), Kraehenmann (2003: 26), and Topintzi (2008: 180), the mora cannot count for weight purposes by any measure of analysis when not associated to higher level prosodic structure. Finally, Topintzi’s (2008) proposal for Trukese and Pattani Malay – and moraic onsets in general – is the one illustrated in (8d), in which the geminate mora is directly associated with the syllable node to which the following vowel belongs. This tautosyllabic representation is striking for at least two reasons. First, it treats the geminate like a regular onset consonant by making it entirely the initial constituent within the first syllable. There is no ambisyllabic double linking, which is characteristic of geminates in all other proposals. Second, the geminate carries a mora equal in all respects to the mora of the syllable nucleus. The implications of such structures are, on the one hand, that moraic onsets are expected to occur not only word-initially but also word-medially and, on the other hand, that, to an even greater extent, moras play the double role of a unit of weight and length. In support of the first implication, Topintzi (2008: 170–172) presents some intriguing data from Marshallese stress patterns and Trique compensatory lengthening processes which suggest an analysis of medial geminates with tautosyllabification (see also chapter 55: onsets; Topintzi 2010). In the majority of cases, though, medial geminates must be analyzed ambisyllabically as illustrated in (8), cases like Marshallese being the exception rather than the rule. Regarding the length issue, Topintzi replaces the basic tenet of traditional moraic theory that onset consonants are never moraic with another, based on findings by Hubbard (1994) and Ham (1998), namely “that moras are allocated a minimum target duration” (2008: 174), the controversial proposal being that surface phonetic duration is in direct relation to underlying prosodic length. To summarize this section, there are a number of different challenges that initial geminates pose for the existing theoretical models of representation. Like medial geminates, initial geminates do not present a uniform behavior across languages. In some cases it is clear that they fully participate in weight-sensitive processes (e.g. Trukese), in other cases weight is not involved at all (e.g. Leti). If, in the interest of a universal geminate representation, geminates are not underlyingly moraic, how do we explain the Trukese cases? And vice versa, if geminates are underlyingly moraic, how do we explain the Leti cases? Moreover, how can such Leti-type geminates occupy the syllable onset, the non-weight position? It is evident that strict adherence to the traditional concepts of the two theories is not very fruitful. A combination of the two approaches seems rather more promising but the proper balance still needs to be established (see also note 10).

4

Phonetic properties and perception

In terms of phonetic properties, initial geminates are a model case for illustrating the interdependence between articulation, acoustics, and perception because not only segmental features, such as place of articulation, manner, etc., are involved, but prosodic features representing duration also play a role. Temporal characteristics have received most attention in phonetic studies of geminates to date. Although there is still some discussion on whether geminates are to be considered long or tense consonants, it is generally accepted that the

Initial Geminates

14

most robust cross-linguistic phonetic correlation with phonological quantity is the relative duration of articulatory constriction, in particular, for voiceless stop (and affricate) consonants, the relative duration of the closure (Abramson 1986a, 1987, 1991, 1999, 2003; Lahiri and Hankamer 1988; Kraehenmann 2001, 2003; Tserdanelis and Arvaniti 2001; Payne and Eftychiou 2006; Mikuteit and Reetz 2007 and references therein; Ridouane 2007 and references therein). Depending on the language and specific phonological contexts within the language, geminates can be between 1.5 and 3 times the duration of singletons (cf. Clements 1986; Lahiri and Hankamer 1988; Ladefoged and Maddieson 1996).15 Closure duration – or consonant duration for non-stop consonants – is thus recognized as the primary phonetic correlate of the quantity distinction. Due to the very nature of the quantity contrast, intervocalic medial geminates are the most studied, covering all existing types from sonorants to obstruents, voiced and voiceless. Investigations of initial geminates, in contrast, have overwhelmingly focused on voiceless stops in just a small handful of languages. One reason for this concentration on voiceless stops is certainly the fact that these sounds are the most frequent ones in initial quantity oppositions, as we saw in §2. Another – and more cogent – reason, however, is that they pose an interesting theoretical conundrum for linguists and a practical conundrum for speakers and listeners, particularly if these sounds are utterance/phrase-initial: in contrast to nasals, continuant consonants (e.g. liquids, glides, and fricatives) and voiced sounds in general, the acoustic duration cue for voiceless stops is a period of silence, the beginning of which cannot be determined by the listeners if no sound precedes it. One of the overarching questions is therefore whether there are additional cues that enable the listeners to distinguish initial voiceless stop geminates from singletons in utterance-initial position, i.e. in absence of the primary duration cue. In a series of meticulous experimental studies on Pattani Malay, Abramson (1986a, 1986b, 1991, 1999, 2003) answers this question in the affirmative. He argues that native listeners rely on secondary, non-durational acoustic cues, namely a combination of amplitude of the first syllable (the louder the syllable, the greater the perceived length of the consonant) and fundamental frequency (F0) of the vowel following the initial consonant (the higher the F0, the greater the perceived length of the consonant). In his acoustic study on geminates in Tashlhiyt Berber, Ridouane (2007) also reports on non-temporal correlates, most of which however involve voiced rather than voiceless stops and are classified into two different types. He calls the first type “concomitant correlates” (2007: 137), because they are largely optional. For examples, for three out of the five speakers, F0 averaged over the first 10 msecs of the postconsonantal vowel of voiced stops is slightly but significantly higher for geminates than for singletons. But no such correlation was found for voiceless stops. The other type is called “secondary correlates” (Ridouane 2007: 137). The author argues that they are contextually limited and/or speaker-specific but are 15

The type of sound can also play a crucial role. For example, Khattab (2007: 156) reports a staggering geminate to singleton ratio of 7.5 : 1 for intervocalic (medial) rhotics after short vowels, along with the comparably modest ratio of 1.6 : 1 for intervocalic voiceless velar stops after a long vowel for Lebanese Arabic.

15

Astrid Kraehenmann

manifestations of the tense articulation of geminates. One example is the finding that the normalized Root Mean Square (RMS) amplitude of the stop release is slightly higher for geminates than for singletons for all but one of the five speakers, who shows the reverse pattern. Another example would be the fact that geminates tend to devoice for aerodynamic reasons (cf. Ohala 1983), depending in degree on factors such as speaker, place of articulation, and position within the word. Also, voiced stop singletons have a weakish propensity to lenite, i.e. become fricatives, while geminates never do. As for durational correlates apart from the primary cue, Ridouane (2007: 129) finds that the duration of the stop release – also called “voice onset time” (VOT) or “after closure time” by other authors – shows no difference in voiceless stops but is longer for voiced geminates than for singletons.16 None of these additional correlates, however, has yet been tested for its significance in perception. The jury is still out on whether Tashlhiyt Berber native listeners can and do rely on these cues when the primary duration cue is not available. While VOT plays no role for voiceless stops of Tashlhiyt Berber, it is a temporal measure correlating with the quantity distinction of these sounds in Cypriot Greek. Parallel to Tserdanelis and Arvaniti’s (2001) findings on medial geminates, Muller (2001, 2003) establishes that VOT is also significantly longer for geminates than for singletons in word-initial position. Like native listeners of Pattani Malay, native listeners of Cypriot Greek can distinguish voiceless stop geminates from singletons at the beginning of an utterance. Muller hypothesizes that the subjects of her very basic perception experiment thus utilize VOT as one of possibly many secondary cues to the phonological quantity distinction. Finally, as well as Pattani Malay, Tashlhiyt Berber, and Cypriot Greek, Swiss German initial geminates have also received some attention in the literature, in particular in the Thurgovian dialect of Swiss German (Kraehenmann 2001, 2003, 2009; Kraehenmann and Lahiri 2008). Kraehenmann (2001, 2003) confirms earlier findings on other dialects, for example by Enstrom and Spörri-Bütler (1981) and Fulop (1994), that VOT does not participate in the quantity opposition: there is no difference in the duration of the stop release for geminates and singletons, be they initial, medial, or final. Since voiceless stops are the only consonants occurring in initial length opposition, the question is again whether the phonological difference is enhanced by cues other than the primary one. While lacking any corroborating perception evidence, Fulop (1994) claims to have found potential secondary nontemporary cues, namely increased intensity, movement, and clarity of post-release sonorant formants above F2 for geminates in comparison to singletons. However, anecdotal evidence in Moulton (1979) and the results of a pilot perception study reported in Kraehenmann (2003) call into question whether listeners can make use of these acoustic differences. Their listeners seemed unable to recover the quantity contrast in utterance-initial context. A crucial fact to verify especially in this case is whether the difference is actually produced by the speakers or whether contrast neutralization occurs. Using electropalatography (EPG), Kraehenmann and Lahiri (2008) set out to do just that in an articulatory study. They find that geminates are indeed articulated significantly longer (by about 85 msecs) than singletons even in absolute utterance/phrase-initial position (see Figure 47.6). 16

Exactly the same finding is reported in Mikuteit and Reetz (2007) for voiceless and voiced stops in East Bengali.

Articulation duration

Initial Geminates 300 250 200

16

geminate singleton

249

195 164 134

150 100

70

64

50 ##_V

V#_V

V_V

Figure 47.6 Articulation duration in ms for initial geminates and singletons in phrase-initial and phrase-medial post-vocalic context in comparison to medial geminates and singletons in intervocalic context (Kraehenmann and Lahiri 2008: 4450)

While, in terms of absolute measures, this duration difference is larger than the 65 msecs for the quantity distinction in a vocalic context (i.e. initial stops after a vowel-final word), in terms of proportional measure, the difference is in fact smaller – namely 1.5 : 1 as opposed to 2 : 1 – because both geminates and singletons are longer on average by about 100 msecs when utterance-initial. This means that although the articulation is heightened (see next section for a more detailed discussion on prosodically conditioned edge phenomena), the phonological contrast is made less salient in production.17 The contrast is produced most clearly in word-medial intervocalic context, where the duration of singletons is one third of that of geminates. Kraehenmann (2009) reports in a controlled follow-up perception experiment, which was conducted with audio data gathered in the articulatory study by Kraehenmann and Lahiri (2008), that native listeners are not able to discriminate phrase-initial geminates from singletons, whereas they have no problem doing so in a phrasal intervocalic context. Unlike Pattani Malay and Cypriot Greek listeners, therefore, Swiss German listeners cannot rely on secondary or enhancement cues. Why not? Although this question leads beyond the scope of this chapter, two points need to be made. First, the speakers do not seem to produce any secondary cues. This is unusual, because, according to work by Stevens and colleagues on Enhancement Theory (e.g. Stevens et al. 1986; Stevens and Keyser 1989; Keyser and Stevens 2006), distinctive properties of sounds are generally supplemented with other phonetic cues in order to make the phonological contrast more salient for perception, the longer VOT of Cypriot Greek geminates being a good example. Second, it is not at all clear why speakers maintain a distinctive articulation for geminates and singletons, when listeners cannot pick up on the primary cue. This clearly goes against functionalist principles of economy and ease of articulation (e.g. Hayes 1999). When a distinction cannot be perceived, why not neutralize it? These are issues that still need to be addressed satisfactorily in future work. In sum, relative consonant/closure duration is the main physical instantiation of the quantity contrast in any word position, even if utterance-initial: geminates are long and singletons are short, relative to each other. Languages differ in the way they do or do not enhance the primary correlate but there is no universal secondary cue that reliably distinguishes geminates from singletons. Muller (2001: 17

In a similar articulatory study, Ridouane (2007: 128–129) reports the same results for Tashlhiyt Berber voiceless stops. See §5 for more details.

17

Astrid Kraehenmann

201) rightly speculates that the segmental inventory of the language in question may affect the “availability” of such acoustic cues. But the complexity of syllable structures and the degree of the functional load of the contrast must also play a significant role.

5

Edge effects

In the field of the phonology–phonetics interface, one area of investigation focuses on how the articulation of speech sounds is influenced by the position within prosodic structure they occur in. A number of effects have been established that involve the edges of prosodic domains such as the syllable (chapter 33: syllableinternal structure), the phonological word (chapter 51: the phonological word), the phonological phrase (chapter 50: tonal alignment), etc. What is most interesting and relevant with respect to the topic of this chapter is the fact that these domain edges – beginnings in particular – have a significant effect on sound duration. More specifically, the length of segments or particular properties of segments is greater the higher up in the prosodic hierarchy their edges are (e.g. Fujimura 1990; Byrd et al. 2005). This has become known in the literature as initial articulatory strengthening or prosodic strengthening. It is important to note at this point, however, that most studies so far have been on sounds that are not in a quantity opposition. For example, in their seminal EPG study, Fougeron and Keating (1997) tested the English dummy syllable no in trisyllabic words with initial, medial, and final stress embedded in a carrier phrase. They found that the amount of linguopalatal contact as well as the acoustic duration of the syllable-initial nasal increased with each increasing prosodic level. Fougeron and Keating conclude that these “more extreme articulations” (1997: 3738) perform an important function for perception. On the one hand, they facilitate segmentation into higher prosodic domains, but particularly into words. On the other hand, they enhance the acoustic cues for identifying sounds and thus assist lexical access. Although the correlation between the amount of articulator contact and acoustic duration is reported as being only weak, it is there nonetheless. Because EPG studies are still few and far between, we will also use this acoustic finding as the main basis for our remaining discussion. In a phonetic study on real words, Cho and McQueen (2005) investigated wordinitial Dutch alveolar stops /t d/ in different phrasal and word-stress contexts. As expected, the closure durations of both voiced and voiceless stops were significantly longer in stronger prosodic positions. In addition, however, the VOTs of voiceless stops were cumulatively shorter in the same prosodic positions, i.e. the opposite of the findings for English /t/. Cho and McQueen argue in essence that the difference is due to the difference in how the phonological encoding of the voicing contrast is phonetically enhanced in the two languages. That is, Dutch /t/, as the phonologically unmarked member of the pair (voiceless vs. voiced), has the phonetic (default) specification [−spread glottis], which is enhanced by shorter aspiration; in contrast, English /t/, as the phonologically marked member of the pair (aspirated vs. non-aspirated), has the phonetic specification [+spread glottis], which is enhanced by longer aspiration. One of their main claims is thus that prosodic strengthening acoustically amplifies the difference between sounds in

249

145

195

94

##_V

V#_V

V_V

120 100

V#_V

V_V

20 0

##_V V#_V V_V Tashlhiyt Berber

##_V V#_V V_V Swiss German

29

40

43

60

47 45

80 43 42

VOT

##_V

114 101

64

V_V

134

164

V#_V

19 20

##_V

70

0

19 20

50

57

100

76

150

141

145

200

49

250

18

geminate singleton

20 19

b.

300 215

a.

Articulation duration

Initial Geminates

##_V V#_V V_V Cypriot Greek

Figure 47.7 Articulation/closure durationa and VOT in msec for initial geminates and singletons in phrase-initial (## __ V) and phrase-medial post-vocalic (V# __ V) context and medial geminates and singletons in intervocalic (V __ V) context in three languages: Tashlhiyt Berber (Ridouane 2007: 128–129), Swiss German (Kraehenmann and Lahiri 2008: 4450 –4451), and Cypriot Greek (Muller 2001: 29, 32) a

The measures of the medial geminates and singletons in Tashlhiyt Berber and Cypriot Greek are acoustic (closure duration); all others are articulatory (linguo-palatal contact)

phonological contrast. With respect to VOT effects, this claim seems to be supported by studies on Korean word-initial voiceless stops (Jun 1993; Cho and Keating 2001; Keating et al. 2003), which found that VOT measures were longest phrase initially, somewhat shorter word-initially within a phrase, and shortest wordmedially within a phrase. The question we want to ask now is whether articulatory strengthening also applies to word-initial geminates and singletons and, if so, whether it shows different characteristics. There are no studies comparable to Fougeron and Keating (1997) and their later work that directly address this issue in any depth. However, some generalizations can be drawn from the studies on voiceless stops in Cypriot Greek (Muller 2001), Tashlhiyt Berber (Ridouane 2007), and Swiss German (Kraehenmann and Lahiri 2008), by comparing medial vs. initial geminates and – in the latter two cases – initial geminates in different phrasal contexts. The measures we are therefore most interested in are articulatory/acoustic closure duration and VOT. To start with the latter, we would not expect VOT to become a differentiating characteristic for initial geminates and singletons if it does not play any role for the medial contrast. This is indeed what we find. As can be seen in Figure 47.7b, Tashlhiyt Berber VOTs are only slightly but not significantly

19

Astrid Kraehenmann

shorter when syllable, word, phrase, and utterance boundaries coincide (## __ V), compared to word-medially (V __ V). For the Swiss German data, the non-effect is even more pronounced, with basically identical measures in all three contexts. If anything had changed, longer measures would have been expected in the vocalic context (V# __ V) for initial stops and the longest ones at the phrase edge (## __ V), at least for geminates, if not for singletons. In contrast to these two languages, there are drastically longer measures in the Cypriot Greek data, where VOT is a secondary cue: geminate VOTs are longer by 71 msecs (265 percent), singleton VOTs by 72 msecs (348 percent). Although the 13 msecs difference in the phraseinitial context is still statistically significant, the differential has substantially decreased from 1.5 : 1 to 1.1 : 1, which goes against Cho and McQueen’s (2005) claim of contrast amplification at increasing prosodic domain edges. Unfortunately, Muller (2001) does not provide any data for the contact/closure duration in utterance-initial or vocalic context. Thus we do not know how the primary cue is affected by prosodic strengthening. This, however, is different for Tashlhiyt Berber and Swiss German (see Figure 47.7a). In Tashlhiyt Berber the trend seems to go in the expected direction in two of the three contexts. The geminate to singleton proportion is 2.5 : 1 word-medially and increases to 3 : 1 for word-initial stops after a vowel-final word. But at the utterance boundary, the highest prosodic domain, it decreases again to 2.8 : 1. In Swiss German we find exactly the opposite of what prosodic strengthening would predict, namely that the contrast magnitude decreases as the prosodic level increases. The difference between geminates and singletons is biggest in word-medial intervocalic context (syllable level, 3 : 1), somewhat smaller in word-initial intervocalic context (word level, 2 : 1), and smallest in utterance-initial context (phrase/utterance level, 1.5 : 1). To conclude, there are some strengthening effects in languages with wordinitial quantity contrast but only to the extent that the primary correlates lengthen substantially at the highest prosodic level, the utterance. However, this lengthening affects both geminates and singletons, resulting in contrast diminishment rather than augmentation. The segmental context, rather than the prosodic level, might therefore be the more reliable predictor of how the quantity contrast is realized, because intervocalic geminates – be they word-medial or phrase-medial word-initial – are actually not at a domain boundary, but contain one, because they straddle two syllables; see (6). The only time initial geminates occur at a domain edge is utterance-initially, which, as we have seen, is a very special context in many respects.

6

Conclusion

The field of word-initial quantity contrasts shows many faces and still awaits unveiling and scrutinizing investigations of a wealth of issues. A start has been made. Yet the existing disputes about phonological representation (length vs. weight), acoustic and articulatory properties, and perception are far from settled and promise to provide discussion material and theoretical arguments for years and decades to come. The historical question has only been hinted at in this chapter but constitutes another fertile ground and worthwhile area of research to be followed.

Initial Geminates

20

REFERENCES Abramson, Arthur S. 1986a. Distinctive length in initial consonants: Pattani Malay. Journal of the Acoustical Society of America 79. 527–527. Abramson, Arthur S. 1986b. The perception of word-initial consonant length: Pattani Malay. Journal of the International Phonetic Association 16. 8–16. Abramson, Arthur S. 1987. Word-initial consonant length in Pattani Malay. In Thomas V. Gamkrelidze (ed.) Proceedings of the 11th International Congress of Phonetic Sciences, vol. 6, 68–70. Tallinn: Academy of Sciences of the Estonian SSR. Abramson, Arthur S. 1991. Amplitude as cue to word-initial consonant length: Pattani Malay. Proceedings of the 12th International Congress of Phonetic Sciences, vol. 3, 98–101. Aix-en-Provence: Université de Provence. Abramson, Arthur S. 1999. Fundamental frequency as cue to word-initial consonant length: Pattani Malay. In Ohala et al. (1999), 591–594. Abramson, Arthur S. 2003. Acoustic cues to word-initial stop length in Pattani Malay. In Solé et al. (2003), 387–390. Baker, Brett. 2008. Word structure in Ngalakgan. Stanford: CSLI. Bender, Byron W., Ward H. Goodenough, Frederick H. Jackson, Jeffrey C. Marck, Kenneth L. Rehg, Ho-min Sohn, Stephen Trussel & Judith W. Wang. 2003. Proto-Micronesian reconstructions I. Oceanic Linguistics 42. 1–110. Broselow, Ellen. 1995. Skeletal positions and moras. In John A. Goldsmith (ed.) The handbook of phonological theory, 175–205. Cambridge, MA & Oxford: Blackwell. Broselow, Ellen, Su-I Chen & Marie Huffman. 1997. Syllable weight: Convergence of phonology and phonetics. Phonology 14. 47– 82. Byrd, Dani, Sungbok Lee, Daylen Riggs & Jason Adams. 2005. Interacting effects of syllable and phrase position on consonant articulation. Journal of the Acoustical Society of America 118. 3860–3873. Cho, Taehong & Patricia Keating. 2001. Articulatory and acoustic studies on domaininitial strengthening in Korean. Journal of Phonetics 29. 155–190. Cho, Taehong & James M. McQueen. 2005. Prosodic influences on consonant production in Dutch: Effects of prosodic boundaries, phrasal accent and lexical stress. Journal of Phonetics 33. 121–157. Clements, G. N. 1986. Compensatory lengthening and consonant gemination in LuGanda. In W. Leo Wetzels & Engin Sezer (eds.) Studies in compensatory lengthening, 37–77. Dordrecht: Foris. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Curtis, Emily. 2003. Geminate weight: Case studies and formal models. Ph.D. dissertation, University of Washington. Davis, Stuart. 1999a. On the representation of initial geminates. Phonology 16. 93–104. Davis, Stuart. 1999b. On the moraic representation of underlying geminates: Evidence from prosodic morphology. In René Kager, Harry van der Hulst & Wim Zonneveld (eds.) The prosody–morphology interface, 39–61. Cambridge: Cambridge University Press. Davis, Stuart. 2003. The controversy over geminates and syllable weight. In Féry & van de Vijver (2003), 77–98. Davis, Stuart. Forthcoming. Quantity. In John A. Goldsmith, Jason Riggle & Alan C. L. Yu (eds.) The handbook of phonological theory. 2nd edn. Malden, MA & Oxford: Wiley-Blackwell. Dmitrieva, Olga. 2009. Geminate typology and perception of consonant length: Experimental evidence from Russian. Paper presented at the 83rd Annual Meeting of the Linguistic Society of America, San Francisco. Enstrom, Daly & Sonja Spörri-Bütler. 1981. A voice onset time analysis of initial SwissGerman stops. Folia Phoniatrica 33. 137–150.

21

Astrid Kraehenmann

Féry, Caroline & Ruben van de Vijver (eds.) 2003. The syllable in Optimality Theory. Cambridge: Cambridge University Press. Fougeron, Cécile & Patricia Keating. 1997. Articulatory strengthening at edges of prosodic domains. Journal of the Acoustical Society of America 101. 3728–3740. Fujimura, Osamu. 1990. Methods and goals of speech production research. Language and Speech 33. 195–258. Fulop, Sean A. 1994. Acoustic correlates of the fortis/lenis contrast in Swiss German plosives. Calgary Working Papers in Linguistics 16. 55–63. Gordon, Matthew. 2005. A perceptually-driven account of onset-sensitive stress. Natural Language and Linguistics 23. 595–653. Ham, William. 1998. Phonetic and phonological aspects of geminate timing. Ph.D. dissertation, Cornell University. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce. 1999. Phonetically driven phonology: The role of Optimality Theory and inductive grounding. In Michael Darnell, Edith Moravcsik, Frederick Newmeyer, Michael Noonan & Kathleen Wheatley (eds.) Functionalism and formalism in linguistics, vol. 1: General papers, 243–285. Amsterdam & Philadelphia: John Benjamins. Hubbard, Kathleen. 1994. Duration in moraic theory. Ph.D. dissertation, University of California, Berkeley. Hume, Elizabeth, Jennifer Muller & Aone van Engelenhoven. 1997. Non-moraic geminates in Leti. Phonology 14. 371–402. Hyman, Larry M. 1985. A theory of phonological weight. Dordrecht: Foris. Jackson, Frederick H. 1984. The internal and external relationships of the Trukic languages of Micronesia. Ph.D. dissertation, University of Hawai’i. Jensen, John T. 1977. Yapese reference grammar. Honolulu: University Press of Hawaii. Jun, Sun-Ah. 1993. The phonetics and phonology of Korean prosody. Ph.D. dissertation, Ohio State University. Keating, Patricia, Taehong Cho, Cécile Fougeron & Chai-Shune Hsu. 2003. Domain initial articulatory strengthening in four languages. In John Local, Richard Ogden & Rosalind Temple (eds.) Phonetic interpretation: Papers in laboratory phonology VI, 145–163. Cambridge: Cambridge University Press. Kennedy, Robert. 2005. Reflexes of initial gemination in Western Micronesian languages. Paper presented at the 12th Meeting of the Austronesian Formal Linguistics Association, University of California, Los Angeles. Keyser, Samuel J. & Kenneth N. Stevens. 2006. Enhancement and overlap in the speech chain. Language 82. 33–63. Khattab, Ghada. 2007. A phonetic study of gemination in Lebanese Arabic. In Jürgen Trouvain & William J. Barry (eds.) Proceedings of the 16th International Congress of Phonetic Sciences, 153–158. Saarbrücken: Saarland University. Kiparsky, Paul. 2003. Syllables and moras in Arabic. In Féry & van de Vijver (2003), 147–182. Kraehenmann, Astrid. 2001. Swiss German stops: Geminates all over the word. Phonology 18. 109–145. Kraehenmann, Astrid. 2003. Quantity and prosodic asymmetries in Alemannic: Synchronic and diachronic perspectives. Berlin & New York: Mouton de Gruyter. Kraehenmann, Astrid. 2009. The perception of the word-initial quantity contrast in voiceless Swiss German stops. Paper presented at the 17th Manchester Phonology Meeting. Kraehenmann, Astrid & Aditi Lahiri. 2008. Duration differences in the articulation and acoustics of Swiss German word-initial geminate and singleton stops. Journal of the Acoustical Society of America 123. 4446–4455. Ladefoged, Peter. 2001. Vowels and consonants: An introduction to the sounds of languages. Malden, MA & Oxford: Blackwell.

Initial Geminates

22

Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’s languages. Oxford & Malden, MA: Blackwell. Lahiri, Aditi & Jorge Hankamer. 1988. The timing of geminate consonants. Journal of Phonetics 16. 327–338. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. Lewis, M. Paul (ed.) 2009. Ethnologue: Languages of the world. 16th edn. Dallas: SIL International. Available (August 2010) at www.ethnologue.com. Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press. Marti, Werner. 1985. Berndeutsch-Grammatik. Bern: Francke. McCarthy, John J. 1979. Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT. Mikuteit, Simone & Henning Reetz. 2007. Caught in the ACT: The timing of aspiration and voicing in East Bengali. Language and Speech 50. 247–277. Morén, Bruce. 1999. Distinctiveness, coercion and sonority: A unified theory of weight. Ph.D. dissertation, University of Maryland at College Park. Moulton, William G. 1979. Notker’s “Anlautgesetz.” In Irmengard Rauch & Gerald F. Carr (eds.) Linguistic method: Essays in honor of Herbert Penzl, 241–251. The Hague: Mouton. Muller, Jennifer. 1999. Geminate markedness: Evidence from Chuukese. Paper presented at the 6th Meeting of the Austronesian Formal Linguistics Association, University of Toronto. Muller, Jennifer. 2001. The phonology and phonetics of word-initial geminates. Ph.D. dissertation, Ohio State University. Muller, Jennifer. 2003. The production and perception of word initial geminates in Cypriot Greek. In Solé et al. (2003), 1867–1870. Ohala, John J. 1983. The origin of sound patterns in vocal tract constraints. In Peter F. MacNeilage (ed.) The production of speech, 189–216. New York: Springer. Ohala, John J., Yoko Hasegawa, Manjari Ohala, Daniel Granville & Ashlee C. Bailey (eds.) 1999. Proceedings of the 14th International Congress of Phonetic Sciences. Berkeley: University of California. Payne, Elinor & Eftychia Eftychiou. 2006. Prosodic shaping of consonant gemination in Cypriot Greek. Phonetica 63. 175–198. Ridouane, Rachid. 2007. Gemination in Tashlhiyt Berber: An acoustic and articulatory study. Journal of the International Phonetic Association 37. 119–142. Ringen, Catherine & Robert M. Vago. 2010. Geminates: Heavy or long? In Charles Cairns & Eric Raimy (eds.) Handbook of the syllable. Leiden: Brill. Selkirk, Elisabeth. 1990. A two root theory of length. University of Massachusetts Occasional Papers 14. 123–171. Sievers, Eduard. 1901. Grundzüge der Phonetik zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. 5th edn. Leipzig: Breitkopf & Härtel. Simpson, Adrian P. 1999. Fundamental problems in comparative phonetics and phonology: Does UPSID help to solve them? In Ohala et al. (1999), 349–352. Solé, M. J., D. Recasens & J. Romero (eds.) 2003. Proceedings of the 15th International Congress of Phonetic Sciences. Barcelona: Causal Productions. Stevens, Kenneth N. & Samuel J. Keyser. 1989. Primary features and their enhancement in consonants. Language 65. 81–106. Stevens, Kenneth N., Samuel J. Keyser & Haruko Kawasaki. 1986. Toward a phonetic and phonological theory of redundant features. In Joseph S. Perkell & Dennis H. Klatt (eds.) Invariance and variability in speech processes, 426–449. Hillsdale, NJ: Lawrence Erlbaum. Swadesh, Morris. 1937. The phonemic interpretation of long consonants. Language 13. 1–10. Thurgood, Graham. 1993. Geminates: A cross-linguistic examination. In Joel Ashmore Nevis, Gerald McMenamin & Graham Thurgood (eds.) Papers in honor of Frederick H. Brengelman on the occasion of the twenty-fifth anniversary of the Department of Linguistics, CSU Fresno, 129–139. Fresno: Department of Linguistics, California State University, Fresno.

23

Astrid Kraehenmann

Topintzi, Nina. 2008. On the existence of moraic onsets. Natural Language and Linguistic Theory 26. 147–184. Topintzi, Nina. 2010. Onsets: Suprasegmental and prosodic behaviour. Cambridge: Cambridge University Press. Tranel, Bernard. 1991. CVC light syllables, geminates and moraic theory. Phonology 8. 291–302. Trubetzkoy, Nikolai S. 1939. Grundzüge der Phonologie. Göttingen: van der Hoeck & Ruprecht. Translated 1969 by Christiane A. M. Baltaxe as Principles of phonology. Berkeley & Los Angeles: University of California Press. Tserdanelis, Georgios & Amalia Arvaniti. 2001. The acoustic characteristics of geminate consonants in Cypriot Greek. Proceedings of the 4th International Conference on Greek Linguistics, 29–36. Thessaloniki: University Studio Press. Yupho, Nawanit. 1989. Consonant clusters and stress rules in Pattani Malay. Mon-Khmer Studies 15. 125–137.

48

Stress-timed vs. Syllabletimed Languages Marina Nespor Mohinish Shukla Jacques Mehler

1

Introduction

Rhythm characterizes most natural phenomena: heartbeats have a rhythmic organization, and so do the waves of the sea, the alternation of day and night, and bird songs. Language is yet another natural phenomenon that is characterized by rhythm. What is rhythm? Is it possible to give a general enough definition of rhythm to include all the phenomena we just mentioned? The origin of the word rhythm is the Greek word osh[óp, derived from the verb oeí, which means ‘to flow’. We could say that rhythm determines the flow of different phenomena. Plato (The Laws, book II: 93) gave a very general – and in our opinion the most beautiful – definition of rhythm: “rhythm is order in movement.” In order to understand how rhythm is instantiated in different natural phenomena, including language, it is necessary to discover the elements responsible for it in each single case. Thus the question we address is: which elements establish order in linguistic rhythm, i.e. in the flow of speech?

2

The rhythmic hierarchy: Rhythm as alternation

Rhythm is hierarchical in nature in language, as it is in music. According to the metrical grid theory, i.e. the representation of linguistic rhythm within Generative Grammar (cf., amongst others, Liberman and Prince 1977; Prince 1983; Nespor and Vogel 1989; chapter 41: the representation of word stress), the element that “establishes order” in the flow of speech is stress: universally, stressed and unstressed positions alternate at different levels of the hierarchy (see chapter 39: stress: phonotactic and phonetic evidence). Two examples of stress alternation are given in (1) and (2), on the basis of Italian and English, respectively. The first level of the grid assigns a star (*) to each syllable, and is meant to represent an abstract notion of time; on the second, third, and fourth level, a star is assigned to every syllable bearing secondary word stress, primary word stress, and phonological phrase stress, respectively. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0048

Marina Nespor, Mohinish Shukla & Jacques Mehler

2 (1)

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Domani mattina partiremo presto con il barcone nuovo di Federico ‘Tomorrow morning we will leave early with the new boat of Federico’

(2)

* * * * * * * * * * * * * * Guinevere will arrive with

* * * * * * * * * * * * * ** * * * * * * * * * * * * Oliver tomorrow morning with a transatlantic

Indeed, these examples clearly show that in the two languages there is a similar alternation of stresses ranging from secondary word stress to primary word stress to phonological phrase stress. The level that is problematic in the metrical grid is the basic level, i.e. the level corresponding to the syllable. This representation does not show any alternation, or any element that establishes order in movement: if we restrict our attention to this level, all syllables are represented with equal prominence. It is clear, however, that grids that are identical at all levels, as in the two following Italian and English sentences, may represent very different rhythms. In particular, the first level – which represents an abstract notion of time for syllables – does not represent important differences between languages, precisely because it is abstract: simple syllables and very complex ones receive identical representations. (3)

a.

* * * * * * * * * * * * * * * * * * Domani Luca tornerà ‘Tomorrow Luca will return’

b.

* * * * * * * * * * * * * * * * * * Tomorrow Albert will return

There are thus empirical differences in rhythm between languages that are not represented in a metrical grid. Long before the metrical grid theory was proposed, phoneticians (e.g. Pike 1945) had proposed the existence of rhythmic classes to account for the rhythmic differences between languages like English or German, on the one hand, and languages like Spanish or Italian, on the other.

3

Linguistic rhythm as isochrony

The idea that languages have different rhythms was first advanced by Lloyd James (1940), who observed that the rhythm of Spanish recalls that of a machine gun

Stress-timed vs. Syllable-timed Languages

3

and that of English that of messages in the Morse code. Indeed, this is the same difference as we hear in sentences like (1) and (2), pronounced by native speakers of Italian and English, respectively. Subsequently, Pike (1945) in an attempt to provide empirical support for this dichotomy, proposed that this difference between Spanish and English was due to the requirement of isochrony at different levels. That is, languages would differ according to which chunks of speech must have similar durations, i.e. must be isochronous. The requirement of isochrony would hold between syllables in Spanish, and between interstress intervals in English. This proposal accounted for the fact that the syllables of Spanish or Italian, but not those of English or Dutch, are similar in quantity. Spanish, and languages with a similar rhythm, were thus referred to as syllable-timed, and languages with a rhythm similar to that of English as stress-timed. In subsequent work along the same lines, Abercrombie (1967) proposed that this was a general pattern of temporal organization for all languages of the world. Like Spanish and Italian, French, Telugu, and Yoruba are syllable-timed. And, like English, Russian and Arabic are stress-timed. A third rhythmic class was later added by Ladefoged (1975) to account for Japanese, whose rhythm differs both from that of English and from that of Spanish. According to Ladefoged, in Japanese, isochrony is maintained at the level of the mora, a sub-syllabic constituent that includes either onset and nucleus, or a coda. Japanese – and languages with a similar rhythm, e.g. Tamil – were thus characterized as mora-timed. In terms of metrical or prosodic phonology, this proposal amounts to establishing that, in the languages of the world, the requirement of isochrony holds at one of three phonological constituents: going from the smallest to the largest of the three, the mora ([), the syllable (q) or the foot. The three different types of isochrony are illustrated in (4). (4)

a. stress-timing Äq Äqq Äq Äq Äqq Äq b. syllable-timing CV CCVC CV CV CVC c. mora-timing [[ [[ [ CV V CV C CV q q

q

q

The different types of isochrony are mutually exclusive: isochrony of both syllables and feet would be possible only in an ideal language – to the best of our knowledge not attested – in which all the syllables were of the same type and in which secondary stresses were maximally alternating. Syllabic isochrony is also incompatible with mora isochrony: it would be compatible only in a language in which all syllables had the same number of moras. This case is attested, to the best of our knowledge, only in Hua, a language spoken in Botswana (Blevins 1995), and the West African language Senufo (Fasold and Connor Linton 2006), both reported to have only CV syllables.

4

Marina Nespor, Mohinish Shukla & Jacques Mehler

It is important to observe that this three-way distinction into rhythmic classes was not meant to deny the relevance of stress for either syllable- or mora-timed languages. Different levels of stress are, in fact, undeniable cross-linguistically. In terms of the phonology of rhythm, the distinction was meant to identify different rhythms exclusively at the basic level. According to this conception of rhythm, as isochrony maintained at one of three different levels, belonging to one or the other group would have consequences for the phonology of a language. For example, if feet are isochronous, the syllables of polysyllabic feet should be reduced in duration, while the only syllable of a monosyllabic foot should be stretched. The basic dichotomy between syllable- and stress-timing was largely taken for granted until various phoneticians, on the basis of measurements in different languages, showed that isochrony was not present in the signal. It was shown that interstress intervals in English vary in duration proportionally to the number of syllables they contain, so that the duration of the intervals between consecutive stresses is not constant (Shen and Peterson 1962; O’Connor 1965; Lea 1974). A similar result was obtained for Spanish: syllable duration was found to vary in proportion to the number of segments they contain. Interstress intervals, instead, were found to have similar durations, an unexplainable fact if isochrony is a characteristic of the syllabic rather than the foot level (Borzone de Manrique and Signorini 1983). Similarly, Dauer (1983), on the basis of an analysis of several syllable-timed languages (Spanish, Greek, and Italian), and of English as an example of a stress-timed language, concluded that the duration of interstress intervals does not differ across the different languages. Rather, the timing of stresses reflects universal properties of rhythmic organization. Similar conclusions are reached by den Os (1988): in a comparative study of Italian and Dutch utterances she showed that if the phonetic material of the two languages is kept similar – by selecting utterances in the two languages with an identical number of segments and syllables – their rhythm in terms of isochrony is similar. The difference in rhythm of “machine-gun” languages and “Morse-code” languages is, however, an undeniable fact. If isochrony between different types of constituents is not at the basis of this clear rhythmic difference, what are the factors responsible for it? Dauer (1983) observed that various phonological properties distinguish the two groups of languages: for example, “syllable-timed” languages have a smaller variety of syllable types than “stress-timed” languages, and they do not display vowel reduction (see chapter 79: reduction). These two characteristics are responsible for the fact that syllables in syllable-timed languages are more similar to each other in duration. In Spanish and French, for example, more than half of the syllables (by type frequency) consist of a consonant followed by a vowel (CV) (Dauer 1983). In Italian, 60 percent of the syllable types are CV (Bortolini 1976). The illusion of isochrony thus finds its origin in different phonological characteristics of languages and not in different temporal organizations. These considerations, together with the existence of languages that are neither clearly classifiable as syllable-timed nor as stress-timed, such as Catalan, European Portuguese, and Polish, led Nespor (1990) to draw the conclusion that there is no rhythm parameter. Establishing different rhythms as the cause – rather than the effect – of various phonological phenomena would in addition not account for the fact that very similar phenomena apply to eliminate arhythmic configurations in, for example, English and Italian. In both

Stress-timed vs. Syllable-timed Languages

5

languages, adjacent primary word stresses constitute a stress clash and are eliminated in much the same way (Liberman and Prince 1977; Nespor and Vogel 1979, 1989). That languages vary in their rhythm is a fact. However, from these studies it can be concluded that it is not different rhythms that trigger different phonological phenomena. Rather, different rhythms arise as a consequence of a series of independent phonological properties (cf. also Dasher and Bolinger 1982).

4

Infants’ sensitivity to rhythmic classes

Linguists were not alone in investigating rhythmic classes. The discovery in developmental psychology that newborns are capable of discriminating a switch from one language to another (Mehler et al. 1987; Mehler et al. 1988) triggered further experiments to explore which cues were responsible for this early human ability. In particular, the grouping of languages into different rhythmic classes attracted the attention of cognitive scientists interested in understanding how language develops in the infant’s brain. Mehler et al. (1996) relied on the classification of languages into syllable-timed, stress-timed, and mora-timed to advance a proposal as to how infants may access the phonological system of the language they are exposed to. In particular, they proposed that the rhythmic class of the language of exposure determines the unit exploited in the segmentation of connected speech: infants exposed to stress-timed languages would use the stress foot (that is, the interstress interval), those exposed to syllable-timed language the syllable and those exposed to a mora-timed language the mora (Cutler et al. 1986; Otake et al. 1993; Mehler et al. 1996). Most convincing are a number of experiments carried out with French newborns, which show that they are able to discriminate English from Japanese, but not English from Dutch, in low-pass filtered sentences, that is, in sentences whose segmental information is reduced, while prosodic information is largely preserved (Nazzi et al. 1998). In order to show that rhythm – rather than any other property of the test languages – is responsible for this discrimination ability, Nazzi et al. also tested newborns on a set of randomly intermixed English and Dutch sentences, and showed that they discriminate these from a set of randomly intermixed Spanish and Italian sentences. However, the discrimination ability disappeared when the newborns were tested on a set of English and Spanish sentences vs. a set of Italian and Dutch sentences. Thus the intuitions that many phoneticians shared about different rhythms in English and Italian, for example, are confirmed by newborns’ sensitivity to this distinction. It is thus clear that some physical property must be present in the signal to account for this difference, but until recently it has not been clear what this property was.

5

Rhythm as alternation at all levels

If isochrony is not responsible for the machine-gun and Morse-code effects, we should ask which characteristics in the signal are responsible for it. That is, what is there in the signal that would account for the clear rhythmic differences of languages belonging to different classes? Or what is the element that establishes

6

Marina Nespor, Mohinish Shukla & Jacques Mehler

order at this level? Ramus et al. (1999) answered this question starting from the hypothesis that newborns hear speech as a sequence of vowels interrupted by unanalyzed noise, i.e. consonants; this hypothesis is known as the Time-Intensity Grid Representation (TIGRE; Mehler et al. 1996). Ramus et al. (1999) proposed that, at the basic level, the perception of different rhythms is created by the way in which vowels alternate with consonants. It is thus the regularity with which vowels recur that establishes alternation at this level: vowels alternate with consonants. Starting from the observation that as we go from stress-timed to syllable-timed and then to mora-timed languages the syllabic structure tends to get simpler – and the observation that simple syllables imply the presence of proportionately greater vocalic spaces – vowels would occupy less time in the flow of speech in stress-timed languages than syllable-timed languages. Likewise, in syllabletimed languages vowels would occupy less time than in mora-timed languages, which have the largest amount of time per utterance occupied by vowels. This difference is clear from the rough division into Vs and Cs in the three sentences in (5)–(7). Notice that, in agreement with Ramus et al. (1999), glides are treated as Cs if prevocalic and as Vs if post-vocalic. (5)

English The next local elections will take place during the winter CVCVCCCCVVCVCVCVCCCVCCCVCCVVCCVVCCVCVCCV

(6)

Italian Le prossime elezioni locali avranno luogo in inverno CVCCVCCVCVVCVCCVCVCVCVCVVCCVCCVCCVCVVCVCCVCCV

(7) Japanese Tsugi no chiho senkyo wa haruni okonawareru daro CVCVCVCVCVCVCCVVCVCVCVCVVCVCVCVCVCVCV Ramus et al. (1999) tested this idea on a corpus of eight languages: English, Polish, and Dutch, representatives of the stress-timed category, French, Italian, Spanish, and Catalan, representatives of the syllable-timed category (Abercrombie 1967), and Japanese, representative of the mora-timed languages (Ladefoged 1975). They observed that languages from the same rhythmic class had similar values for %V – i.e. a similar amount of time occupied by vowels in the speech stream – as compared to languages from different rhythmic classes. The computation of %V was carried out on the basis of a careful segmentation, on the basis of both auditory and visual cues from the spectrogram (cf. Ramus et al. 1999). Given the assumption that newborns do not retain the difference between individual Cs and individual Vs, for each sentence only the vocalic and consonantal intervals were measured. Adjacent vowels and adjacent consonants are thus treated as vocalic and consonantal chunks, respectively. A second measure that clusters the languages into three groups is the standard deviation of the duration of consonantal intervals (DC), i.e. a broad measure of the regularity with which vowels recur (see Figure 48.1). Both measures are related to syllable structure. A high %V implies that the repertoire of the possible syllable types is restricted, thus also that the consonantal intervals do not vary a great deal, given that there are no languages in which all

Stress-timed vs. Syllable-timed Languages

7

0.06 English Dutch Turkish

Italian

0.05 DC

Polish Spanish

Marathi

French

0.04

Catalan Hungarian Finnish Basque

0.03 38

40

42

44

46 48 %V

50

Tamil Japanese

52

54

Figure 48.1 DC, the standard deviation of the consonantal intervals, vs. %V, the amount of time per utterance spent in vowels, for 14 languages. The widths of the ellipses along the two axes represent standard errors of the mean along the axes. Dark ellipses represent head-initial languages, and light ellipses head-final languages. Turkish, Hungarian, Basque, Finnish, Marathi, and Tamil are from unpublished results. Data for the remaining languages are from Ramus et al. (1999)

syllables are complex. Rather, even in languages with the greatest variety of syllable types, the basic syllable type – CV – is the most unmarked (Blevins 1995; Rice 2007). Thus, according to this proposal, rhythm is alternation at all levels: of consonants and vowels at the basic level and of stressed and unstressed syllables, feet, and words at subsequent levels. This order in the flow of speech is always established by the alternation of more and less audible elements.

6

Other proposals

The analysis proposed by Ramus et al. (1999) is not the only one to rely on a purely acoustic-phonetic description of the speech stream in trying to understand the basis of linguistic rhythm. In Ramus et al. the %V and DC variables do not consider the relative ordering of long and short intervals inside an utterance. That is, sequences like CCV(.CCV(.CV.CV and CCV(.CV.CV.CCV( (where V( is a long vowel, as opposed to VV, which denotes two different adjacent vowels) will yield identical values for their two variables. Grabe and colleagues therefore chose to examine the pairwise variability indices (PVI) in the vocalic and intervocalic intervals in speech (e.g. Low et al. 2000; Grabe and Low 2002). This measure is meant to capture a little more of the (local) temporal patterns in speech by considering the variability of all pairs of vocalic or intervocalic intervals.

Marina Nespor, Mohinish Shukla & Jacques Mehler

8

The raw pairwise variability index is given by the formula in (8): (8)

m−1 G rPVI = ∑ |dk − dk+1 |/(m − 1)J I k=1 L

where m is the number of intervals, and dk is the duration of the k th interval. Thus, for example, in a language with only simple syllable types, the average durational variability between successive consonantal intervals will be less than in a language with many syllable types. The former language will therefore have a lower consonantal PVI than the latter. These authors showed that also the PVI of the vocalic and intervocalic intervals in speech captures some of the rhythmic distinctions between languages. Thus, as in Ramus et al., these authors too find a measure – the pairwise variability of the vocalic intervals – that separates stress-timed languages (higher PVI) from syllable-timed languages (lower PVI). However, there remain some discrepancies between these measures and those of Ramus et al. For example, while %V and DC clearly separate Japanese (“mora-timed”) from the “syllable-timed” languages, this difference is not apparent with PVIs. Since one of the purposes of a theory of grammar in general and of rhythm in particular is to account for first-language acquisition (cf. §7 below), infants’ discrimination or lack of it between a “syllabletimed” and a “mora-timed” language would help decide which of the two theories makes the correct prediction. The fact that Nazzi et al. (2000) found that 5-monthold infants raised in an American English environment discriminate Japanese from Italian leads us to prefer the %V and DC proposal. Both the analysis proposed by Ramus et al. and that proposed by Grabe and Low start with an initial categorization of the speech stream into vocalic and intervocalic intervals. A different approach is proposed by Galves et al. (2002), who try to avoid any analysis into categories (e.g. vowels and consonants). Instead, they use a measure of sonority, as estimated directly from the spectrogram. In particular, they use a procedure that maps regions of the spectrum as more or less stable (i.e. constant) over short periods of time, as measured by the entropy from one time-slice to the next. This is a fully automatic procedure, and the value for each set of time-slices of the spectrogram goes from 0 to 1. Values close to 1 correspond to a regular spectrum with little variation (low entropy), typical of sonorants, and values close to 0 reflect noisy spectra (high entropy), as might be expected for obstruents. These authors then consider measures of mean and variation in the “sonority” of time-slices as analogues of Ramus et al.’s measures, i.e. %V and DC. This proposal has the great advantage of being based on an automated method. And the authors succeed in roughly replicating the observation of Ramus et al. on their corpus. They can segregate the stress-, syllable-, and moratimed languages in a similar, though less precise manner. An alternative attempt to automatically determine the rhythmic class of a language has been elaborated in Singhvi et al. (in progress). It consists of algorithms for vowel and consonant recognition based on a variety of acoustic features, that allow one to compute %V and DC directly from the speech stream. As noted earlier (§3), linguistic rhythm does not correspond to isochrony at the level of different phonological units in speech. In order to nevertheless capture the intuition of rhythmicity, O’Dell and Nieminen (1999) propose a coupled-oscillator model for speech rhythm. In this model, a lack of overt isochrony is seen as a result of a tension between two rhythmic oscillators, one for stress-groups (roughly, feet)

Stress-timed vs. Syllable-timed Languages

9

and one for syllables. In a general mathematical model of two coupled oscillators, a single parameter determines which of the two oscillators (e.g. the foot or the syllable oscillator) is dominant. It turns out that this parameter can be directly estimated from speech: for stress-timed languages it is large (>1, corresponding to a dominant foot oscillator), and for syllable-timed languages it is small (≤1, corresponding to a dominant syllable oscillator). Thus, in this approach, the idea is that simple, overt isochrony in speech might be constrained by the requirement for temporal coordination across hierarchically organized phonological units in speech (see also Cummins and Port 1998).

7

Rhythm and related properties of grammar: Implications for language acquisition

Given infants’ sensitivity to basic rhythmic properties of languages, as proposed above, the issue must be addressed of the sort of implications that this sensitivity could have for language acquisition. The questions that should be addressed are: do the different rhythmic cues reflect specific grammatical properties? And, might the infant be able to use such cues to bootstrap the related properties? Although there is no one-to-one mapping between phonology and syntax, the two are correlated (cf., amongst many others, Selkirk 1984; Nespor and Vogel 1986, 2008; Morgan and Demuth 1996). On the one hand, it has been shown that the acoustic correlates of prosodic phenomena that signal phonological constituency can allow disambiguation of otherwise ambiguous sequences of words (Cooper and Paccia-Cooper 1980; Nespor and Vogel 1986, 2008; Price et al. 1991). On the other hand, typologists have documented several aspects of phonology and syntax that go together (e.g. Whaley 1997). Indeed, typological studies have revealed a wealth of correlations between different aspects of language, such as morphology, phonology, and syntax. Thus, Greenberg (e.g. 1963) observes that whether a language is VO (verb–object) or OV (object–verb) is correlated with many other grammatical properties of the language. For example, he notes that verb-final languages almost always have a case system. In addition, Koster (1999) observes that most OV languages have flexible word order. In a typological study, Donegan and Stampe (1983) suggest that languages with simple syllabic structures tend to be verb-final. Similarly, Fenk-Oczlon and Fenk (2005) find that languages with simple syllables tend to have postpositions (i.e. to be OV) and richer case systems. Several authors have proposed functional explanations for these observed correlations (e.g. Comrie 1981; Cutler et al. 1985; DuBois 1987; Hawkins 1988; Fenk-Oczlon and Fenk 2004, amongst others). From the point of view of acquisition, these correlations suggest that any cues in the input that lead to the acquisition of a single property might also provide cues and biases for acquiring all the (functionally) related properties. Thus, a cue to a phonological property might also provide cues to morphology and syntax. Indeed, there have been some concrete proposals for how phonology might allow infants to bootstrap a basic syntactic property like word order (Nespor et al. 1996; Christophe et al. 1997; Nespor et al. 2008). Given that newborns show great sensitivity to rhythmic classes, we can speculate that this ability might be useful in bootstrapping various properties of their target language. As we saw in §5, languages from the different rhythmic

10

Marina Nespor, Mohinish Shukla & Jacques Mehler

classes differ in their syllabic structure: going from a low %V to a high %V, languages go from having more complex to having simpler syllabic structures. Typologists have in fact observed that various morphosyntactic properties are correlated with the complexity of syllables in a language (Gil 1986; Fenk-Oczlon and Fenk 2004) and, in addition, with its rhythmic patterns (Donegan and Stampe 1983). The computation of %V might therefore offer cues to very different properties of the language of exposure. Shukla et al. (in progress) hypothesize that the correlates of linguistic rhythm, %V and DC, have consequences for acquiring correlated morphosyntactic properties like agglutination and word order. These researchers extend the results from Ramus et al. to a larger and more varied set of languages. The results indicate that there is a tendency for languages with a low %V to differ from languages with a high %V in head direction, degree of agglutination, richness of the case system, and flexibility of word order (see Figure 48.1). Thus, it is proposed that a simple syllabic structure is correlated with agglutination: if many suffixes can be attached to a word, complex syllabic structure would make these words excessively long and possibly hard to parse. The question remains why agglutination is found almost exclusively in head-final languages. Two different reasons, both syntactic in nature, have been given. In van Riemsdijk (1998), the explanation for the correlation between headfinality and agglutinative morphology is based on head adjunction, the syntactic device that assembles independent, phonetically realized morphemes in complex words. A principle states that head adjunction can take place only between linearly adjacent heads; since heads are adjacent in head-final languages, while they are separated by intervening specifiers in head-initial languages, head adjunction – and thus agglutination – is expected to take place in OV languages only. More recently, Cecchetto (forthcoming) assumes that morphological conflation, responsible for fusional morphology, requires that a direct syntactic dependency be established between a selecting head and a selected one. However, in headfinal languages this dependency would go backwards, since the selecting head linearly follows the selected one, and backward dependencies are disfavored, for processing reasons (e.g. Fodor 1978). As a consequence, in head-final languages affixes cannot be fused, and result in agglutination instead. If there is indeed a syntactic explanation for the correlation between head direction and agglutination, the identification of the rhythmic class of the language of exposure would be one of the mechanisms that would assist the infant in the bootstrapping of both the favored morphological operations and word order in the language to which they are exposed.

8

Concluding remarks

In conclusion, linguists’ intuitive notion of stress-timed and syllable-timed rhythm is most likely a consequence of the phonological organization of different languages, which can be captured by two relatively simple acoustic-phonetic cues, such as %V and DC. Languages appear to be grouped into three rhythmic classes: one corresponding to so-called stress-timed languages, one to so-called syllable-timed languages, and one to so-called mora-timed languages. The rhythmic class to which a language belongs appears also to determine the segmentation unit used by its

Stress-timed vs. Syllable-timed Languages

11

native speakers. Rhythm tends also to be correlated with a constellation of phonological, morphological, and syntactic properties of the language. The observation that newborns segregate languages on the basis of their rhythmic class suggests that this ability may be utilized in the acquisition of various such properties of their target language directly from the input.

REFERENCES Abercrombie, David. 1967. Elements of general phonetics. Edinburgh: Edinburgh University Press. Blevins, Juliette. 1995. The syllable in phonological theory. In John A. Goldsmith (ed.) The handbook of phonological theory, 206–244. Cambridge, MA & Oxford: Blackwell. Bortolini, Umberta. 1976. Tipologia sillabica dell’Italiano: Studio statistico. In Studi di Fonetica e Fonologia, vol. 1, 2–22. Roma: Bulzoni Editore. Borzone de Manrique, Ana Maria & Angela Signorini. 1983. Segmental durations and the rhythm in Spanish. Journal of Phonetics 11. 117–128. Cecchetto, Carlo. Forthcoming. Backwards dependencies must be short: A unified account of the Final-over-Final and the Right Roof Constraints and its consequences for the syntax/morphology interface. In Theresa Biberauer & Ian G. Roberts (eds.) Challenges to linearization. Berlin & New York: Mouton de Gruyter. Christophe, Anne, Marina Nespor, Maria-Teresa Guasti & Brit van Ooyen. 1997. Reflections on phonological bootstrapping: Its role in lexical and syntactic acquisition. In Gerry T. M. Altmann (ed.) Cognitive models of speech processing: A special issue of language and cognitive processes, 585 –612. Mahwah, NJ: Lawrence Erlbaum. Comrie, Bernard. 1981. Language universals and linguistic typology. Oxford: Blackwell. Cooper, William E. & Jeanne Paccia-Cooper. 1980. Syntax and speech. Cambridge, MA: Harvard University Press. Cummins, Fred & Robert F. Port. 1998. Rhythmic constraints on stress timing in English. Journal of Phonetics 26. 145–171. Cutler, Anne, John A. Hawkins & Gary Gilligan. 1985. The suffixing preference: A processing explanation. Linguistics 23. 723 –758. Cutler, Anne, Jacques Mehler, Dennis Norris & Juan Segui. 1986. The syllable’s differing role in the segmentation of French and English. Journal of Memory and Language 25. 385–400. Dasher, Richard & David Bolinger. 1982. On pre-accentual lengthening. Journal of the International Phonetic Association 12. 58 –71. Dauer, Rebecca M. 1983. Stress-timing and syllable-timing reanalyzed. Journal of Phonetics 11. 51– 62. Donegan, Patricia J. & David Stampe. 1983. Rhythm and the holistic organization of language structure. Papers from the Annual Regional Meeting, Chicago Linguistic Society: Parasession on the interplay of phonology, morphology and syntax, 337–353. DuBois, John. 1987. The discourse basis of ergativity. Language 64. 805 –855. Fasold, Ralf & Jeffrey Connor Linton. 2006. An introduction to language and linguistics. Cambridge: Cambridge University Press. Fenk-Oczlon, Gertraud & August Fenk. 2004. Systemic typology and crosslinguistic regularities. In Valery Solovyev & Vladimir Polyakov (eds.) Text processing and cognitive technologies, 229 –234. Moscow: MISA. Fenk-Oczlon, Gertraud & August Fenk. 2005. Crosslinguistic correlations between size of syllables, number of cases, and adposition order. In Gertraud Fenk-Oczlon & Christian Winkler (eds.) Sprache und Natürlichkeit: Gedenkband für Willi Mayerthaler, 75–86. Tübingen: Gunther Narr. Fodor, Jerry. 1978. Parsing strategies and constraints on transformations. Linguistic Inquiry 9. 427–473.

12

Marina Nespor, Mohinish Shukla & Jacques Mehler

Galves, Antonio, Jesus Garcia, Denise Duarte & Charlotte Galves. 2002. Sonority as a basis for rhythmic class discrimination. In Bernard Bel & Isabel Marlien (eds.) Proceedings of the speech prosody 2002 conference, 323–326. Aix-en-Provence: Laboratoire Parole et Langage. Gil, David. 1986. A prosodic typology of language. Folia Linguistica 20. 165 –231. Grabe, Esther & Ee Ling Low. 2002. Durational variability in speech and the rhythm class hypothesis. In Carlos Gussenhoven & Natasha Warner (eds.) Laboratory phonology 7, 377–401. Berlin & New York: Mouton de Gruyter. Greenberg, Joseph H. 1963. Some universals of grammar with particular reference to the order of meaningful elements. In Joseph H. Greenberg (ed.) Universals of language, 73–113. Cambridge, MA: MIT Press. Hawkins, John A. 1988. Explaining language universals. In John A. Hawkins (ed.) Explaining language universals, 3 –28. Oxford: Blackwell. Koster, Jan. 1999. The word orders of English and Dutch: Collective vs. individual checking. In Werner Abraham (ed.) Groninger Arbeiten zur germanistischen Linguistik, 1–42. Groningen: University of Groningen. Ladefoged, Peter. 1975. A course in phonetics. New York: Harcourt Brace Jovanovich. Lea, Wayne A. 1974. Prosodic aids to speech recognition, vol. 4: A general strategy for prosodicallyguided speech understanding. St Paul, MN: Sperry Univac. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. Lloyd James, Arthur. 1940. Speech signals in telephony. London: Pitman & Sons. Low, Ee Ling, Esther Grabe & Francis Nolan. 2000. Quantitative characterizations of speech rhythm: Syllable-timing in Singapore English. Language and Speech 43. 377–401. Mehler, Jacques, Ghislaine Lambertz, Peter Jusczyk & Claudine Amiel-Tison. 1987. Discrimination de la langue maternelle par le nouveau-né. C. R. Academie des Sciences de Paris 303, 15. 637–640. Mehler, Jacques, Peter Jusczyk, Ghislaine Lambertz, Nilofar Halsted, Josiane Bertoncini & Claudine Amiel-Tison. 1988. A precursor of language acquisition in young infants. Cognition 29. 143 –178. Mehler, Jacques, Emmanuel Dupoux, Thierry Nazzi & Ghislaine Dehaene-Lambertz. 1996. Coping with linguistic diversity: The infant’s viewpoint. In Morgan & Demuth (1996), 101–116. Morgan, James L. & Katherine E. Demuth (eds.) 1996. Signal to syntax: Bootstrapping from speech to grammar in early acquisition. Mahwah, NJ: Lawrence Erlbaum. Nazzi, Thierry, Josiane Bertoncini & Jacques Mehler. 1998. Language discrimination by newborns: toward an understanding of the role of rhythm. Journal of Experimental Psychology: Human Perception and Performance 24. 756 –766. Nazzi, Thierry, Peter Jusczyk & Elizabeth K. Johnson. 2000. Language discrimination by English learning 5-month-olds: Effects of rhythm and familiarity. Journal of Memory and Language 43. 1–19. Nespor, Marina. 1990. On the rhythm parameter in phonology. In Iggy Roca (ed.) The logical problem of language acquisition, 157–175. Dordrecht: Foris. Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Nespor, Marina, Maria-Teresa Guasti & Anne Christophe. 1996. Selecting word order: The rhythmic activation principle. In Ursula Kleinhenz (ed.) Interfaces in phonology, 1–26. Berlin: Akademie Verlag. Nespor, Marina, Mohinish Shukla, Ruben van de Vijver, Cinzia Avesani, Hanna Schraudolf & Caterina Donati. 2008. Different phrasal prominence realization in VO and OV languages. Lingue e Linguaggio 7(2). 1–28. Nespor, Marina & Irene Vogel. 1979. Clash avoidance in Italian. Linguistic Inquiry 10. 467–482. Nespor, Marina & Irene Vogel. 1989. On clashes and lapses. Phonology 6. 69 –116. Nespor, Marina & Irene Vogel. 2008. Prosodic phonology. Berlin & New York: Mouton de Gruyter. 1st edn, 1986. Dordrecht: Foris.

Stress-timed vs. Syllable-timed Languages

13

O’Connor, J. D. 1965. The perception of time intervals. London: University College London. O’Dell, Michael L. & Tommi Nieminen. 1999. Coupled oscillator model of speech rhythm. In John J. Ohala, Yoko Hasegawa, Manjari Ohala, Daniel Granville & Ashlee C. Bailey (eds.) Proceedings of the 14th International Congress of Phonetic Sciences, 1075–1078. Berkeley: Department of Linguistics, University of California, Berkeley. Os, Els den. 1988. Rhythm and tempo in Dutch and Italian: A contrastive study. Ph.D. dissertation, University of Utrecht. Otake, Takashi, Giyoo Hatano, Anne Cutler & Jacques Mehler. 1993. Mora or syllable? Speech segmentation in Japanese. Journal of Memory and Language 32. 258 –278. Pike, Kenneth L. 1945. The intonation of American English. Ann Arbor: University of Michigan Press. Plato. The Laws, book II. 93. Price, P. J., M. Ostendorf, S. Shattuck-Hufnagel & C. Fong. 1991. The use of prosody in syntactic disambiguation. Journal of the Acoustical Society of America 90. 2956–2970. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19 –100. Ramus, Franck, Marina Nespor & Jacques Mehler. 1999. Correlates of linguistic rhythm in the speech signal. Cognition 73. 265 –292. Rice, Keren. 2007. Markedness in phonology. In Paul de Lacy (ed.) The Cambridge handbook of phonology, 79 –97. Cambridge: Cambridge University Press. Riemsdijk, Henk van. 1998. Head movement and adjacency. Natural Language and Linguistic Theory 16. 633 –678. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Shen, Yao & Giles G. Peterson. 1962. Isochronism in English. Occasional Papers, University of Buffalo Studies in Linguistics 9. 1–36. Shukla, Mohinish, Marina Nespor & Jacques Mehler. In progress. Grammar on a language map. Singhvi, Abhinav & David M. Gomez. In progress. Automatic recognition of language rhythm. Whaley, Lindsay J. 1997. Introduction to typology: The unity and diversity of language. Thousand Oaks, CA: Sage.

49

Sonority Steve Parker

1

Introduction

If an interface between phonetics and phonology really exists (pace Ohala 1990b), then one topic having a long and controversial history in that domain is sonority. Sonority can be defined as a unique type of relative, n-ary (non-binary) featurelike phonological element that potentially categorizes all speech sounds into a hierarchical scale. For example, vowels are more sonorous than liquids, which are higher in sonority than nasals, with obstruents being the least sonorous of all segments. In terms of traditional phonetic systems for categorizing natural classes of sounds, then, the feature encoded by sonority most closely corresponds to the notion manner of articulation (see chapter 13: the stricture features). In this sense, sonority is like most other features: it demarcates groups of segments that behave similarly in cross-linguistically common processes. At the same time, however, sonority is unlike most features in that it exhaustively encompasses all speech sounds simultaneously, i.e. every type of segment has some inherent incremental value for this feature. Sonority is also unique in that it has never been observed to spread (assimilate), in and of itself. A major function of sonority is to organize (order) segments within syllables. Specifically, more sonorous sounds, such as vowels, tend to occur in the nucleus, while less sonorous sounds normally appear in the marginal (non-peak) positions – onsets and codas. This concept has engendered several chronic and frequently discussed research questions: (1)

a. b.

What role, if any, does sonority play in Universal Grammar? How many and what kinds of natural class distinctions need to be made in the sonority hierarchy? c. Are its rankings fixed or permutable (reversible)? d. Which distinctions in the sonority scale, if any, are universal and which, if any, are language-particular? e. Is sonority an abstract phonological mechanism only, or does it also have a consistent, measurable phonetic basis?

To answer (1e) briefly, the main acoustic correlate of sonority is intensity. As Ladefoged (1975: 219) notes, “The sonority of a sound is its loudness relative to The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0049

Sonority

2

that of other sounds with the same length, stress, and pitch.” Nevertheless, although much progress has been made in addressing the issues in (1), little consensus has emerged in understanding many of them. This chapter touches on each of the questions in (1), although not necessarily in the same order or to the same degree. The goal is to summarize the debates and document the types of empirical data that have been presented in arguing for the different positions. This chapter is organized as follows: §2 reviews the cross-linguistic phonological evidence for sonority. Thus §2.1 discusses the Sonority Sequencing Principle, §2.2 Minimum Sonority Distance effects, §2.3 the Syllable Contact Law, and §2.4 the contribution of sonority to the relative weight of the rhyme. §3 describes the Sonority Dispersion Principle, while §4 presents several desirable characteristics that a complete sonority hierarchy should ideally display. Finally, §5 examines the physical basis of sonority, as demonstrated experimentally.

2

Phonological evidence for sonority

This section describes various phonological phenomena that demonstrate that sonority is active in many languages. The exposition summarizes works discussing the issues in more depth, such as Parker (2002) and Cser (2003). Phonotactic constraints and morphophonemic alternations provide the most compelling evidence for establishing the divisions in the sonority hierarchy. Consequently, most of the argumentation here relies on these two factors. Several patterns are attested in enough languages to motivate the hypothesis that some notion of sonority should be considered part of Universal Grammar (UG; the innate linguistic faculty shared by all humans; Chomsky and Halle 1968; Kenstowicz 1994). How sonority is best expressed in UG is a separate topic, not discussed in detail here. In many cases, opposing points of view exist, with some linguists denying that sonority is actually involved in these phenomena. See e.g. Ohala (1974, 1990a, 1990b) and Kawasaki (1982) for arguments against appealing to sonority as an explanation for these data.

2.1

The Sonority Sequencing Principle

The domain in which sonority is most often invoked is the syllable, and related notions such as permissible consonant clusters in onset or coda position. This reflects the analogy that the syllable is like a wave of energy (Sievers 1893; Pike 1943). Specifically, syllables universally tend to abide by the following constraint: (2)

Every syllable exhibits exactly one peak of sonority, contained in the nucleus.

This is known as the Sonority Sequencing Principle (SSP) or the Sonority Sequencing Generalization. Key works assuming this principle as a basis for analysis include Hooper (1976), Selkirk (1984), Blevins (1995), and, in Optimality Theory, Cho and King (2003) and Zec (2007). Ohala (1990a, 1990b) and Wright (2004) note that a rudimentary notion of the SSP is observed in the work of de Brosses (1765). For the purpose of formally encoding and testing the SSP, the most frequently cited sonority scale is the following:

Steve Parker

3

Table 49.1 Cross-linguistic variation in syllabic segments based on sonority Vowels

Liquids

Nasals

Obstruents

✓ ✓ ✓ ✓

– ✓ ✓ ✓

– – ✓ ✓

– – – ✓

Bulgarian, Hawaiian, Kabardian, Latin, Spanish Lendu, Sanskrit, Slovak English (Central) Carrier, (Imdlawn) Tachelhit (Berber)

(3)

Modal sonority hierarchy (e.g. Clements 1990; Kenstowicz 1994; Smolensky 1995) vowels

> glides > liquids > nasals

higher in sonority

> obstruents

lower in sonority

In terms of sonority, the five natural classes in (3) are the easiest ones to motivate and the most useful ones to employ. Assuming the SSP and the hierarchy in (3), hypothetical syllables like [ta], [k7u], [wos], and [phlænt] are well formed, since their sonority slope uniformly rises from the beginning of the syllable to the nuclear vowel, and falls from the nucleus to the end of the syllable. Conversely, syllables containing “sonority reversals” such as [lpa] and [odm] violate (2) and are therefore illicit in most languages. One argument for the SSP is that cross-linguistically the inventory of [+syllabic] segments in particular languages normally forms a continuous range based on a scale like (3). Thus, the propensity for a sound to occur in nuclear position is correlated with how sonorous it is. The typology of permissible syllabic segments across languages is illustrated in Table 49.1, adapted from Blevins (1995) and Zec (2007).1 The generalization is that if a language permits syllabic segments from a lower sonority class, it also allows nuclei from all higher sonority classes. In Tachelhit even voiceless stops occur in nuclear position. However, glides are omitted here since by definition they are non-nuclear. The following example lists forms containing syllabic consonants from two of these languages, where [.] marks a syllable boundary (Parker 2002; Zec 2007; Ridouane 2008).2 (4)

a.

Slovak [ky.vi] [v=.ka]

b. ‘blood’ ‘wolf’

Tachelhit [t».dxt] [ty.g=t] [tv.tw.tztt]

‘gather wood’ ‘lock’ ‘you sprained it (fem)’

Nevertheless, while the pattern in Table 49.1 is a strong tendency, it is not universally obeyed. For example, many languages (especially in Africa) attest syllabic

1

Language names and genetic affiliations follow the Ethnologue (Lewis 2009). In the online version of this chapter, the appendix provides more details about the languages cited here: country, linguistic phylum, primary source of data, etc. 2 The online version of this chapter contains more illustrative data throughout.

Sonority

4

nasals but not syllabic liquids: e.g. Djeebbana, Lele (Chad), and Swahili (Blevins 2006). Thus, factors other than sonority must also be appealed to in some cases. Besides syllabic consonants, another reason to adopt the SSP is that it accounts for tautosyllabic consonant clusters in most languages. The following example illustrates three languages that strictly follow the SSP in onsets and codas (Blevins 2006): (5)

a.

b.

c.

Cheke Holo [kai.ka.fli] [kmai.kma.ji] Djeebbana [è.ka.la] [kalk.bet] Spanish [plan] [t7ans.k7i.ßi7]

‘flash on and off’ ‘eat a varied meal (reduplicated)’ ‘fork’ ‘northern black wallaroo’ ‘plan’ ‘to transcribe’

Again there are exceptions; (6) shows data from two languages in which the underlined consonant clusters apparently violate the SSP (Blevins 2006): (6)

a.

Leti (Indonesia) [pni.nu] [sra(t] [rka(.lu] [rstp.le]

‘fool’ ‘main road’ ‘they shout’ ‘they sail’

b.

Yir Yoront [melt] [patl]

‘animal, bird’ ‘clean, bald’

Counterexamples to the SSP also occur in some Indo-European languages, such as Czech, Romanian, and Russian. Extreme cases are found in Georgian, a Kartvelian language attesting the word-initial clusters /zrd/, /mkrt/, /msøv/ and /m[vrtn/ (Blevins 2006). Such exceptions have led some researchers to conclude that analyses based on sonority are circular in nature (Ohala 1990b: 160). However, no studies exist in which the proportion of languages with sonority reversals is tabulated among a statistically reliable and balanced sample. Therefore, based on available data, it seems safe to conclude that a large percentage of the world’s languages do conform to the SSP. Furthermore, purported counterexamples like Georgian /m[vrtn/ are dubious if the onset cluster in question occurs only word-initially, but not word-internally. This is crucial since, in a given language, a consonant cluster should appear in a position other than at a word edge in order to count as a canonical syllable type. Otherwise, when a greater number of consonants show up next to a word boundary, it is debatable whether this constitutes a true syllable margin. A more principled explanation is to analyze the SSP-violating segment(s) as a degenerate syllable or an extrasyllabic appendix licensed by the prosodic word. See Cho and King (2003) for further discussion. Finally, some alleged SSP violations cannot withstand further scrutiny. For example, Blevins (2006) lists Leti [rka(lu] in (6) above. However, van Engelenhoven (2004) states that word-initially before another consonant the trilled /r/, nasals, /l/ and /s/ are lengthened and “syllabic.” Thus, a more accurate transcription of this word is [y.ka(.lu]. Since the /r/ and the /k/

5

Steve Parker

Table 49.2 Typological range of languages containing sC clusters s+

Spanish

French, Western Keres

Greek

English

Dutch

German

Russian

– – – – –

✓ – – – –

✓ ✓ (–) – –

✓ – ✓ ✓ –

✓ ✓ ✓ ✓ (–)

✓ – ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓

stop fricative nasal lateral rhotic

are not tautosyllabic, this is not a counterexample to the SSP. Rather, it confirms it. Cross-linguistically, the most frequent exceptions to the SSP involve initial /s/ followed by a plosive, as in the English words spill, still, and skill. Morelli (2003) and Goad (chapter 38: the representation of sc clusters) focus on this phenomenon. Table 49.2 is adapted from the latter. Summarizing this table, Goad observes that the lower a consonant is in sonority, the more preferred it is after an initial /s/. She thus posits that s + stop › s + nasal › s + lateral › s + rhotic, where “›” = “is more harmonic than.” This scale (minus the s) follows many sonority hierarchies that posit more natural class distinctions than (3), such as the maximally detailed scale in §4. Morelli (2003) reaches a similar conclusion. She notes that many languages have onset cluster inventories comprising three main types: (1) stop + sonorant, (2) fricative + sonorant and (3) fricative + stop. These first two satisfy the SSP, while the third reverses it (assuming that fricatives are more sonorous than stops; see below). Illustrative languages include Haida, Hindi, Hungarian, Isthmus Zapotec, Italian, Mohave, Swedish, Telugu, Yecuatla Totonac, and Yuchi. To her knowledge, however, no language exists that is analogous to these, yet completely follows the SSP: hypothetically, (1) stop + sonorant, (2) fricative + sonorant, and, crucially, (3) stop + fricative (not counting affricates). Consequently, she posits that among onset clusters consisting of two obstruents, the unmarked type is fricative + stop, where “unmarked” = phonologically default and most common (Kenstowicz and Kisseberth 1973: 3; de Lacy 2006). Summarizing thus far, the SSP is a strong universal tendency but has exceptions. In some cases the reversals in the sonority slope are of the largest possible degree: an onset consisting of a glide followed by a voiceless stop. To illustrate, Santa María Quiegolani Zapotec exhibits many words like the following (Regnier 1993): (7) [wkìt] [wtò(?] [jk7]

‘game’ ‘sell (completive)’ ‘buy (potential)’

Nevertheless, typological generalizations can still be made. For example, most languages have more consonant cluster types and tokens obeying the SSP than violating it. Furthermore, more languages attest obstruent + liquid (OL) onset clusters, for instance, than the opposite (LO). This can be stated even more forcefully as an implicational universal: if a language allows complex onsets of the type

Sonority

6

LO, it must permit OL clusters as well, whereas the inverse is not necessarily true (Greenberg 1978). However, this kind of observation cannot be extended to include all possible natural class combinations. For instance, Texistepec Popoluca does not permit obstruent + nasal (ON) syllable initially, yet it does allow NO clusters: [mbak] ‘my bone’ (Wichmann 2002).3 Consequently, absolute claims about the SSP tend to break down given enough languages. Nevertheless, one apparently exceptionless statement is the following: (8)

No language exists in which all tautosyllabic consonant clusters reverse the SSP.

Returning to the five-category sonority hierarchy in (3), many phonologists expand this by making subdivisions within three of the natural classes: vowels, liquids, and obstruents (cf. (27) in §4). For example, fricatives are often claimed to be more sonorous than stops (Hankamer and Aissen 1974; Steriade 1982, 1988; Kager 1999). To illustrate, in Sanskrit reduplication, when a verb base begins with a consonant cluster, the prefix retains the less sonorous of these two sounds. Thus, in (9), when a stop and a continuant are adjacent in either order, the reduplicant invariably surfaces with the stop: (9)

Sanskrit (from Whitney 1889) /praŒ h/ /swar/ /tsar/ /st ha(/ /tja–/ /œrat h/ /druw/ /mluŒ/ /rdh/

→ → → → → → → → →

[pa-praŒ h] [sa-swar] [ta-tsar] [ta-st ha(] [ta-tja–] [œa-œrat h] [du-druw] [mu-mluŒ] [a(r-di-dham]4

‘ask’ ‘sound’ ‘approach stealthily’ ‘stand’ ‘forsake’ ‘slacken’ ‘run’ ‘set’

In (9) the obvious generalization is that the reduplicant copies the less sonorous consonant from the onset of the base, regardless of its relative position within the cluster. Otherwise, if all obstruents are equal in sonority, the analysis of this process is more complicated to express (Benua 1997; Hironymous 1999). For further data and discussion of Sanskrit reduplication, see chapter 119: reduplication in sanskrit. While the full details are complex, the pattern whereby the least sonorous segment emerges in the prefix is very regular. For a mathematical explanation of this effect, see §3. When underlying representations juxtapose sounds violating the SSP, these are repaired in four different ways cross-linguistically: (1) vowel epenthesis, (2) deletion, (3) syllabic consonants, and (4) metathesis. First, a vowel can be inserted to rescue the unsyllabifiable consonant, a process called stray epenthesis (Itô 1986). This occurs in Serbo-Croatian (Kenstowicz 1994): 3

Prenasalized stops (common in African languages) do not violate the SSP, since they are single phonemic units, not true sequences. Syllabic nasals, such as in hypothetical [è.da], do not constitute tautosyllabic onsets either. 4 Whitney does not gloss this root, but notes that the form is aorist.

Steve Parker

7 (10) a. b. c.

masculine

neuter

pust zelen dobar jasan bogat kriÚan

pusto zeleno dobro jasno bogato kriÚano

‘empty’ ‘green’ ‘good’ ‘clear’ ‘rich’ ‘cross’

In the adjective paradigm in (10), the neuter is marked by the suffix /-o/. The masculine column displays no overt morphological marking. In (10b), [a] alternates with Ø. This vowel surfaces phonetically in the final syllable of the masculine forms, between the last two consonants, but is absent in the neuter column. The contrasting forms in (10c) contain an [a] in the second syllable in both columns. This demonstrates that the alternation in (10b) involves epenthesis of [a] in the masculine forms, not syncope of underlying /a/ in the neuter column. The underlying representations of the roots in (10b) are /dobr/ and /jasn/. These underlying representations end with a consonant cluster consisting of an obstruent followed by a sonorant. If these were syllabified directly into a complex coda, they would violate the SSP. In contrast, the root /pust/ in (10a) ends with a cluster in which sonority falls. Therefore stray epenthesis is not needed since the sequence /pust/ can be exhaustively syllabified while respecting the SSP. The sonority profile of two of these contrasting roots is displayed in the following metrical-like grids (Jespersen 1904; Zec 1988; Clements 1990; Kenstowicz 1994): (11)

vowel glide liquid nasal fricative stop

* * * * * * * * * * p u s t

* * * * * * * * * * * * d o b r

These grids employ the five-category sonority hierarchy in (3), supplemented by obstruents being split between fricatives and stops, motivated by the Sanskrit data in (9). As these figures show, the morpheme /pust/ in isolation (in the masculine column) contains one peak of sonority (the /u/), whereas /dobr/ contains two (the /o/ and the /r/). Consequently, the motivation for inserting a vowel in the second case ([dobar]) is to rescue the /r/, which cannot be incorporated into the same syllable as the /b/ without violating the SSP. A second process used to fix SSP violations is the deletion of an unlicensed (unsyllabifiable) consonant, known as stray erasure (Itô 1986). This process is illustrated by Ancient Greek. The following data show that complex onset and coda clusters are permitted, including word-medially (Steriade 1982; Kenstowicz 1994): (12)

kleph smerd.nos am.blus

‘to steal’ ‘power, force’ ‘dull’

Sonority as.tron a.elp.tos t helk.tron pemp.tos

8

‘star’ ‘unhoped for’ ‘charm’ ‘sent’

The form [t helk.tron] demonstrates that up to four consonants can be concatenated intervocalically, providing the SSP is respected. However, in the reduplicated form /CV-graph-st hai/ → [gegraph t hai] ‘to have been written’, the underlying /s/ at the beginning of the infinitival suffix occurs between two stops. If this word were assigned a sonority profile as in (11), the /s/ would constitute a peak of sonority. However, this /s/ is not syllabic, nor can it be incorporated into a syllable with the preceding /ph/ or the following /t h/ without violating the SSP. Consequently, since it is prosodically unparsable, it is elided. According to Steriade (1982) and Itô (1986), this phenomenon is a default universal mechanism automatically applying at the end of the derivation to clean up any remaining problems (see chapter 68: deletion). A third strategy for dealing with SSP violations is to simply retain the offending consonant, in which case it is automatically realized phonetically as syllabic. English illustrates this with unstressed sonorant consonants in word-final clusters: prism, button , pickle, manner. Another example is Chamicuro (Parker 1989): (13)

[w-usm-i] ‘I sing’ 1sg-sing-epenthetic

[w-usx-kati] ‘I sang’ 1sg-sing-past

Fourth, and most rarely, SSP violations are resolved by metathesis. The most convincing case of this to date is Western Farsi. When a final vowel is deleted by a general process of apocope, an obstruent or nasal in a potential coda cluster metathesizes with a following liquid. Otherwise (without metathesis), the final liquid would constitute a separate sonority peak, which this language does not allow (Hock 1985): (14)

Œaxra suxra vafra asru vazra *namra5

→ → → → → →

Œarx surx barf ars gurz narm

‘wheel’ ‘red’ ‘snow, ice’ ‘tear’ ‘club’ ‘soft’

(cf. suhr-ab ‘ruddy goose’)

Hock (1985) attributes this alternation to the SSP. However, from his description this is primarily a historical process, so it may no longer be synchronically active.

2.2

Minimum Sonority Distance

While the SSP rules out many of the prohibited syllable types in most languages, it is not the full story. For example, the three syllables [kna], [kla], and [kwa] equally

5

The form /namra/ is preceded by *, since it is reconstructed.

Steve Parker

9

Table 49.3 Minimal Sonority Distance language types

MSD MSD MSD MSD

= = = =

0 1 2 3

Maximal inventory of permissible onset clusters

Languages

OO, ON, OL, OG, NN, NL, NG, LL, LG, GG ON, OL, OG, NL, NG, LG OL, OG, NG OG

Bulgarian, Leti Chukchee Gizrra, Kurdish, Spanish Mono, Panobo, Japanese (?), Mandarin Chinese (?)

satisfy the SSP. Nevertheless, although many languages permit onset clusters such as [kl] and/or [kw], syllables like [kna] are much less common. One explanation for this asymmetry is a language-specific parametric requirement that the members of a tautosyllabic consonant cluster be separated by a minimum number of ranks on the sonority scale (Steriade 1982; Selkirk 1984). For example, /k/ and /l/ are sufficiently distinct in relative sonority and may therefore be combined. However, /k/ and /n/ are too close along this scale, and this is not tolerated in many languages. Conversely, a language like Russian, which permits words like /kniga/ ‘book’, has a lower threshold on this parameter. This condition is captured by the following principle: (15)

Minimal Sonority Distance (MSD) Given an onset composed of two segments, C1 and C2, if a = Sonority Index of C1 and b = SI(C2), then b − a ≥ x, where x ∈ {0, 1, 2, 3}.

Assuming the sonority hierarchy in (3), the typology of possible languages shown in Table 49.3 is generated (cf. Zec 2007). The generalization is that if a language permits clusters with a lower sonority distance, it allows clusters of all higher sonority distances as well, ceteris paribus. The inverse of this is not true. The reversed counterparts of these onsets, such as *LO, can be excluded by the independently motivated SSP when necessary (§2.1). The data in (16) illustrate typical consonant sequences from each of the four language types in Table 49.3. Naturally, not every cluster type is fully productive for all phoneme combinations in these languages. Nevertheless, enough representative examples occur to justify the general trends. (16)

a.

Leti (van Engelenhoven 2004) [ptu.na] ‘star’ [tmu.ra] ‘tin’ [kru.ki] ‘crab (sp.)’ [x.kwo.ri] ‘you (sg) lift’

b.

Chukchee (Kämpfe and Volodin 1995) [plHtkuk] ‘end, finish, conclude’ [qlikkin] ‘twenty’ [t7eŒejwH:?e] ‘I will go’ [lju7] ‘suddenly’

Sonority c.

10

Gizrra (van Bodegraven and van Bodegraven 2005) [gles] ‘dew’ [ta.p7az.dH] ‘on (his) fangs’ [djao] ‘palm (sp.)’ [u7.mjao] ‘tree (sp.)’

d. Panobo (Parker 1992) [hwhn.ti] ‘heart’ [ßwi.ni.k]1] ‘they are taking, carrying’ [pja.ka] ‘nephew, niece’ [wa.ta.tjan] ‘last year’ As Table 49.3 displays, a significant implication of the MSD approach is that the ideal onset cluster consists of an obstruent plus a glide, all else being equal. Thus, if a language allows complex onsets and has glides in its phonemic inventory, it must permit stop + glide clusters. This is the only onset sequence occurring in all four language types in Table 49.3. An explanation for this is that these two natural classes (stops and glides) are maximally separated in terms of their relative sonority, since they occupy the extreme ends of the scale (among consonants). Baertsch (2002) proposes one way to capture MSD effects in Optimality Theory (OT: Prince and Smolensky 1993). The corresponding prediction is that some languages should exist which permit OG but no other onsets. Two such cases are Mono (Democratic Republic of the Congo; Olson 2005) and Panobo. Other possibilities are Japanese (Vance 1987, 2008) and Mandarin Chinese (Yuan 1989). The latter two are listed in Table 49.3, followed by question marks to highlight their controversial status. Also, in Hindi (Ohala 1983) and Koluwawa (Guderian and Guderian 2005), the only initial clusters are OG and NG, but not *OL. There is, however, a problem. In a sequence like [kwa], the [w] is potentially ambiguous since it allows different phonological interpretations. A priori it could pertain to a diphthongal nucleus rather than the onset: [k\]. Alternatively, it might be a secondary articulation (labialization) of the preceding /k/: [kwa]. If so, then there really is no consonant cluster, just a single complex phonemic unit. The third possibility is that [kw] simply constitutes a true onset cluster, as in Panobo. Teasing apart these different conclusions is complicated, and often the language-specific evidence is not compelling either way. Unfortunately, then, when no other canonical onset clusters (such as OL) exist in a language, the argumentation is in danger of circularity regardless of which segmentation is posited. See §3 for an alternative model that claims that the unmarked initial cluster is not OG but OL. Finally, in the MSD approach, the sonority distance between the second onset consonant and the vowel is not crucial, because phonotactic restrictions rarely obtain across onset–nucleus junctures (Blevins 1995). However, see §3 for an approach in which the nature of this sequence (C2 + V) does matter. See also chapter 15: glides for further discussion of glides, and chapter 55: onsets for an expanded treatment of onsets.

2.3

The Syllable Contact Law

Another sonority-based principle active in many languages is the Syllable Contact Law (SCL). Some seminal references are Hooper (1976), Murray and Vennemann (1983), and Clements (1990). More recent treatments of the SCL as a

Steve Parker

11

Table 49.4 Alternations motivated by the Syllable Contact Law Process

Illustration Language

Example

coda weakening

g.n → w.n

Hausa

/hagni/ → [haw.ni] ‘left side’

Kyrgyz Kazakh Kazakh Kazakh

/konok-lar/ → [konok.tar] ‘guest-pl’ /kol-lar/ → [kol.dar] ‘hand-pl’ /koIQz-lar/ → [koIQz.dar] ‘bug-pl’ /koIQz-ma/ → [koIQz.ba] ‘bug-int’ [tek-.l>ç] ‘daily’, [e.kl>ç] ‘disgusting’

onset strengthening k.l → k.t (desonorization) l.l → l.d z.l → z.d z.m → z.b tautosyllabification

k.l → .kl

Germanic

gemination

b.r → b.br

Latin > Italian /lab7um/ → [lab.b7o] ‘lip’

epenthesis

n.r → n.dr

Spanish

regressive assimilation

k.m → I.m Korean

/kuk-mul/ → [kuI.mul] ‘broth’

progressive total assimilation

g.n → g.g

Pali

/lag-na/ → [lag.ga] ‘attach (past part)’

regressive total assimilation

n.l → l.l

Korean

/non-li/ → [nol.li] ‘logic’

anaptyxis

p.r → pV.r

Ho-Chunk

/hipres/ → [hiperes] ‘know’

metathesis

d.n → n.d

Sidamo

/gud-nonni/ → [gun.donni] ‘they finished’

/beni7-a/ → [ben.d7a] ‘(s/he) will come’

family of OT constraints include Davis (1998), Gouskova (2004), and Zec (2007). The following are two typical formulations of the SCL: (17) Syllable Contact Law a. b.

A heterosyllabic juncture of two consonants A.B is more harmonic (ideal) the higher the sonority of A and the lower the sonority of B. In any heterosyllabic sequence of two consonants A.B, the sonority of A is preferably greater than the sonority of B.

By (17), for example, the sequence [l.k] is inherently less marked than [k.l]. Vennemann (1988: 50) provides a list of sample repair strategies that languages employ to improve satisfaction of the SCL. These are summarized in Table 49.4, as annotated by Davis (1998: 183) and supplemented by Seo (chapter 53: syllable contact), which offers more data and discussion. Based on a survey of 31 languages with SCL effects, her results give a better idea of the range of typological generalizations and their relative robustness. For example, Kazakh tolerates a /j-l/ juncture, as in [mandaj.lar] ‘foreheads’, since sonority drops slightly from /j/ to /l/. Kyrgyz, nevertheless, requires a greater fall in sonority and maps /aj-lar/ to [aj.dar] ‘moons’ (Davis 1998). Examining the details of SCL phenomena in particular languages allows us to establish subtle differences in sonority ranks. For instance, Spanish attests words such as [pe7.la] ‘pearl’ and [al.re.Ïe.Ïo7] ‘around’, yet the hypothetical sequence *[l.7] systematically does not occur. When such a juncture would be created, an intrusive stop appears instead. This happens when the future tense is derived by dropping the infinitival theme vowel: /sali7/ ‘to leave’ → *[sal.7a] → [sal.d7a] ‘(s/he)

Sonority

12

will leave’. These facts motivate the following sonority hierarchy among Spanish liquids, based on the second of the two definitions in (17): flap > lateral > trill (Bonet and Mascaró 1997; Parker 2008). Nevertheless, there are problems with the SCL too. For example, it predicts that obstruent + sonorant junctures should be “fixed” more often than sonorant + sonorant clusters, ceteris paribus. The opposite in fact is true (chapter 53: syllable contact). Furthermore, in Akan both /O-N/ and /N-O/ sequences result in phonetic [NN]. The latter is a mirror image of Korean nasal assimilation (Table 49.4), even though this makes syllable contact worse: /óU-dú/ → [óU.nú?] ‘he should arrive’ (Schachter and Fromkin 1968).

2.4

Rhyme weight

It is well known that the heavier a syllable is, the more it tends to attract stress (Hayes 1980; Prince 1990). For example, open syllables are light, but closed syllables are usually bimoraic. Thus, they may be obligatorily stressed (see also chapter 57: quantity-sensitivity). Also, in some languages, rhymes headed by /e/ or /o/ attract stress more than those with /i/ and /u/, indicating that mid vowels have more weight than high vowels in these systems. Furthermore, the propensity for a coda consonant to project a mora is correlated with how sonorous it is (Zec 1988, 1995). An adequate theory of phonology should provide a unified (non-accidental) explanation for these facts. Appealing to a scalar feature like sonority allows us to do that. Based on case studies examining the relationship between segmental quality and syllable weight effects, the following hierarchy of vowel sonority has been posited (Kenstowicz 1997; de Lacy 2002, 2004, 2006, 2007a). Specific languages may choose to exploit different subsets among these natural classes: (18)

Relative sonority of vowels a > e, o > i, u > H > q

To illustrate, Kobon vowels are divided into four groups in terms of stress assignment: /a/ > /e o/ > /i u/ > /H q/. In this case the potential distinction between /q/ and /H/ is underexploited. In unaffixed Kobon words, stress predictably falls on the most sonorous nucleus within a disyllabic window at the right edge (Davies 1980, 1981): (19)

a>e a>i a>H a>q o>u o>i o>q i>H i>q

[haI’gaße] [kh i.’a] [kh HßH’ja] [’aJqm-’aJqm] [’mo.u] [si.’oIkh] [gq’7o-gq’7o] [ga’ÈinHI] [’jimbqÈ]

‘blood’ ‘tree (sp.)’ ‘rat’ ‘lightning’ ‘thus’ ‘bird (sp.)’ ‘talk (mother pig to piglet)’ ‘bird (sp.)’ ‘very’

The generalization in (18) and (19) is that vowels which are more peripheral in the acoustic space are more sonorous than central ones. Furthermore, within

Steve Parker

13

these two sets, segments involving a lower jaw configuration outrank their higher counterparts (Coetzee 2006). At the bottom of this hierarchy, /H/ is higher in sonority than /q/. This is due to languages such as Lushootseed, in which /H/ can be stressed (unlike English) when it is the only nucleus in a root. Stress falls on the first “full” vowel of the stem; otherwise on the first schwa (Urbanczyk 2006): (20)

Stressable /H/ in Lushootseed [’?itut] [dzH’xixw ] [’Œuqwu-d] [ŒH’gwas] [k’H’daju] [’–HsHd] [’bHŒ]

‘sleep’ ‘creek’ ‘to whittle something’ ‘wife’ ‘rat’ ‘foot’ ‘fall down’

Other languages in which sonority is crucial to stress assignment include Pichis Ashéninka (Payne 1990; Hayes 1995), Komi (Hayes 1995; de Lacy 1997), and Finnish (Anttila 1995; Hayes 1995). However, a reviewer notes that contrastive /H/ in languages like Lushootseed may be different in quality from phonetic [H] resulting from reduction in English and analogous languages. For example, the former is probably longer in duration than the latter. This is a valid point that must be controlled for in cross-linguistic comparisons of this sort. For Lushootseed, Urbanczyk posits a phonemic /H/ in underlying forms. In motivating a constraint against stressed schwas she writes: *p has the distributional hallmarks of a markedness constraint because there are languages which never stress schwa, languages which avoid stressing schwa, and languages which permit schwa, along with other vowels, to be stressed, but no language enforces the stressing of schwa in preference to other vowels. (Urbanczyk 2006: 210)

She then lists other Salishan languages in which this prohibition is active (2006: 211, fn. 24). See chapter 26: schwa for further discussion of schwa in general. Concerning the relative weight of coda consonants, the typological range of languages is also dependent on sonority. Table 49.5 is adapted from Zec (1995, 2007). The generalization is that if a lower-sonority class is moraic in a particular language, then all higher-sonority categories are also moraic in syllable-final position. Zec (2007) knows of no language in which coda liquids count as heavy but nasals do not; she considers this an accidental gap. In addition to stress attraction, other diagnostics for consonant moraicity are: (1) the ability to bear a contrastive tone (Tiv; Zec 1995), (2) prosodic minimality (Fijian; Dixon 1988), and

Table 49.5 Inventories of moraic segments Natural classes contributing to syllable weight

Languages

vowels only vowels and liquids vowels, liquids, and nasals vowels, liquids, nasals, and obstruents

Fijian, Halh Mongolian, Lardil, Yidiny ? Gonja, Kwakiutl, Lithuanian, Tiv Egyptian Arabic, English, Latin, Maithili

Sonority

14

(3) blocking of processes such as vowel reduction (Maithili; Hayes 1995) (see also chapter 33: syllable-internal structure). (21) gives sample data from the three attested language types in Table 49.5: (21)

a.

Fijian (Dixon 1988) Stress the syllable containing the penultimate mora (closed syllables do not occur). [’siIa] ‘day’ m [ bu’ta?o] ‘steal’ m [ buta’?oÏa] ‘steal-trans’ [?i’laa] ‘know-trans’ [’raiÏa] ‘see-trans’ [’lu.a] ‘vomit (vb)’

b.

Tiv (Zec 1995) Only sonorant consonants occur in codas, where they bear tone. [bág] ‘salt’ [fá!3] ‘rainy season’ [rùmù5] ‘agreed, confessed’

c.

Egyptian (Cairene) Arabic (Hayes 1995) Stress the ultima if superheavy (trimoraic), otherwise the penult if heavy, otherwise the antepenult. [ka’tabt] ‘I wrote’ [ha–’–a(t] ‘pilgrimages’ [’be(tak] ‘your (masc sg) house’ [ka’tab.ta] ‘you (masc sg) wrote’ [mu’dar.ris] ‘teacher’ [?in’kasara] ‘it got broken’ [’kataba] ‘he wrote’

If fricatives are more sonorous than stops (§2.1), by implication some languages should exist in which fricatives occur in coda position but stops do not. This is exemplified by Panobo. Syllable-final consonants include glides, nasals, and fricatives. However, this is complicated by the fact that the flap /7/ occurs in onsets, yet not in codas. Evidence that coda consonants are moraic in Panobo is that in word-final position they attract stress. Otherwise, the default quantity-sensitive foot type, a moraic trochee, assigns stress to the penultimate syllable (Parker 1992): (22)

Heavy final syllables in Panobo [’atsa] [ka’noti] [ja’wiœ] [tah’põI] [pih’k]1]

3

‘manioc’ ‘bow (weapon)’ ‘opossum’ ‘root’ ‘(they) will eat’

The Sonority Dispersion Principle

Clements (1990) proposes an approach to syllable phonotactics that is also based on sonority. In his model, syllables are divided into two parts. The initial

Steve Parker

15

demisyllable consists of any onset consonants (if present) plus the nucleus, and the final demisyllable contains the nucleus plus the coda, i.e. the rhyme. The term demisyllable is borrowed from Fujimura and Lovins (1978). The nucleus crucially resides in both demisyllables simultaneously. For example, the word /plom/ contains the demisyllables /plo/ and /om/. The essence of the Sonority Dispersion Principle (SDP) is that initial demisyllables are preferred when their constituents are maximally and evenly dispersed in sonority; e.g. /ta/. The same tendency is inverted for final demisyllables, favoring open rhymes (those ending with a vowel). More precisely, initial demisyllables of the same length (number of segments) are more harmonic to the degree that they minimize D in (23) below. Conversely, final demisyllables are more harmonic to the degree that they maximize D, all else being equal. The formula for D comes from the realms of physics and geometry, where it governs the distribution of mutually repelling forces in potential fields (like electrons). Its linguistic use originates with the work of Liljencrants and Lindblom (1972) on perceptual distance between segments in the acoustic vowel space. Hooper (1976) and Vennemann (1988: 13–14) anticipate its application to sonority and syllable structure. (23)

Sonority Dispersion Principle m

D=

1

∑d i=1

where

2 i

d = distance between the sonority indices of each pair of segments, and m = number of pairs of segments (including non-adjacent ones), where m = n(n − 1)/2, and where n = number of segments.

Clements (1990: 304) paraphrases (23) as follows: “D . . . varies according to the sum of the inverse of the squared values of the sonority distances between the members of each pair of segments within” a demisyllable. D, then, is the reciprocal of dispersion. To illustrate the application of (23), Clements assumes a sonority scale with the five categories from (3): sonority index

(24) vowels glides liquids nasals obstruents

(V) (G) (L) (N) (O)

5 4 3 2 1

When D is computed for demisyllables containing exactly one or two consonants, it yields the following values (ignoring types that violate the SSP): (25)

Sonority Dispersion demisyllable values a.

OV NV LV GV

= = = =

.06 .11 .25 1.00

most natural onset

least natural onset

b.

OLV OGV ONV NGV NLV LGV

= = = = = =

.56 1.17 1.17 1.36 1.36 2.25

Sonority c.

VO VN VL VG

= = = =

.06 .11 .25 1.00

least natural coda

most natural coda

d. VLO VGO VNO VGN VLN VGL

= = = = = =

16

.56 1.17 1.17 1.36 1.36 2.25

In (25a) the SDP favors a single syllable-initial consonant that maximizes the sonority slope between the onset and the vowel. Therefore, it correctly predicts that the preferred syllable of type CV has an onset consonant as low in sonority as possible (D = .06). This results in the following scale of relative unmarkedness: /ta/ > /na/ > /la/ > /ja/. This is foreshadowed in the discussion of Sanskrit reduplication (§2.1). Recall that the pattern there demonstrates that fricatives can outrank stops in sonority. Therefore, not all obstruents are necessarily equal in sonority (see §4). Summarizing thus far, a unique consequence of the formula for D is that in an initial demisyllable of two segments (CV), the segments should differ from each other in sonority as much as possible. This is somewhat analogous to the MSD approach for onset clusters (§2.2). However, when the initial demisyllable contains three segments (CCV), what matters (given D) is that the aggregate total of the sonority distances between all of these together be maximized. This is accomplished by spacing apart the segments as evenly as possible (in sonority). This results in the best evaluation for OLV in (25b), since it has the lowest obtained value (.56). This is because liquids fall precisely midway between obstruents and vowels in terms of their sonority indices in Clements’s five-category scale in (24). As evidence for the SDP, Clements notes that underlyingly French permits complex demisyllables of the type OLV only. However, in surface forms, some instances of OGV also exist. This raises an important typological question: which onset cluster is universally unmarked, OL or OG? On the one hand (as noted in §2.2), the MSD principle predicts that some languages should permit OG but not *OL. This seems to be correct, but see the caveats in §2.2. On the other hand, the SDP claims that OL is preferred. One piece of evidence that could help resolve this would be an alternation mapping underlying OLV to OGV, or vice versa. Unfortunately, no such process has yet been observed. A cross-linguistic survey documenting the number of languages with one type of cluster but not the other would also be enlightening. It may be that both kinds of onsets (OG and OL) need to be optimal simultaneously, i.e. in the grammars of different languages. Rod Casali and Ken Olson (personal communication) note that apparent OG-only languages are especially common in Africa. Finally, the SDP assigns an equal evaluation to the two demisyllable types OGV and ONV in (25b). This may be problematic, since many languages exhibit the initial sequence OG, but not *ON. For example, Table 49.3 mentions Gizrra, Kurdish, Mono, Panobo, and Spanish. However, Clements does not claim that demisyllables of the same rank necessarily co-occur in any language containing one of them. At the same time, no languages appear to allow ONV but not *OGV. If this is a systematic gap, it is troubling for the predictions of the SDP.

Steve Parker

17

4

The complete sonority hierarchy

Perhaps no issue in phonological theory has led to more competing proposals than the internal structure of the sonority hierarchy, i.e. the numbers and types of natural classes, and their corresponding ranks. Parker (2002) notes that more than 100 distinct sonority scales are found in the literature. The purpose of this section is to lay out several desirable characteristics that a full and final sonority hierarchy should possess, and then present one specific model that arguably comes closest to fulfilling those goals. Briefly, an adequate sonority scale should display the a priori traits in (26). In principle these criteria apply not just to sonority, but to all phonological features, that is, classic binary features like [±voice], [±round], etc. (26)

All else being equal, an ideal sonority scale would have these characteristics: Universal: It potentially applies to all languages. Exhaustive: It encompasses all categories of speech sounds. Impermutable: Its rankings cannot be reversed (although they may be collapsed or ignored). d. Phonetically grounded: It corresponds to some consistent, measurable physical parameter shared by all languages.

a. b. c.

Each of the points in (26) will now be discussed. First, ideally we can establish a single, unique sonority hierarchy to analyze all known languages. This is not to say that any particular language actually exploits every one of the natural class rankings in the sonority scale. On the contrary, it would be quite amazing (although fortuitous) to discover such a case. Nevertheless, the explanatory power of sonority is maximized if we ascribe it to UG, making it equally available to all humans. Second, an adequate theory of sonority should include every known type of phonological segment. Many hierarchies omit recalcitrant natural classes such as glottal consonants (/h/ and /?/), affricates, etc., perhaps because of their inherent complications. Such scales then cannot apply to all languages. This undermines their universality. Third, the rankings in the sonority scale should be impermutable. This is a beneficial characteristic since it is the most restrictive hypothesis possible, i.e. it severely limits the types of processes directly attributable to sonority. In addition to avoiding overgeneration of non-attested language types, impermutability makes claims about sonority easier to falsify. This in turn reduces the danger of circular argumentation. For example, once it is established that laterals, for instance, are more sonorous than nasals, the entailment is that there is no language in which nasals pattern as more sonorous than laterals by the same criteria. At the same time, however, potential divisions between sonority ranks are frequently underexploited in many languages. See de Lacy (2002, 2004, 2006, 2007a) for a formal approach to “underspecification” of sonority classes in OT. Fourth, in an ideal world we can show that sonority is based on concrete articulatory gestures and/or their acoustic counterparts. This is the topic of the next section.

Sonority

18

Although no sonority scale to date perfectly fulfills every desideratum sketched above, one that perhaps comes closest is that of Parker (2008), reproduced below: (27) Final hierarchy of relative sonority Natural class low vowels mid peripheral vowels (not [H]) high peripheral vowels (not [q]) mid interior vowels ([H]) high interior vowels ([q]) glides rhotic approximants ([P]) flaps laterals trills nasals voiced fricatives voiced affricates voiced stops voiceless fricatives (including [h]) voiceless affricates voiceless stops (including [?])

Sonority index 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Space does not permit a detailed justification of (27). Nevertheless, to highlight a few positive aspects of this scale, the evidence for most of the natural classes is fairly robust and secure. Parker (2002, 2008) summarizes the debates and provides at least one argument to motivate every ranking in this hierarchy (every pair of adjacent categories). Much of this is reviewed in §2 of this chapter. To give another example, Koine (Ancient) Greek permits the clusters /pn/ and /kn/ (/pniktos/ ‘(things) strangled’, /kne.o/ ‘have itching, tickled’; Mounce 1993), but proscribes */bn/ and */dn/. This can be explained as an MSD effect if voiced stops are closer to nasals (in sonority) than voiceless stops are. Furthermore, flaps are higher in sonority than trills in Spanish, as established by the SCL in §2.3. Three other facts confirm this. First, in word-initial position the contrast between /r/ and /7/ is neutralized to /r/ ([rana] ‘frog’). Second, in codas it is neutralized in favor of /7/: [Œa7la7] ‘to chat’. These two points follow from the SDP (Bonet and Mascaró 1997). Third, /7/ and /l/ appear as the second member of complex onsets, yet /r/ does not (see (5)). This is another MSD effect if /r/ is less sonorous than /7/ and /l/ (Bakovio 1994). These distributional facts indicate that liquids do not always pattern as a monolithic class in Spanish. Finally, the rhotic approximant /P/ is higher in sonority than /l/ in English: (1) the contrast between Carl (one syllable) vs. caller (two syllables) follows from the SSP if /P/ outranks /l/ (Borowsky 1986); (2) /P/ is the default epenthetic coda in Eastern Massachusetts speech (McCarthy 1993). This follows from the SDP if /P/ is the most sonorous consonant available in this position. And (3) syllabic /P/ may bear stress (bird, curtain), but /l/ never does (Zec 2003). Another strength of the scale in (27) is its breadth, i.e. the number of different types of segment classes it encompasses. Nevertheless, it is still not exhaustive,

19

Steve Parker

because it leaves out a few rarer kinds of sounds, such as clicks and implosives. Since no known sonority hierarchy includes these, more research in this area would be welcome. A third advantage of (27) is that it has been rigorously tested and found to provide a good fit with all phonetic segments in seven specific languages. This is discussed in §5. However, a few problems persist. For example, the placement of affricates between stops and fricatives is a controversial issue, remaining open to disagreement. Many scales either leave affricates out entirely or group them with plosives, using a term such as stops. Such proposals may simply assume that affricates behave phonologically like stops, as they do in many languages. For instance, Kang (chapter 95: loanword phonology) concludes that affricates are just stops with a special place specification, e.g. [strident]. See chapter 16: affricates for more discussion of affricates in general. Similarly, the ranking of voiced stops over voiceless fricatives is harder to justify than most aspects of this hierarchy. A major reason for this is that many languages require consonant clusters to agree in voicing. Therefore, crucial diagnostic examples are rare, but one such token is the English word midst. Since this form is monosyllabic, /d/ is higher in sonority than /s/ by the SSP since /d/ is closer to the nucleus. Finally, the question of whether glottal consonants are sonorants or obstruents is also contested. Clearly /h/ and /?/ pattern phonologically with prototypical sonorants in some languages, yet behave like obstruents in others. In (27) they are classified as obstruents. One piece of evidence supporting this is that in Panobo, /h/ groups with /ß p t/ in exclusively obstruent + glide clusters (see (16)). Also, in many languages [?] is inserted as a default onset, where segments of low sonority are preferred by the SDP (Lombardi 2002). Finally, in the P-base sample of 549 languages, there are 65 distinct phonological processes in which /h/ and/or /?/ pattern solely with consonants that are unambiguously obstruents. In 21 other cases they group with sonorants (Mielke 2008).6 In the scale in (27) the tendency is obviously to “split” rather than to “lump together” natural classes. The motivation for this is as follows. There is ample evidence that fine-grade distinctions in sonority need to be made in some languages, such as between fricatives and stops in Sanskrit (§2.1). UG then must allow for these options, and hence the potential exists for other languages to exploit them as well. If we start with a hierarchy that assigns a unique rank to every distinct manner of articulation (like (27)), it is a trivial matter to formally compress (conflate) ranks together in order to analyze languages not invoking those splits. This procedure applies to every language in one way or another. However, if we assume a maximal sonority hierarchy with just the five groups in (3), no mechanism exists to “decompose” these, making more narrow distinctions in the scale when necessary. Consequently, only a fully detailed hierarchy such as (27) is flexible enough to generate the range of variation attested among the languages of the world with respect to processes involving sonority. Nevertheless, based on acoustic studies of many languages, Zhang (2001) and Gordon (2006) deny that a universal sonority scale is theoretically the most parsimonious option. Rather, they reject the existence of invariant sound classes. Gordon, for instance, concludes that syllable weight effects are not a unified

6

Thanks to Jeff Mielke (personal communication) for these counts.

Sonority

20

phenomenon. He claims that nasals, for example, display slightly different phonetic behavior from one language to another. This can then influence their phonological patterning in terms of sonority.

5

The physical substance of sonority

As summarized in §2, the function of sonority in phonological systems is fairly well understood. Nevertheless, these phenomena raise an important, related question that has provoked much contention and speculation: is there any coherent notion of sonority grounded in evidence external to the phonotactic facts that sonority is assumed to account for? In other words, what is the articulatory, acoustic, and/or perceptual source of sonority in the speech signal? To date at least 98 different correlates of sonority have been posited, documented in Parker (2002). The most frequently proposed phonetic definition of sonority is probably openness (of the vocal tract) or (supralaryngeal) aperture (Bloomfield 1914; Jespersen 1922; Goldsmith 1990; Kirchner 1998), or its inverse, (supraglottal) stricture, closure, impedance, etc. (Halle and Clements 1983; Kenstowicz 1994; Hume and Odden 1996). However, notions such as impedance are difficult to quantify. A more promising correlate of sonority is amplitude/intensity, or its perceptual counterpart, loudness (Bloomfield 1914; Laver 1994; cf. §1 and see also chapter 98: speech perception and phonology). Recently a major instrumental study was carried out measuring relative sound levels or RMS (root mean square) intensity of all phonemes of Peruvian Spanish, Cusco Quechua, and Midwestern American English (Parker 2008). See Jany et al. (2007) for a similar investigation of four other languages (Egyptian Arabic, Hindi, Mongolian, and Malayalam). In Parker (2008), the obtained intensity values for all segments yield an overall mean Spearman’s correlation of .91 with the sonority indices proposed in (27). Given those results, it is proposed there that the best way to characterize the physical basis of sonority is via a linear regression equation such as (28) below. This is calculated from the mean intensity measurements of all English coda consonants pertaining to nine of the natural classes from (27). These were pronounced five times each by five male native speakers of English. (28)

sonority = 13.9 + .48 × dB (dB = decibel; r 2 = .95)

This formula predicts an estimated sonority index based on a hypothetical intensity value. It characterizes the best-fitting line corresponding to the relative sonority of English coda consonants in phrasally stressed words. The obtained intensity value of each of these segments was compared to that of a stressed, utterance-initial low vowel (/A/) in a fixed carrier sentence. In (28) the Y intercept is 13.9. This is the projected value of Y (sonority) when X (dB) equals 0. Here it is significantly higher than the theoretical null value. This is because the obtained intensity of the reference vowel /A/, whose sonority index is 17, was subtracted from that of the target consonant for each utterance measured. This is a type of normalization procedure often performed to control for random fluctuations in loudness across speakers and tokens. The slope in (28), .48, indicates the rate of change in the dependent variable Y (sonority) per unit change in the independent variable X (dB). Its obtained value allows us to approximate the mathematical

21

Steve Parker

nature of the relationship between intensity and sonority. Specifically, for every decibel by which the relative sound level is increased, the corresponding sonority rank increases by about .48 units (for this sample of five English speakers). Also in (28), r 2 = .95. This is the coefficient of determination. It indicates the proportional reduction of error gained by using the linear regression model. Given this r 2 value, we can conclude that the single factor sonority accounts for (predicts) about 95 percent of the systematic variability in the intensity measurements of that dataset. Compared with previous accounts of sonority, the definition in (28) has several advantages (Parker 2008): (1) it is precise; (2) it is non-arbitrary; (3) it is phonetically grounded; (4) it is empirically verifiable and replicable; (5) it can be calculated for other speakers and languages; and (6) the underlying methodology (regression) is compatible with different (competing) sonority scales. However, while studies of this type represent progress, some problems remain. For example, in Parker (2008) the majority of the mismatches between sonority ranks and segmental intensity values (in all three languages) involve the sonorant consonants, particularly the approximants (laterals and glides). The reason for this is not clear at this point and merits further investigation. Finally, other researchers appeal to more functional aspects of the speech signal to avoid invoking sonority altogether. For example, building on the phonetically based work of Mattingly (1981) and Silverman (1995), Wright (2004: 35) reformulates the SSP as “a perceptually motivated and scalar constraint in which an optimal ordering of segments is one that maximises robustness of encoding of perceptual cues to the segmental make-up of the utterance.” Similarly, Ohala (1990a) claims that what drives the phonological phenomena discussed here is not really sonority but simply a need for adequate modulation in the acoustic wave.

6

Conclusion

Despite its problems, sonority makes sense. If it did not exist, it would be invented (Parker 2008). In this chapter a number of important issues have been examined. Nevertheless, certain topics need to be left for future work. For example, in §3 a possible contradiction between the claims of the MSD approach and the SDP is noted, involving the relative unmarkedness of OGV vs. OLV demisyllables. An in-depth typological study of these clusters would be helpful. Also, more attention should be given to the phonetic and/or functional bases of principles such as the SSP, the SCL, and the MSD. The question of why these hold true is potentially intriguing. Finally, another interesting point not discussed here is whether sonority scales are necessarily the same across different domains, such as phonotactics vs. the calculation of syllable weight.

ACKNOWLEDGMENTS Thanks to the following people for helpful comments: Marc van Oostendorp, Keren Rice, Michael Boutin, Mike Cahill, Ken Olson, two anonymous reviewers, and the students in my Frontiers in Phonology course taught at the Graduate Institute of Applied Linguistics in the fall of 2009. This work was partially funded by NSF grant #0003947.

Sonority

22

REFERENCES Anttila, Arto. 1995. Deriving variation from grammar: A study of Finnish genitives. Unpublished ms., Stanford University (ROA-63). Baertsch, Karen. 2002. An optimality theoretic approach to syllable structure: The split margin hierarchy. Ph.D. dissertation, Indiana University. Bakovio, Eric. 1994. Strong onsets and Spanish fortition. MIT Working Papers in Linguistics 23. 21–39. Benua, Laura. 1997. Transderivational identity: Phonological relations between words. Ph.D. dissertation, University of Massachusetts, Amherst (ROA-259). Blevins, Juliette. 1995. The syllable in phonological theory. In John A. Goldsmith (ed.) The handbook of phonological theory, 206–244. Cambridge, MA & Oxford: Blackwell. Blevins, Juliette. 2006. Syllable: Typology. In Keith Brown (ed.) Encyclopedia of language and linguistics. 2nd edn. vol. 12, 333–337. Amsterdam: Elsevier. Bloomfield, Leonard. 1914. An introduction to the study of language. New York: Henry Holt & Co. Bodegraven, Nico van & Elly van Bodegraven. 2005. Phonology essentials: Gizrra language. In Parker (2005), 191–210. Bonet, Eulalia & Joan Mascaró. 1997. On the representation of contrasting rhotics. In Fernando Martínez-Gil & Alfonso Morales-Front (eds.) Issues in the phonology and morphology of the major Iberian languages, 103–126. Washington, DC: Georgetown University Press. Borowsky, Toni. 1986. Topics in the lexical phonology of English. Ph.D. dissertation, University of Massachusetts, Amherst. Brosses, Charles de. 1765. Traité de la formation méchanique des langues, et des principes physiques de l’étymologie, vol. 1. 130 –133. Paris: Saillant, Vincent, Desaint. Cho, Young-mee Yu & Tracy Holloway King. 2003. Semisyllables and universal syllabification. In Féry & van de Vijver (2003), 183–212. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Coetzee, Andries W. 2006. Variation as accessing “non-optimal” candidates. Phonology 23. 337–385. Cser, András. 2003. The typology and modelling of obstruent lenition and fortition processes. Budapest: Akadémiai Kiadó. Davies, John. 1980. Kobon phonology. (Pacific Linguistics B68.) Canberra: Australian National University. Davies, John. 1981. Kobon. Amsterdam: North-Holland. Davis, Stuart. 1998. Syllable contact in Optimality Theory. Korean Journal of Linguistics 23. 181–211. de Lacy, Paul. 1997. Prosodic categorisation. M.A. thesis, University of Auckland (ROA-236). de Lacy, Paul. 2002. The formal expression of markedness. Ph.D. dissertation, University of Massachusetts, Amherst. de Lacy, Paul. 2004. Markedness conflation in Optimality Theory. Phonology 21. 145 –199. de Lacy, Paul. 2006. Markedness: Reduction and preservation in phonology. Cambridge: Cambridge University Press. de Lacy, Paul. 2007a. The interaction of tone, sonority, and prosodic structure. In de Lacy (2007b), 281–307. de Lacy, Paul (ed.) 2007b. The Cambridge handbook of phonology. Cambridge: Cambridge University Press.

23

Steve Parker

Dixon, R. M. W. 1988. A grammar of Boumaa Fijian. Chicago: University of Chicago Press. Engelenhoven, Aone van. 2004. Leti, a language of Southwest Maluku. Leiden: KITLV Press. Féry, Caroline & Ruben van de Vijver (eds.) 2003. The syllable in Optimality Theory. Cambridge: Cambridge University Press. Fujimura, Osamu & Julie B. Lovins. 1978. Syllables as concatenative phonetic units. In Alan Bell & Joan B. Hooper (eds.) Syllables and segments, 107–120. Amsterdam: North-Holland. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Gordon, Matthew. 2006. Syllable weight: Phonetics, phonology, typology. London: Routledge. Gouskova, Maria. 2004. Relational hierarchies in Optimality Theory: The case of syllable contact. Phonology 21. 201–250. Greenberg, Joseph H. 1978. Some generalizations concerning initial and final consonant clusters. In Joseph H. Greenberg, Charles A. Ferguson & Edith A. Moravcsik (eds.) Universals of human language, vol. 2: Phonology, 243–279. Stanford: Stanford University Press. Guderian, Brad & Toni Guderian. 2005. Organised phonology data supplement, Koluwawa language. In Parker (2005), 75 –86. Halle, Morris & G. N. Clements. 1983. Problem book in phonology. Cambridge, MA: MIT Press. Hankamer, Jorge & Judith Aissen. 1974. The sonority hierarchy. Papers from the Annual Regional Meeting, Chicago Linguistic Society: Parasession on natural phonology, 131–145. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Published 1985, New York: Garland. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hironymous, Patricia. 1999. Selection of the optimal syllable in an alignment-based theory of sonority. Ph.D. dissertation, University of Maryland at College Park. Hock, Hans Henrich. 1985. Regular metathesis. Linguistics 23. 529 –546. Hooper, Joan B. 1976. An introduction to natural generative phonology. New York: Academic Press. Hume, Elizabeth & David Odden. 1996. Reconsidering [consonantal]. Phonology 13. 345–376. Itô, Junko. 1986. Syllable theory in prosodic phonology. Ph.D. dissertation, University of Massachusetts, Amherst. Published 1988, New York: Garland. Jany, Carmen, Matthew Gordon, Carlos M. Nash & Nobutaka Takara. 2007. How universal is the sonority hierarchy? A cross-linguistic acoustic study. Paper presented at the 16th International Congress of Phonetic Sciences, Saarbrücken. Jespersen, Otto. 1904. Lehrbuch der Phonetik. Leipzig & Berlin: Teubner. Jespersen, Otto. 1922. A Modern English grammar on historical principles. Part 1: Sounds and spelling. 3rd edn. Heidelberg: Carl Winter’s Universitätsbuchhandlung. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kawasaki, Haruko. 1982. An acoustical basis for universal constraints on sound sequences. Ph.D. dissertation, University of California, Berkeley. Kämpfe, Hans-Rainer & Alexander P. Volodin. 1995. Abriß der tschuktschischen Grammatik auf der Basis der Schriftsprache. Wiesbaden: Harrassowitz. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Kenstowicz, Michael. 1997. Quality-sensitive stress. Rivista di Linguistica 9. 157–187. Kenstowicz, Michael & Charles W. Kisseberth. 1973. Unmarked bleeding orders. Studies in the Linguistic Sciences 1(1), 8–28. Reprinted in Charles W. Kisseberth (ed.) 1973. Studies in generative phonology, 1–12. Champaign, IL: Linguistic Research. Kirchner, Robert. 1998. An effort-based approach to consonant lenition. Ph.D. dissertation, UCLA (ROA-276).

Sonority

24

Ladefoged, Peter. 1975. A course in phonetics. New York: Harcourt Brace Jovanovich. Laver, John. 1994. Principles of phonetics. Cambridge: Cambridge University Press. Lewis, M. Paul (ed.) 2009. Ethnologue: Languages of the world. 16th edn. Dallas: SIL International. Liljencrants, Johan & Björn Lindblom. 1972. Numerical simulation of vowel quality systems: The role of perceptual contrast. Language 48. 839 –862. Lombardi, Linda. 2002. Coronal epenthesis and markedness. Phonology 19. 219 –251. Mattingly, Ignatius G. 1981. Phonetic representation and speech synthesis by rule. In Terry Myers, John Laver & John Anderson (eds.) The cognitive representation of speech, 415–420. Amsterdam: North Holland. McCarthy, John J. 1993. A case of surface constraint violation. Canadian Journal of Linguistics 38. 169 –195. Mielke, Jeff. 2008. The emergence of distinctive features. Oxford: Oxford University Press. Morelli, Frida. 2003. The relative harmony of /s+stop/ onsets: Obstruent clusters and the Sonority Sequencing Principle. In Féry & van de Vijver (2003), 356–371. Mounce, William D. 1993. The analytical lexicon to the Greek New Testament. Grand Rapids, MI: Zondervan Publishing House. Murray, Robert W. & Theo Vennemann. 1983. Sound change and syllable structure in Germanic phonology. Language 59. 514 –528. Ohala, John J. 1974. Phonetic explanation in phonology. Papers from the Annual Regional Meeting, Chicago Linguistic Society: Parasession on natural phonology, 251–274. Ohala, John J. 1990a. Alternatives to the sonority hierarchy for explaining segmental sequential constraints. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26(2). 319–338. Ohala, John J. 1990b. There is no interface between phonology and phonetics: A personal view. Journal of Phonetics 18. 153–171. Ohala, Manjari. 1983. Aspects of Hindi phonology. New Delhi: Motilal Banarsidass. Olson, Kenneth S. 2005. The phonology of Mono. Dallas: SIL International & University of Texas at Arlington. Parker, Steve. 1989. The sonority grid in Chamicuro phonology. Linguistic Analysis 19. 3–58. Parker, Steve. 1992. Datos del idioma huariapano. Yarinacocha, Pucallpa, Peru: Ministerio de Educación & Instituto Lingüístico de Verano. Parker, Steve. 2002. Quantifying the sonority hierarchy. Ph.D. dissertation, University of Massachusetts, Amherst. Parker, Steve (ed.) 2005. Data papers on Papua New Guinea languages, vol. 47: Phonological descriptions of PNG languages. Ukarumpa, Papua New Guinea: Summer Institute of Linguistics. Parker, Steve. 2008. Sound level protrusions as physical correlates of sonority. Journal of Phonetics 36. 55 –90. Payne, Judith. 1990. Asheninca stress patterns. In Doris L. Payne (ed.) Amazonian linguistics: Studies in lowland South American languages, 185 –209. Austin: University of Texas Press. Pike, Kenneth L. 1943. Phonetics: A critical analysis of phonetic theory and a technic for the practical description of sounds. Ann Arbor: University of Michigan Press. Prince, Alan. 1990. Quantitative consequences of rhythmic organization. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26(2). 355–398. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Regnier, Sue. 1993. Quiegolani Zapotec phonology. Work Papers of the Summer Institute of Linguistics, University of North Dakota Session 37. 37– 63. Ridouane, Rachid. 2008. Syllables without vowels: Phonetic and phonological evidence from Tashlhiyt Berber. Phonology 25. 321–359.

25

Steve Parker

Schachter, Paul & Victoria Fromkin. 1968. A phonology of Akan: Akuapem, Asante, Fante. UCLA Working Papers in Phonetics 9. Selkirk, Elisabeth. 1984. On the major class features and syllable theory. In Mark Aronoff & Richard T. Oehrle (eds.) Language sound structure, 107–136. Cambridge, MA: MIT Press. Sievers, Eduard. 1893. Grundzüge der Phonetik zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. Leipzig: Breitkopf & Härtel. Originally published 1881. Silverman, Daniel. 1995. Phasing and recoverability. Ph.D. dissertation, University of California, Los Angeles. Published 1997, New York: Garland. Smolensky, Paul. 1995. On the structure of the constraint component Con of UG (ROA-86). Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Ph.D. dissertation, MIT. Published 1990, New York: Garland. Steriade, Donca. 1988. Reduplication and syllable transfer in Sanskrit and elsewhere. Phonology 5. 73–155. Urbanczyk, Suzanne. 2006. Reduplicative form and the root-affix asymmetry. Natural Language and Linguistic Theory 24. 179 –240. Vance, Timothy J. 1987. An introduction to Japanese phonology. Albany: State University of New York Press. Vance, Timothy J. 2008. The sounds of Japanese. Cambridge: Cambridge University Press. Vennemann, Theo. 1988. Preference laws for syllable structure and the explanation of sound change: With special reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. Whitney, William Dwight. 1889. Sanskrit grammar, including both the classical language, and the older dialects, of Veda and Brahmana. 2nd edn. Cambridge, MA: Harvard University Press. Wichmann, Søren. 2002. Diccionario analítico del popoluca de Texistepec. México, D.F.: Universidad Nacional Autónoma de México. Wright, Richard. 2004. A review of perceptual cues and cue robustness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.) Phonetically based phonology, 34–57. Cambridge: Cambridge University Press. Yuan, Jiahua. 1989. Hanyu fangyan gaiyao [An introduction to Chinese dialects]. 2nd edn. Beijing: Wenzi Gaige Chubanshe. Zec, Draga. 1988. Sonority constraints on prosodic structure. Ph.D. dissertation, Stanford University. Published 1994, New York: Garland. Zec, Draga. 1995. Sonority constraints on syllable structure. Phonology 12. 85 –129. Zec, Draga. 2003. Prosodic weight. In Féry & van de Vijver (2003), 123–143. Zec, Draga. 2007. The syllable. In de Lacy (2007b), 161–194. Zhang, Jie. 2001. The effects of duration and sonority on contour tone distribution: Typological survey and formal analysis. Ph.D. dissertation, University of California, Los Angeles. Published 2002, New York: Routledge.

50

Tonal Alignment Pilar Prieto

1

Introduction

In recent decades, the issue of tonal alignment has been at the forefront of several phonological and phonetic debates in the analysis of intonation. Since the groundbreaking work of Bruce (1977), the autosegmental metrical approach to intonation proposed that intonational patterns were to be represented as autosegmental tone melodies (Pierrehumbert 1980; Beckman and Pierrehumbert 1986; Ladd 1996; and others). Given that melodies are independent from the segments which realize them in this theory (chapter 45: the representation of tone; chapter 14: autosegments), and since the tones are realized potentially over quite long strings, it is a central research issue to find a set of principles for mapping tones to segments. The term tonal alignment thus refers to the temporal implementation of fundamental frequency (F0) movements with respect to the segmental string. Tonal alignment has not only been used in crucial ways as an argument in favor of a given phonological framework, but has also been the focus of debate in itself. This notion has played an important role in current theories of intonational phonology, since relative alignment of tones with the segmentals has been shown to be a crucial piece of information when describing the phonological make-up of the melodic contour. This chapter reviews four important topics in the recent history of phonology in the discussion of which tonal alignment has been a crucial component. One of the important issues in intonational phonology is the investigation of the acoustic correlates that encode intonational categories. Since the beginning of the autosegmental metrical approach to intonation, tonal alignment has been claimed to play a central role in encoding intonational contrasts. Pierrehumbert (1980) and Pierrehumbert and Steele (1989) showed that the timing of F0 peaks or valleys with segments functions contrastively in English, and that early-aligned pitch accents are phonologically distinct from late-aligned pitch accents. In the decades since the publication of these studies, a body of experimental research has shown that tonal alignment cues semantic distinctions in a number of languages and that it can be perceived in a near-categorical fashion (e.g. Kohler 1987; Niebuhr 2007 for German; D’Imperio and House 1997 and D’Imperio 2000 for Neapolitan Italian; Gili-Fivela 2009 for Pisa Italian; Pierrehumbert and Steele The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0050

2

Pilar Prieto

1989 and Dilley 2007 for English). In §2 we will review recent experimental evidence that elucidates the role of tonal alignment in encoding intonational distinctions in a number of languages. The relationship between tonal association and tonal alignment has been a central issue in the tonal representation debates within the autosegmental metrical theory of intonation. Though the autosegmental metrical representational proposal has met with considerable success in accounting for melodic patterns in a variety of languages, the literature on tonal representation has identified a few phenomena that resist transparent analysis. Two such phenomena have to do with the metrical part of the model and the standard interpretation of the relationship between phonological association and phonetic alignment. It has recently been claimed that the theoretical concept of starredness is somewhat unclear and that its definition cannot be based solely on phonetic alignment (Arvaniti et al. 2000; Prieto et al. 2005). In §3 we describe the standard view of the relationship between phonological association of tones and phonetic alignment and then review some recent proposals on the topic. Another important goal of several models of intonation has been to develop a phonetic model of tonal alignment. Within these models, it is a central issue to determine what part of the variation in the realization of the tune-to-text mapping is due to phonetic implementation and what part is phonological and is accounted for in a phonological representation (either of the tone melodies or of prosodic or segmental anchors for tones). A body of work on tune–text alignment has shown that, apart from phonological distinctions in alignment, a variety of phonetic factors, such as tonal crowding, speech rate and syllable structure influence the fine-grained patterns of F0 location in predictable ways. For example, it has been demonstrated that time pressure from the right-hand prosodic context (i.e. the proximity of an upcoming accent or boundary tones) is crucial in determining the location of H peaks (see e.g. Silverman and Pierrehumbert 1990 for English and Prieto et al. 1995 for Spanish). Recent work has shown that when such right-hand prosodic effects are excluded (i.e. when the tonal features under investigation are not in the vicinity of pitch accents or boundary tones), the alignment of F0 peak targets is consistently governed by segmental anchoring (Arvaniti et al. 1998 for Greek; Ladd et al. 1999 for English). Similarly, other work on production and perception supports the hypothesis that prosodic structure must play an essential part in our understanding of the coordination of pitch gestures with the segmentals and that listeners are able to employ these fine details of H tonal alignment due to syllable structure or within-word position to identify lexical items (D’Imperio et al. 2007b; Prieto et al., forthcoming). In §4 we review recent proposals regarding phonetic models of tonal alignment and the role of prosodic structure in the implementation of F0 tonal alignment patterns. Finally, tonal alignment studies have also been used to test specific predictions by different phonological models of prosody and intonation. Arvaniti and Ladd (2009) provide a useful example of how a production study on alignment can be used to test specific predictions by target-based vs. configuration-based models of intonation (chapter 32: the representation of intonation). As we will see below, Arvaniti and Ladd undertook a very detailed phonetic study of the intonation of Greek wh-questions and tested different predictions about tonal implementation. The F0 alignment data showed predictable adjustments in alignment depending on the location of adjacent tonal targets. The authors conclude

Tonal Alignment

3

that models that specify the F0 of all syllables, and models that specify F0 by superposing contour shapes for shorter and longer domains, cannot account for predictable variation without resorting to ad hoc tonal specifications, which, in turn, do not allow for phonological generalizations about contours applying to utterances of different lengths. In §5 we review the evidence coming from a variety of tonal alignment studies that test specific predictions from different phonological models of intonation. In the following sections, we present and discuss each of these four topics, providing the relevant data and highlighting some of the unresolved issues.

2

The role of tonal alignment in distinguishing intonational categories

One of the key discoveries within work on intonation is the fact that tones in intonational languages are associated with either metrically prominent syllables (pitch accents) or prosodic edges (boundary tones). Many theories of intonational phonology thus draw a clear distinction between the two sorts of tonal units, i.e. tonal entities associated with prominent or metrically strong syllables and tonal entities associated with edges of prosodic domains. Within the autosegmental metrical (AM) approach to intonation initially developed by Pierrehumbert (1980), she argues that the English intonation system consists of an inventory of tonal units, each consisting of either one or two tones, which can be High (H) or Low (L) (see chapter 14: autosegments; chapter 116: sentential prominence in english). These tones can either be associated with metrically strong syllables (and represented with a *, i.e. H* and L*) or be associated with prosodic edges (and represented with a %, i.e. H% and L%). Tonal units can be monotonal or bitonal. In the case of tonal units associated with prominent syllables, or pitch accents, Pierrehumbert proposed a phonological inventory of six pitch-accent shapes for English (H*, L*, H*+L, H+L*, L*+H, L+H*), some of them encoding alignment differences. Crucially, the AM model started to make use of the star notation (*) in bitonal pitch accents to indicate tonal association with metrically strong syllables and relative alignment – see §3 for a review of the starredness concept. The autosegmental representations in (1) capture the fact that the LH shape is aligned differently in the two contrastive pitch accents exemplified in figure 50.1. While L*+H has a low tone (L) on the stressed syllable and a high tone (H) trailing it, L+H* has a high tone on the stressed syllable with a low tone leading it: (1)

a.

Only a millionaire L*+H

b. Only a millionaire L+H*

In sum, an important proposal of the AM model of intonation, based on Bruce’s (1977) analysis of the tonal alignment contrast between Accent I and Accent II in

4

Pilar Prieto

F0

(a) O n l y

450 400 350 300 250 200 150

m i l l i o n a i r e

a

L* + H 0

0.5 Seconds

1

F0

(b) O n l y

450 400 350 300 250 200 150

a

m i l l i o n a i r e

L 0

+

H*

0.5 Seconds

1

Figure 50.1 The fundamental frequency contour of the utterance Only a millionaire spoken with two different pitch accents on millionaire: the late-aligned pitch accent, which indicates incredulity or uncertainty (a), and the early-aligned pitch pattern, which indicates assertion (b). The vertical cursor is placed at the [m] release in millionaire. Figure reproduced from Pierrehumbert and Steele (1989: 182)

Swedish, is that pitch accent types can be phonologically distinguished by their relative alignment with the metrically prominent syllable. Pierrehumbert (1980) shows that tonal alignment functions contrastively in English and that earlyaligned pitch accents are phonologically distinct from late-aligned pitch accents. Figure 50.1 shows two intonation patterns of the utterance Only a millionaire spoken with two different pitch accents on millionaire: the late-aligned pitch accent, which indicates incredulity or uncertainty (a), and the early-aligned pitch pattern, which indicates assertion (b). In their seminal paper, Pierrehumbert and Steele (1989) performed an imitation task with the two intonation patterns of the abovementioned utterance Only a millionaire (see Figure 50.1). They created a synthesized continuum of several steps of alignment between the two, and asked subjects to imitate the utterance. The results of the imitation task revealed the existence of two separate phonological categories. The authors argued that if the subjects had been able to reproduce the full range of the continuum in their imitation, peak alignment differences could be regarded as gradient. However, since they found that by and large the distribution

Tonal Alignment

5

of peak alignment was bimodal in the imitation data, they therefore concluded that the distinction between early and late peak alignment was categorically distinct. Pierrehumbert and Steele’s paper represented an important first step in a series of experimental investigations on the perception of tonal alignment (see chapter 98: speech perception and phonology). Since then, a body of experimental research has demonstrated that tonal alignment cues intonational meaning distinctions in a number of languages (e.g. Kohler 1987 and Niebuhr 2007 for German; D’Imperio and House 1997 and D’Imperio 2000 for Neapolitan Italian; Gili-Fivela 2009 for Pisa Italian; Dilley 2007 for English). The issue of whether a certain pair of intonational contrasts can be accompanied by categorical differences in meaning and whether these contrasts are perceived in a discrete or a gradient fashion has been an important research question in the field of intonation. A number of experimental methods have been used to study what is categorical or linguistic in intonation and what is paralinguistic and gradient (see the reviews in Gussenhoven 2004, 2006; also chapter 89: gradience and categoricality in phonological theory). In what follows we review recent studies that have provided evidence from a number of languages on the role of tonal alignment in encoding intonational distinctions. All in all, these articles provide robust experimental evidence for the claim that changes in F0 alignment of peaks and valleys are especially salient and cue phonological distinctions across languages. This evidence has been generally interpreted as direct support for AM theory, as tonal alignment differences in this model are encoded phonologically at the pitch accent level. Kohler’s (1987) paper was the first to apply the Categorical Perception paradigm to alignment data and to show that alignment contrasts can be perceived categorically. The Categorical Perception paradigm involves firstly an identification/classification task in which the listeners have to categorize stimuli taken from a continuum, and secondly a discrimination task in which listeners are asked to judge pairs of stimuli as being either the same or different. For perception to be considered categorical, a peak of discrimination is expected at the point in the acoustic domain that separates the two categories (for a review, see Dilley 2007). Kohler (1987) employed the complete paradigm to investigate the perception of a set of F0 contours in German involving rises with a continuum created between early and medial peaks. He found that the early peak was associated with finality (“knowing,” “coming to the end of an argument”), and the medial peak with openness (“observing,” “starting a new argument”). The results of both tasks of the paradigm revealed categorical changes in the identification of early vs. medial peaks, with a discrimination maximum across the category boundary. More recently, Niebuhr (2007) carried out a series of perception experiments with the same German alignment contrasts and showed that the function-based identification of the peak categories is influenced not only by peak synchronization, but also by peak shape and height. In general, though, his findings corroborate the existence of the two categories in German intonation and support the idea that the timing of the peak movements with regard to the accented vowel is important for their perceptual differentiation. Similar results have been obtained for American English tonal alignment contrasts. Following Pierrehumbert and Steele’s (1989) investigation, a number of studies have examined the distinction between an early-aligned pitch accent (L+H*) and a late-aligned pitch accent (L*+H) in American English. In the most comprehensive study, Dilley (2007) conducted a series of perception experiments

6

Pilar Prieto

with the two pairs of accents attested in American English (H* and H+L*, and L* and L+H*), an identification task, two types of discrimination tasks and an imitation task. Evidence of discrimination maxima that aligned well with identification crossover points in the identification task demonstrated categorical perception for intonation and provided converging evidence with earlier results by Kohler (1987). Moreover, converging evidence for the categorical perception of intonation categories was obtained from the imitation study. Though Kohler (1987) and Dilley (2007) are advocates of the application of the Categorical Perception paradigm to intonation, few other studies have shown clear evidence of categorical perception, i.e. with a clear discrimination peak in the expected position. The discrimination functions observed differ between studies, and in the majority of cases no discrimination peaks appear at the category crossover point revealed by the identification test. One such example is described in Gili-Fivela’s (2009) article. She investigated the contrast between narrow focus and narrow contrastive focus in Pisa Italian, represented as H* and H*+L. In Pisa Italian, as in other languages, narrow contrastive focus is expressed through the use of retracted pitch peaks and an increase in pitch height. Gili-Fivela applied the Categorical Perception paradigm to the data, with both identification and discrimination tasks being performed, and also an imitation task. She manipulated both the alignment and scaling patterns of a rising pitch accent in narrow focus and a rising-falling pitch accent in contrastive narrow focus. The results showed that while there is a clear difference between a narrow focus pattern and a contrastive focus pattern in production, the contrast might not be categorically perceived, as the identification and discrimination functions do not correspond to an abrupt shift in identification aligned with a discrimination peak. Other studies have shown that the slope of the rise and the shape of the peak also contribute to tonal contrast identification. D’Imperio and House (1997) and D’Imperio (2000) investigated the distinction between questions and statements in Neapolitan Italian. In Neapolitan Italian, questions and statements are characterized by a rise in pitch that occurs in the vicinity of the accented syllable. The materials in D’Imperio and House (1997) consisted of a series of stimuli in which the F0 peak of a rising-falling pitch accent was shifted forward and backwards within the accented syllable. Neapolitan listeners performed an identification task in which they listened to the stimuli and then classified each of them as either a question or a statement. The results showed that questions and statements are primarily distinguished by the relative alignment of the rise in a rise-fall pattern in the accented syllable. In subsequent experiments using this same contrast, D’Imperio (2000) showed that both details of the temporal alignment of target tones and the shape of the peak contribute to the identification of the contrast between questions and statements in this language. Moreover, she found that syllable structure detail modifies acoustic target alignment but does not modify the crossover point between the two categories (for more details, see §4). New experimental paradigms have been recently applied to study the role of tonal alignment in spoken language processing. Chen et al. (2007) adopted the eye-tracking paradigm to investigate the role of pitch accent type and deaccentuation in online processing of information status in British English.1 It was found that 1

For a review of the eye-tracking paradigm applied to prosody research, see Watson et al. (2006) and Watson et al. (2008).

Tonal Alignment

7

two types of pitch accents (H*L and L*HL) create a strong bias toward newness, whereas deaccentuation and the L*H pitch accent create a strong bias toward givenness. Watson et al. (2008) also used the eye-tracking paradigm to investigate whether the presence of a pitch accent difference between L+H* and H* in English biases listeners toward interpreting a temporarily ambiguous noun as referring to either a discourse-given or a discourse-new entity. Participants had to perform a word-recognition task (for example, candle vs. candy) and pick up one of the two competing objects, while their eye movements were being monitored. They found that although listeners interpreted these accents differently, their interpretive domains overlapped. L+H* created a strong bias toward contrast referents, whereas H* was compatible with both new and contrast referents. The electro-encephalography (EEG) technique, a procedure which measures electrical activity of the brain and which allows for the non-invasive measuring of brain activity during cognitive processing, has also been used to study pitch processing. For example, Fournier et al. (2010) used this technique to investigate the tonal and intonational pitch processing of some tonal contrasts (some of them alignment contrasts) by native speakers of the tonal dialect of Roermond Dutch as compared to a control group of speakers of Standard Dutch, a non-tone language. A set of words with identical phoneme sequences but distinct pitch contours, which represented different lexical meanings or discourse meanings (e.g. statement vs. question), were presented to both groups. The stimuli were arranged in a mismatch paradigm, under several experimental conditions: in the first condition (lexical), the pitch contour differences between stimuli reflected differences between lexical meanings; in the second condition (intonational), the stimuli differed in their discourse meaning. In these two conditions, both native and non-native responses showed a clear magnetic mismatch negativity in a time window from 150 to 250 msecs after the divergence point of standard and deviant pitch contours. In the lexical condition, a stronger response was found over the left temporal cortex of speakers of standard as well as non-standard Dutch. Crucially, in the intonational condition, the same activation pattern was observed in the control group, but not in the group of Roermond Dutch speakers, who showed a right-hemisphere dominance instead. Thus the lateralization of pitch processing was condition-dependent in the Roermond Dutch group only, suggesting that processes are distributed over both temporal cortices according to the functions available in the grammar. Finally, researchers have begun to consider the role of potential articulatory landmarks and the coordination or alignment between tonal gestures (measured as F0 turning points) and oral constriction gestures. Recent work by Mücke et al. (2006), D’Imperio et al. (2007a), and Mücke et al. (2009) has investigated alignment patterns for three different languages (Italian, German, and Catalan) by using electromagnetic mid-sagittal articulography (EMMA) for capturing oral constriction gestures alongside acoustic recordings. The end of pitch movements in bitonal pitch accents co-occurs with the minima and maxima of the closing gesture of C2 in C1V.C2 and C1VC2 sequences. In all these studies, such pitch targets were seen to be more closely aligned in time with articulatory landmarks than with acoustic ones. However, there was some variation as to the articulatory landmark which served as an anchor for the tonal target. For example, in German nuclear LH accents, the H peaks co-occurred with the intervocalic C target, whereas in pre-nuclear accents peaks co-occurred with the target for the following vowel (what is called “accent shift”; Mücke et al. 2009). In Catalan it was the consonantal

8

Pilar Prieto

peak velocity rather than the consonantal target which served as the landmark. Such an apparently small alignment difference in the articulatory anchor type may be used by speakers to make (or contribute toward making) phonological distinctions, as in Neapolitan, where H in L*+H (questions) aligns with the maximum constriction, and H in L+H* (statements) aligns with peak velocity (see D’Imperio et al. 2007a).

3

Phonological encoding: Tonal association and tonal alignment

The topic of this section is the relation between phonological association and phonetic alignment of tones and how it is encoded in a representational system. The starting point is provided by the autosegmental metrical approach to intonation, which has developed an explicit phonological representational approach that has been applied to a variety of languages (Pierrehumbert 1980; Pierrehumbert and Beckman 1988; Ladd 1996; Gussenhoven 2004; among others). Though the AM representational proposal can account for melodic patterns in a variety of languages, there are a number of areas that remain unresolved. Two of these issues relate to how to interpret the relationship between tones and metrically strong syllables in the AM model, namely the concept of starredness on the one hand and the interpretation of the relationship between phonological association and phonetic alignment on the other. The AM phonological representation of pitch accents encodes “autosegmental” information (or pitch accent shapes, LH or HL) and “metrical” information, i.e. information about the association of tones with metrical constituents and the relative alignment of tones with the metrically prominent syllable. The surface alignment of tones is basically derived from the use of the star notation (*). The star notation encodes two complementary things: (i) phonological association between pitch accent shapes and stressed syllables – in other words, a tone gets a star when it is associated to a metrically strong position; (ii) relative alignment in bitonal pitch accents – i.e. the tone that gets the star is the one that is directly linked to the metrically strong position. In bitonal accents, the question of which tone in LH or HL accent shapes should be assigned a star is not completely straightforward. On this issue, Pierrehumbert’s original definition states that “a strength relationship is defined on the two tones of bitonal accents and that it is the stronger tone which lines up with the accented syllable” (Pierrehumbert 1980: 76–77). According to this definition, it is ambiguous whether the star notation * indicates phonetic alignment between the tonal unit and the stressed syllable or just a “looser” phonological association. Similarly, Pierrehumbert and Beckman (1988: 234) note that “the * diacritic marks which tone of a bitonal accent is aligned with stress.” Arvaniti et al. (2000: 120) state that “phonetically this use of the star is to be interpreted as signifying that the starred tone is aligned in time with the stressed syllable.” In subsequent work, one of the most common interpretations of the star notation is that the starred tone is phonetically aligned with the stressed syllable, and thus a strict temporal alignment between the tone and its tone-bearing unit is expected. Recently, attention has been drawn to the various problems created by the representational ambiguity of the star notation. One of them is that it can be difficult to decide between competing AM analyses of bitonal accents, because the same

Tonal Alignment

9

contours can be transcribed in different ways (Prieto et al. 2005). For example, let us compare the surface alignment of the tones described by the English and Spanish L+H* – L*+H contrasts according to, respectively, Pierrehumbert (1980) and Sosa (1999). Even though the two phonological units capture the two-way phonological contrast present in both languages, the same labels L+H* and L*+H refer to different phonetic realizations (or alignment patterns) in the two languages. In fact, English L+H* corresponds to Spanish L*+H. This difference between the notational systems is caused by different interpretations of the star notion: while in the English notation the star is interpreted as an indication of phonological association between the tone and the prominent syllable, in Spanish it is interpreted as phonetic alignment, that is, the star is indicating whether the H peak is aligned (H*) or not aligned (L*) with the stressed syllable. (2)

Schematic representation of L+H* and L*+H a.

English (after Pierrehumbert 1980) L+H* L*+H

b.

Spanish (after Sosa 1999) L+H* L*+H

In addition, some authors have pointed out that the theoretical concept of starredness is ill-defined and cannot be based solely on phonetic alignment (Arvaniti et al. 2000). Arvaniti et al. present evidence from Greek of the types of problems that arise when phonetic alignment to the accented syllable is taken to be the exponent of association of tones with segments. As they note: we show that there exist pitch accents that are clearly bitonal but neither tone is, strictly speaking, aligned with the accented syllable. We argue from this fact that association cannot be based on phonetic alignment in any straightforward way and that a more abstract and rigorously defined notion of starredness is required.

In Greek rising pitch accents in pre-nuclear position, typically, neither L nor H is phonetically aligned with the stressed syllable: in most cases, the L is consistently aligned before the beginning of the accented syllable (5 msecs on average before the onset), and H displays more variability and is typically located in the posttonic. Thus, these authors conclude that “if alignment is the sole exponent of the association of tones to segments, phonetic variability in this domain becomes a crucial issue when the phonological structure of a bitonal accent is in question” (Arvaniti et al. 2000: 121). We take it as essentially correct that a one-to-one relationship between phonological association and phonetic alignment is difficult to maintain in the current AM model. In a recent proposal, Prieto et al. (2005) describe the contrastive possibilities of alignment of rising accents in three Romance languages, Central Catalan, Neapolitan Italian, and Pisa Italian (see also chapter 2: contrast). According to these authors, these Romance languages provide evidence that small differences in alignment in rising accents must be encoded phonologically. To account for such facts within the AM model, they develop the notion of “phonological anchoring” as an extension of the concept of secondary association originally proposed by Pierrehumbert and Beckman (1988). They propose that the

10

Pilar Prieto

phonological representation of pitch accents needs to include two independent mechanisms to encode alignment properties with metrical structure: (i) encoding of the primary phonological association (or affiliation) between the tone and its tone-bearing unit; and (ii), for some specific cases, encoding of the secondary phonological anchoring of tones to prosodic edges (i.e. moras, syllables, and prosodic words). (3) shows the schematic representation of the primary and secondary associations of a phrasal H within the accentual phrase in Japanese (Pierrehumbert and Beckman 1988: 129). The solid line indicates primary association to the accentual phrase a and the dashed line secondary association to the second sonorant mora [ within the accentual phrase. (3)

Japanese (after Pierrehumbert and Beckman 1988: 129)

[

a

accentual phrase

[

mora

H

tone tier [+son] [+son]

phoneme tier

The Romance data provide crucial evidence of mora-edge, syllable-edge, and wordedge H tonal associations and suggest that not only peripheral edge tones seek secondary associations. In this way, the specification of metrical anchoring points in the phonological representation offers a more transparent analysis of the alignment contrasts found in Romance languages and, ultimately, can help in the task of defining a more transparent pitch accent typology. Finally, Prieto et al. (2005) argue that such an approach makes the mapping from phonological representation to surface alignment patterns more explicit, and that it thus allows for more straightforward cross-linguistic comparisons. The evidence described above shows that even though AM representations are adequate when it comes to characterizing the minimal contrasts in pitch accent types found in different languages, the proper procedures by which to map phonological representations and the surface alignment of tones (through the use of the star notation) are still somewhat unclear. This is because the specific details of the coordination between tones and the segments that are linked to the structural unit are not part of the phonological representation itself. We thus agree with Arvaniti et al.’s (2000: 130) suggestion “that the task for the future is to refine the notion of the phonological association of tones in intonational systems.” In the near future, the contrastive possibilities of alignment found cross-linguistically need to be explored. This will provide firm ground from which to advocate a further refinement of the metrical side of the AM model.

4

Phonetic models of tonal alignment

Apart from changes in tonal alignment which have phonological effects, i.e. which encode a difference in meaning (see §2 and §3), tonal alignment is influenced by

Tonal Alignment

11

a variety of phonetic factors, such as tonal crowding, speech rate, segmental composition, and syllable structure composition. These fine-grained F0 alignment differences do not affect meaning or representation, and are instead considered to arise from differences in phonetic implementation rather than phonological representation. In this section we review some of the production studies that have investigated the influence of such factors on tonal alignment patterns and the perception studies that have demonstrated that some of these effects are employed by native speakers in lexical access tasks. Cross-linguistically, the location of fundamental frequency peaks (or H values) has been shown to be greatly affected by the right-hand prosodic context, in such a way that the peak is retracted before upcoming pitch accents and boundary tones (see Silverman and Pierrehumbert 1990 for English and Prieto et al. 1995 for Spanish, for example). Prieto et al. (1995) examined the peak placement patterns in rising accents in Spanish and found the following: (i) the location of the start of the F0 rise is fairly constant (generally at the onset of the accented syllable); (ii) as in English, the duration of the rising gesture is highly correlated with syllable duration. These results show that the slope and/or duration of a speech F0 movement are not constant, as claimed by the fixed rise-time hypothesis (Fujisaki 1983; ’t Hart et al. 1990; and others), but are instead governed by the coordination of the movement with the segmental string. Both studies demonstrated that a successful quantitative model of peak placement must contain at least two factors, namely the duration of the accented syllable and the distance in syllables to upcoming pitch accents or boundary tones. The Segmental Anchoring Hypothesis (henceforth SAH), as articulated by Ladd et al. (1999) on the basis of work by Prieto et al. (1995) and Arvaniti et al. (1998), refers to the idea that the slope of tonal movements is not invariant, but rather is specifically related to segmental anchors. Arvaniti et al. (1998) found an unexpected and consistent stability effect when little or no tonal pressure was exerted on the pitch accent. In a Greek word such as [pa’ranoma] ‘illegal’, the H target in the LH pitch accent associated with the test stressed syllable [’ra] was consistently aligned over – or “anchored to” – the frontier between the post-accentual onset and the following vowel ([n] and [o]). This clearly contradicts the traditional “constant slope” and “constant duration” hypotheses (i.e. the fixed rise-time hypothesis: Fujisaki 1983; ’t Hart et al. 1990; and others). The SAH says that both the beginning and the end of a rising or falling F0 movement are anchored to specific points in the segmental string, such as the beginning of the stressed syllable or the following unstressed vowel, and consequently the duration of the F0 movement is strongly dependent on the duration of the segmental interval between the anchor points. As we will see below, work on the effects of lower prosodic structure levels such as the syllable or the prosodic word on tonal alignment shows that we need to refine the SAH to incorporate these findings. Recent work on tonal alignment in different languages has shown that the position of the peak tends to change across syllable structure types (e.g. Rietveld and Gussenhoven 1995 for Dutch; D’Imperio 2000 for Neapolitan Italian; Prieto and Torreira 2007 for Peninsular Spanish; Prieto 2009 for Catalan). For example, D’Imperio (2000) found that the peak was located closer to the vowel offset in closed syllables in Neapolitan Italian. While in open syllables the peak was aligned with the end of the accented vowel, in closed syllables the peak was somewhat retracted and located within the coda consonant. This same effect of coda

12

Pilar Prieto

consonants on alignment has been detected in both rising accents in various languages (see citations above) and falling nuclear accents in Catalan (Prieto 2009). The results indicate that while the beginning of the falling accent gesture (H) is tightly synchronized with the onset of the accented syllable, the end of the falling gesture (L) is more variable and is affected by syllable structure: in general, while in open syllables the end of the fall is aligned roughly with the end of the accented syllable, in closed syllables it is aligned well before the coda consonant. D’Imperio et al. (2007b) hypothesized that Neapolitan listeners might capitalize on the alignment regularity for the perception of lexical contrast. Specifically, their hypothesis was that listeners of Neapolitan Italian might identify more closed syllable items when tonal alignment details are congruent with those for this type of syllable structure (see also Petrone 2008). In order to test this hypothesis, two natural productions of the words nono ‘ninth’ and nonno ‘grandfather’, both carrying a yes/no question nuclear accent, were manipulated in two ways. First, the researchers modified the length of the stressed vowel and the following consonant in five steps, in order to shift the perception of each item from nono to nonno and vice versa. Then, tonal alignment was shifted earlier, in four steps, without changing the percept of the question to that of a statement but merely creating question patterns that would be more or less congruent with the syllabic structure of the base. Thirteen Neapolitan listeners identified the stimuli as either nono or nonno. Significantly, the results showed that the alignment manipulation produced a category boundary shift in the nonno base stimulus series, but no effect in the open syllable series, supporting the hypothesis that fine detail of tonal alignment not only is employed to signal pragmatic contrast but may also be stored as part of the phonological specification of lexical items. Similarly, acoustic work on a variety of languages has shown that H peaks are consistently affected by the position of the accented syllable within the word (for English, see Silverman and Pierrehumbert 1990, and for Spanish Prieto et al. 1995). In general, peaks tend to shift backwards as their associated syllables approach the end of the word: in other words, the distance from the beginning of the accented syllable to the peak is longer in words with antepenultimate stress than in words with penultimate stress, which in turn show a longer distance than in words with final stress. In order to correct for the potentially confounding effects of stress clash (or distance to the next accented syllable), Prieto et al. (1995) analyzed a subset of the data obtained from test syllables in different positions in the word (e.g. número ‘number’, numero ‘I number’, numeró ‘(s)he numbered’). Their materials consisted of word sequences in which there was a distance of two unstressed syllables between one accented syllable and the next (e.g. número rápido ‘quick number’, numero nervioso ‘I number nervously’, and numeró regular ‘(s)he numbered in a regular way’). The three diagrams in Figure 50.2 show a schematic representation of the difference in F0 timing patterns in the three conditions, número rápido, numero nervioso, and numeró regular. A significant effect of word position on different measures of peak alignment was found in all the comparisons. Similarly, in Silverman and Pierrehumbert’s (1990) model of F0 peak location, the dropping of the variable “Word-Boundary” (while leaving the variable “Stress Clash” as a main predictor) significantly worsened the fit of the model. Prosodic word effects seem to suggest the possibility that the end of the word (and not only the presence of upcoming accents or boundary tones) is acting as a kind of prosodic boundary that exerts prosodic pressure on H tonal targets and

Tonal Alignment

n

ú m

e

r

o

m e r o

n

e r

r

ó

r

e

g

13

u

Figure 50.2 A schematic representation of the difference in F0 timing patterns in the three conditions, número rápido, numero nervioso, and numeró regular

that this effect can be exploited in word boundary identification tasks. Prieto et al. (forthcoming) performed a set of production and perception experiments that dealt with potentially ambiguous utterances distinguished by word boundary location in Catalan and Spanish (e.g. Catalan mirà batalles ‘(s)he looked at battles’ vs. mirava talles ‘I/(s)he used to look at carvings’; Spanish da balazos ‘(s)he fires shots’ vs. daba lazos ‘I/(s)he gave ribbons’). For the perception experiments, they hypothesized that relative peak location would help Catalan and Spanish listeners in terms of lexical access. The results of the production experiments showed that the prosodic word domain has a significant shifting effect on F0 peak location, and the results of the perception experiments showed that these alignment patterns are actively used by listeners in word identification tasks. In general, the results of studies on lexical access (D’Imperio et al. 2007b; Prieto et al., forthcoming) support the hypothesis that listeners are able to employ fine allophonic details of H tonal alignment due to syllable structure or within-word position to identify lexical items. This empirical evidence demonstrates that prosodic structure must play an essential role in our understanding of the coordination of pitch gestures with the segmentals and argues in favor of a view supported by other work that prosodic structure is manifested in details of articulation.

5

Tonal alignment: Evidence for target- vs. configuration-based theories of intonation

As pointed out in §2, work on tonal alignment has provided robust experimental evidence that changes in the synchronization of peaks and valleys with segmental landmarks are key perceptual cues for phonological distinctions across languages. This evidence has been interpreted as direct support for AM theory, which is widely held to afford a number of advantages over other discrete tone theories, as tonal alignment differences in this model are encoded phonologically in pitch accent units. Alignment studies have also been used to test specific predictions about different phonological models of prosody and intonation. For example, one of the old controversies in intonation studies surrounds the relative merits of the target-based vs. configuration-based theories of intonational primitives (see Ladd 1996: §1.2 for a review; also Arvaniti and Ladd 2009). The target-based model (also called target-and-interpolation model by Arvaniti and Ladd 2009) is the phonetic basis of AM intonational phonology, which has become the dominant phonological framework for analyzing intonation. This model assumes that certain points

14

Pilar Prieto

in the contour (e.g. local targets or F0 maxima and F0 minima) reflect phonologically specified targets and thus derive the intonational contour by defining the tonal targets and then connecting those through an interpolating F0 curve that goes from one target to the next. In recent years there has been accumulating evidence from tonal alignment studies that L and H tones behave as static targets and that they align with the segmental string in extremely consistent ways. Typically, in a variety of languages, the L valley of pre-nuclear rises is precisely aligned with the beginning of the accented syllable (see Prieto et al. 1995 for Spanish, Arvaniti et al. 1998 for Greek, and Ladd et al. 1999 for English, for example). Moreover, some studies have shown that this precise L intonational alignment with word or syllable boundaries is used by listeners in lexical identification tasks. For example, Ladd and Schepman (2003) showed that the different alignment of L in minimal pairs like Norman Elson/Norma Nelson is a useful cue to the word-boundary distinction between them. If L alignment was modified experimentally in such ambiguous phrases, this affected the listeners’ judgments in the identification task. Similarly, a recent study on the tonal marking of the French Accentual Phrase (AP) by Welby (2003) showed that the L tone associated with the left edge of the first content word of the AP is aligned at the boundary between the last function word and the first syllable of the first content word. Welby’s results for perception showed that French listeners use the alignment of the L tone as a cue for lexical access (in pairs such as mes galops ‘my gallops’ and mégalo ‘megalomaniac’). All in all, these alignment results, as well as many scaling results, have been interpreted in favor of the target-based hypothesis (for a review, see Ladd 1996). On the other hand, configuration-based theories (also called concatenation models by Arvaniti and Ladd 2009) treat the contour as the result of stringing together entire tonal sequences (not necessarily straight lines) of various lengths. Traditional intonational descriptions of the so-called “British school” (e.g. Crystal 1969; O’Connor and Arnold 1973) and the approach adopted by the Eindhoven-based Instituut voor Perceptie Onderzoek (IPO; e.g. ’t Hart et al. 1990) are of this sort, as is the more recent syllable-concatenation model proposed by Xu and colleagues (e.g. Xu and Wang 2001; Xu and Xu 2005). There have been several results reported in the literature that provide support for a configuration-based theory of intonation. For example, as mentioned above, D’Imperio and House (1997) undertook a perception experiment that investigated the contrast between questions and statements in Neapolitan Italian. They wanted to determine whether the major perceptual cue to this category distinction involved only the temporal alignment of the high-level target with the syllable or if instead the category percept also depended on the presence of a rising or falling melodic movement within the syllable nucleus. The results showed that the primary perceptual cue for questions is a rise through the vowel, while the primary cue for statements is a fall through the vowel. D’Imperio and House claimed that their results confirmed the second hypothesis, in that perceptually a rise in the vowel was the most important cue for the question, while a fall in the vowel was the most important cue for the statement, thus supporting the notion that pitch movements through areas of stability are perceptually important for identifying tonal categories. Contrasting results were obtained by Arvaniti and Ladd (2009), who carried out a production study in which they used acoustic alignment measures to test specific predictions about different phonological models of intonation. This involved

Tonal Alignment

apopu

na

NV PNV

me ton

menelo

PNSV

FV

frequency (Hz)

275

milane

15

75

IL

NL NH 0.2

0.4

L1 0.6

0.8 time (sec)

L2FH 1.0

1.2

1.4

Figure 50.3 The waveform, spectrogram, and F0 contour of [apo’pu na mi’lane me ton ’menelo] ‘Where could they be speaking to Menelos from?’ (speaker KP), illustrating the measurements taken on the F0 contour and relevant segmental onsets. Figure reproduced from Arvaniti and Ladd (2009: 55)

undertaking a very detailed phonetic study of the Greek wh-question melody. According to their results, certain points in the Greek wh-question melody show little variability in scaling and predictable variability in alignment. A close analysis of the F0 alignment data showed that (i) the exact contour shape depended on the length of the question, and (ii) the position of the first peak and the low plateau depended on the position of the prominent anchor syllables. The study also showed predictable adjustments in alignment depending on the proximity of adjacent tonal targets. Figure 50.3 shows the F0 contour of a long wh-question. In long wh-questions, the contour starts with a rise from a low F0 point, the fall from the peak is relatively shallow, and the following low F0 stretch is long. By contrast, short wh-questions consist of a high tone associated in time with the stressed syllable of the wh-word, followed by a rapid fall to a stretch of low F0, followed by a small rise. Arvaniti and Ladd (2009) argue that the Greek wh-question data strongly argue in favor of a target-based model of intonational phonology like that proposed by the autosegmental metrical framework of intonational phonology, and in particular in favor of the notion of sparse tonal specification. This is because one key assumption of the autosegmental metrical framework is that there is not necessarily any role for the syllable in modeling utterance contours. Rather, F0 targets can be temporally anchored to the segmental string in a variety of ways. This is exactly what we find in the wh-contour data in Greek, as the alignment and scaling adjustments observed in the contour are totally predictable, and depend on the length and tonal crowding manipulations in the target utterance. Arvaniti and Ladd claim that these predictable effects cannot be explained by superposition models of intonation, such as Fujisaki’s (1983) command–response model, or by configuration-based models that specify F0 by superposing contour shapes for shorter and longer domains, since both of them lack the mechanisms to account

16

Pilar Prieto

for effects such as the truncation of targets or asymmetrical adjustments to the larger tonal domains. Similarly, models that specify the F0 of all syllables (like Xu and colleagues’ model), and thus assume that all syllables are specified for tone, cannot account for lawful variation except by using ad hoc tonal specifications, which, in turn, do not allow for phonological generalizations about contours applying to utterances of different lengths.

6

Conclusion

In recent decades, the issue of tonal alignment has been a key focus of phonological research in intonational phonology. We now have solid evidence coming from different languages that F0 alignment differences can convey intonational contrasts, and that these alignment differences can be perceived in a near-categorical way. In this chapter, we have reviewed this work, and the use of several techniques in the investigation of tonal alignment processing (§2). As we have seen, a wide range of methodological paradigms have been applied to alignment research, including acoustic and articulatory analyses of speech productions, judgments and reaction times obtained during identification and discrimination tasks, measurements of brain activity, and eye movements. A recent debate within the autosegmental metrical approach to intonation has been how to represent these phonological contrasts in tonal alignment. As has been reported before, this theory does an especially good job of accounting for why tone alignment differences can convey intonational contrasts. In the AM framework, the star notation encodes both phonological association of the tones with a stressed syllable and the relative alignment in bitonal pitch accents. However, though the AM representations can adequately characterize the minimal contrasts in pitch accent types found in different languages, the procedures for mapping the surface alignment of tones through the use of the star notation onto phonological representations are still somewhat unclear. This chapter has reviewed some recent proposals regarding this issue which highlight the need to further investigate the contrastive possibilities of alignment found cross-linguistically. Apart from the phonological contrasts induced by tonal alignment, F0 tonal patterns are influenced by a variety of phonetic factors, such as prosodic crowding, speech rate, segmental composition, upcoming syllable structure, and prosodic word boundaries. In this case these fine-grained F0 alignment differences do not affect intonational meaning. This chapter has reviewed some of the production and perception studies that have informed the current phonetic models of tonal alignment. This work has highlighted principles of stability and also of adaptation to neighboring prosodic structure as basic pillars of phonetic models of tonal alignment. Importantly, some of these alignment patterns have been shown to be actively used by listeners in word identification tasks and lexical access. Finally, tonal alignment issues have historically been used as arguments to test the predictions of phonological models of intonation and to bear upon current theories of intonational phonology. The last section of this chapter has offered a selection of the arguments put forth in favor of the target-based model of intonation. As a final note, we believe that the full exploitation of recent methodological advances will provide important answers to the role of tonal alignment in phonological and phonetic models of intonation.

Tonal Alignment

17

ACKNOWLEDGMENTS Preparation of this chapter was supported by two grants awarded by the Spanish Ministerio de Ciencia e Innovación: projects FFI2009-07648/FILO and CONSOLIDER-INGENIO 2010 “Bilingüismo y Neurociencia Cognitiva CSD2007–00012.” I am grateful to Rafèu Sichel, two anonymous reviewers, and the editors of this volume, Colin Ewen, Beth Hume, Marc van Oostendorp, and Keren Rice, for all their help and very useful comments on an earlier version of the manuscript.

REFERENCES Arvaniti, Amalia & D. Robert Ladd. 2009. Greek wh-questions and the phonology of intonation. Phonology 26. 43–74. Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 1998. Stability of tonal alignment: The case of Greek prenuclear accents. Journal of Phonetics 26. 3–25. Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 2000. What is a starred tone? Evidence from Greek. In Michael B. Broe & Janet B. Pierrehumbert (eds.) Papers in laboratory phonology V: Acquisition and the lexicon, 119–131. Cambridge: Cambridge University Press. Beckman, Mary E. & Janet B. Pierrehumbert. 1986. Intonational structure in Japanese and English. Phonology Yearbook 3. 255–309. Bruce, Gösta. 1977. Swedish word accents in sentence perspective. Lund: Gleerup. Chen, Aoju, Els den Os & Jan Peter de Ruiter. 2007. Pitch accent type matters for online processing of information status: Evidence from natural and synthetic speech. Linguistic Review 24. 317–344. Crystal, David. 1969. Prosodic systems and intonation in English. Cambridge: Cambridge University Press. Dilley, Laura C. 2007. The role of F0 alignment in distinguishing categories in American English intonation. Unpublished ms., Bowling Green State University. D’Imperio, Mariapaola. 2000. On defining tonal targets from a perception perspective. Ph.D. dissertation, Ohio State University. D’Imperio, Mariapaola & David House. 1997. Perception of questions and statements in Neapolitan Italian. Proceedings of the 5th European Conference on Speech Communication and Technology (Eurospeech 1997), vol. 1, 251–254. Rhodes, Greece. D’Imperio, Mariapaola, Robert Espesser, Hélène Lœvenbruck, Caroline Menezes, Noël Nguyen & Pauline Welby. 2007a. Are tones aligned with articulatory events? Evidence from Italian and French. In Jennifer Cole & José Ignacio Hualde (eds.) Laboratory phonology 9, 577–608. Berlin & New York: Mouton de Gruyter. D’Imperio, Mariapaola, Caterina Petrone & Noël Nguyen. 2007b. Effects of tonal alignment on lexical identification in Italian. In Carlos Gussenhoven & Tomas Riad (eds.) Tones and tunes, vol. 2: Experimental studies in word and sentence prosody, 79–106. Berlin & New York: Mouton de Gruyter. Fournier, R., C. Gussenhoven, O. Jensen & P. Hagoort. 2010. Lateralization of tonal and intonational pitch processing: An MEG study. Brain Research 1328. 79–88. Fujisaki, Hiroya. 1983. Dynamic characteristics of voice fundamental frequency in speech and singing. In Peter F. MacNeilage (ed.) The production of speech, 39–55. New York: Springer. Gili-Fivela, Barbara. 2009. From production to perception and back: An analysis of two pitch accents. In Susanne Fuchs, Hélène Lœvenbruck, Daniel Pape & Pascal Perrier (eds.) Some aspects of speech and the brain, 363–405. Frankfurt am Main: Peter Lang. Gussenhoven, Carlos. 2004. The phonology of tone and intonation. Cambridge: Cambridge University Press.

18

Pilar Prieto

Gussenhoven, Carlos. 2006. Experimental approaches to establishing discreteness of intonational contrasts. In Sudhoff et al. (2006), 321–334. Hart, Johan ’t, René Collier & Antonie Cohen. 1990. A perceptual study of intonation: An experimental-phonetic approach. Cambridge: Cambridge University Press. Kohler, Klaus J. 1987. Categorical pitch perception. In Thomas V. Gamkrelidze (ed.) Proceedings of the 11th International Congress of Phonetic Sciences, vol. 5, 331–333. Tallinn: Academy of Sciences of the Estonian SSR. Ladd, D. Robert. 1996. Intonational phonology. Cambridge: Cambridge University Press. Ladd, D. Robert & Astrid Schepman. 2003. Sagging transitions between high pitch accents in English: Experimental evidence. Journal of Phonetics 31. 81–112. Ladd, D. Robert, D. Faulkner, H. Faulkner & A. Schepman. 1999. Constant “segmental” anchoring of F0 movements under changes in speech rate. Journal of the Acoustical Society of America 106. 1543–1554. Mücke, Doris, Martine Grice, Johannes Becker, Anne Hermes & Stefan Baumann. 2006. Articulatory and acoustic correlates of prenuclear and nuclear accents. In Rüdiger Hoffmann & Hansjörg Mixdorff (eds.) Proceedings of Speech Prosody 2006, 297–300. Dresden: TUDpress. Mücke, Doris, Martine Grice, Johannes Becker & Anne Hermes. 2009. Sources of variation in tonal alignment: Evidence from acoustic and kinematic data. Journal of Phonetics 37. 321–338. Niebuhr, Oliver. 2007. The signalling of German rising-falling intonation categories: The interplay of synchronization, shape, and height. Phonetica 64. 174–193. O’Connor, J. D. & G. F. Arnold. 1973. Intonation of colloquial English. London: Longman. Petrone, Caterina. 2008. From targets to tunes: Nuclear and prenuclear contribution in the identification of intonation contours in Italian. Ph.D. dissertation, Université de Provence. Pierrehumbert, Janet B. 1980. The phonetics and phonology of English intonation. Ph.D. dissertation, MIT. Pierrehumbert, Janet B. & Mary E. Beckman. 1988. Japanese tone structure. Cambridge, MA: MIT Press. Pierrehumbert, Janet B. & Shirley Steele. 1989. Categories of tonal alignment in English. Phonetica 46. 181–196. Prieto, Pilar. 2009. Tonal alignment patterns in Catalan nuclear falls. Lingua 119. 865–880. Prieto, Pilar & Francisco Torreira. 2007. The segmental anchoring hypothesis revisited: Syllable structure and speech rate effects on peak timing in Spanish. Journal of Phonetics 35. 473–500. Prieto, Pilar, Jan van Santen & Julia Hirschberg. 1995. Tonal alignment patterns in Spanish. Journal of Phonetics 23. 429–451. Prieto, Pilar, Mariapaola D’Imperio & Barbara Gili-Fivela. 2005. Pitch accent alignment in Romance: Primary and secondary associations with metrical structure. Language and Speech 48. 359–396. Prieto, Pilar, Eva Estebas-Vilaplana & Maria del Mar Vanrell. Forthcoming. The relevance of prosodic structure in tonal articulation: Edge effects at the prosodic word level in Catalan and Spanish. Journal of Phonetics. Rietveld, Toni & Carlos Gussenhoven. 1995. Aligning pitch targets in speech synthesis: Effects of syllable structure. Journal of Phonetics 23. 375–385. Silverman, Kim E. A. & Janet B. Pierrehumbert. 1990. The timing of prenuclear high accents in English. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 72–106. Cambridge: Cambridge University Press. Sosa, Juan Manuel. 1999. La entonación del español: Su estructura fónica, variabilidad y dialectología. Madrid: Cátedra.

Tonal Alignment

19

Sudhoff, Stefan, Denisa Lenertová, Roland Meyer, Sandra Pappert, Petra Augurzky, Ina Mleinek, Nicole Richter & Johannes Schließer (eds.) 2006. Methods in empirical prosody research. Berlin & New York: Mouton de Gruyter. Watson, Duane G., Christine A. Gunlogson & Michael K. Tanenhaus. 2006. Online methods for the investigation of prosody. In Sudhoff et al. (2006), 259–282. Watson, Duane G., Michael K. Tanenhaus & Christine A. Gunlogson. 2008. Interpreting pitch accents in online comprehension: H* vs. L+H*. Cognitive Science 32. 1232–1244. Welby, Pauline. 2003. The slaying of Lady Mondegreen, being a study of French tonal association and alignment and their role in speech segmentation. Ph.D. dissertation, Ohio State University. Xu, Yi & Emily Wang. 2001. Pitch targets and their realization: Evidence from Mandarin Chinese. Speech Communication 33. 319–337. Xu, Yi & Ching X. Xu. 2005. Phonetic realization of focus in English declarative intonation. Journal of Phonetics 33. 159–197.

51

The Phonological Word Anthi Revithiadou

1

Introduction

In the past few decades, the field of phonology has witnessed the development of an assortment of phonological theories and their offshoots. Seminal among them is the theory of Prosodic Phonology, which explores how prosodic structure is built in relation to morphosyntactic structure. Prosodic Phonology employs mapping rules that aim at organizing chunks of structure (e.g. strings smaller or larger than the grammatical word) into hierarchically ordered layers of prosodic units which, in turn, form the domains within which phonological rules apply. Such phonological domains need not be isomorphic to morphosyntactic constituents. More importantly, the existence of a mapping mechanism entails that rules of phonology proper (i.e. rules inducing changes in the phonological shape and pattern of a string of elements) do not make direct reference to morphosyntactic constituents.1 In general, the basic tenet of Prosodic Phonology is that phonological rules cannot see nor refer to any structure other than the units of the Prosodic Hierarchy (Selkirk 1978b, 1980, 1981a, 1981b, 1984, 1986, 1995; Nespor and Vogel 1982, 1986; Hayes 1989; see also chapter 33: syllable-internal structure, chapter 40: the foot, chapter 48: stress-timed VS. syllable-timed languages, chapter 54: the skeleton, chapter 104: root–affix asymmetries, and chapter 56: sign syllables for some other aspects of the Prosodic Hierarchy):

1

In this sense, therefore, Prosodic Phonology sharply contrasts with direct reference approaches to the interface, most articulately expressed in the work of Kaisse (1983, 1985) and Odden (1987, 1990). The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0051

The Phonological Word

2

(1) Prosodic Hierarchy Utterance (U) Intonational Phrase (IP) Phonological Phrase (PPh) (CG)2 Phonological Word (PW) Foot (F) Syllable (q) The Phonological Word (PW) is one of the best-established constituents of the Prosodic Hierarchy, and has received a great deal of attention. Despite its wide acceptance, certain aspects of the PW are still under debate in the literature. Part of the controversy surrounding its name pertains to the fact that the PW has broadly been established as the constituent that mediates the interface of phonology with morphology (i.e. lexical component) and syntax (i.e. post-lexical component), even though originally it was not intended to encapsulate both aspects of the interface. Moreover, it is the prosodic constituent which, roughly, corresponds to a grammatical word, and hence, naturally, cannot escape the conflicts and ambiguities associated with this notion. This chapter, therefore, aims to address all the major issues pertaining to the PW. More specifically, I will first explore the exact nature of the mapping mechanism(s) involved in the construction of this prosodic constituent. The pivotal questions center on the ways the mapping rules operate to group a specific chunk of morphosyntactic structure into a PW and the way this is formally expressed. It will become apparent from the discussion that, despite the thirty or so years of research in this area, certain aspects of the mapping mechanism are still poorly understood. Second, I will present the methodology and, in particular, the diagnostic criteria employed by researchers in identifying the domain of the PW. This survey will provide the opportunity to investigate and, more importantly, assess the amount of knowledge that has been accumulated over the years regarding the nature of rules that identify this prosodic constituent. Third, I will review the empirical situation and, more specifically, whether cross-linguistic evidence renders the PW universally viable or not. To this end, alternative proposals – which range from the formation of extended versions of the PW to the introduction of smaller or larger reincarnations of it (e.g. the small word, etc.) – will also be reviewed. Such enriched versions of the Prosodic Hierarchy have been proposed to accommodate 2

The CG is the most debatable prosodic constituent, and is usually not included in the Prosodic Hierarchy. It was originally introduced by Hayes (1989) and later adopted and further established by Nespor and Vogel (1986) (see also chapter 84: clitics).

3

Anthi Revithiadou

complex constructions involving function elements (e.g. particles, clitics, etc.), compounds, and complex predicates, which yield somewhat “looser” versions of the notion “word” and hence raise challenging questions for the mode in which the mapping is performed. The remainder of this chapter is organized as follows: §2 sketches out the main advancements that led to the development of a hierarchical model of prosodic constituency, an integral part of which is the PW. §3 examines how prosodic units relate to constituents of morphosyntactic structure and the places where this mapping is performed. §4 sets out the diagnostic criteria for identifying phonological wordhood. §5 focuses on the properties of an extended version of the PW, and §6 concludes this chapter.

2

The birth of the Prosodic Hierarchy: From boundaries to prosodic domains

In the Sound Pattern of English (SPE; Chomsky and Halle 1968) two concepts of syntactic surface structure are acknowledged: (a) output of the syntactic component, and (b) input to the phonological component. Re-adjustment rules are employed to handle discrepancies between these two types of structure. Their main task is to convert the syntactic string into a form that can later be read off and interpreted by phonology. More specifically, syntactic information is encoded in phonology by rules that insert boundary symbols at the edges of syntactic constituents. Such boundaries are considered to be segment-like elements that lack any phonetic manifestation. Two are relevant for the discussion here: (a) the syntactic boundary, #, used to indicate edges of major syntactic categories (N, V, A, etc.), phrasal categories (e.g. NP, VP, etc.) and stress-neutral affixes (Chomsky and Halle 1968: 12, 366), and (b) the morphological boundary, +, which indicates lexically assigned morphological boundaries (Chomsky and Halle 1968: 94). Essentially, the introduction of boundaries initiates an indirect mode of interaction between the components of grammar: phonological rules can either refer to boundaries in their structural description or be blocked by them, but they can never refer directly to syntactic edges. Much of the post-SPE era has been devoted to defining the exact number of boundaries and their relative strength prominence (e.g. McCawley 1968; Selkirk 1972). Gradually, however, the focus of attention shifted from linearly ordered boundary-defined domains to hierarchically organized prosodic ones.3 Based on the seminal work of Liberman (1975) and Liberman and Prince (1977), Selkirk (1978b, 1980, 1981a) proposes that phonological representations, like syntactic ones, are hierarchical in nature. She provides compelling arguments in favor of a “suprasegmental, hierarchically arranged organization” of the utterance (Selkirk 1981a: 111). Selkirk aptly points out the “problem of nestedness”: a phonological rule that applies across a “stronger” boundary can also apply across all “weaker” ones. Analogously, a phonological rule that applies before or after a certain

3

The increasing debate against boundaries was also stimulated by their clearly diacritic character. Boundaries do not constitute linguistic objects and, as such, lack any formal existence in mental representations (see e.g. Pyle 1972; Rotenberg 1978).

The Phonological Word

4

boundary can do so with all stronger ones. For instance, in Sanskrit a rule of voicing affects stops in intervocalic position (Selkirk 1980: 115). Crucially, the rule applies at the domain of the Utterance and affects only consonants at # boundaries (2b), ignoring the ones residing at a + boundary (2a). (2) a. #marut+i# → b. #parivraÍ# #ajam# →

maruti parivra“ ajam

In the boundary theory, the typological predictions in the domain of application of phonological rules described above are a mere stipulation. However, they come for free in a theory that assumes a hierarchically organized set of prosodic categories, i.e. sub-units of prosodic structure such as the syllable, the foot, the phonological word, the phonological phrase, the intonational phrase, and the utterance. This hierarchical constellation, the Prosodic Hierarchy, signals the birth of Prosodic Phonology. The advantage of the new theory is that the principle in (3) (Selkirk 1984; Hayes 1989) can easily capture the “nested” effect in the application of phonological rules mentioned above. (3) Strict Layer Hypothesis (SLH) (Hayes 1989: 204) The categories of the Prosodic Hierarchy may be ranked in a sequence C 1, C2 , . . . Cn , such that a. all segmental material is directly dominated by the category C n , and b. for all categories C i , i ≠ n, C i directly dominates all and only constituents of the category C i+1. Selkirk strengthens the argumentation in favor of prosodic constituency by showing that prosodic domains have independent motivation, besides the interface (Selkirk 1980: 110, 126–129; 1981a: 125; 1984: 8ff.). For instance, the PW in English is a category with an internal organization that yields patterns of relative (i.e. strong vs. weak) stress prominence, e.g. (‘/r/)Fw (’spekt/v)Fs , while at the same time constituting the domain of application of various phonological rules (e.g. nonlow vowels are tensed in PW-final position). After an exploratory period (Nespor and Vogel 1982, 1983; Nespor 1985, 1986; Vogel 1985, 1986), Nespor and Vogel (1986) extended Selkirk’s work and further enriched it with novel data from a wide array of languages (e.g. Hungarian, Greek, Turkish, and Italian). The focus of research in the following years was on defining the basic premises of Prosodic Phonology (e.g. the nature of the mapping algorithm, the way it is performed, etc.) and lending further support to the theory with empirical evidence from cross-linguistic research (e.g. Booij 1983, 1985a, 1985b; Booij and Rubach 1984; Hayes 1989; Itô and Mester 2003). In the Prosodic Phonology literature, however, the main motivation for prosodic constituency is non-isomorphism, most commonly expressed as a mismatch between phonological and morphosyntactic boundaries (e.g. Selkirk 1981a; Nespor and Vogel 1982). Nespor and Vogel (1986) argue that in Hungarian, for instance, vowel harmony takes place within a stem and a string of suffixes – (4a) and (4b) – but fails to apply in strings that contain sequences of stems (4c) or a prefix plus a stem (4d) (see chapter 123: hungarian vowel harmony). The reason for the disparity in the application of the process relies on the nature of

Anthi Revithiadou

5

the phonological domain formed by the elements involved. Vowel harmony applies within the domain of the PW, and a stem + suffix string is mapped into one PW. Significantly, compounds and prefixed constructions form two separate PWs, as shown in (4c) and (4d), despite the fact that they constitute single grammatical words. (4) Hungarian vowel harmony (Nespor and Vogel 1986) ölelés hajó b. ölelés-nek hajó-nak c. Buda-Pest könyv-tar book-collection d. be-utazni in-commute oda-menni there-go a.

[ölelés]PW [hajó]PW [ölelésnek]PW [hajónak]PW [Buda]PW [Pest]PW [könyv]PW[tar]PW

‘embracement’ ‘ship’ ‘embracement-dat sg’ ‘ship-dat sg’ ‘Budapest’ ‘library’

[be]PW [utazni]PW

‘to commute in’

[oda]PW [menni]PW ‘to go there’

The reference to prosodic constituents such as the PW allows us to describe the rule of vowel harmony in a unified way. The issue of (non-)isomorphism is central in defining and exploring the nature of phonological wordhood and is addressed in detail in subsequent sections of this chapter. To summarize, each one of the prosodic categories in the Prosodic Hierarchy is governed by its own principles of internal constituency, and each one forms a domain for the application of phonological rules. Moreover, certain prosodic domains are introduced in order to capture the non-isomorphic character of the interface. Boundaries, on the other hand, are entities that “assist” the interface but have a strong diacritic flavor. The ensuing sections explore the mode in which a particular prosodic constituent, namely the PW, relates to constituents of the morphosyntactic structure and the place this mapping occurs.

3

The Phonological Word at the interface

The classical Prosodic Phonology theory (Selkirk 1981a, 1981b, 1984, 1986, 1995; Nespor and Vogel 1982, 1986) assumes the organization of grammar in (5). Thus, the rules that undertake the mapping of morphosyntactic structure into a PW are part of the post-lexical phonology. Selkirk (1984: 82) dubs this a syntax-first model. (5)

syntactic surface structure → mapping rules → prosodic representations → phonological rules → phonetic representations

This view, however, has been challenged by several researchers (e.g. Booij 1983, 1988; van der Hulst 1984; Inkelas 1989; Booij and Lieber 1993). Nowadays it is widely accepted that constituents up to the level of the PW are built lexically, and that phonological and morphological structure is constructed at the same time, from bottom up. This issue is addressed in detail in §3.3.

The Phonological Word

6

3.1 Mapping rules under the Strict Layer Hypothesis Nespor and Vogel (1986) recognize that there are several options available for the definition of PW. They attribute the attested variation to an assortment of morphological notions that mapping rules can be sensitive to (e.g. stems, roots, sequences of stems, and suffixes, etc.). There are languages, like Greek for instance, in which a lexical word, i.e. the terminal element of the syntactic tree, whether a compound or simple word, is mapped into a PW. On the other hand, there are languages like Hungarian, in which mapping rules are sensitive to smaller elements such as stems, prefixes, or sequences of stems and suffixes. Finally, there are languages such as Dutch, in which certain elements (e.g. suffixes or prefixes) are endowed with a diacritic that grants them independent PW status regardless of the general dictates of the mapping mechanism. For instance, in Italian vowel-final prefixes are assigned independent PW status due to language-specific syllable well-formedness conditions, a restriction that the mapping rule must read and comply with. In Nespor and Vogel’s (1986) model of mapping there is yet another source of phonological wordhood, namely the SLH in (3). The principle of proper nesting requires all “unparsed” elements to prosodify into a PW in the absence of a neighboring host. The definition of PW in (6) encapsulates the aforementioned mapping possibilities. (6) PW domain (Nespor and Vogel 1986: 141) A. The domain of PW is Q (= any terminal element of the syntactic tree). or B. I. The domain of PW consists of a. a stem; b. any element identified by specific phonological and/or morphological criteria; c. any element marked with the diacritic [+W]. II. Any unattached elements within Q form part of the adjacent PW closest to the stem; if no such PW exists, they form a PW on their own. The above set of statements on the application of mapping rules makes some predictions. First, PWs cannot be larger than a terminal element in the syntactic tree. Second, no single stem can be mapped into more than one PW, and, third, affixes will always be part of the PW of their base unless they bear a diacritic. Interestingly, the definition in (6) encompasses diacritic information as well as special restrictions imposed by the phonological and/or morphological components of the grammar. Such a mapping algorithm, therefore, is far too powerful to be insightfully implemented in an interface theory which aspires at advancing the understanding of the factors involved in PW formation. Selkirk (1986), based on Chen (1987), proposes an end-based mapping theory, which operates on the edges of X-bar constituents. The basic idea is that a prosodic constituent is demarcated by the right or the left edge of selected syntactic constituents (i.e. X 0, X′, X″) (Selkirk and Shen 1990: 319): (7) The syntax–phonology mapping For each category C n of the prosodic structure of a language there is a two-part parameter of the form C n: {Right/Left; X m}, where X m is a category type in the X-bar hierarchy.

Anthi Revithiadou

7

It is a parametric choice of a language as to which edge of the X 0 (= word), for instance, will serve as the beginning or the end point of the PW domain. An advantage of the model is that the mapping algorithm is cross-categorical: the end rules can apply at some level of the X-bar hierarchy in order to form the appropriate prosodic domains. For instance, they apply at the X 0 to derive PWs and at the X′ or X″ to derive PPhs. A comparison between the two algorithms reveals a significant difference in their descriptive and predictive power. The prosodic patterns established between a PW and a morphological word (MW)4 by Nespor and Vogel’s (1986) algorithm are given in (8). Recall that this algorithm grants independent PW status to subminimal elements such as function words (fnc), prefixes (prf), etc., either as a result of a diacritic or due to the SLH. Crucially, a PW can never be larger than an MW. (8)

a. b. c.

PW = MW or x PW < MW *PW > MW

where x = fnc/prf

Selkirk’s (1986) end-based mapping algorithm, on the other hand, assigns a PW status to MWs but, importantly, permits a PW larger than an MW as a prosodic output. This domain can be derived when function elements are trapped in between two MWs, as shown in (9). Selkirk (1986, 1995) explicitly states that only lexical categories – not functional ones – and their projections are visible to the mapping rules. This entails, therefore, that, depending on the end rule parameter setting, the function words will prosodify together with the preceding or the following MW (= X 0). In the following abstract example, the PW will extend from the left (9b) or the right (9c) edge of one lexical item to the left or the right edge, respectively, of the next, incorporating in the process any intervening function words. (9)

a. b. c.

X 0 fnc fnc X 0 [X 0 fnc fnc]PW [X 0]PW [X 0]PW [fnc fnc X 0]PW

end rule: left end rule: right

Importantly, the original formulation of the end-based mapping rule does not allow elements that are smaller than the MW to constitute independent PWs. As a consequence, clitics, prefixes, and suffixes are deprived of this prosodic possibility. Given the SLH, only one option is available: incorporation of the sub-minimal element into the PW of a neighboring MW. The parsing options for PWs derived by the end-based algorithm are summarized in (10): (10)

4

a. b. c.

PW = MW *PW < MW PW > MW

In linguistics, the notion “morphological or grammatical word” is not uncontroversial. Dixon and Aikhenvald (2002: 18ff.) provide different types of criteria to define it and also distinguish it from PW. The most important three are that the elements a grammatical word consists of must always occur together, in a fixed order, and have a conventionalized coherence and meaning.

The Phonological Word

8

Interestingly, the two models diverge in the cases of non-isomorphism: Nespor and Vogel’s algorithm allows a PW to be smaller but not larger than an MW, whereas the end-based mapping sanctions the opposite. Furthermore, each theory makes different predictions with respect to the prosodization of sub-minimal elements. Due to the SLH, the end-based algorithm will force such elements to incorporate into an adjacent PW, whereas Nespor and Vogel’s algorithm will allow them to form an independent PW. The end-based algorithm is silent regarding the prosodization of non-terminal syntactic elements such as stems, prefixes, and suffixes. The empirical facts provide only partial support for each model. Let us start with the pattern PW < MW, predicted only by Nespor and Vogel’s algorithm. In northern Italian, prefixes and the stems of compounds, but not suffixes, form separate PW-domains for the rule of intervocalic s-voicing (Nespor and Vogel 1986: 125). As evidenced by the examples in (11), this rule applies in monomorphemic and inflected words, (11a) and (11b), which constitute one PW. It is blocked, however, between a prefix and stem (11c) or between the stems of a compound (11d), suggesting that such constructions are parsed into two PWs. (11)

Northern Italian: Intervocalic s-voicing a. b. c. d.

a/s/ola ca/s/-e a-/s/ociale tocca-/s/ana

a[z]ola ca[z]e a[s]ociale tocca[s]ana

[azola]PW [caze]PW [a]PW [sociale]PW [tocca]PW [sana]PW

‘button hole’ ‘house-pl’ ‘asocial’ ‘cure all’

During the exploratory period of Prosodic Phonology, a growing body of crosslinguistic evidence revealed that affixes may form independent PWs (e.g. Booij and Rubach 1984 for Polish and English; Hannahs 1991, 1995a, 1995b for French; amongst many others). Cross-linguistic evidence, therefore, gives an empirical advantage to the Nespor and Vogel mapping algorithm compared to the end-based one. However, several researchers have shown that the latter can easily handle these facts if end rules are appropriately modified so that they can read off edges of elements smaller than the word, such as stems (Kang 1992), or even edges of functional categories, e.g. agreement (Rice 1993). Furthermore, the end-based algorithm seems to have the advantage on the empirical side in cases where the PW is larger than the MW. Booij (1988) draws attention to clitic constructions from Latin and Dutch where clitics – even though they are independent grammatical words – are phonologically dependent on an adjacent word. This mapping is predicted by the end-based algorithm, but not by the Nespor and Vogel one. It is important to emphasize here that Nespor and Vogel (1986) posit a different prosodic constituent for the prosodic organization of such clitic constructions, namely the Clitic Group (CG). Within this larger constituent, however, the Nespor and Vogel mapping algorithm, under the dictates of SLH, is forced to elevate each independent element, i.e. clitic, into a PW. Below I exemplify from Greek the results of each model of mapping. In Greek, weak object pronouns precede the non-imperative verb form (12a) but follow the imperative (12b) (Revithiadou and Spyropoulos 2008, and references cited therein).

Anthi Revithiadou

9 (12)

Greek object clitics a.

o ’petr-os to ’Ïjava-se the Peter-nom.sg clt.3n.sg.acc read-pst.3sg ‘Peter read it.’

b. ‘Ïjava-’se to read-2sg.imp clt.3n.sg.acc ‘Read it!’

(< ’Ïjava-se)

Under a Nespor and Vogel-type mapping, each clitic will be granted PW status, as shown in (13). The problem with this structure, however, is that the PW of the clitic exhibits a different behavior from that of the PW of the host element. A clitic may “trigger” stress on the final syllable of the preceding PW (14a) or carry stress itself (14b). More importantly, examples like (14a) pose a serious threat to the SLH. The addition of the clitic causes the form to be further footed so that the threesyllable restriction imposed by the language can be salvaged: (‘Ïjava)(’se to) mu. The problem now is that the newly formed foot cuts across two PW boundaries: [(‘Ïjava)(’se]PW [to)]PW [mu]PW, in total disrespect of proper containment.5 (13)

(14)

PW

PW

PW

‘Ïjava’se

to

mu

a. ‘Ïjava-’se to mu read-2sg.imp clt.3n.sg.acc clt.1sg.gen ‘Read it to me!’ b. ‘par-e ’mu to take-2sg.imp clt.1sg.gen clt.3n.sg.acc ‘Take it for me!’

Furthermore, there is compelling evidence that postverbal clitics are incorporated into the PW of the verb (see footnote 5 and Revithiadou and Spyropoulos 2008 for more evidence), as correctly predicted by the end-based algorithm PW: {Left; X 0}. Thus, the comparison so far gives a descriptive advantage to the end-based model. A more careful examination of the facts, however, reveals that this algorithm is not unproblematic either. Given the left-end orientation of the rule, pre-verbal clitics are expected to encliticize to the preceding PW, but they do not, as shown in (15). If the clitics incorporated into the PW of the NP /o ’.oÏoros/ ‘the Theodorenom.sg’, the window restriction would have forced the development of a new stress on the last syllable of the noun, yielding [o .oÏo’ros tu to]PW ‘the Theodore-nom.sg clt.3sg.gen clt.3n.sg.acc.’ Clearly, this expectation is not confirmed by the data.

5

Nespor and Vogel (1986: 154–155) escape this problem by employing a (grid-based) Stress Readjustment rule. There is independent evidence, however, that clitics in Greek are footed. For instance, imperatives opt to incorporate the enclitic by deleting their final vowel, e.g. /:rapse ton/ (’:rapston)F ‘write-imp.2sg clt.3m.sg.acc’. Alternatively, the clitic may be augmented so that it can form its own foot, e.g. (‘:rapse)F (’tone)F. Similar augmentation phenomena are independently enforced by foot well-formedness conditions, e.g. /ro’ta-n/ (with inherent accent) ‘ask-3pl’ ro(’tane) ~ ro(’tan).

The Phonological Word (15)

10

o ’.oÏor-os tu to Ïja’vaz-i the Theodore-nom.sg clt.3sg.gen clt.3n.sg.acc read-3sg ‘Theodore reads it to him.’

Similar cases have been found in a variety of languages, underlining, among other things, the asymmetric behavior often witnessed in the prosodic organization of constructions with function elements (e.g. particles, clitics, etc.), compounds, and other complex expressions (e.g. Inkelas 1989; Booij 1996; Leben and Ahoua 1997; Peperkamp 1997; Vigário 1999, 2003).6 These findings therefore challenged the descriptive and explanatory efficiency of both mapping algorithms, and opened up new directions for research in this area.

3.2 The Weak Layer Hypothesis: Mapping as constraint interaction A number of researchers have sought a solution to the problems mentioned above in the architecture of the Prosodic Hierarchy. In particular, they questioned the necessity of the SLH as a well-formedness principle of the arboreal prosodic structure. The proposed modifications involve the introduction of (a) recursive structures, in which a prosodic constituent of a certain type dominates another constituent of the same type (16a), and (b) non-exhaustively parsed structures, in which a constituent of one type is permitted to skip intermediate levels and dominate a constituent of more than one level lower in the Prosodic Hierarchy (16b). The relaxed version of the SLH is referred to as the Weak Layer Hypothesis (e.g. Booij 1988, 1995, 1996, 1999; Itô and Mester 2003). (16)

Weak layering: Recursion and non-exhaustivity a.

b.

Cn

Cn Cn−1

Cn Cn−1

Cn−1

Cn−1

Cn−2

Cn−2

Cn−2

Under the influence of Optimality Theory (Prince and Smolensky 1993), Selkirk (1995) goes one step further and proposes that the SLH should be decomposed into its primitive components, which take the form of the prosodic domination constraints in (17). (17) Constraints on prosodic domination (Selkirk 1995: 443) (where C n = some prosodic category) a. Layeredness: No C i dominates a C j, j > i. b. Headedness: Any C i must dominate a C i−1. c. Exhaustivity: No C i immediately dominates C j, j < i–1. d. Non-Recursivity: No C i dominates C j, j = i. 6

The asymmetry in the prosodization of weak elements constitutes one of the main arguments against the CG as a prosodic constituent.

Anthi Revithiadou

11

Layeredness and Headedness are argued to be universally inviolable and therefore undominated in the constraint ranking of all languages. The structures in (16) result from the relative ranking of Exhaustivity and Non-Recursivity to the other constraints of the system. A significant role in the construction of prosodic constituents has been played by the alignment family of constraints, which basically undertakes the mapping of morphosyntactic constituents to prosodic structure (McCarthy and Prince 1993a, 1993b): (18)

Alignment constraints a. MW-Constraint (WCon): Align(MW, L/R; PW, L/R) b. PW-Constraint (PCon): Align(PW, L/R; MW, L/R)

These interface constraints are translations of Selkirk’s (1986) parameterized endbased theory of mapping into Optimality Theory (OT) constraint-based terminology. They have a uniform general scheme, which can easily account for cross-categorical mappings, since units from different levels of the Prosodic Hierarchy can coincide with morphosyntactic units. For instance, the constraint in (18a) requires the left or right edge of every MW to coincide with the left or right edge of some PW. Differences in the ranking of the relevant constraints and/or the morphosyntactic structure of an input string are taken to be responsible for the attested variation in the prosodization of weak elements (e.g. function words, prefixes, etc.) cross-linguistically. The interaction of the above constraints yields the following prosodic patterns for the abstract string /x x V/ (where x is a weak element): (19) a. b. c. d.

structures

typology

rankings

[[V x]PW [V]PW]PPh [x x [V]PW]PPh [x x [V]PW]PW [x x V]PW

PW free recursive internal

Exh, NonRec, MCon >> PCon MCon, PWCon >> NonRec >> Exh Exh, MCon >> NonR, PCon Exh, NonRec >> MCon, PWCon

It remains an open question whether all predicted typologies receive empirical support or whether they lead to vast overgeneration of PW patterns. There is, however, ample cross-linguistic evidence in support of the typology of function words in (19). In a cross-dialectal study of Italian clitics, for example, Peperkamp (1997) shows that the patterns in (19b) and (19d) correspond to specific Italian dialects. Similarly, Revithiadou (2008) provides evidence from a cross-dialectal survey of object clitics in Greek that all four prosodic patterns are empirically attested. Moreover, she demonstrates on the basis of diachronic evidence that there is a transition from “looser” types of constructions – e.g. patterns (19a) and (19b) – to nested ones – pattern (19c) – and from there to total integration of the clitic to its host, pattern (19d) (see also chapter 84: clitics).

3.3 Split between two worlds After discussing various approaches as to how a mapping rule assigns a PW make-up to a portion of the morphosyntactic string, I now turn to addressing where and when the mapping takes place. Such questions are of course meaningful only

The Phonological Word

12

within a theory that assumes a division of labor between the components of grammar and a procedural view of the interface.7 Selkirk takes a clear stance on this issue and argues that both word structure and sentence structure are sensitive to the same mechanism, and must therefore be treated alike, i.e. after syntax at the post-lexical component (Selkirk 1984: 415). Any additional tools would simply lead to an unnecessary proliferation of the machinery phonology has at its disposal. Nespor and Vogel (1986), on the other hand, take a more moderate approach. They draw a distinction between two sets of rules: (a) those that refer to phonological domains only and (b) those that make direct reference to morphological information. By doing so, they implicitly adopt a compartmentalized view of phonology. First, there is a section of phonology that may access directly “morphological structure and/or specific morphological elements” (Nespor and Vogel 1986: 18). Its rules are handled by a different mechanism, possibly Lexical Phonology (Kiparsky 1982a, 1982b; Mohanan 1982; chapter 94: lexical phonology and the lexical syndrome). Second, there is another section that accesses the interface indirectly, via the constituents of the Prosodic Hierarchy, and operates strictly on Prosodic Phonology rules proper. The question as to how these blocks of rules are ordered is left open to further research (see also chapter 103: phonological sensitivity to morphological structure). Inkelas (1989) attempts to salvage the indirect nature of the interface alluded to by Nespor and Vogel (1986) with the introduction of a radical move: she eliminates the q and the F from the Prosodic Hierarchy and introduces prosodic constituency below the PW in the Lexicon. The ranks below the PW in (1), which roughly correspond to levels in Lexical Phonology, can now accommodate strings of structure into prosodic units that are smaller than the PW. They basically provide the appropriate domain within which phonological rules can apply without having to directly refer to morphological constituents. Booij and Rubach (1984) and Booij (1988) concur that a lexically constructed PW is necessitated on empirical grounds, e.g. the prefix a- in Italian; see example (11). They further argue that it can offer solutions to bracketing paradoxes, special rules of allomorphy, and so on. For instance, the word [[un[grammatical]A]Aity]N poses a problem, because the prefix un- is stress-neutral (i.e. level 2 in Lexical Phonology terminology), and hence should be added after the stress-shifting suffix -ity (i.e. level 1). This implies, however, that the prefix un- should attach to grammaticality and not to grammatical, which is also problematic, because un- attaches only to adjectival bases. At the phonological level, the suffix -ity shifts the stress of the base ungrammatical, which nevertheless contains a stress-neutral prefix. This is problematic too, because only stress-neutral suffixes can be added after that prefix (i.e. a level 1 affix may not follow a level 2 affix). This bracketing paradox receives a straightforward explanation once a prefix is assigned an independent PW status: [un]PW [grammaticality]PW. Each PW constitutes an independent domain of stress assignment. As a result, the shifting property of the suffix can never interfere with the prefix, simply because the latter belongs to a different PW. Elements that impose such prosodic restrictions are often called non-cohering.

7

In non-serial theories of phonology such as OT, where mapping is the job of interface constraints that operate simultaneously on both modules, such questions are almost redundant, unless one embraces the distinction between a lexical and a post-lexical component in phonology.

13

Anthi Revithiadou

If this property is not derivable from the size and the segmental shape of the morpheme in question, it must be specified in the subcategorization frame of the relevant lexical item (Booij and Lieber 1993). Clitics may have similar prosodic selection requirements. For instance, ie ‘he’ in Dutch subcategorizes for a right PW boundary: ]PW __ (Booij and Lieber 1993: 39). However, there is a major drawback in assigning PW status to morphological elements in the lexicon: it reintroduces via the back door the problem of the PW boundary as a diacritic, thus posing a serious threat to the very nature of the Prosodic Hierarchy. Another major consequence of moving the construction of PWs into the lexicon is that the mapping now applies twice: once in the lexicon and again after syntax, i.e. post-lexically. This suggests that there are two types of PWs, lexical and post-lexical, and, consequently, the interface between phonology and morphology is of a different nature than that between phonology and syntax. If we admit this, we need to jettison the idea of a unified phonology that Prosodic Phonology – especially through the work of Selkirk – originally pursued. Yet another problem with implementing prosodic rules in the lexicon is that it blurs the division between Prosodic Phonology and Lexical Phonology (see chapter 94: lexical phonology and the lexical syndrome). The former theory is designed to deal with the interface of phonology and morphosyntax postlexically; the latter relies on a procedural mode of interaction of phonology and morphology in order to account for “idiosyncratic,” morphology-dependent phonological behavior. If the two models meet in the lexicon, then one may wonder which type of phenomenon each model targets and, moreover, on which grounds the division of labor is decided. Do their rules apply simultaneously or in an ordered fashion? Even though the model of Lexical Phonology has gradually lost headway, the distinction between lexical and post-lexical PW has endured over time (Booij and Lieber 1993; see also Booij 1999 for Dutch; Nespor 1990, Vogel 1991, Peperkamp 1997 for Italian; amongst others). The lasting nature of this division emphasizes all the more the truly interface nature of PW and further establishes it as the prosodic constituent that intersects the interface of phonology with morphology and syntax. However, more work needs to be done in order to acquire a better understanding of the factors that dictate the formation of PWs, and the workspace where this is done.

4

Diagnostics for the Phonological Word as a prosodic constituent

In the previous sections, we examined ways and places within which PWs are built. Here, the focus will be on what type of phonological evidence has been put forward to substantiate the PW as an integral part of the Prosodic Hierarchy. Nespor and Vogel (1986: 58ff.) identify certain diagnostics for establishing the concept of constituent in phonology. Phonological rules must refer to a string of elements in their formulation and have this string as their domain of application. The same piece of structure may also serve as the domain of phonotactic restrictions and stress prominence relations. Vogel (2009) aptly remarks that in establishing a prosodic constituent a number of phenomena must cluster together in using a particular string of elements as their domain. In the almost thirty years of research

The Phonological Word

1217

on Prosodic Phonology, various diagnostic criteria have been put forward to identify the domain of the PW. In the ensuing paragraphs, we will closely examine the most important ones, at the same time drawing attention to the existence of conflicting evidence and its repercussions for the status of PW as part of Universal Grammar.

4.1 Segmental rules In previous sections, we have seen that it has been proposed that both vowel harmony in Hungarian and intervocalic s-voicing in Italian apply within the domain of the PW. The Prosodic Phonology literature is replete with examples of segmental rules used as diagnostics for the definition of the PW domain in a variety of languages (see e.g. Hannahs 1991, 1995a, 1995b; Kang 1992; Peperkamp 1997; Kleinhenz 1998; Hall and Kleinhenz 1999; Vigário 1999, 2003). To illustrate with an example from French, Hannahs (1995a, 1995b) argues that glide formation and vowel nasalization can be safely used as diagnostics of the PW domain. For instance, underlying high vowels such as /i y/ semi-vocalize intervocalically to [j] and [8], respectively. Crucially, this rule applies only in stem + suffix strings (20a) but never when the vowel in question occurs at the end of a prefix (20b) or the first element of a compound (20c) (Hannahs 1995b: 1131). The blocking of the rule in the latter environments is taken as evidence that the stem plus suffix string constitutes a different PW from the remainder of the word. This is a typical example of non-isomorphism between prosodic and morphological structure. (20)

b.

colonie colonial anti-alcoolique

c.

tissue-éponge

a.

[kDlDni] [kDlDnjal] [ftialkDlik] *[ftjalkDlik] [tisyep6Ú] *[tis8ep6Ú]

‘colony’ ‘colonial’ ‘anti-alcohol’ ‘terry-cloth’

Similarly, Vigário (2003) employs a great variety of both lexical and post-lexical rules to substantiate the existence of the PW in European Portuguese. One such rule is vowel reduction, which affects all stressless vowels of a word except for the word-initial one, e.g. promo’ver → promu’ver ‘(to) promote’. A more careful examination of the data nevertheless reveals that the vowel is protected even when it is not in absolute word-initial position (21b). For Vigário (1999: 272–273), this constitutes evidence for the presence of a PW boundary at the left of the base ocu’par. The exact prosodic structure of such prefixed formations will be discussed in §5. (21)

a. ocu’par b. desocu’par

‘(to) occupy’ ‘(to) not occupy’

4.2 Stress and tone Stress is prototypically considered to be an infallible diagnostic for the PW domain: a PW must bear only one primary word stress (see chapter 41: the representation

Anthi Revithiadou

1218

of word stress). In Greek, for example, the phonological wordhood of lexical words and stem–word compounds is signaled by the presence of one primary stress: (22)

a.

/an.rop-os/ man-nom.sg b. /pali-o-ma:az-o/ bad-linkV-shop-nom.sg

[’an.ropos]

‘man’

[paljo’ma:azo]

‘lousy shop’

Dixon (2002) reports that in Jarawara, a Madi dialect of the Arawá family of southern Amazonian, primary stress is on the penultimate syllable, and rhythmic stress occurs on every other syllable to the left of the main-stressed foot (23a). Curiously, compounds and reduplicative formations display two stress peaks, which, crucially, are not on the second and the fourth mora from the end of the complex word but rather on the penultimate syllable of each of their constituents; (23b) and (23c). The attested stress patterns, therefore, suggest that the relevant formations form two PWs, e.g. [’bani]PW [ka’sako]PW (Dixon 2002: 128). (23)

a. b. c.

to-’wa-ka-’tima-’maro away-applic-in.motion-upstream-FPef ’bani-ka’sako / *ba’nika’sako ’kete-ke’tebe / *ke’teke’tebe < ke’tebe ‘run, follow’

‘took upstream’ ‘wild dog species’ ‘run a lot’

Tonal information may also serve as a criterion for delimiting the PW boundaries (see chapter 42: pitch accent systems). Leben and Ahoua (1997) provide an instructive example from Baule, a Bia language of the central Tano group. The language has both a High and a Low tone. Interestingly, a sequence of High tones shows an upstepping pattern, which involves a gradual rise in pitch from a level phonetically close to Low to a Super-High level. The domain of the upsweep, as this rule is commonly referred to, is the PW. The examples in (24) demonstrate that the rule operates in monomorphemic words (24a) and noun–noun compounds (24b), but it is blocked in possessor–possessed (24c) and subject–predicate phrases (24d) (Leben and Ahoua 1997: 117–118). The difference between these two sets of formations is attributed to a difference in their respective prosodic constituencies. The former constitute a single PW, e.g. [bólí nFnnFn]PW (24b), whereas the latter are organized into two separate PWs, e.g. [bólí]PW [mángún]PW (24c). The proposed prosodic structures are enhanced with additional evidence from segmental processes (Leben and Ahoua 1997: 122ff.). (24)

a.

Ákísí [– – – ] b. b ó l í n F n n F n – ] [ – – – c. b ó l í m á n g ú n – ] [ – – ][ – d. Á y á b ó l í [– – ][ – –]

‘Akisi’ ‘goat milk’ ‘goat’s friend’ ‘Aya is a goat’

The Phonological Word

1219

4.3 Phonotactics and syllabification The PW has been considered an important domain for phonotactic constraints, together with the syllable and the foot. In Dutch, for instance, the schwa as a featureless vowel cannot be preceded and followed by the same consonant; /a(lHl/, for example, is an impossible morpheme. Across PW boundaries, however, as in compounds, such sequences are tolerated, e.g. formu/ lH-l/ijst ‘formula list’ suggesting that the relevant Obligatory Contour Principle restriction is confined by the PW (Booij 1999: 56–57). The boundaries of PWs can also be subject to certain restrictions. In German, a lax vowel constraint (i.e. *[–tense, –low, –long]) limits the distributional freedom of vowels that occur at the right edge of the PW (Hall 1999). In European Portuguese (Vigário 2003), a PW cannot begin with a flap [7]. Furthermore, Sanskrit shrinks its consonant clusters at the right edge of the PW (Selkirk 1980), sharply contrasting in this respect with Dutch, which tolerates the occurrence of extra consonants and heavier rhymes in the same position (Booij 1995, 1999). In several, but not all, languages syllabification is taken to be a reliable diagnostic for establishing the PW domain (e.g. Kang 1992 for Korean; Booij 1995, 1999 for Dutch; Hall 1999 for German; Peperkamp 1997 for Italian; Raffelsiefen 1999 for English). Kang (1992) argues that in Korean, aspirated consonants, which normally neutralize to stops in coda position, syllabify together with derivational suffixes (25a). This option, however, is not available when the relevant segment is in between the two stems of a compound (25b), suggesting that resyllabification is blocked by an intervening PW-boundary. (25)

a. [[kiph]V i]N [ki.ph i] *[kip.i] deep-nml b. [[[aph]N [aph]N i]Adv [ap.a.ph i] *[a.pha.ph i] front front advs

[kiph i]PW ‘depth’ [ap]PW [aph i]PW ‘to each person’

4.4 Other diagnostics Given that PWs dominate feet and feet are usually branching constituents, PWs should be minimally either bimoraic or disyllabic, depending on whether they are composed of moraic or syllabic feet. Thus, foot binarity together with Headedness (17b) derive the notion minimal word (McCarthy and Prince 1986, 1993a, 1993b, 1998). Prosodic minimality is also enforced in morphological processes such as nickname truncations and various types of clippings (see McCarthy and Prince 1998 and references cited therein). In Greek, for instance, name truncations are minimal words (Topintzi 2003): (26)

a. Ni’kolaos b. Poli’kseni c. Aspa’sia

’nikos ’poli, ’kseni ’aspa

An often cited example in the literature for the identification of the PW is reduction under coordination (e.g. Booij 1985b for Dutch; Kleinhenz 1998 for German). For instance, Booij (1985b) demonstrates, on the basis of the examples in (27) that parts of complex words – i.e. compounds (27a) and complex formations with the so-called non-cohering affixes (27b) – can be cut off in coordinate constructions:

Anthi Revithiadou

1220

natuurkunde en scheikunde ‘nature knowledge and analysis knowledge’ b. zichtbaar en tastbaar ‘visible and tangible’

(27) a.

The condition is that this rule applies to trim off only parts of grammatical words which constitute separate PWs. That is, it can never eliminate an affix that is included in the PW of the stem it combines with: *rodig of groenig ‘reddish or greenish’. All that the rule “sees” is phonological structure, i.e. PW boundaries, and not the internal morphonological structure of words. Language games constitute a resourceful supplier of evidence for the PW. Henderson (2002) reports on a play language called “Rabbit Talk” in Arrernte, a central Australian language. In Rabbit Talk, the first syllable of the word is removed from its original position, and transposed to the end of the PW (28a). Monosyllabic words skip the transposition rule, and the syllable /ej/ is prefixed instead (28b). For the purposes of this discussion, I follow Henderson in assuming that the syllable structure is VC(C). Rabbit Talk suggests that disyllabic case clitics constitute a separate PW. In (28c), the two elements of the construction count as separate domains: the first element is treated as a monosyllabic word and hence receives the /ej/ prefix. The clitic /-akerte/, on the other hand, constitutes a PW by itself and, as such, it is subjected to the transposition rule. (28) a.

ampangk+eme moan+pres b. artwe man c. irlpe-akerte ear-comit

ordinary speech

Rabbit Talk

/amp.angk.em/

/angk.em.amp/

/artw/

/ej.artw/ /ej.irlp.ert.ak/

4.5 Conflicting evidence and methodological issues in defining the Phonological Word domain Within the Prosodic Phonology framework, the methodology applied in establishing prosodic constituents in general and the PW in particular involves four basic steps: first, discovering the domain within which a given phonological process applies or is blocked; second, establishing that more rules have the same domain as their locus of application; third, matching the string of elements with a particular unit of the Prosodic Hierarchy; and fourth, giving the particulars of the mapping mechanism that determines which chunk of morphosyntactic structure is organized into that particular unit. A good theory should have a mapping mechanism with a certain degree of descriptive power. In practice, however, the weight of investigation falls primarily on substantiating a specific prosodic constituent on the basis of the phonological rules alone, with much less attention paid to the specifics of the morphosyntax. This proves to be quite problematic when the phonological component sends conflicting signals. Cetnarowska (2000) reports that the distribution of secondary (rhythmic) stress in proclitic + host strings in Polish presumes the existence of a foot that straddles

The Phonological Word

1221

a PW boundary which is established on the basis of segmental evidence (Rubach and Booij 1990). In this case, the stress facts of the language conflict head on with the evidence provided by segmental rules. Likewise, Raffelsiefen (1999) argues that assimilation rules in English do not constitute a reliable criterion for establishing PW structure in English. This lack of agreement among different types of diagnostic criteria has led some researchers to propose units smaller or even larger than the PW in order to accommodate the “problematic” or “misbehaving” data. For instance, Rice (1993) acknowledges the existence of a small word in Slave, which is subject to certain rhymal constraints, as opposed to the PW, which is the domain of foot-based processes and various segmental rules. What is clear from the discussion so far is that the proponents of Prosodic Phonology have been confronted with challenging issues, and have opted for different solutions to deal with them. The first solution is to acknowledge that different types of criteria can serve as diagnostics of phonological wordhood in different languages and that each language decides on how to prioritize these criteria or even discard some of them on the basis of some sort of “scale” of relative importance. The question that naturally arises in this case is the way and the context in which this decision is made. The second solution resorts to the proliferation of prosodic domains by inserting pieces of structure with analogous prosodic behavior to distinct slots below the PW. The problem in this case is whether there is an upper bound to the proliferation of domains and, more importantly, whether these domains are universal. There is a third approach to (partially) tackling the problem described above. Recall from §3.2 that the Weak Layering Hypothesis offers the option of constructing a recursive PW. Such an extended version of the PW has been employed to capture non-isomorphic aspects of the interface from where some (but not all) of the challenging data stem, as we will see in the following section.

5

Extending the Phonological Word

The relaxation of the SLH allowed the emergence of recursive structures. It is doubtful whether constituents lower than the PW can be recursive (Itô and Mester 2009; Kabak and Revithiadou 2009). However, recursion is commonly assumed for higher levels of the Prosodic Hierarchy, such as the PW and the PPh. Selkirk (1995), in a study on the typology of clitics, motivates PW recursion as resulting from morphosyntactic recursion. In (29), phonology “mimics” the nested structure of the morphosyntactic representation by assigning PW boundaries to the edges of constituent X (see also Kabak and Revithiadou 2009 for more examples and argumentation). X

(29)

PW

X [[a

PW

→ b]

c]

[[a

b]

c]

A recursive PW (PW-Rec) comes in two shapes: as a result of adjunction (30a) and as a prosodic compound (30b). The latter is typically associated with word–word

Anthi Revithiadou

1222

compounds and sequences of function words, the first of which may be inherently stressed or stressed due to binarity. (30)

a.

adjunction

b.

prosodic compound PW

PW

[a fnc prf stem

b] word word word

PW

PW

PW c]]

[[a b] ’fnc fnc word

[c

d]] word word

In the Prosodic Phonology literature, PW recursion does not always have a morphosyntactic motivation. It is often assumed to arise either as a parsing choice of a particular language or as the result of the subcategorization requirements of individual elements (e.g. Booij and Lieber 1993 on certain Dutch clitics and prefixes). Evidence in support of recursion is primarily drawn from phonology and, specifically, the fashion in which phonological rules apply within the lower and upper PW. The main motivation for the existence of a PW-Rec is the blocking8 or the optional application of a PW-level phonological process (e.g. Booij 1995, 1996; Peperkamp 1997; Vigário 1999, 2003). In Greek, for example, proclitics and certain prefixes are subject to the same segmental rules that typically apply within the PW (31a), such as s-voicing before a nasal or a voiced fricative: (31)

a.

/:eras-’menos/ old-part b. /mas ’Ïinis/ clt.1pl.gen give-2sg c. /Ïis-’mirii/ twice-ten thousand-pl

:era’zmenos ‘aged’ maz.’Ïinis ‘you give us’ Ïiz.’mirii ‘twenty thousand ones’

The fact that such sub-minimal elements are part of the extended PW and not of the PW is evidenced by the blocking of resyllabification, e.g. :era.’zme.nos vs. maz.’Ïi.nis, which indicates the existence of a boundary at the left edge of the word. The PW boundary prevents the proclitic/prefix from fully incorporating into the PW of its host/base, suggesting that the sub-minimal element adjoins recursively to the PW of the word: [cl/prf [X0]PW]PW. The recursive PW has proved to be extremely useful in accounting for attested asymmetries in the degree of cohesion that clitics, affixes, and other dependent elements show in relation to their host. For instance, in many languages enclitics incorporate to their host, whereas proclitics adjoin recursively to it (e.g. Booij 1996 for Dutch; Peperkamp 1997 for Italian; Vigário 2003 for Portuguese; Revithiadou 8

The blocking of rule application is considered as an immediate result of adjunction. Elements of the outer layer of the PW inherit the properties of the mother constituent, but, because they are not dominated by all of its segments (Chomsky 1986), they can escape (some of) the rules (cf. Booij 1996).

The Phonological Word

1223

and Spyropoulos 2008 for Greek). Similarly, languages may choose to incorporate suffixes but not prefixes into the PW of their base. Despite its broad use, however, the recursive PW has been called into question as a legitimate prosodic constituent mainly because it is inherently incompatible with the non-recursive nature of phonology (see e.g. Neeleman and Koot 2006; Scheer 2008; Vogel 2009).

6

Conclusions

It is clear from the discussion so far that the PW, as a constituent that lies at the heart of the interface of phonology with morphology and syntax, cannot fully escape the problems naturally associated with the complex nature of the mapping. Four key aspects of PW have been the focus of attention in this chapter: (a) the mechanism that maps a string of elements into a PW and the principles that govern it, (b) the distinction between lexical and post-lexical PW, which essentially reflects the split nature of the mapping itself, (c) the type of criteria used to motivate the PW domain, and (d) the solutions proposed to account for conflicting evidence. The discussion has revealed that all of these issues are surrounded by a number of sometimes thorny problems, and has left numerous questions open for further research. For instance, it is still undecided whether the distinction between lexical and post-lexical PW can be dispensed with or not, or whether certain types of processes are universally associated with the PW. On the other hand, PW, as a theoretical construct, has been shown to play an important role in language acquisition (e.g. Fikkert 1994; Gerken 1994) and in language change (e.g. Lahiri 2000). Furthermore, psycholinguistic research has established a strong relation between the PW and units of production and perception (Wheeldon and Lahiri 1997, 2002), thus lending further support to the PW as a functionally useful constituent of the Prosodic Hierarchy. Future research will hopefully shed light on less clear aspects of the properties of the PW and the mode in which it is constructed and hence advance our understanding of this pivotal constituent of the Prosodic Hierarchy and, by extension, of the prosodic organization of grammatical elements.

ACKNOWLEDGMENTS I wish to thank two anonymous reviewers, as well as the editors of the Companion to Phonology, especially Marc van Oostendorp and Keren Rice, for their insightful comments. All errors are of course my own.

REFERENCES Andersen, Henning (ed.) 1986. Sandhi phenomena in the languages of Europe. Berlin: Mouton de Gruyter. Booij, Geert. 1983. Principles and parameters in prosodic phonology. Linguistics 21. 249–280. Booij, Geert. 1985a. The interaction of phonology and morphology in prosodic phonology. In Edmund Gussmann (ed.) Phono-morphology: Studies in the interaction of phonology and morphology, 23 –34. Lublin: Katolicki Uniwersytet Lubelski.

1224

Anthi Revithiadou

Booij, Geert. 1985b. Coordination reduction in complex words: A case for prosodic phonology. In van der Hulst & Smith (1985), 143–160. Booij, Geert. 1988. Review of Nespor & Vogel (1986). Journal of Linguistics 24. 515 –525. Booij, Geert. 1995. The phonology of Dutch. Oxford: Clarendon Press. Booij, Geert. 1996. Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13. 219 –242. Booij, Geert. 1999. The role of the prosodic word in phonotactic generalizations. In Hall & Kleinhenz (1999), 47–72. Booij, Geert & Rochelle Lieber. 1993. On the simultaneity of morphological and prosodic structure. In Sharon Hargus & Ellen M. Kaisse (eds.) Studies in Lexical Phonology, 23–44. San Diego: Academic Press. Booij, Geert & Jerzy Rubach. 1984. Morphological and prosodic domains in Lexical Phonology. Phonology Yearbook 1. 1–27. Booij, Geert & Jerzy Rubach. 1990. Edge of constituent effects in Polish. Natural Language and Linguistic Theory 8. 427–463. Cetnarowska, Bozena. 2000. On the (non)recursivity of the prosodic word in Polish. ZAS Papers in Linguistics 19. 1–21. Chen, Matthew Y. 1987. The syntax of Xiamen tone sandhi. Phonology Yearbook 4. 109–149. Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Dixon, R. M. W. 2002. The eclectic morphology of Jarawara, and the status of word. In Dixon & Aikhenvald (2002). 125 –152. Dixon, R. M. W. & Alexandra Y. Aikhenvald (eds.) 2002. Word: A cross-linguistic typology. Cambridge: Cambridge University Press. Fikkert, Paula. 1994. On the acquisition of prosodic structure. Ph.D. dissertation, University of Leiden. Gerken, LouAnn. 1994. Young children’s representation of prosodic phonology: Evidence from English speakers; Weak syllable omissions. Journal of Memory and Language 33. 19–38. Grijzenhout, Janet & Barı; Kabak (eds.) 2009. Phonological domains: Universals and deviations. Berlin: Mouton de Gruyter. Hall, T. A. 1999. Phonotactics and the prosodic structure in German function words. In Hall & Kleinhenz (1999), 99 –131. Hall, T. A. & Ursula Kleinhenz (eds.) 1999. Studies on the phonological word. Amsterdam & Philadelphia: John Benjamins. Hannahs, S. J. 1991. Prosodic structure and French morphophonology. Ph.D. dissertation, University of Delaware. Hannahs, S. J. 1995a. Prosodic structure and French morphophonology. Tübingen: Niemeyer. Hannahs, S. J. 1995b. The phonological word in French. Linguistics 33. 1125–1144. Hayes, Bruce. 1989. The prosodic hierarchy in meter. In Paul Kiparsky & Gilbert Youmans (eds.) Rhythm and meter, 201–260. San Diego: Academic Press. Henderson, John. 2002. The word in Eastern/Central Arrernte. In Dixon & Aikhenvald (2002), 100–124. Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Hulst, Harry van der & Norval Smith (eds.) 1982. The structure of phonological representations. 2 parts. Dordrecht: Foris. Hulst, Harry van der & Norval Smith (eds.) 1985. Advances in nonlinear phonology. Dordrecht: Foris. Inkelas, Sharon. 1989. Prosodic constituency in the lexicon. Ph.D. dissertation, Stanford University. Inkelas, Sharon & Draga Zec (eds.) 1990. The phonology–syntax connection. Chicago: University of Chicago Press.

The Phonological Word

1225

Itô, Junko & Armin Mester. 2003. Weak layering and word binarity. In Takeru Honma, Masao Okazaki, Toshiyuki Tabata & Shin-Ichi Tanaka (eds.) A new century of phonology and phonological theory: A festschrift for Professor Shosuke Haraguchi on the occasion of his sixtieth birthday, 26 –65. Tokyo: Kaitakusha. Itô, Junko & Armin Mester. 2009. The extended prosodic word. In Grijzenhout & Kabak (2009), 135–194. Kabak, Barı; & Anthi Revithiadou. 2009. An interface account to prosodic word recursion. In Grijzenhout & Kabak (2009), 105 –134. Kaisse, Ellen M. 1983. The syntax of auxiliary reduction in English. Language 59. 93 –122. Kaisse, Ellen M. 1985. Connected speech: The interaction of syntax and phonology. New York: Academic Press. Kang, Ongmi. 1992. Korean prosodic phonology. Ph.D. dissertation, University of Washington. Kiparsky, Paul. 1982a. Lexical morphology and phonology. In Linguistic Society of Korea (ed.) Linguistics in the morning calm, 3 –91. Seoul: Hanshin. Kiparsky, Paul. 1982b. From Cyclic Phonology to Lexical Phonology. In van der Hulst & Smith (1982: Part I), 131–175. Kleinhenz, Ursula. 1998. On words and phrases in phonology: A comparative study with focus on German. Ph.D. dissertation, University of Tübingen. Lahiri, Aditi. 2000. Hierarchical restructuring in the creation of verbal morphology in Bengali and Germanic: Evidence from phonology. In Aditi Lahiri (ed.) Analogy, leveling, markedness: Principles of change in phonology and morphology, 70–123. Berlin & New York: Mouton de Gruyter. Leben, William R. & Firmin Ahoua. 1997. Prosodic domains in Baule. Phonology 14. 113–132. Liberman, Mark. 1975. The intonational system of English. Ph.D. dissertation, MIT. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. McCarthy, John J. & Alan Prince. 1993a. Generalized alignment. Yearbook of Morphology 1993. 79–153. McCarthy, John J. & Alan Prince. 1993b. Prosodic morphology I: Constraint interaction a nd satisfaction. Unpublished ms., University of Massachusetts, Amherst & Rutgers University. McCarthy, John J. & Alan Prince. 1998. Prosodic morphology. In Andrew Spencer & Arnold M. Zwicky (eds.) The handbook of morphology, 283–305. Oxford & Malden, MA: Blackwell. McCawley, James D. 1968. The phonological component of a grammar of Japanese. The Hague & Paris: Mouton. Mohanan, K. P. 1982. Lexical Phonology. Ph.D. dissertation, MIT. Distributed by Indiana University Linguistics Club. Neeleman, Ad & J. van de Koot. 2006. On syntactic and phonological representations. Lingua 116. 1524–1552. Nespor, Marina. 1985. The phonological word in Italian. In van der Hulst & Smith (1985), 193–204. Nespor, Marina. 1986. The phonological word in Greek and Italian. In Andersen (1986), 65–74. Nespor, Marina. 1990. Vowel deletion in Italian: The organization of the phonological component. The Linguistic Review 7. 375 –398. Nespor, Marina & Irene Vogel. 1982. Prosodic domains of external sandhi rules. In van der Hulst & Smith (1982: Part I), 225 –255. Nespor, Marina & Irene Vogel. 1983. Prosodic structure above the word. In Anne Cutler & D. Robert Ladd (eds.) Prosody: Models and measurements, 123 –140. Berlin: Springer.

1226

Anthi Revithiadou

Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Odden, David. 1987. Kimatuumbi phrasal phonology. Phonology Yearbook 4. 13 –36. Odden, David. 1990. Syntax, lexical rules and postlexical rules in Kimatuumbi. In Inkelas & Zec (1990), 259 –277. Peperkamp, Sharon. 1997. Prosodic words. Ph.D. dissertation, University of Amsterdam. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Pyle, Charles. 1972. On eliminating BM’s. Papers from the Annual Regional Meeting, Chicago Linguistic Society 8. 516 –532. Raffelsiefen, Renate. 1999. Diagnostics for prosodic words revisited. In Hall & Kleinhenz (1999), 133–201. Revithiadou, Anthi. 2008. A cross-dialectal study of cliticisation in Greek. Lingua 118. 1393 –1415. Revithiadou, Anthi & Vassilios Spyropoulos. 2008. Greek object clitic pronouns: A typological survey of their grammatical properties. STUF: Language Typology and Universals 61. 39 –53. Rice, Keren. 1993. The structure of the Slave (Northern Athabaskan) verb. In Sharon Hargus & Ellen M. Kaisse (eds.) Studies in Lexical Phonology, 145–171. San Diego: Academic Press. Rotenberg, Joel. 1978. The syntax of phonology. Ph.D. dissertation, MIT. Scheer, Tobias. 2008. Why the prosodic hierarchy is a diacritic and why the interface must be direct. In Jutta M. Hartman, Veronika Hegedüs & Henk van Riemsdijk (eds.) Sounds of silence: Empty elements in syntax and phonology, 145 –192. Amsterdam: Elsevier. Selkirk, Elisabeth. 1972. The phrase phonology of English and French. Ph.D. dissertation, MIT. Selkirk, Elisabeth. 1978a. The French foot: On the status of “mute” e. Studies in French Linguistics 1. 141–150. Selkirk, Elisabeth. 1978b. On prosodic structure and its relation to syntactic structure. Paper presented at the Conference on Mental Representation in Phonology. Selkirk, Elisabeth. 1980. Prosodic domains in phonology: Sanskrit revisited. In Mark Aronoff & Mary-Louise Kean (eds.) Juncture, 107–129. Saratoga: Anma Libri. Selkirk, Elisabeth. 1981a. On prosodic structure and its relation to syntactic structure. In Thorstein Fretheim (ed.) Nordic prosody II, 111–140. Trondheim: Tapir. Selkirk, Elisabeth. 1981b. On the nature of phonological representation. In Terry Myers, John Laver & John Anderson (eds.) The cognitive representation of speech, 379 –388. Amsterdam: North Holland. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. Phonology Yearbook 3. 371– 405. Selkirk, Elisabeth. 1995. The prosodic structure of function words. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 439–469. Amherst: GLSA. Selkirk, Elisabeth & Tong Shen. 1990. Prosodic domains in Shanghai Chinese. In Inkelas & Zec (1990), 313 –338. Topintzi, Nina. 2003. Prosodic patterns and the minimal word in the domain of Greek truncated nicknames. Proceedings of the 6th International Conference of Greek Linguistics. Available (April 2010) at: www.philology.uoc.gr/conferences/6thICGL. Vigário, Marina. 1999. On the prosodic status of stressless function words in European Portuguese. In Hall & Kleinhenz (1999), 253–293. Vigário, Marina. 2003. The prosodic word in European Portuguese. Berlin & New York: Mouton de Gruyter. Vogel, Irene. 1985. On constraining prosodic rules. In van der Hulst & Smith (1985), 217–233.

The Phonological Word

1227

Vogel, Irene. 1986. External sandhi rules operating between sentences. In Andersen (1986), 55–64. Vogel, Irene. 1991. Level ordering in Italian Lexical Phonology? In Pier Marco Bertinetto, Michael Kenstowicz & Michele Lopocaro (eds.) Certamen Phonologicum II: Papers from the 1990 Cortona Phonology Meeting, 81–101. Turin: Rosenberg & Sellier. Vogel, Irene. 2009. The status of the Clitic Group. In Grijzenhout & Kabak (2009), 15–46. Wheeldon, Linda R. & Aditi Lahiri. 1997. Prosodic units in speech production. Journal of Memory and Language 37. 356 –381. Wheeldon, Linda R. & Aditi Lahiri. 2002. The minimal unit of phonological encoding: Prosodic or lexical word. Cognition 85. B31–B41.

52

Ternary Rhythm Curt Rice

1

The facts

Most languages with iterative stress patterns show a simple rhythmic alternation between stressed and unstressed syllables (chapter 39: stress: phonotactic and phonetic evidence; chapter 41: the representation of word stress). But in a few cases, stress appears not on every second syllable, but rather on every third one. Patterns of this nature reveal the phenomenon of ternary rhythm. Ternary rhythm is most easily seen in a language with a stress system that ignores the internal structure of syllables, i.e. a quantity-insensitive system (see chapter 57: quantity-sensitivity). Cayuvava, now extinct, but formerly spoken in parts of Bolivia, is a language well documented in the work of Key (e.g. Key 1961, 1967). It is classified as an isolate, with no established genetic relationship to other languages. Key’s fieldwork documents ternary rhythm in Cayuvava and no relevant syllable quantity. Stress in this language appears on every third syllable counting from the right edge of the word. To see the pattern schematically, consider the representations in (1). Each number represents a syllable: “0” represents a syllable with no stress, “1” represents a syllable with primary stress, and “2” represents secondary stress. The pattern is claimed to emanate from the right edge of the word and the representations here are therefore right-justified. The pattern clearly emerges from this schematic representation. (1)

Ternary alternation patterns of Cayuvava a. 10 b. 100 c. 0100 d. 00100 e. 200100 f. 0200100 g. 00200100 h. 200200100 i. 0200200100

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0052

Ternary Rhythm

2

The transcribed data from the literature on Cayuvava flesh out the schematic patterns. We can see forms from Key’s work in (2), which correspond to the patterns already sketched in (1). (2)

Cayuvava a. b. c. d. e. f. g.

’da.pa ’to.mo.ho a.’ri.po.ro a.ri.’pi.ri.to ‘a.ri.hi.’hi.be.e ma.‘ra.ha.ha.’e.i.ki i.ki. ‘ta.pa.ra.’re.pe.ha

‘canoe’ ‘small water container’ ‘he already turned around’ ‘already planted’ ‘I have already put the top on’ ‘their blankets’ ‘the water is clean’

To see a more complex instance of ternary rhythm, we turn to Tripura Bangla. Das (2001) describes Tripura Bangla as a dialect of Bangla, resulting from a complicated sociolinguistic situation in the small Indian state of Tripura, where it is a commonly used lingua franca. One complication in the pattern of Tripura Bangla when compared with Cayuvava is the relevance of syllable structure for stress assignment. Before illustrating this, we can discern the default pattern through a consideration of words consisting only of light syllables. In such strings, we find main stress on the initial syllable and secondary stress emanating rightward in a ternary rhythm. However, a final light syllable cannot bear stress. When the pattern would place stress on a final syllable – e.g. in strings of four or seven syllables – that stress is not realized. This means, for example, that a word consisting of exactly four light syllables will have only one stress, namely the main stress on the word-initial syllable. (3)

Tripura Bangla default pattern a. b. c. d. e. f. g. h. i. j.

’ra.za ’go.ra.li ’ne.ta ’boi.ra.gi ’be.na.ro.œi ’bi.ße.sD.na ’œD.ma.lD.‘sD.na ’o.nu.kD.‘ro.ni.jD ’D.no.nu.‘da.ßo.ni.jD ’D.no.nu. ‘kD.ro.ni.‘jD.ta

‘king’ ‘ankle’ ‘leader’ ‘mendicant’ ‘Benaras silk’ ‘consideration’ ‘criticism’ ‘imitable’ ‘unintelligible’ ‘inimitability’

These patterns can be perturbed by closed syllables. Closed syllables can under certain circumstances tolerate stress in word-final position and they can also draw stress off from a word-initial open syllable. A simple generalization is that a stressed light syllable cannot be immediately followed by a closed syllable. When this would happen, the closed syllable bears stress instead. Further complications assign stress to the third syllable when it is heavy and to a word-final closed syllable, unless immediately preceded by a stressed syllable. Providing analyses at this level of detail is not the aspiration here, but both Das (2001) and Houghton (2006) discuss these patterns in detail. The data in (4) have closed syllables in

Curt Rice

3

various positions. In positions where we expect stress anyway, the patterns are as in (3). In other cases, the heavy syllable interrupts the default pattern. (4)

Tripura Bangla quantity effects a. b. c. d. e. f. g. h. i. j. k. l.

’mal.œa ’za.til ’œDr.kar D.’hDI.kar ’œDI.rDk.‘kDn o.’big.ga.‘zDn ’o.nu.‘bik.kDn ’baz.zaiÍ.Ía.mi ’D.no.‘nu.kD.‘rDn ’zD–.–a.lD.‘sD.na ’œDI.rDk.ko.‘ni.jD.ta ’za.rD.‘doœ.œi.kD.ta

‘big metal bowl’ ‘earthen pot’ ‘government’ ‘pride’ ‘reservation’ ‘intimation’ ‘microscope’ ‘adamancy’ ‘non-imitation’ ‘deliberation’ ‘preservability’ ‘expertness’

Ternary rhythm is also visible in Chugach Alutiiq, a Yup’ik language spoken by a small number of individuals in Alaska. This language is most extensively documented in a series of important works by Leer (e.g. 1985a, 1985b). These data and Leer’s discussion figure prominently in the literature on ternary rhythm, including many of the theoretical works cited in the present chapter. Quantity is also relevant for the placement of stress in Chugach Alutiiq, but unlike in Tripura Bangla, a syllable must have a long vowel to perturb the pattern. Closed syllables with short vowels – except when in word-initial position – do not attract stress. Assuming that the default stress pattern is revealed in strings with no relevant quantity distinctions, stress in Chugach appears on the second syllable and then every third syllable thereafter. A word in Chugach Alutiiq with five or six light syllables will have stress on the second and fifth. A word with four, however, will have stress on the second and fourth. Syllables with long vowels always attract stress. (5)

Chugach Alutiiq a. b. c. d. e. f. g. h. i. j. k. l.

mu.’lu.kan a.’ku.ta.’mek ta.’qa.ma.lu.’ni a.’ku.tar.tu.’nir.tuq ma.’Iar.su.qu.’ta.qu.’ni ’taa.’taa ’taa.ta.’qa ’naa.ma.ci.’quq ’naa.qu.ma.’lu.ku ’naa.ma.’ci.’qua mu.’u.’kuut u.’lu.te.ku.’ta.’raa

‘if she takes a long time’ ‘akutaq (a food) (abl sg)’ ‘apparently getting done’ ‘he stopped eating akutaq’ ‘if he (refl) is going to hunt a porpoise’ ‘her father’ ‘my father’ ‘it will suffice’ ‘apparently reading it’ ‘I will suffice’ ‘if you take a long time’ ‘he’s going to watch her’

The three cases presented above are the clearest examples of ternary rhythm that have been uncovered to this point. The languages include some very long words, and even in the quantity-sensitive languages, there are words consisting of long

Ternary Rhythm

4

strings of light syllables. Leer’s transcriptions of those strings indicate stressed syllables that are separated by two unstressed syllables. This is the empirical basis for the claim that ternary rhythm is a real phenomenon and that a metrical theory of stress assignment must have formal tools that can generate such patterns. While the clearest cases are presented above, there are other languages which have been analyzed as having ternary rhythm, at least in some subset of the data. Most familiar among these are Ho-Chunk, Sentani, and Munster Irish. Having established the basis for the claim that ternary rhythm is an empirical fact, we turn now to metrical theory and the major strategies that that literature offers for the analysis of these data.

2

Theory and analysis

The preceding section has established that a plausible theory of metrical structure must offer a strategy for modeling ternary rhythm. We turn now to a brief review of the emergence of this issue in the literature and the general tendencies that can be identified. Hints about the treatment of ternary rhythm can be found very early in the development of a generative theory of stress assignment. As chapter 40: the foot discusses in detail, early work in generative phonology treated stress as the realization of a phonological feature [stress]. In this way, stress was analyzed with tools parallel to those used in the analysis of place of articulation – e.g. [coronal] or [dorsal] – or manner of articulation – e.g. [voice] or [continuant]; cf. Chomsky and Halle (1968) (see chapter 17: distinctive features). A breakthrough in the study of stress systems came with Liberman’s (1975) proposal that stress should be characterized not as a feature with absolute values, but rather as a relation in which two elements differ in their relative prominence. Along with this proposal came hierarchical representations and the introduction of the metrical foot into the generative literature, further developed in Liberman and Prince (1977). The foot naturally invited a more extensive theory of prosodic structure, incorporating segments into syllables, syllables into feet, and so on up the prosodic tree to the phrase or utterance; cf. Nespor and Vogel (2008) (see also chapter 40: the foot; chapter 51: the phonological word; chapter 84: clitics; chapter 50: tonal alignment). This is the context in which any proposed modifications of metrical theory find themselves today. The first extensive typological work on stress systems is found in Hayes (1980). Hayes studies the stress systems of many languages, and identifies a number of parameters that can be used to characterize the variation shown in these languages. Parameters specify points of variation, such as the direction of foot construction, sensitivity to syllable-internal quantitative structure, trochaic or iambic headedness of the feet, whether feet are binary or unbounded, the edge of the word that hosts main stress, and whether or not peripheral material can be excluded from the initial parse through extrametricality. And, indeed, it is precisely the discussion leading up to the proposal of extrametricality that includes the earliest considerations of ternary rhythm. Before turning to the treatment of iterative ternary rhythm in metrical theory and Optimality Theory (OT), the relevance of extrametricality and its competitor are discussed. For a more thorough overview of metrical theory, see van der Hulst (1999) or Hammond (1995).

5

3

Curt Rice

Extrametricality vs. ternary feet

Extrametricality as a theoretical tool arose in response to apparent ternary rhythm at the edges of words. The stress pattern of English nouns offers a relevant illustration. In sufficiently long words, we can see that English displays alternating, binary stress assignment, in words such as Apalachicola, Minnesota, candelabra. But when we examine the right edges of words more closely, we quickly find that stress is sometimes found not on one of the final two syllables, but rather on the antepenultimate syllable, as in America, cinema, analysis. In this way, we identify a fundamentally binary system that has a ternary component, namely a three-syllable window at the right edge of the word. A model that only constructs binary feet over an entire word would not be able to generate this pattern. Specifically, the construction of feet from right to left in English nouns would always result in penultimate stress. How can a binary foot “reach in” far enough to position primary stress on the antepenult? To model antepenultimate stress, two possible enhancements of the theory were entertained early on. One of these is extrametricality. Extrametricality is a theoretical tool that does not explicitly entail enhancement of the inventory of feet. Instead, it provides a particular strategy for foot construction, or parsing a string of syllables. In particular, extrametricality excludes a peripheral syllable from the string to be parsed into feet. In the case of English nouns, exclusion of the final syllable, followed by construction of a binary leftheaded foot, will place stress on the antepenultimate syllable. Extrametricality is also illustrated in chapter 43: extrametricality and non-finality. As we will see below, some later work on iterative ternary rhythm relativizes the peripherality requirement, such that syllables can be excluded from the string not only when they are word-peripheral but also, for example, when they are foot-peripheral. A conceptually different approach from extrametricality would be to enhance the model such that it also includes ternary feet. Data of the type described for English nouns would then be modeled by building a ternary foot at the edge, followed by the construction of binary feet iterating leftward. Since the stress that is found in the three-syllable window is the primary stress, this amounts to a proposal that primary stress can be modeled through the use of one kind of foot while secondary stress requires another. Such proposals can be found for other points of parametric variation for foot construction, as well. For example, primary stress may require the use of a quantity-sensitive foot, while iterative secondary stress seems to be quantity-insensitive (van der Hulst 1984, 1999). One strategy for modeling an edgemost ternary domain – extrametricality – enhances the parsing strategies available in the theory, while the other strategy – a ternary foot – enhances the inventory of feet available in the theory.

4

Modeling iterative ternary rhythm

The parameters of metrical theory specify the nature of feet and control their construction across words in languages. The feet that are constructed are constituents that create a domain for the assignment of relative prominence. Prince (1983) offers an alternative approach without internal constituency, representing relative prominence instead only with a grid, and reconstruing some of Hayes’s

Ternary Rhythm

6

parameters such that their effects can be replicated without binary constituents (see also chapter 41: the representation of word stress). The debate about constituency includes argumentation based on sensitivity of non-stress phenomena to feet, as reviewed in Kenstowicz (1993). This debate is present in much of the subsequent literature, finding one of its most extensive and significant considerations in Halle and Vergnaud (1987). Halle and Vergnaud’s constituentized grid representation integrates grids and feet. Grids are built but the gridmarks are grouped and these groupings represent constituents. The construction process implements parameter settings here, too, also in pursuit of a typology of stress systems. And it is here, in Halle and Vergnaud’s opus, that we find the first discussion of iterative ternary stress presented in a major work on stress system typology. Halle and Vergnaud of course draw on papers and presentations regarding ternarity that were floating about in the immediately preceding years, with some issues already nascent in McCarthy (1979). The discussion of iterative ternary rhythm and its implications for the typologies under consideration in the relevant literature was initiated by Levin (1985), which was ultimately published in a significantly modified form as Levin (1988). Levin’s work drew on the data from Cayuvava in (2). Halle and Vergnaud (1987) discuss neither Tripura Bangla nor Chugach Alutiiq. Regarding the latter, Leer’s (1985a, 1985b) careful and important results would soon influence the details of the constituentized grid theory. Leer’s work was picked up on in Rice (1988), where an analysis in the spirit of Halle and Vergnaud (1987) is advanced. This, in turn, influenced subsequent revisions of the theory, as presented in Halle (1990). The theory developed by Halle and Vergnaud models ternary rhythm through the construction of ternary feet, extending to the problem of iterative ternary rhythm the spirit of the approach discussed above in the context of word-final three-syllable stress windows. A competing approach also reflects that earlier debate. This competitor maintains a size limit such that feet are maximally binary. Ternary rhythm is achieved with a parsing strategy that leaves occasional syllables unincorporated into feet, extending the basic notion of extrametricality; cf. Hammond (1990) and Hayes (1995). These two general approaches, to be illustrated presently, form the heart of the theoretical debate occasioned by ternary stress patterns. As we will see in the discussion of ternary rhythm and OT below, the debate persists there, too. We turn now to the chronologically first approach, namely an analysis of ternary rhythm using ternary feet.

5 5.1

Ternary feet Amphibrachs

At first glance, the Cayuvava stress patterns in (1) and (2) suggest an analysis with dactylic feet (strong–weak–weak), built from right to left. If we maintain a parametric strategy for constructing feet, then the independently established presence in the theory of a parameter placing heads at the left or right edge of the foot means that the admission of dactyls to the inventory of derivable feet would imply

Curt Rice

7

the introduction of anapests (weak–weak–strong) as well. Allowing a ternary foot with its head at the left edge implies via the relevant parameter the possibility of constructing a ternary foot with its head at the right edge. With Cayuvava as the only known case of iterative ternary rhythm at the time of this theoretical work, generating dactyls would lead to the phenomenon of overgeneration, i.e. being able with the tools of the theory to generate patterns not known to exist. In pursuit of a restrictive theory, Levin (1988) therefore takes a different tack, relaxing metrical theory just enough to allow for exactly one type of ternary foot, instead of two; dactyls and anapests are disallowed, but the theory now permits amphibrachs, i.e. ternary feet with prominence on the middle syllable, employing a strategy described below. When combined with final extrametricality – which can be overridden when necessary to build at least one foot on the (minimal) disyllabic words – the construction of amphibrachs will yield a footing of the schematic patterns in (1) that correctly locates stress, as seen in (6). Parentheses indicate feet and angled brackets mark extrametricality. In longer words, initial lone syllables are left unfooted, by stipulation. (6)

Ternary alternations parsed into amphibrachs a. b. c. d. e. f. g. h. i.

(10) (10)〈0〉 (010)〈0〉 0(010)〈0〉 (20)(010)〈0〉 (020)(010)〈0〉 0(020)(010)〈0〉 (20)(020)(010)〈0〉 (020)(020)(010)〈0〉

Halle and Vergnaud (1987) adopt Levin’s strategy, and also limit Universal Grammar (UG) to this one type of ternary foot. They parameterize the requirement that the head of a constituent be at its edge. When this parameter is set such that the head is not required to be at the edge of a constituent, the only ternary foot that can emerge is an amphibrach; cf. Rice (1988, 1990) for related discussion. The approach developed by Levin and widely discussed in publications by Halle and Vergnaud effectively views iterative ternary rhythm as evidence for expanding the inventory of feet. Constituents may have one head, but as many as two non-heads. For them, there is no hierarchical structure within the foot, so that this approach generates flat ternary feet. Another main thrust of the literature also sees a proposal with ternary feet, but now with internal hierarchical structure. Early proponents of this include Dresher and Lahiri (1991) and Rice (1992), building on Rice (1990). Leer’s (1985b) article offers the leading idea, namely identifying the quantitative equivalence of two light syllables with a single heavy syllable, allowing either of those configurations to be the head of a foot. Taking into account a non-head consisting of a light syllable, a foot might consist of three light syllables, two of which are themselves a subconstituent. Hence, ternary feet become an option. In the foot typology of Hayes (1980), some languages were identified in which the heads of feet must be heavy, a foot type dubbed the obligatory branching foot.

Ternary Rhythm

8

Rice (1992) in particular draws a parallel between Hayes’s obligatory branching feet and the analysis of ternarity under consideration, since the head consisting of one heavy syllable or two light ones could be construed as obligatory branching. That analysis is also relevant to the contrast between (6e) and (6d): in the former case, two word-initial syllables concluding the right-to-left parse are sufficient for a foot, while the single syllable in the latter case is not. The analysis in Rice (1992) suggests that degenerate feet must have a head, and in the case of the ternary feet constructed for Cayuvava, two lights are required to constitute a head, hence the minimal foot (and word) is binary. In the approach with flat ternary feet, it is unclear why a minimum of two syllables is necessary for a degenerate foot. Additional discussion related to this approach can be found in Everett (1988), Hewitt (1992), Rice (1993), Blevins and Harrison (1999), van der Hulst (1999), Rifkin (2003), and other references mentioned below.

5.2

Weak local parsing

The appearance of iterative ternary stress patterns in the literature on metrical phonology triggered, as noted above, a second strategy. Instead of increasing the set of possible feet, this second strategy increased the set of possible parsing strategies. This approach is developed in Hayes (1995), drawing on earlier work by Hammond (1990). In Hayes’s approach, universal grammar allows only three kinds of feet, as in (7) (see also chapter 44: the iambic–trochaic law). (7)

The Hayesian foot typology a. b. c.

Syllabic trochee (x q Moraic trochee (x L Iamb (. L

.) q .) L x) q

or or

(x) H (x) H

No exhaustive parsing of a string with any of these feet will give an iterative ternary pattern. But non-exhaustive parsing can do that. Hayes proposes that UG include a weak local parsing parameter that creates the possibility of leaving an unparsed syllable between each foot. The unparsed syllable can by stipulation only be a light one. Having an unparsed light syllable between each foot yields a ternary pattern using only binary feet, as in (8). (8)

Ternary alternations parsed into non-exhaustive binary feet a. b. c. d. e. f. g. h. i.

(10) (10)〈0〉 0(10)〈0〉 00(10)〈0〉 (20)0(10)〈0〉 0(20)0(10)〈0〉 00(20)0(10)〈0〉 (20)0(20)0(10)〈0〉 0(20)0(20)0(10)〈0〉

9

Curt Rice

For the Cayuvava pattern, trochees are constructed from right to left; whether they are moraic or syllabic trochees is irrelevant, since there is no quantity distinction in this language. The parsing also uses final extrametricality and, of course, weak local parsing. The final syllable in (8c) is extrametrical and the initial syllable is unfooted, because it is too little to be a foot, since no degenerate feet are allowed. In (8d) the influence of weak local parsing is seen; in this form, there is in fact sufficient material at the left edge of the word to form a foot. Since doing so would result in adjacent feet – which is not allowed with weak local parsing – no foot can be formed. Not until we have six syllables, as in (8e), is there sufficient space to build two non-adjacent – i.e. weakly local parsed – feet. A detail beyond the scope of this chapter is that adjacency can be tolerated in the case of adjacent heavy syllables – as in Chugach – suggesting that the requirement to incorporate heavy syllables into feet has priority over the prohibition on adjacent feet under weak local parsing.

5.3

Summary

At the level of analysis, we have seen that there are two primary strategies for constructing constituents across strings when the goal is to achieve iterative ternary alternations. One strategy is to expand the inventory of constituents, and here there are also two approaches. In the approach developed by Levin (1988) and Halle and Vergnaud (1987), flat ternary feet are allowed. Any non-head in a foot must be adjacent to its head. This allows exactly one kind of ternary foot, namely an amphibrach, where the head is not found at the edge of the foot, but rather is flanked by two non-heads. The second inventory-expanding approach, as seen primarily in Dresher and Lahiri (1991) and Rice (1992), relaxes the requirement that the head of a foot can span only one syllable. Feet that require heavy heads can draw their material either from one heavy syllable or from two light ones. The alternative to expanding the foot inventory is expanding the strategies available for constructing binary feet, and the primary representative of this approach is Hayes (1995). In this approach, binary feet are constructed in a new way. There are two necessary properties to the weak local parsing of a string: feet must be non-adjacent and they must be minimally non-adjacent. These requirements lead to iterative construction of feet that are separated by one light syllable. There was, as noted, an early debate in the metrical literature regarding the need for constituency; perhaps stress systems can be modeled simply with a theory of prominence as represented with a grid, and feet are superfluous. This debate has not shown itself in the context of ternary rhythm, insofar as the literature lacks a grid-only analysis of iterative ternary rhythm.

6

Ternary rhythm in Optimality Theory

The typological enterprise in generative grammar has enjoyed enhanced prominence in the era of OT (McCarthy and Prince 1993; Prince and Smolensky 1993). One of the core foci in the OT literature is typology (Archangeli and Langendoen 1997; Roca 1997; Kager 1999). Classic OT achieves typological insights by having

Ternary Rhythm

10

a universally fixed set of constraints. Variation is modeled with constraint reranking. The factorial typology of constraint rankings defines the range of possible grammars. Stress patterns and metrical theory have played an important role in the construction and exploration of optimality-theoretic approaches to modeling grammar. Indeed, one of the important early discussions of the power of violable constraints was built around the pursuit of a parallelist strategy for achieving the effects of directionality. The insight in this discussion is that minimal violation of a requirement that all feet be at one edge of the word (AllFt-L or AllFt-R), when combined with the force of a requirement that all syllables be parsed into feet (Parse-q), will yield as optimal a parse identical with serial foot construction from one edge of a string to the other. Less present in the OT literature, however, has been discussion of ternary rhythm. There has been almost no debate about the contrast between analyses using ternary feet and those using non-exhaustive parsing with binary feet. Indeed, Rice (2007) is to the best of my knowledge the only publication in which the issue is even mentioned, although Hyde (2002) also offers relevant perspectives on the nature of ternary parsing. The most prominent discussions of ternary rhythm in OT mimic the weak local parsing approach, as in Ishii (1996) and Elenbaas and Kager (1999). Elenbaas and Kager take an important principled position on methodology. In particular, they articulate and adopt the goal of deriving iterative ternary rhythm with tools that are already necessary to account for other phenomena. This laudable position of theirs contrasts with the too frequent practice in OT analyses of positing new constraints to give new analyses. That practice has substantial implications, in light of the methodology of the factorial typology noted above; introducing a new constraint introduces many new grammars, and the restricted typological enterprise as construed in classic OT is substantially challenged with every new constraint that is introduced. The analysis of ternary rhythm in OT based on underparsing has two crucial components. First, Parse-q – which requires that syllables be incorporated into feet – must be relatively low-ranked. This will be important in allowing optimization of an incomplete parse along the lines seen in (8). But simply ranking Parse below a requirement that all feet align with the right edge of the word will yield a parse with only one foot. Note that AllFt-R awards a violation for every syllable intervening between the right edge of a foot and the right edge of the word, for each foot. (9)

Underparsing with low-ranked Parse qqqqqq a. (’qq)(’qq)(’qq) b. q(’qq)q(’qq) ☞ c. qqqq(’qq)

AllFt-R Parse *!***** *!**

** ****

This brings us to the second crucial component. To counter the pressure of AllFt-R, parsing of at least some of the other syllables must be rewarded. The solution offered builds on well-established insights that lapses in long parses

11

Curt Rice

should be avoided; cf. Selkirk (1984). There are various *Lapse constraints in the literature (e.g. Kager 1994; Green 1995; Gordon 2002), where the leading idea is that a string of more than two unstressed syllables is disfavored. With *Lapse ranked above AllFt-R, the optimal parse will show only as much parsing as is necessary to minimize violations of *Lapse, and will favor options in which the feet are relatively toward the right. (10)

Ternary rhythm with *Lapse qqqqqq

*Lapse AllFt-R Parse

a. (’qq)(q’q)(q’q) b. (q’q)q(q’q)q ☞ c. q(q’q)q(q’q) d. qq(q’q)(q’q)

*!

****!** ****!*

**

***

**

**

**

The OT analysis of ternary stress requires consideration of many more details and much more discussion, which is to be found in the cited works. For present purposes, it is sufficient to note that an analysis akin to Hayes’s weak local parsing strategy is achieved through the interaction of *Lapse and AllFt-R. Although the methodology of pursuing an analysis built simply on the reranking of independently motivated constraints is commendable, that goal has not yet been achieved. For example, careful study reveals that multiple versions of *Lapse will be necessary, one of which is specifically designed for the ternary cases; this is made laudably explicit in Houghton (2006). While the use of ternary-specific tools is not an a priori flaw of these analyses, it nonetheless keeps them from clearing the high bar set in pursuit of an analysis by pure reranking of constraints that are not ternary-specific. In addition to facilitating an illustration of the strategy that has been most thoroughly pursued in providing an analysis of ternary rhythm within OT, the patterns under consideration here raise another important methodological point. Future work in OT that considers the relative merits of the two main types of analyses illustrated above – ternary feet or underparsing – must consider related issues about the division of labor among the modules of the theory. Consider, for example, the possibility that a particular analysis intends to optimize a parse that does not use flat ternary feet, or amphibrachs. How will such feet and their optimization be avoided? If it is possible that prosodic structure is present in inputs – a possibility required by the richness of the base methodology – then amphibrachs are possibly present in inputs. One way in which these can be prevented from being selected as optimal is with a constraint that rules them out. Introducing an anti-amphibrach constraint, however, implicitly raises the possibility that it could be ranked relatively low, which in turn could open the door to the optimization of such structures. If one’s position is that amphibrachs are never optimal, then it is unfortunate to achieve this universal exclusion with a constraint. The alternative is to provide structure to Gen, such that the output of Gen cannot include amphibrachs. The tension between these possibilities is the focus of Rice (2007), and it is one of the general theoretical issues raised in OT by the study of ternary rhythm.

Ternary Rhythm

7

12

Implications and directions for future research

As noted earlier, one strategy for modeling an edgemost ternary domain – extrametricality – enhances the parsing strategies available in the theory, while the other strategy – a ternary foot – enhances the inventory of feet available in the theory. On what basis can a theoretician select the model to be pursued? Are these two options really different from one another in some meaningful sense? Does one approach allow for the description of some kind of situation that the other one does not? If the approaches do not make different predictions, are there other strategies available for selecting among them? One possibility would be to appeal to general principles of theory construction or to findings in other realms of cognitive science. Such principles or findings may have implications for selecting among competing theories of phonology. To see one example of argumentation for selecting among competing theories, we can turn to Hayes’s (1980) argumentation for extrametricality over ternary feet. This argument is based primarily on identifying differences in the types of systems the competing theories predict. In the cited work, Hayes compares his foot inventory at that point with a proposal made by Morris Halle, attributed to “class notes” (Hayes 1980: 114ff.). Halle had introduced ternary feet into his version of the theory, while Hayes had developed the approach using extrametricality. Hayes begins his argument against Halle’s inventory by stating the following: “I know of no languages whose stress patterns could simply be described using feet of the [ternary] form” (Hayes 1980: 115). This quotation suggests that a gap in the typology – namely the absence of languages with iterative ternary rhythm – can be used as an argument against a theory that could in fact model that. At the conclusion of the section, we are again encouraged to adopt an inventory with binary feet and peripheral extrametricality, in part because it provides “an explanation for why feet which have [ternary] surface forms . . . are never assigned iteratively” (Hayes 1980: 122). This example from the early literature on ternary rhythm illustrates an argument based on overgeneration. The theories are evaluated, and the one that generates a pattern not known to exist is dispreferred on those grounds. Gaps in the typology become a criterion for theory selection. However, as we now know from the sections above, this gap was soon revealed to be accidental. This is not the only example in the generative literature of argumentation based on gaps in the empirical record. We might use this occasion to ask whether such experiences are relevant as we hone our methodology for identifying the properties of UG. The following paragraphs present some of the broader implications that may be explored on the basis of our discussion of ternary rhythm. The typological enterprise as widely practiced in generative phonology aspires to model grammatical variation through simple formal manipulations of various components of the theory: reranking of constraints, different rules or rule orders, or the setting of parameters to different values. Generative linguists often see themselves as doing work grounded in a reliable typology of the structures found in natural languages. Our goal – cf. Odden (forthcoming) – is to identify the limits of the human linguistic capacity and to model that knowledge. What is a possible grammar, and what is not? What

13

Curt Rice

cognitive structures must be posited to restrict the outcome of the language acquisition process to possible grammars, thereby rendering unattainable the impossible ones? And regarding the case at hand, what does the fact of ternary rhythm force us to posit in our theory? These questions are sometimes studied from “above.” Researchers could take as their starting point a theory of cognitive capacity and build a model of linguistic knowledge within the context of that theory. Demonstrated incompatibility of a conceivable linguistic structure with a known fact about our cognitive system would be an argument for genuine universal ungrammaticality: cf. Odden (2008) and Reiss (2008). Alternatively, one could develop a theory from “below.” In this case, one would approach the matter through a deep study of one language or one family of languages, or through a carefully selected set of unrelated languages. Regardless of the starting point one adopts, any predictive model of linguistic knowledge will be held accountable to facts about natural languages. We work to enhance the empirical foundation for developing theories of grammars by studying individual languages. The very act of documenting, describing, and analyzing the multifarious properties and subsystems that are found in natural language entails engagement in a typological enterprise; cf. Newmeyer (2005) and Hyman (2009). A model of linguistic knowledge that allows the construction of a particular grammar gains credibility vis à vis its competitors when a specific language is identified that requires precisely that grammar. We might call this the positive typological enterprise: certain structures are attested in the set of well-studied natural languages, and any theory of grammar must sanction the generation of such structures. But there is also a negative typological enterprise. Work on this side of the program aspires to model the absence of unattested structures. If we imagine competing models of linguistic knowledge, all of which satisfy the positive side of the enterprise insofar as all of them generate those structures that are known to exist, then we need some criterion for choosing among them. Proposed models of linguistic knowledge are therefore routinely criticized from the negative side, i.e. for allowing the construction of a grammar that is not known to be instantiated by any natural language; this is the state of overgeneration, of which we saw an example above. A model that overgenerates is in principle inferior to one that does not. A model that completely fails to overgenerate matches the systems that it cannot generate with those that are unattested. While this seems at first glance to be an important goal, the danger we must guard against when attempting to eliminate aspects of a theory that overgenerates is the equation of unattested with impossible. How can we know whether a structure that is absent from the empirical database is merely unattested – as was the case with iterative ternary rhythm – or genuinely beyond the grasp of UG? Finding an answer to this question seems insurmountably forbidding when we realize that distinguishing the merely unattested from the cognitively impossible would be no easier if all languages were thoroughly documented, studied, and analyzed. Those activities, of course, may fill gaps in our knowledge, and they will certainly generate a richer base for our research enterprise, thereby contributing to a deeper understanding of our cognitive capacity. But even if all languages were deeply understood, any linguist would still be able to posit conceivable albeit unattested structures, and all theories that overgenerate would make this

Ternary Rhythm

14

challenge easy. Such conceivable but unattested structures can be assumed to be universally ungrammatical only if we assume that all possible structures in fact do appear somewhere in the set of human languages. This assumption, alas, is no more plausible than an assumption that all possible structures for eyes, for example, are attested somewhere in the animal kingdom. This realization presents a significant confrontation to the bottom-up approach to linguistic theory construction. Building theories on the basis of what is attested and unattested tends to confuse does not exist with cannot exist. The work of linguists is not to explain linguistic structures that do exist, but rather to explain linguistic structures that can exist. While the starting point may be the empirical record, that cannot be the ending point. The empirical record does not and cannot show us the limits of human linguistic capacity. The empirical record cannot reveal what is necessarily beyond the realm of our grammatical competence; cf. Isac and Reiss (2008).

8

Conclusions

Although it is not the purpose of this chapter to go deeply into methodological issues, it is essential nonetheless to highlight the importance of doing so. The study of ternary rhythm and the history of analyses of this phenomenon in the generative literature raise issues of the kind presented here. The remaining challenge is to take these discussions further, not only in this context, but whenever we enter into discussions arguing for the selection of one theoretical model over another. There is more work to be done in metrical theory and on ternary rhythm. This must address several issues, both in the context of a specific analysis and with respect to the theory with which that analysis is built. Naturally, any analysis has to satisfy the positive typological enterprise by allowing the generation of those patterns that are attested. An analysis must take a position on the kind of constituents that are constructed. Are they binary or ternary? If there are ternary feet, i.e. feet with three terminals, is there any internal structure to those constituents, or are they flat ternary structures? If there are not ternary feet, how are binary feet constructed with iterative non-exhaustivity, i.e. periodic non-parsing as with the earlier theory of weak local parsing? What is the fate of those syllables that are left unparsed? How are they incorporated into the prosodic structure of the word, and what implications does this have for a theory of layering in prosodic structure (Selkirk 1984)? Beyond the basic matter of developing an analysis, future work on this topic should also address the role of typological evidence in the theory being offered, specifically the extent to which patterns absent from the empirical record are considered to be not merely unattested but unattestable. In this way, the typological enterprise will remain an important topic of discussion, and the languages showing ternary rhythm will play their significant role in future refinements of metrical theory and the methodologies of research in generative linguistics.1 1

Works relevant to the study of ternary rhythm that have not been mentioned in this article, but which students of this topic should consult, include the following: Crowhurst (1992); Idsardi (1992); Kager (1993); Green and Kenstowicz (1995); Rowicka (1996); van de Vijver (1998); Elenbaas (1999); Hyde (2001); Gordon (2002); McCartney (2003); Karttunen (2006); Buckley (2009).

15

Curt Rice

ACKNOWLEDGMENTS For feedback on an early draft of this chapter, I thank Sylvia Blaho, Peter Jurgec, Bruce Morén-Duolljá, Dave Odden, Charles Reiss, Bridget Samuels, and two anonymous reviewers. I also thank Sylvia Blaho for extensive support during the editing process. I am indebted to phonologists too many to name, with whom I have had fruitful discussions about metrical theory and ternarity over the years. Some of the perspectives advanced here owe an intellectual debt to Hale and Reiss (2008), which is hereby gratefully acknowledged.

REFERENCES Archangeli, Diana & D. Terence Langendoen. 1997. Optimality Theory: An overview. Cambridge, MA & Oxford: Blackwell. Blevins, Juliette & Sheldon P. Harrison. 1999. Trimoraic feet in Gilbertese. Oceanic Linguistics 38. 203–230. Buckley, Eugene. 2009. Locality in metrical typology. Phonology 26. 389–435. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Crowhurst, Megan J. 1992. Minimality and foot structure in metrical phonology and prosodic morphology. Ph.D. dissertation, University of Arizona. Das, Shyamal. 2001. Some aspects of the phonology of Tripura Bangla and Tripura Bangla English. Ph.D. dissertation, Central Institute of English and Foreign Languages, Hyderabad (ROA-493). Dresher, B. Elan & Aditi Lahiri. 1991. The Germanic foot: Metrical coherence in Old English. Linguistic Inquiry 22. 251–286. Elenbaas, Nine. 1999. A unified account of binary and ternary stress: Considerations from Sentani and Finnish. Ph.D. dissertation, Utrecht University. Elenbaas, Nine & René Kager. 1999. Ternary rhythm and the lapse constraint. Phonology 16. 273–329. Everett, Daniel L. 1988. On metrical constituent structure in Pirahã phonology. Natural Language and Linguistic Theory 6. 207–246. Gordon, Matthew. 2002. A factorial typology of quantity-insensitive stress. Natural Language and Linguistic Theory 20. 491–552. Green, Thomas. 1995. The stress window in Pirahã: A reanalysis of rhythm in Optimality Theory. Unpublished ms., MIT (ROA-45). Green, Thomas & Michael Kenstowicz. 1995. The lapse constraint. Proceedings of the 6th Annual Meeting of the Formal Linguistics Society of the Midwest. 1–14 (ROA-101). Hale, Mark & Charles Reiss. 2008. The phonological enterprise. Oxford: Oxford University Press. Halle, Morris. 1990. Respecting metrical structure. Natural Language and Linguistic Theory 8. 149–176. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: MIT Press. Hammond, Michael. 1990. Deriving ternarity. Unpublished ms., University of Arizona, Tucson. Hammond, Michael. 1995. Metrical phonology. Annual Review of Anthropology 24. 313– 342. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Published 1985, New York: Garland. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press.

Ternary Rhythm

16

Hewitt, Mark. 1992. Vertical maximization and metrical theory. Ph.D. dissertation, Brandeis University. Houghton, Paula. 2006. Ternary stress. Unpublished ms., University of Massachusetts, Amherst (ROA-836). Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Hulst, Harry van der. 1999. Word accent. In Harry van der Hulst (ed.) Word prosodic systems in the languages of Europe, 3–115. Berlin & New York: Mouton de Gruyter. Hyde, Brett. 2001. Metrical and prosodic structure in Optimality Theory. Ph.D. dissertation, Rutgers University (ROA 476). Hyde, Brett. 2002. A restrictive theory of metrical stress. Phonology 19. 313–359. Hyman, Larry M. 2009. How (not) to do phonological typology: The case of pitch-accent. Language Sciences 31. 213–238. Idsardi, William J. 1992. The computation of prosody. Ph.D. dissertation, MIT. Isac, Daniela & Charles Reiss. 2008. I-language: An introduction to linguistics as cognitive science. Oxford: Oxford University Press. Ishii, Toru. 1996. An optimality theoretic approach to ternary stress systems. UCI Working Papers in Linguistics 2. 95–111. Kager, René. 1993. Alternatives to the iambic-trochaic law. Natural Language and Linguistic Theory 11. 381–432. Kager, René. 1994. Ternary rhythm in alignment theory. Unpublished ms., Utrecht University (ROA-35). Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Karttunen, Lauri. 2006. The insufficiency of paper-and-pencil linguistics: The case of Finnish prosody. Unpublished ms., Stanford University (ROA-818). Kenstowicz, Michael. 1993. Evidence for metrical constituency. In Kenneth Hale & Samuel J. Keyser (eds.) The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger, 257–273. Cambridge, MA: MIT Press. Key, Harold H. 1961. Phonotactics of Cayuvava. International Journal of American Linguistics 27. 143–150. Key, Harold H. 1967. Morphology of Cayuvava. The Hague: Mouton. Krauss, Michael (ed.) 1985. Yupik Eskimo prosodic systems: Descriptive and comparative studies. Fairbanks: Alaska Native Language Center. Leer, Jeff. 1985a. Prosody in Alutiiq (the Koniag and Chugach dialects of Alaskan Yupik). In Krauss (1985), 77–133. Leer, Jeff. 1985b. Toward a metrical interpretation of Yupik prosody. In Krauss (1985), 159–172. Levin, Juliette. 1985. Evidence for ternary feet and implications for a metrical theory of stress rules. Unpublished ms., University of Texas, Austin. Levin, Juliette. 1988. Generating ternary feet. Texas Linguistic Forum 29. 97–113. Liberman, Mark. 1975. The intonational system of English. Ph.D. dissertation, MIT. Liberman, Mark & Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249–336. McCarthy, John J. 1979. Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT. McCarthy, John J. & Alan Prince. 1993. Prosodic morphology I: Constraint interaction and satisfaction. Unpublished ms., University of Massachusetts, Amherst & Rutgers University. McCartney, Steven J. 2003. Ternarity through binarity. Ph.D. dissertation, University of Texas, Austin. Nespor, Marina & Irene Vogel. 2008. Prosodic phonology. Berlin & New York: Mouton de Gruyter. 1st edn. 1986, Dordrecht: Foris. Newmeyer, Frederick J. 2005. Possible and probable languages: A generative perspective on linguistic typology. Oxford: Oxford University Press. Odden, David. 2008. Ordering. In Vaux & Nevins (2008), 61–120.

17

Curt Rice

Odden, David. Forthcoming. Rules v. constraints. In John A. Goldsmith, Jason Riggle & Alan C. L. Yu (eds.) The handbook of phonological theory. 2nd edn. Malden, MA & Oxford: Wiley-Blackwell. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19–100. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Reiss, Charles. 2008. Constraining the learning path without constraints, or the OCP and NoBanana. In Vaux & Nevins (2008), 252–301. Rice, Curt. 1988. Stress assignment in the Chugach dialect of Alutiiq. Papers from the Annual Regional Meeting, Chicago Linguistic Society 24. 304–315. Rice, Curt. 1990. Pacific Yupik: Implications for metrical theory. Coyote Papers 3. Tucson: Department of Linguistics, University of Arizona. Rice, Curt. 1992. Binarity and ternarity in metrical theory: Parametric extensions. Ph.D. dissertation, University of Texas, Austin. Rice, Curt. 1993. A note on ternary stress in Sentani. University of Trondheim Working Papers in Linguistics 17. 67–71. Rice, Curt. 2007. The roles of Gen and Con in modeling ternary rhythm. In Sylvia Blaho, Patrik Bye & Martin Krämer (eds.) Freedom of analysis?, 233–255. Berlin & New York: Mouton de Gruyter. Rifkin, Jay I. 2003. Ternarity is prosodic word binarity. In Jeroen van de Weijer, Vincent J. van Heuven & Harry van der Hulst (eds.) The phonological spectrum, vol. 2: Suprasegmental structure, 127–150. Amsterdam & Philadelphia: John Benjamins. Roca, Iggy (ed.) 1997. Derivations and constraints in phonology. Oxford: Clarendon Press. Rowicka, GraTyna. 1996. 2+2=3: Stress in Munster Irish. In Henryk Kardela & Bogdan Szymanek (eds.) A Festschrift for Edmund Gussmann, 217–238. Lublin: University Press of the Catholic University of Lublin. Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Vaux, Bert & Andrew Nevins (eds.) 2008. Rules, constraints, and phonological phenomena. Oxford: Oxford University Press. Vijver, Ruben van de. 1998. The iambic issue: Iambs as a result of constraint interaction. Ph.D. dissertation, University of Leiden.

53 Syllable Contact Misun Seo

1

Introduction

Syllable contact is a notion introduced to describe the sonority relation between adjacent segments across a syllable boundary, that is, between heterosyllabic coda and onset segments. According to Hooper (1976), Murray and Vennemann (1983), and Vennemann (1988), among others, there is a cross-linguistic preference to avoid rising sonority across a syllable boundary; this tendency is formulated as the Syllable Contact Law (SCL). This law states, for example, that al.ta, with falling sonority, is preferred to at.la, with rising sonority. This law has been adduced to account for both diachronic and synchronic sound alternations in coda–onset clusters. In this chapter, I will review the notion of syllable contact by referring to sound alternations involving coda–onset clusters observed in different languages. The chapter is organized as follows. In §2, I offer an overview of sonority and sonority scales on which the SCL is crucially based. In addition, I review previous proposals on syllable contact and different types of diachronic and synchronic phonological changes analyzed as repair strategies for bad syllable contact. In §3, I present the issues and debates on syllable contact such as the necessity of syllable contact, the categorical vs. gradient nature of syllable contact, language-specific variation, and problems of syllable contact (see also chapter 49: sonority).

2

Syllable contact

The notion of syllable contact is essentially based on the relative sonority of segments. Thus, in §2.1, I first review sonority and then the SCL. In §2.2, I provide examples of different types of diachronic and synchronic phonological changes which have been analyzed using the notion of syllable contact.

2.1

Sonority and the Syllable Contact Law

The SCL is based on the concept that speech sounds can be classified into different categories according to their relative sonority. There have been many proposals for the definition of sonority. For example, Ladefoged (1982) defines sonority as The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0053

2

Misun Seo

the loudness of a sound relative to that of other sounds with the same length, stress, and pitch, and Clements (1990) defines it in terms of a set of major class features. In Venneman (1988), sonority is described as an inverse restatement of strength, which is based on “degree of deviation from unimpeded (voiced) air flow.” In addition, Parker (2002, 2008) claims that sonority has a phonetic basis in intensity measured as “sound level differences in decibels between a target segment and a constant reference segment in the environment.” There exist many competing sonority scales, obtained on the basis of different definitions of the sonority. Different sonority scales result from the controversial relative sonority of laterals and rhotics, voiced and voiceless obstruents, stops, fricatives, and affricates, and high, mid, and low vowels. However, it is generally agreed among most researchers that the uncontroversial sonority hierarchy is as follows (Bell and Hooper 1978; Harris 1983; van der Hulst 1984; Clements 1987, 1990; Kenstowicz 1994; Smolensky 1995; Holt 1997; van Oostendorp 1999): (1)

Sonority hierarchy more sonorous vowels glides liquids nasals obstruents less sonorous

For extensive discussion of sonority and sonority scales, see chapter 49: sonority. With reference to the relative sonority of segments, Hooper (1976) proposes a constraint on heterosyllabic coda–onset clusters. According to the principle, the sonority of a coda consonant must exceed that of a following onset consonant. Murray and Vennemann (1983) and Vennemann (1988) extend the principle as follows: (2)

Syllable Contact Law (Vennemann 1988: 40) A syllable contact A.B is the more preferred, the less the consonantal strength of the offset A and the greater the consonantal strength of the onset B.

The SCL can be rephrased as in (3), using the concept of sonority, which is the reverse of strength and more commonly used in current phonology. (3)

Syllable Contact Law (sonority version) (Davis and Shin 1999: 286) A syllable contact A.B is the more preferred, the greater the sonority of the offset A and the less the sonority of the onset B.

The SCL has also been invoked as a family of related OT constraints (Bat-El 1996; Davis 1998; Ham 1998; Davis and Shin 1999; Rose 2000; Baertsch 2002; Gouskova 2004; Holt 2004; Zec 2007).

Syllable Contact

2.2

3

Phonological change as optimization of syllable contact

2.2.1 Diachronic change The notion of syllable contact has been employed in motivating various diachronic changes attested in coda–onset sequences (see also chapter 93: sound change). For example, Hooper (1976) claims that various phonological processes have applied to the /nr/ sequence diachronically in different Spanish dialects to remedy the sequence violating the SCL, as illustrated below: (4)

venirá > venrá > vendrá verná verrá

(by stop insertion) (by metathesis) (by assimilation)

‘(it) will come’

In the example above, after the occurrence of syncope, the resulting form venrá is unacceptable since the onset /r/ in the /nr/ sequence is more sonorous than the coda /n/, thus violating the SCL. Therefore, phonological processes such as stop insertion, metathesis, and assimilation apply in different Spanish dialects to improve the less preferred heterosyllabic /nr/ sequence. Stop insertion also applied to various heterosyllabic consonant–liquid clusters in Old Spanish and Old French, as shown below. (5)

a.

Stop insertion in Old Spanish (Martínez-Gil 2003) Latin fem(i)na hum(e)ru trem(u)lare ingen(e)rare mel(io)rare

b.

Old Spanish (f)embra hombro tremblar engendrar me(l)drar

‘female’ ‘shoulder’ ‘to shake, shiver’ ‘to engender, beget’ ‘to grow’

Stop insertion in Old French (Walker 1978; Morin 1980; Wetzels 1985; Picard 1987; Martínez-Gil 2003) Latin cam(e)ra sim(u)lare ten(e)ru mol(e)re laz(a)ru ess(e)re spin(u)la

(> (> (> (> (> (> (>

chamre) semler) tenre) molre) lazre) esre) espinle)

> > > > > > >

Old French chambre sembler tendre molder la(z)dre estre espingle

‘room’ ‘to resemble’ ‘tender’ ‘to grind’ ‘beggar’ ‘to be’ ‘pin’

According to Martínez-Gil (2003), all the heterosyllabic consonant–liquid clusters in the examples above violate the SCL by showing more sonorous onsets than codas, and thus the stop is inserted within the clusters as a strategy to improve bad syllable contact. Holt (2004) analyzes the metathesis observed in the heterosyllabic /dn/ and /dl/ sequences of Old Spanish as a repair strategy to optimize syllable contact. In Old Spanish, the heterosyllabic /dn/ and /dl/ sequences brought about by

Misun Seo

4

syncope in Late Spoken Latin or by morpheme concatenation underwent many repair strategies, such as dissimilation, palatalization, stop insertion, deletion, and strengthening, e.g. antenatu (Latin) > adnado ~ andado ~ andrado ~ alnado ~ anado ~ annado (Old Spanish) ‘stepchild’. The examples in (6) illustrate that metathesis is optional in Old Spanish forms, while only metathesized forms occur in Modern Spanish. (6)

a.

b.

/dn/ Latin cat(e)natu antenatu legitimu retina

Old Spanish cadnado ~ candado adnado ~ andado lidmo ~ lindo riedna ~ rienda

Modern Spanish candado andado lindo rienda

‘padlock’ ‘stepchild’ ‘pretty’ ‘rein’

/dl/ Latin spatula capitulu foliatile titulo

Old Spanish espadla ~ espalda cabidlo ~ cabildo hojalde ~ hojaldre tidle ~ tilde

Modern Spanish espalda cabildo hojaldre tilde

‘back’ ‘town council’ ‘puff pastry’ ‘written accent’

According to Holt, metathesis applied to /dn/ and /dl/ in Old Spanish, since the sequences have rising sonority over a syllable boundary and thus show bad syllable contact. (See also chapter 59: metathesis for a discussion of synchronic metathesis processes in response to phonotactic requirements more generally.) In addition, West Germanic word-medial gemination after a short vowel and before *j, *r, and *l has been analyzed by Murray and Vennemann (1983) as being driven by the need to avoid an onset which is more sonorous than the coda. (7) Gemination in West Germanic (Murray and Vennemann 1983; Braune and Eggers 1987; Ham 1998)1 a.

b.

East Germanic Go. skapjan Go. bidjan Go. hafjan ON framja Go. halja ON bitr Go. akrs VL facla ON epli

West Germanic OS skeppian, OE scieppan OS biddian, OE biddan OHG heffan, OE hebban OHG fremmen, OE fremman OHG hella, OS hellia OHG bittar (rut ‘fruit’ pHfaizHr ‘Pfizer’ Bengali gelaœ ‘glass’ Central Pahari silet ‘slate’ Sinhalese tijage Sanskrit tjage ‘gift’ Wolof kalas ‘class’ Uyghur kulub Russian klub ‘club’

As shown in (12a), a vowel is inserted before the onset consonants with falling or flat sonority, since the resulting coda–onset clusters do not violate the SCL. However, as illustrated in (12b), a vowel is inserted between the two consonants of the onset with rising sonority to avoid the coda–onset clusters violating the SCL. According to Rose (2000), syllable contact plays a role in determining the position of an epenthetic vowel in Chaha (see chapter 67: vowel epenthesis). For example, when [q] is epenthesized in /VCCCV/ for structural reasons, either [VCCqCV] or [VCqCCV] can surface with the aim of achieving good syllable contact. Thus, /t-n-k’rHt’m-nH/ ‘while we are cutting’ surfaces as [tqnk’qrHt’qmnH], since the alternative form *[tqnqk’rHt’qmnH] has the coda–onset cluster [k’.r], with bad syllable contact. On the other hand, /t-n-msHkr-nH/ ‘while we are testifying’ is realized as [tqnqmsHkqnnH] rather than *[tqnmqsHkqnnH], with a coda–onset cluster with flat sonority, which is avoided in the language. In addition, epenthesis in Picard clitics (Auger 2003) is said to be influenced by the preference for achieving good syllable contact, and Pons (2005) argues that regressive manner assimilation, rhotacism, gliding, onset strengthening, epenthesis, and deletion attested in Romance languages are strategies triggered to avoid bad syllable contact.

3 3.1

Issues and debates on syllable contact Sonority Dispersion Principle and the Syllable Contact Law

As seen in §2.2, the notion of syllable contact has been employed in motivating different types of diachronic and synchronic sound changes. However, there have been debates regarding whether or not it is necessary to posit a separate law just for coda–onset sequences. Clements (1990) claims that the Sonority Dispersion Principle makes the SCL dispensable. According to the Sonority Dispersion Principle, sonority rise is required to be maximal from the onset to the nucleus and sonority drop is required to be minimal from the nucleus to the coda. Thus, for example, the principle says that [ta] is more preferred than [la] as the onset and [al] is more preferred than [at] as the coda. From the Sonority Dispersion Principle, it can be predicted that maximal sonority drop is preferred across syllable boundaries, as stated in the SCL. Thus, Clements proposes that the SCL follows from the more general Sonority Dispersion Principle. However, based on the data from Kazakh in (13), Davis (1998) and Gouskova (2004) argue that the SCL cannot be reduced to the Sonority Dispersion Principle.

Misun Seo

8 (13)

Syllable contact in Kazakh (Davis 1998) a.

b.

No onset desonorization /alma-lar/ [al.ma.lar] /mandaj-lar/ [man.daj.lar] /kijar-lar/ [kijar.lar] /kol-ma/ [kol.ma] /kijar-ma/ [ki.jar.ma]

‘apples’ ‘foreheads’ ‘cucumbers’ ‘hand-interrog’ ‘cucumber-interrog’

Onset desonorization /kol-lar/ [kol.dar] /murin-lar/ [mu.rin.dar] /koIQz-lar/ [ko.IQz.dar] /murin-ma/ [mu.rin.ba] /koIQz-ma/ [ko.IQz.ba]

‘hands’ ‘noses’ ‘bugs’ ‘nose-interrog’ ‘bug-interrog’

In Kazakh, consonants of any sonority can be onsets if they are preceded by vowels or by consonants of higher sonority. Thus, no onset desonorization is attested in the examples of (13a). However, when the onset is preceded by a consonant with lower or the same sonority, it desonorizes, as shown in (13b). Therefore, to explain the sound alternations of onsets in the Kazakh examples, it is essential to refer to the sonority relation of the coda and the following onset. For this reason, both Davis and Gouskova claim that the Sonority Dispersion Principle cannot completely replace the SCL.

3.2

Nature of syllable contact: Categorical vs. gradient

Since Hooper (1976), the SCL has been extended to account for more crosslinguistic data, and the nature of syllable contact is now viewed as gradient rather than categorical (see chapter 89: gradience and categoricality in phonological theory). Hooper (1976) originally proposed the SCL for Spanish, where the sonority of a syllable-final consonant must exceed that of a following syllable-initial consonant. According to this, the syllable contact is categorical in nature. Different types of coda–onset clusters are equally fine as long as the sonority of the coda exceeds that of the onset. Thus, although the sonority distance between the coda and the onset is greater in al.ta than in al.na, both are equally fine clusters with respect to syllable contact since the coda is more sonorous than the onset. Likewise, both at.la and an.la are equally non-optimal since the coda is less sonorous than the onset. Murray and Vennemann (1983) and Vennemann (1988) propose an extended version of the SCL whose nature is gradient: two adjacent heterosyllabic segments are more preferred, the greater the sonority of the first segment and the less the sonority of the second segment. Clements (1990) paraphrases the extended SCL as follows: (14)

The extended Syllable Contact Law (Clements 1990: 319) The preference for a syllabic structure A.B, where A and B are segments and a and b are the sonority values of A and B respectively, increases with the value of a minus b.

Syllable Contact

9

Davis and Shin’s (1999) version of the SCL in (3) is also a gradient one. In the gradient view of the syllable contact phenomenon, coda–onset clusters with falling sonority are not considered to exhibit the same degree of optimality. In addition, not all violations of the SCL by clusters with rising sonority are considered equally serious. Depending on the relative sonority distance between the coda and the onset, one cluster is more optimal or less non-optimal than the others. For example, al.ta is more optimal than al.na, since the sonority distance between the coda and onset is greater in al.ta than in al.na. In addition, a sequence such as at.la constitutes a more severe violation of the law than a sequence such as an.la. In line with this, Clements (1990) provides the following aggregate complexity score of each of the coda–onset sequences by summing the complexity values of each of the demisyllables (nucleus–coda and onset–nucleus sequences) that constitute it. (15)

C2 C1 obstruent nasal liquid glide

obstruent

nasal

liquid

glide

5 4 3 2

6 5 4 3

7 6 5 4

8 7 6 5

According to Clements (1990), obstruent–glide sequences such as at.wa violate the SCL most seriously and glide–obstruent sequences such as aw.ta satisfy the law most faithfully. A more fine-grained syllable contact scale is proposed by Gouskova (2004), based on Jespersen’s (1904) sonority scale. (16)

a.

Sonority scale (Jespersen 1904) glides > rhotics > laterals > nasals > voiced fricatives > voiced stops > voiceless fricatives > voiceless stops (abbreviated as: w > r > l > n > z > d > s > t)

b.

Syllable contact scale (Gouskova 2004) more harmonic 1 2 3 4 5 6 7 8 w.t w.s w.d w.z w.n w.l w.r w.w r.t r.s r.d r.z r.n r.l r.r l.t l.s l.d l.z l.n l.l n.t n.s n.d n.z n.n z.t z.s z.d z.z d.t d.s d.d s.t s.s t.t −7 −6 −5 −4 −3 −2 −1 0 drop flat

9 r.w l.r n.l z.n d.z s.d t.s

10 l.w n.r z.l d.n s.z t.d

+1 +2

11 n.w z.r d.l s.n t.z

+3

less harmonic 12 13 14 15 z.w d.w s.w t.w d.r s.r t.r s.l t.l t.n

+4

+5 +6 +7 rise

In Gouskova’s syllable contact scale in (16b), glide + voiceless stop clusters (represented as w.t), for example, are characterized as sequences with a sonority drop of −7, since the sonority of the voiceless stop is lower than that of the glide by

10

Misun Seo

seven steps according to Jespersen’s sonority scale in (16a). As can be seen in (16b), Gouskova proposes that the syllable contact scale has 15 strata, with glide + voiceless stop sequences (represented as w.t) on stratum 1 the most harmonic, and voiceless stop + glide sequences (represented as t.w) on stratum 15 the least harmonic. Clements and Gouskova employ different sonority hierarchies, and thus propose different gradient versions of the sonority contact scale. Gouskova’s more fine-grained syllable contact scale can account for the different patterning of /ll/ and /rl/ in Kazakh, while this is not possible within Clements’ syllable contact scale. According to Davis (1998), coda–onset sequences with flat or rising sonority undergo desonorization in Kazakh in order to improve syllable contact, as in /murin-lar/ → [murindar] ‘noses’. On the other hand, no desonorization applies to coda–onset sequences with falling sonority, as can be seen from /mandaj-ga/ → [mandajga] ‘forehead + direct’. In the case of /ll/ with flat sonority, desonorization applies, as in /kol-lar/ → [kol.dar] ‘hands’, while /rl/ surfaces as [rl], as in /kijar-lar/ → [kijarlar] ‘cucumbers’. The different phonological patterning of /ll/ and /rl/ in Kazakh suggests that /r/ is more sonorous than /l/, and that /rl/ is the sequence with falling sonority, as in Gouskova’s syllable contact scale. Note that Clements’ syllable contact scale cannot account for the different patterning of /ll/ and /rl/ in Kazakh, since /r/ is assumed to be as sonorous as /l/. The gradient nature of syllable contact makes it possible to explain languagespecific patterns of the syllable contact phenomenon. According to Gouskova (2004), languages differ in the level of complexity they tolerate, and thus language-specific patterns of syllable contact are attested. For example, Kirghiz and Kazakh require sonority to drop from the coda to the onset but the two languages are different in the thresholds of acceptable sonority drop. In Kazakh, sonority is merely required to drop from the coda to the onset and need not be maximal. Thus, as can be seen in (13b), onset desonorization applies to coda–onset sequences with flat or rising sonority, but not to coda–onset sequences with falling sonority, as in (13a). On the other hand, as can be seen from the examples in (17), sonority drop alone is not sufficient in Kirghiz, which requires maximal sonority drop. (17) Syllable contact in Kirghiz (Hebert and Poppe 1964; Kasymova et al. 1991; Gouskova 2004) /konok-lar/ /taœ-lar/ /konok-nu/ /taœ-nu/ /atan-lar/ /rol-lar/ /atan-nu/ /kar-lar/ /rol-nu/ /aj-lar/ /aj-nu/ /kar-du/ /too-lar/ /too-nu/

+5 +4 +4 +3 +1 0 0 −1 −1 −2 −3 −4 — —

konok.tar taœ.tar konok.tu taœ.tQ atan.dar rol.dar atan.dQ kar.dar rol.du aj.dar aj.dQ kar.dQ too.lar too.nu

0 −1 0 −1 −2 −3 −2 −4 −3 −5 −5 −4 — —

‘guest (pl)’ ‘stone (pl)’ ‘guest (obj)’ ‘stone (obj)’ ‘gelded camel (pl)’ ‘role (pl)’ ‘gelded camel (obj)’ ‘snow (pl)’ ‘role (obj)’ ‘moon (pl)’ ‘moon (obj)’ ‘snow (obj)’ ‘mountain (pl)’ ‘mountain (obj)’

Syllable Contact

11

As can be seen from the examples above, desonorization in Kirghiz applies to suffix-initial sonorants after any consonant, in order for sonority to drop maximally from the coda to the onset. Note that the Kazakh and Kirghiz data above cannot be explained properly within the categorical version of the SCL, which predicts that all coda–onset clusters with falling sonority are equally acceptable.

3.3

Syllable contact as a language-specific constraint on minimal sonority distance

With respect to syllable contact phenomena, language-specific variations can be observed. First, some languages respecting the SCL allow coda–onset clusters with flat sonority while others disallow them. For example, as shown in (8), Korean prohibits coda–onset clusters with rising sonority. However, a coda can be as sonorous as a following onset, as can be seen from /sallim/ → [sallim] ‘housekeeping’, /simmun/ → [simmun] ‘interrogation’, etc. On the other hand, as shown in (13), coda–onset clusters with flat sonority as well as rising sonority are not sanctioned in Kazakh. Thus, /kol-lar/ is realized as [kol.dar] ‘hands’, to improve syllable contact. Such syllable contact variation of coda–onset sequences with equal sonority has been accounted for by positing two versions of the SCL. (18)

a.

Syllable Contact Law (strict) (Rose 2000: 401) The first segment of the onset of a syllable must be lower in sonority than the last segment in the immediately preceding syllable.

b.

Syllable Contact Law (loose) (Bat-El 1996; Davis and Shin 1999) The first segment of the onset of a syllable must not be of greater sonority than the last segment of the immediately preceding syllable.

Coda–onset sequences of equal sonority are not allowed within the SCL in (18a), while they are permitted within the loose version of the SCL in (18b). Syllable contact variations that cannot be explained by the two versions of the SCL are also found. The SCL has generally been employed to constrain coda– onset clusters with rising sonority. However, languages which allow only falling sonority from a coda to a following onset can vary with respect to minimal sonority distance. According to Gouskova (2004), both Kirghiz and Sidamo allow coda–onset sequences with falling sonority, forbidding sequences with flat or rising sonority. However, the thresholds of acceptable sonority drop are different in the two languages. As can be seen from the examples in (17), minimal sonority drop required in Kirghiz is −4, and coda–onset sequences with sonority drop of less than −4 undergo phonological change of desonorization to remedy bad syllable contact. Thus, desonorization applies to /aj-nu/ ‘moon (obj)’, which is realized as [aj.dQ], with a sonority drop of −3. On the other hand, /kar-du/ ‘snow (obj)’, with a sonority drop of −4, is not targeted by any phonological process in order to remedy bad syllable contact, and surfaces as [kar.dQ]. In the case of Sidamo, unlike Kirghiz, the acceptable minimal sonority drop is −2.

Misun Seo

12 (19)

Syllable contact in Sidamo (Moreno 1940; Gouskova 2004) a.

Sonority rises: Metathesis /duk-nanni/ +4 duI.kanni /huŒ-nanni/ +4 hun.Œanni /has-nemmo/ +3 han.seemo /hab-nemmo/ +2 ham.bemmo

−4 −4 −3 −2

‘they carry’ ‘they pray/beg/request’ ‘we look for’ ‘we forget’

b.

Sonority drops less than −2 or is flat: Gemination /af-tinonni/ −1 affinonni — ‘you (pl) have seen’ /lelliœ-toti/ −1 lelliœœoti — ‘don’t show!’ /ful-nemmo/ −1 fullemmo — ‘we go out’ /um-nommo/ 0 ummommo — ‘we have dug’

c.

Sonority drops more than −2: Place assimilation only /ma7-toti/ −5 ma7.toti −5 ‘don’t go’ /ful-te/ −5 ful.te −5 ‘your having gone out’ /qaram-tino/ −4 qaran.tino −4 ‘she worried’

In Sidamo, coda–onset clusters with rising sonority are repaired by metathesis, as in (19a), and the clusters with flat sonority or sonority drop of less than −2 are modified by gemination, as in (19b). On the other hand, when coda–onset sequences have a sonority drop of −2 or above, no modification applies to improve syllable contact, although the sequences undergo place assimilation, as in (19c). In addition, different languages can permit different degrees of minimal sonority rise, which cannot be explained within the SCL in (18). For example, Faroese and Icelandic set different acceptable sonority distances between a coda and a following onset. According to Gouskova (2004), Faroese permits a sonority rise of +4 or below, as illustrated in (20). (20)

Syllable contact in Faroese (Gouskova 2004) a.

Sonority rise of a(.hkvamar>n vea(.hkr>r ai(.h trant> -ea(.h pr>r mi(.hkl>r e(.h pl>

5 points or more: Complex onsets +7 ‘beryl’ +6 ‘beautiful (masc pl)’ +6 ‘poisonous’ +6 ‘sad’ +5 ‘great (masc pl)’ +5 ‘potato’

b.

Sonority rise of fewer than 5 points: Heterosyllabic coda–onset clusters s>—.r> +4 ‘further south’ h +a t.na +4 ‘to improve’ >-.la +3 ‘or’ ves.na +3 ‘to worsen’ Áar.na +2 ‘gladly’ rDhk.t> 0 ‘smoked (sg)’ ves.tÁr −1 ‘west’ hen.-Ár −2 ‘hands’ Áœr.-> −4 ‘did (sg)’ no-.-> — ‘approached (sg)’

Syllable Contact

13

In Faroese, where initial syllables are always stressed and heavy, syllable contact plays a crucial role in syllabification. As can be seen in (20a), when two intervocalic consonants show a sonority rise of +5 or more, the two consonants are syllabified into a complex onset and the preceding vowel is long. On the other hand, when two intervocalic consonants exhibit a sonority rise of +4 or less, the two consonants are syllabified as heterosyllabic and the preceding vowel is short, as in (20b). Thus, the examples in (20) show that the maximal sonority rise from a coda to a following onset permitted in Faroese is +4 or less. In Icelandic, unlike Faroese, the acceptable sonority distance between a coda and a following onset is +5 or less. (21)

Syllable contact in Icelandic (Gouskova 2004) a.

b.

Sonority rise of 6 v>:.t(h)ja +7 vœ:.k(h)va +7 a:.k(h)rar +6 t h >:.t(h)ra +6

points or more: ‘to visit’ ‘to water’ ‘fields’ ‘to vibrate’

Complex onsets skD:.p(h)ra +6 tv>:.svar +6 e:.sja +6

Sonority rise of fewer than 6 points: Heterosyllabic eh p.l> +5 ‘apple’ t hem.ja +3 h e k.la +5 ‘lack’ vel.ja +2 h †Œ t.la +5 ‘to intend’ ver.ja +1 +>Ï.ra +4 ‘to ask’ t hev.ja 0 stœÏ.va +4 ‘to stop’ hes.tõr −1 h†Œ:.r> +3 ‘right’ ev.r> −1 +laÏ.ra +3 ‘balloon’ av.la:a −2 s>—.la +3 ‘to sail’ -ver.—õr −4 v>s.na +3 ‘to wither’

‘to roll’ ‘twice’ ‘the mountain Esja’ coda–onset clusters ‘to domesticate’ ‘to choose’ ‘to defend’ ‘to delay’ ‘horse’ ‘upper’ ‘to bend out of shape’ ‘dwarf’

As can be seen from the examples in (21a), two consonants with a sonority rise of +6 or more are syllabified as a complex onset and the preceding vowel is lengthened to make the first stressed syllable heavy. On the other hand, when two consonants show a sonority rise of +5 or less, as in (21b), they are syllabified as heterosyllabic, and the preceding vowel is realized as a short vowel since the first stressed syllable is heavy, due to the coda. According to Gouskova (2004), the language-specific patterns of syllable contact discussed above suggest that languages differ in the thresholds of acceptable sonority distance between a coda and a following onset and that the SCL is not a single constraint forbidding rising sonority. To account for such language variations within Optimality Theory, Gouskova reformulates the SCL as a hierarchy of constraints derived from sonority scales, which target negative, flat, or positive sonority distance across a syllable boundary.

3.4

Problems with the Syllable Contact Law

Seo (2003) points out that the SCL has two problems. First of all, the SCL is problematic, since it cannot explain why a consonant + liquid cluster is targeted by a phonological process when a syllable boundary cannot be assumed between two segments in the cluster, for example, in word-initial position. Recall that the SCL

Misun Seo

14

crucially refers to a syllable boundary in explaining phonological change of a consonant + liquid cluster by assuming that the change is motivated to avoid cases where the onset B has higher sonority than the coda A in a heterosyllabic A.B sequence. As illustrated below, in Leti an /nl/ sequence is realized as [ll], whether it occurs word-initially or intervocalically. Note that vowel deletion occurs in the examples given. (22)

Leti (van Engelenhoven 1995; Hume et al. 1997) /na+losir/ /a(na+leti/

→ →

[llosir] [a(lleti]

‘3sg-to follow’ ‘Alety (clan)’ (child + Leti)

In line with the view that the change of a consonant + liquid cluster is motivated due to the SCL, the word-initial geminate [ll] in [llosir] might be considered to be bisyllabic, as represented below (see also chapter 37: geminates): (23)

q

q

[

[

l

o

However, as discussed in Hume et al. (1997), there is a problem with the representation in (23). Leti has a minimal word requirement that lexical words must be minimally bimoraic. With the representation of the geminate in (23), the minimal word requirement is satisfied in words consisting of an initial geminate and a vowel, for example [p.pe]. Thus, it is expected that there will exist words with such a structure in Leti. However, none are attested. On the other hand, if word-initial geminates are assumed to be part of the onset of a single syllable, words containing initial geminates such as [ppu.na] ‘nest’ are bisyllabic, conforming to the dominant tendency that lexical words are made up of two syllables in Leti. Thus, Hume et al. (1997) propose that geminates in Leti have phonological structure with a root node multiply linked to two timing slots, as in (24). (24)

Geminates and long vowels in Leti (Hume et al. 1997)

x

q

q

q

[

[

[

x

x

x

x

x

l o [llo]

a

l [alle]

e

x

In view of the minimal word requirement, it cannot be assumed that an initial geminate is bisyllabic in Leti. Therefore, the modification of the word-initial

Syllable Contact

15

tautosyllabic /nl/ sequence in Leti cannot be motivated by the SCL since a syllable boundary cannot be referred to. On the other hand, phonological change of the intervocalic /nl/ sequence in Leti could be accounted for by relying on the avoidance of rising sonority over a syllable boundary. Thus, even though both heterosyllabic and tautosyllabic /nl/ sequences show the same pattern in Leti, the syllable contact account cannot provide a unified account of the modification. The SCL is also argued to be problematic in Seo (2003), since it cannot provide a unified account of the same types of phonological changes found in nasal + liquid and liquid + nasal sequences in Korean. (25)

Modifications of nasal/liquid sequences in Korean (Davis and Shin 1999) /nl/ /ln/ cf. /lm/

→ → →

[ll] [ll] [lm]

/non-li/ /sHl+nal/ /pul+mjHn/

→ → →

[nolli] [sHllal] [pulmjHn]

‘logic’ ‘New Year’s Day’ ‘insomnia’

As illustrated in (25), in Korean /nl/ surfaces as [ll]. According to the SCL, this change is expected since the sequence violates the law. However, the same change applies to /ln/, while /lm/ (with the same sonority distance as /ln/) does not undergo any change. Thus it must be assumed within the syllable contact account that the modification of a nasal + liquid sequence and that of a liquid + nasal sequence result from different factors, although the same types of phonological changes occur in both types of sequence. Korean is not the only language that has the same types of alternations in nasal/liquid sequences. As shown in (26), the same type of phonological change applies to both /nl/ and /ln/ sequences in Leti, Toba Batak, and Boraana Oromo. (26)

Modifications of nasal/liquid sequences a.

Leti (van Engelenhoven 1995; Hume et al. 1997) /nl/ → [ll] /a:na+leti/ → [a(lleti] ‘Alety (clan)’ /ln/ → [ll] /vulan/ → [vulla] ‘moon’

b.

Toba Batak (Nababan 1981) /nl/ → [ll] /laDn+laDn/ /ln/ → [ll] /bal+na/

c.

→ [laDllaDn] ‘eventually’ → [balla] ‘his ball’

Boraana Oromo (Stroomer 1995) /nl/ → [ll] /hin+lool+a/ → [hilloola] ‘I/he will fight’ /ln/ → [ll] /kofl+na/ → [kofalla] ‘we smile’

Within the syllable contact account, the modification of /ln/ would have to be motivated by a factor other than syllable contact, although both /ln/ and /nl/ undergo the same type of phonological change. Seo (2003) proposes that the phonological processes found in phonological modifications of heterosyllabic coda–onset sequences could be viewed as resulting from a segment contact phenomenon which is closely related to speech perception, rather than from the SCL (see chapter 98: speech perception and phonology). According to Seo, contrasts of weak perceptibility triggered by phonetic similarity between two members of a cluster are a key factor in motivating the alternations in the cluster. Thus it is expected that phonological modifications will apply when

16

Misun Seo

two consonants are perceptually similar to each other, and when they occur in a sequence, regardless of the presence or absence of a syllable boundary and the order of the two consonants.

4

Conclusion

In this chapter, I have reviewed four major issues concerning the SCL. First, different opinions regarding the necessity of the SCL were discussed – while Clements (1990) claims that the SCL is not required with the Sonority Dispersion Principle, Davis (1998) and Gouskova (2004) argue that the SCL cannot be reduced to the Sonority Dispersion Principle based on the Kazakh data. Second, the change in viewpoint on the nature of syllable contact was reviewed, and it was shown that a gradient of syllable contact makes it possible to explain languagespecific syllable contact patterns while a categorical view does not. Third, after providing examples illustrating cross-linguistic variations of syllable contact, I introduced Gouskova’s (2004) proposal of the SCL as a language-specific constraint on minimal sonority distance. Finally, based on Seo (2003), after pointing out problems associated with the SCL, I argued that phonological modifications of heterosyllabic coda–onset sequences could be viewed as a segment contact phenomenon, which is closely related to speech perception.

REFERENCES Auger, Julie. 2003. Pronominal clitics in Picard revisited. In Núñez-Cedeño et al. (2003), 3–20. Baertsch, Karen. 2002. An optimality theoretic approach to syllable structure: The split margin hierarchy. Ph.D. dissertation, Indiana University. Baruch, Kalmi. 1930. El judeo-español de Bosnia. Revista de Filología Española 17. 113–154. Bat-El, Outi. 1996. Selecting the best of the worst: The grammar of Hebrew blends. Phonology 13. 283–328. Bell, Alan & Joan B. Hooper. 1978. Issues and evidence in syllabic phonology. In Alan Bell & Joan B. Hooper (eds.) Syllable and segments, 3–22. Amsterdam: North-Holland. Bradley, Travis G. 2007. Constraints on the metathesis of sonorant consonants in JudeoSpanish. Probus 19. 171–207. Braune, Wilhelm & Hans Eggers. 1987. Althochdeutsche Grammatik. 14th edn. Tübingen: Niemeyer. Clements, G. N. 1987. Phonological feature representation and the description of intrusive stops. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23(2). 29–50. Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Davis, Stuart. 1998. Syllable contact in Optimality Theory. Korean Journal of Linguistics 23. 181–211. Davis, Stuart & Seung-Hoon Shin. 1999. The syllable contact constraint in Korean: An optimality-theoretic analysis. Journal of East Asian Linguistics 8. 285 –312. Engelenhoven, Aone van. 1995. A description of the Leti language as spoken in Tutukei. Ph.D. dissertation, University of Leiden. Gouskova, Maria. 2001. Falling sonority onsets, loanwords and syllable contact. Papers from the Annual Regional Meeting, Chicago Linguistic Society 37. 175 –186.

Syllable Contact

17

Gouskova, Maria. 2004. Relational hierarchies in Optimality Theory: The case of syllable contact. Phonology 21. 201–250. Hall, T. A. 2007. German glide formation and its theoretical consequences. The Linguistic Review 24. 1–31. Ham, William. 1998. A new approach to an old problem: Gemination and constraint reranking in West Germanic. Journal of Comparative Germanic Linguistics 1. 225 –262. Harris, James W. 1983. Syllable structure and stress in Spanish: A nonlinear analysis. Cambridge, MA: MIT Press. Hebert, Raymond & Nicholas Poppe. 1964. Kirghiz manual. The Hague: Mouton. Holt, D. Eric. 1997. The role of the listener in the historical phonology of Spanish and Portuguese: An optimality-theoretic account. Ph.D. dissertation, Georgetown University (ROA-278). Holt, D. Eric. 2004. Optimization of syllable contact in Old Spanish via the sporadic sound change metathesis. Probus 16. 43–61. Hooper, Joan B. 1976. An introduction to natural generative phonology. New York: Academic Press. Hudson, Grover. 1975. Suppletion in the representation of alternations. Ph.D. dissertation, University of California, Los Angeles. Hudson, Grover. 1995. Phonology of Ethiopian languages. In John A. Goldsmith (ed.) The handbook of phonological theory, 782–797. Cambridge, MA & Oxford: Blackwell. Hulst, Harry van der. 1984. Syllable structure and stress in Dutch. Dordrecht: Foris. Hume, Elizabeth. 1999. The role of perceptibility in consonant/consonant metathesis. Proceedings of the West Coast Conference on Formal Linguistics 17. 293–307. Hume, Elizabeth, Jennifer Muller & Aone van Engelenhoven. 1997. Non-moraic geminates in Leti. Phonology 14. 371–402. Iverson, Gregory K. & Hyang-Sook Sohn. 1994. Liquid representation in Korean. In Young-Key Kim-Renaud (ed.) Theoretical issues in Korean linguistics, 1–19. Stanford: CSLI. Jespersen, Otto. 1904. Lehrbuch der Phonetik. Leipzig & Berlin: Teubner. Kasymova, Bella, Kurmanbek Toktonalijev & Asan Karybajev. 1991. Izucajem Kyrgyzskii jazyk. Frunze: Mektep. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA & Oxford: Blackwell. Ladefoged, Peter. 1982. A course in phonetics. 2nd edn. New York: Harcourt Brace Jovanovich. Lamouche, L. 1907. Quelques mots sur le dialecte espagnol parlé par les Israélites de Salonique. Romanische Forschungen 23. 969 –991. Martínez-Gil, Fernando. 2003. Consonant intrusion in heterosyllabic consonant–liquid clusters in Old Spanish and Old French: An optimality theoretical account. In NúñezCedeño et al. (2003), 39–58. Moreno, Martino Mario. 1940. Manuale di Sidamo. Milan: Mondadori. Morin, Yves-Charles. 1980. Morphologisation de la épenthèse en ancient français. Canadian Journal of Linguistics 32. 365 –375. Murray, Robert W. & Theo Vennemann. 1983. Sound change and syllable structure in Germanic phonology. Language 59. 514 –528. Nababan, P. W. J. 1981. A grammar of Toba Batak. (Pacific Linguistics D37.) Canberra: Australian National University. Núñez-Cedeño, Rafael, Luis López & Richard Cameron (eds.) 2003. A Romance perspective on language knowledge and use. Amsterdam & Philadelphia: John Benjamins. Oostendorp, Marc van. 1999. Syllable structure in Esperanto as an instantiation of universal phonology. Esperantologio/Esperanto Studies 1. 52–80. Parker, Steve. 2002. Quantifying the sonority hierarchy. Ph.D. dissertation, University of Massachusetts, Amherst. Parker, Steve. 2008. Sound level protrusions as physical correlates of sonority. Journal of Phonetics 36. 55 –90.

18

Misun Seo

Picard, Marc. 1987. On the general properties of consonant epenthesis. Canadian Journal of Linguistics 32. 133–142. Pons, Claudia. 2005. It is all downhill from here: The role of syllable contact in romance languages. Paper presented at the 13th Manchester Phonology Meeting (ROA-802). Rice, Keren. 1992. On deriving sonority: A structural account of sonority relationships. Phonology 9. 61–99. Rice, Keren & Peter Avery. 1991. On the relationship between laterality and coronality. In Carole Paradis and Jean-François Prunet (eds.) The special status of coronals: Internal and external evidence, 101–124. San Diego: Academic Press. Rose, Sharon. 2000. Epenthesis positioning and syllable contact in Chaha. Phonology 17. 397– 425. Seo, Misun. 2003. A segment contact account of the patterning of sonorants in consonant clusters. Ph.D. dissertation, Ohio State University. Smolensky, Paul. 1995. On the structure of the constraint component Con of UG. (ROA-86.) Stroomer, Harry. 1995. A grammar of Boraana Oromo (Kenya): Phonology, morphology, vocabularies. Cologne: Rüdiger Köppe Verlag. Vennemann, Theo. 1988. Preference laws for syllable structure and the explanation of sound change: With special reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. Walker, Douglas C. 1978. Epenthesis in Old French. Canadian Journal of Linguistics 23. 66–83. Wetzels, W. Leo. 1985. The historical phonology of intrusive stops: A non-linear description. Canadian Journal of Linguistics 30. 285 –333. Zec, Draga. 2007. The syllable. In Paul de Lacy (ed.) The Cambridge handbook of phonology, 161–194. Cambridge: Cambridge University Press.

54

The Skeleton Péter Szigetvári

Only theories of phonology that attach significance to representations of phonological objects and, in addition, subscribe to an autosegmental version of these representations face the question of what the phonological skeleton looks like. Therefore, this chapter presupposes an autosegmental view of phonological representations. The motivation for the autosegmental model is the fact that the segmentation of the speech signal can never result in absolutely discrete segments. Here segmentation is taken to mean practically the conversion of the continuous speech signal into the alphabetical symbols of the IPA. Some of these symbols pertain to more than one segment: for example, the stress mark to the syllable after it, tones potentially to even longer stretches. Take the question you live by the sea? Its last word, carrying the most prominent stress in the sentence, the tonic, might be transcribed as [ ↑si(]. In this transcription, the tone mark has a scope lasting all through the word (basically its only vowel): the pitch rises on steadily until the end of the utterance. The same holds if the string after the tone mark is longer, for example, as in you live by the | ↑seaside, Martin? It would take a very complicated mechanism to maintain that pitch was a property of individual segments and in some cases this rising pitch was realized on a single vowel, while in others it was split into low, higher, even higher, and highest pitch and added to several other vowels following. Tone is clearly not an immanent property of a vowel; it is an ephemeral phenomenon (from the point of view of a vowel) controlled by syntactic and pragmatic factors. If so, it is useful to represent it separately from the rest of the properties of the sound string. Such autonomous sound properties came to be known as autosegments. If the phonological shape of an utterance is represented as a string of discrete feature bundles, the only option of representing the rising pitch in ↑sea includes a feature [rising tone] (here R) in the set of features corresponding to the vowel, as in (1a). In ↑seaside, Martin, on the other hand, a set of features [low tone], [higher tone], [even higher tone], etc. (here 0H, 1H, etc.) has to be assigned to the vowels following the tonic, as in (1b). (1)

a. b.

si(R si(0H sa>1Hd mA(2H t> 3H n

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0054

2

Péter Szigetvári

One flaw of such representations is obvious: there appears to be nothing in common between the two rising tones, i.e. nothing to indicate their relationship. It is clear that the same tone is spread over the available vowels, but this is not shown in (1b). Not only tone but many other sound properties turn out to be similarly promiscuous, with the potential of simultaneously belonging to several segments, and being manipulable independently of the segment(s) they belong to. (For further discussion, see chapter 45: the representation of tone and chapter 14: autosegments.) The more sound properties extracted from their feature bundles, the fewer there remain. There are two widespread views on how many: according to one – historically the earlier – one feature, [syllabic], remains in the “bundle” (e.g. McCarthy 1979; Halle and Vergnaud 1980; Clements and Keyser 1983); according to the other, no feature remains (e.g. Levin 1985; Lowenstamm and Kaye 1986). The string of segmental positions thus vacated is called the phonological skeleton (a name suggested by Halle and Vergnaud 1980: 83) – or, alternatively, the timing tier or skeletal tier. The former type, in which the skeletal positions hold the feature [syllabic], is the CV skeleton (discussed in §2); the latter one, which is completely empty, is the X skeleton (discussed in §3). A non-segment-based framework involving only syllables and moras is introduced in §4. I will then argue that there is a way of incorporating moras in the old CV skeleton with clear advantages over the moraic framework (§5). To begin with, let us examine the types of relation that may exist between skeletal positions and phonetic features associable to them.

1

Melody–skeleton relations

Skeletal positions represent the presence of a segment, and serve as an anchoring site for the phonetic properties associated with that portion of the speech signal. If the relationship of feature bundles – referred to as melody, again following Halle and Vergnaud (1980), among many others – and skeletal positions were always one to one, the latter would be superfluous. But, as we have already seen in the case of tone, this is not so. Let us take the non-trivial options one by one.

1.1

One-to-many relations

The standard textbook examples for this type of skeleton–melody relationship are affricates and prenasalized plosives (see chapter 29: secondary and double articulation). With respect to the former, there has been much debate among phonologists about the feature [±delayed release] (and the marginal oppositions it creates), which Chomsky and Halle (1968) introduced to distinguish affricates from plosives. An alternative approach, that affricates are bisegmental (discussed by Gimson 1989: 172f. and Roca 1994: 3, among others), as suggested by the IPA symbols used to represent them, is undermined by many facts. In most cases, the distribution of affricates shows that they are not clusters, but single segments. It is even possible that an affricate in a system does not contrast with a homorganic fricative (e.g. Castilian Spanish has [Œ], but not [œ]), rendering a cluster analysis improbable.

The Skeleton

3

The separation of quantity (skeleton) and quality (melody) offers an opportunity for handling the quantitatively simplex, but qualitatively complex, affricates in an intuitive way, as so-called contour segments. (2) depicts the view of the affricate [ts] along these lines. (The skeletal slot is represented as “x”, but this is not meant to indicate a standpoint in the CV vs. X skeleton debate.) (2)

An affricate as a contour segment x t

s

The representation in (2), however, incorporates a misconception, namely, that the melody of segments, without the slot they attach to, forms some kind of unit, two of which are here associated with a single skeletal slot. In reality, the symbols “t” and “s” above have no theoretical status. What exist in an autosegmental framework (or for that matter any other phonological theory since the middle of the twentieth century) are features, many of which occur in both parts of the affricate (e.g. place of articulation, laryngeal properties). Another difficulty with the contour model of affricates lies in the interpretation of autosegmental representations. Any melody linked to a slot of the skeleton – also known as the timing tier – is interpreted simultaneously. Temporal sequencing is managed by the skeleton, i.e. what is linked to an earlier slot is interpreted earlier than what is linked to a later slot. Associating the stop part of the affricate to the left leg of the contour segment and the fricative part to the right is then no more than a graphical trick, which cannot have any realizational consequences. The standard solution to this problem, involving root nodes, is discussed in §4. As Clements (1999) and Lin (chapter 16: affricates) argue, affricates are best thought of as non-contour segments (strident stops), as Jakobson et al. (1952) have proposed. It seems then that we are left without one-to-many relations between the skeleton and melodic material. In fact, such relations are the most common occurrences in representations, since it is not segments but features that are associated with the slots of the skeleton. Thus most segments embody the one-to-many relation, as the partial representations of two very common segments, [d] and [A], show in (3). (3)

Partial autosegmental representations a. [d] x …

[+voiced] [+ coronal] [− continuant] b. [A] x [+back]

… [+low] [−round]

Péter Szigetvári

4

1.2

Many-to-one relations

The long–short contrast of a vowel could be encoded in a feature [long], so that the long vowel is [+long] and the short one [−long]. It is evident, however, that this is not an adequate way of modeling length contrasts. Vowel length (or consonant length for that matter, on which see chapter 37: geminates) is not a property like vowel height (or the voicing of obstruents): it does not harmonize or trigger or undergo assimilation of any type (see chapter 20: the representation of vowel length). Furthermore, changes in segmental length are usually unlike common assimilatory changes. Take, for example, the Rhythmic Law of Slovak, which shortens a suffixal long vowel after a long vowel in the stem. The agentive -ník (the acute accent marks length) inherently contains a long vowel (e.g. rol-ník [rolJi(k] ‘farmer’), which shortens when added to a long-vowelled stem (e.g. stráN-nik [stra(ÚJik] ‘guard’; Kenstowicz and Rubach 1987). The rule could be categorized as a dissimilatory process. What is conspicuously missing in languages is any assimilation of this type: i.e. changes where a short vowel would lengthen in the vicinity of a long vowel, and, crucially, because of that long vowel, or a long vowel would shorten purely because of the shortness of a neighboring vowel. An even more telling phenomenon is compensatory lengthening (see also chapter 64: compensatory lengthening).1 A synchronic comparison of the forms of the 1st singular copula in two varieties of Ancient Greek, Attic [e(mi] and Aeolic [em(i] ‘I am’, suggests a simple shift in the host of the alleged feature [+long]. In light of the reconstructed Proto-Greek etymon *[esmi], however, a different analysis is called for. The loss of the [s] triggers the lengthening of one of the neighboring segments, the preceding vowel in Attic and the following consonant in Aeolic. If length were encoded by a feature, the change could only be described by a pair of rules applying simultaneously, one deleting the coda consonant, the other lengthening the segment next to the deletion site. It is clear that the two rules are interrelated: spontaneously, open syllable lengthening is not attested in Attic, nor is intervocalic gemination in Aeolic – these changes only occur in tandem with the loss of the coda consonant. It is therefore difficult to understand why these two rules so commonly co-occur. If the quantity of segments is stored separately from their quality, this process, and any similar one, has a very neat explanation: it is only the quality (melody) of the coda [s] that is lost (more precisely, only its association with the skeleton); its place, that is, the time it occupied in the string of sounds, is retained (cf. e.g. Ingria 1980; Steriade 1982; Hock 1986; Hayes 1989). It is this empty place that one of the neighboring segments fills in, as shown in (4). (4)

Compensatory lengthening: The stability of the skeleton a.

1

Attic

b. Aeolic

x

x

x

x

x

x

x

x

e

s

m

i

e

s

m

i

Much of the literature limits the term compensatory lengthening to cases involving the lengthening of a vowel. The lengthening of a consonant is called inverse compensatory lengthening by Hayes (1989: 280–281). Here I will refer to both processes by the same name.

The Skeleton

5

While cases like the above could also be analyzed as the total assimilation of the [s] to the preceding vowel or the following consonant, there are more complicated types of compensatory lengthening, for which such an analysis is not at all viable. Cases in point include Middle English tale [talH] > [ta(l] (Minkova 1982), Old Church Slavonic bóg§ > pre-Serbo-Croatian bôg [bóog] ‘god’, bob? > bób [boób] ‘bean’ (Hock 1986: 435), and Old Hungarian [hida] > híd [hi(d] ‘bridge’, [levele] > [leve(l] (Modern Hungarian levél [leve(l] ‘leaf’, with subsequent closing of the second vowel; E. Abaffy 2003: 331). The English case is controversial (see Lahiri and Dresher 1999 for a different analysis), but for the others there is evidence that they are not cases of open syllable lengthening ensued by apocope. In Slavonic the original bisyllabic stress pattern is preserved on the long vowel of the monosyllabic forms. In Hungarian no lengthening takes place before suffixes that retain the stem-final vowel: Modern Hungarian hidam [hidAm] ‘my bridge’, levelek [levelek] ‘leaves’. Lengthening due to a minimal word constraint is also excluded by the last example: the process takes place in monosyllabic and polysyllabic words alike. The bipositional analyis of long vowels is also made likely by the fact that they behave similarly to “vowel clusters,” i.e. diphthongs. In English, for example, neither category occurs before non-coronal consonant clusters, and both occur word-finally, unlike short monophthongs (Fudge 1969: 272f.; Harris 1994: 37; Gussmann 2002: 20–23; see Prince 1984 for the same conclusion in Finnish for both vowels and consonants). Accordingly, there is a general consensus that long vowels ought to be represented as in (5a), and long (i.e. geminate) consonants as in (5c). The representation of diphthongs and other consonant clusters is given in (5b) and (5d), for comparison. (5)

The autosegmental representation of vowel and consonant clusters a. x

x

b. x

x

A

u

A

c. x

x

d. x

x

n

t

t

It is not only complete segments that may be linked to more than one skeletal position. The standard situation in fact is that features (autosegments) are multiply linked. Take, for example, the Hungarian word különbség ['yl«mpœe(g] ‘difference’, depicted in (6). (The features only serve illustrative purposes; their exact identity and location is irrelevant here.) (6)

Multiply linked features in the representation of ['yl«mpœe(g] [rounded] x

x

[mid, unrounded]

[front] x

x

x

x

x

x

x

x

[nasal] [voiceless] [voiceless] [voiced] [velar] [coronal, lateral] [palatal] [velar] [labial] Chaotic as it seems, the diagram in (6) does not contain all the relevant features specifying the segmental content of the string ['yl«mpœe(g] – manner of articulation features, for example, are lacking. Nevertheless, it can clearly be seen that it is more common for a feature to be associated with several skeletal slots than

Péter Szigetvári

6

to be associated exclusively with one. In (6) this is because of voicing, place of articulation assimilations, vowel nasalization, and consonant fronting, as well as vowel harmony. Even when feature sharing is not a result of such phonological processes, i.e. in monomorphemic items, the multiple association of a single feature is dictated by the Obligatory Contour Principle (Leben 1973; McCarthy 1986; Odden 1986).

1.3

One-to-zero and zero-to-one relations

As we have seen, many-to-one and one-to-many relations between the skeleton and melody are very common. Two further options are discussed in this section. It is possible that a skeletal position is not associated with any melodic material. The opposite case may also occur: features unlinked to any point on the skeleton. French liaison exemplifies both of these possibilities (see chapter 112: french liaison). In this phenomenon, a word-final consonant is pronounced when the next word begins with a vowel, but not when it begins with a consonant. (The syntactic conditions on liaison need not concern us here.) Thus in the phrase petit garçon ‘little boy’ the first element ends in a vowel ([pHti garsõ]), in petit enfant ‘little child’ a [t] is pronounced ([pHtit √f√]). According to one analysis (Prunet 1987: 226) petit comes with only four skeletal slots, but five segments; enfant, on the other hand, has an extra skeletal slot – it begins with an initial consonantal slot which is empty. The situation is shown in (7). (7) Liaison [x

x

x

x]

p

H

t

i

[x t

x

x

x]

√

f

√

The [t] at the end of petit is not associated to the skeleton; it is said to be floating. Floating melody fails to be pronounced unless it is able to associate to the skeleton. Vowel-initial words supply an empty skeletal position that the floating melody can associate to. The floating [t] at the end of petit must be lexically determined: there are other liaison consonants besides [t], whose identity is unpredictable (e.g. gros enfant [groz √f√] ‘fat child’, mon enfant [mõn √f√] ‘my child’, gentil enfant [Ú√tij √f√] ‘nice child’, long article [lõg artikl] ‘long article’, etc., where the consonant before the space appears only if the next word begins with a vowel). Therefore this consonant must be included in the lexical representation. It is also not unjustified to suppose that vowel-initial words carry an empty skeletal slot at their left side. It is true for all languages that at least some words (and syllables) begin with a consonant (chapter 55: onsets). For some languages this is not optional, but obligatory; crucially, though, there are no languages where an onset is not possible. One may argue that a syllable-initial consonantal position is in fact obligatory in all languages, the optionality lying in whether this position may or may not be left empty (see e.g. Kaye 1989: 134). Thus consonant-initial words do not carry an empty skeletal slot to their left, while vowel-initial words do, and as a result, the latter can host the floating consonantal melody at the end of the preceding word. Apparently, even languages that allow syllable-initial consonantal positions to be empty prefer them to be filled.

The Skeleton

7

Hypothesizing that there is an empty skeletal position between two vowels in hiatus and that languages make an effort to fill it also explains the common phenomenon of hiatus filling (chapter 61: hiatus resolution). Unless a language manages to get rid of this consonantal position (often together with one of the neighboring vowels), an intervocalic consonantal position is filled by some melody associating to it from one of the vowels. (8) illustrates this with English skier and Hungarian síel [œi(el] ‘s/he skies’. (8)

Hiatus filling a. x

x

x

s

k

i

x

x

x H

b.

x

x

œ

i

x

x

x

x

e

l

The hiatus between [i(] and [H] or [e] is filled by the melody of the first vowel, resulting in the forms [ski(jH] (Gimson 1989: 215, 2001: 213) and [œi(jel] (Siptár and Törkenczy 2000: 283). The possibility of vocalic positions being empty is considerably more controversial; this issue will be taken up in §5.

2

The CV skeleton

The notion of the CV tier was originally developed for the analysis of the nonconcatenative morphology of Classical Arabic by McCarthy (1979, 1981). As in other Semitic languages, a large number of morphological categories are not expressed by linking morphemes after one another, but by fusing individually unpronounceable components into one. A similar, but much less elaborate case is the ablaut found in Germanic languages, e.g. English sing, sang, sung, and song, where the consonants carry the lexical entry and the vowel the grammatical category. (See also chapter 105: tier segregation and chapter 108: semitic templates.) Paradigms in Arabic are classified into groups traditionally called conjugations – or, as McCarthy refers to them, binyans. The prime phonological property of a binyan is the order in which consonants and vowels are arranged. Roots of three (sometimes two or four) consonants contribute a lexical field to the meaning; the vowels are often responsible for grammatical categories like tense and voice. A portion of McCarthy’s (1979: 244) table depicting the forms for the root /ktb/ ‘to write’ is given in (9). (9)

Some forms of /ktb/ binyan I II III ... IX

perf act katab kattab kaatab

perf pass kutib kuttib kuutib

ktabab

—

The CV skeletons of the first three binyans are CVCVC, CVCCVC, and CVVCVC, respectively; that of binyan IX is CCVCVC. The root consonants and the vowels

8

Péter Szigetvári

supplied by the grammatical category are mapped onto this skeleton more or less according to the association conventions elaborated by Goldsmith (1976). Three cases are shown in (10). (10)

a. katab

b.

k

t

b

C

V

C

V

kattab

C

a

c. ktabab

k

t

b

C

V

C

C

V

C

a

k

t

b

C

C

V

C

V

C

a

In (10a), the consonants are linked to the C slots of the skeleton, one by one. It is vital that the consonantal and vocalic skeletal slots be distinguished, since the linking of the root consonants and the vowel(s) can be done as required only thus. The case of (10c) shows that association takes place from left to right: with three consonants to four positions, the last consonant is linked to the surplus position ([ktabab]). (10b) poses a problem in this respect: either association is idiosyncratically edge-in in this case, or some extra mechanism is needed. McCarthy (1979: 256) uses brute force here: he assumes the expected *[katbab] in the first round, with a later rule delinking the first linkage of [b] ([katCab], where C represents the slot from which the melody of the [b] is delinked). This is automatically followed by the spreading of the [t] ([kattab]), much like an instance of compensatory lengthening (see §1.2). A slightly less powerful solution is proposed by Lowenstamm and Kaye (1986: 117–118), who claim that association to the first position is inhibited from the start, thus each consonant of the root occupies its final position in the first round, as shown in (11a). The resulting configuration (empty C followed by filled C) is interpreted as a geminate, as in (11b). I have adapted the original to the previous diagrams of this chapter to aid comparison. We will see below (§3) that Lowenstamm and Kaye use a significantly different scheme. (11)

The mapping of a geminate ([kattab]) a. k

t

b

C

V

C

a

C

V

C

b. k

t

b

C

V

C

C

V

C

a

Note that McCarthy’s second-round-spreading solution cannot apply after Lowenstamm and Kaye’s first-round blocking, since this would yield the unattested form *[kaktab].2 2

Neither analysis gives a reason for delinking or inhibiting the association of the consonant encircled in (11a), so that the unattested form *[katbab] is avoided. Following Hoberman (1988), we may assume that long-distance geminates (those separated by a vowel) are more marked (their inhibition is ranked higher) than local ones and that word-initial geminates are even more marked. This explains why [kattab] is preferred to *[katbab], but [ktabab] to *[kkatab].

The Skeleton

9

In McCarthy’s analysis, the CV skeleton of Arabic words is a morpheme (a prosodic template, in his words), identifying the binyan of the word form, contributing to the semantic elements of the specific binyan (as if the Attic–Aeolic difference between [e(mi] and [em(i] represented a difference in morphological categories). Clements and Keyser (1983: 11) apply the CV skeleton as a universal phonological device, the mediator between the syllable and autosegments, its two types of members, C and V representing “the useful but ill-defined notion of ‘phonological segment’.” The C for them is an anchor for anything [−syllabic] and the V for [+syllabic] segments. Prince (1984) shows that such impoverished representations adequately capture the templates of, for example, verbal person endings in Finnish, which are -C in the singular and -CC[e] in the plural, with the melody [m] in the 1st, and [t] in the 2nd person. The surface forms are thus 1st singular [-n] (by an independently motivated rule turning [m] to [n] word-finally), 2nd singular [-t], 1st plural [-mme] and 2nd plural [-tte].

3

The X skeleton

Simultaneously with the development of theories of the CV skeleton there evolved an alternative view that considered the distinction of C and V slots redundant, and argued that skeletal slots are uniform, usually marked with dots or x’s (e.g. Lowenstamm and Kaye 1986; Levin 1985). Proponents of the X skeleton have put forward a number of arguments against skeletal positions predestined for syllabicity.

3.1

Reduplication in Mokilese

Levin (1985: 35–41) shows some unusual cases of reduplication from Mokilese, which, she believes, are analyzable only with an X skeleton. The point is that the reduplicant is a copy of the first three segments of the first syllable of the stem, irrespective of their being consonants or vowels. Levin argues that the template of the reduplicant must therefore also lack this information. The relevant data are given in (12). (12)

Mokilese reduplication stem a. pDdok b. kasD c. pa d. wia e. ca(k f. onop g. andip

progressive pDdpDdok kaskasD pa(pa wi(wia ca(ca(k onnonop andandip

‘plant’ ‘throw’ ‘weave’ ‘do’ ‘bend’ ‘prepare’ ‘spit’

Levin contends that the reduplicant must be a totally specificationless skeleton, q[xxx], to which the copy of the melody of the stem is associated following universal conventions. The case of (12a), (12b), and (12g) is now straightforward. When the stem is too short, as in [pa], (12c), the last melody is multiply linked. The fact that the reduplicant is a single syllable inhibits the second vowels of [wia] and [onop] from associating to the skeleton (12d) and (12f). As a result, the preceding

10

Péter Szigetvári

vowel or consonant is lengthened again. There is a problem with the stem [ca(k] though, (12e). The melody of the stem comprises three segments, “|c|”, “|a|”, and “|k|”, and the expected reduplicated form is therefore *[cakca(k] rather than the attested [ca(ca(k]. Levin has to stipulate that multiple melodic associations like that of the long [a(] are transferred in reduplication. A further problem of this analysis lies in the interpretation of the reduplicant: it is specified as a syllable, but it is not one in [on.n-onop] or [an.d-andip] (where the hyphen indicates the boundary between the reduplicant and the stem), since a word-internal pre-vocalic consonant forms a syllable with the following vowel, as the universal onset maximalization principle requires. Yet the constraint on the reduplicant being a syllable cannot be relaxed, because if the first three segments were copied without reference to a syllable, undesirable results like *[wi.a-wia] or *[o.no-onop] would emerge. In fact, Moravcsik says that in her survey of reduplication types she has never come across formulations like “reduplicate the first two [or, in our case, three – szp] segments (regardless of whether they are consonants or vowels)” (1978: 307–308). If in a language reduplication copies the first CVC part of the stem for consonant-initial stems, it will copy VC (not VCV) of vowel-initial stems. Actually, a simpler account is available for the data in (12). Theoretically it is no more plausible than Levin’s, but needs fewer stipulations, and thus invalidates her analysis as an argument for the X skeleton. Suppose, as in §1.3 above, that syllable onsets are always represented on the skeleton, either as a filled or as an empty C position. (This immediately explains Moravcsik’s observation.) The reduplicant then is a copy of the first CVC part of the stem, melody and skeleton included. The cases of (12a–c) are obvious. The third slot for (12c) is automatically filled by the vowel of the reduplicant, just as for Levin. The objection (also put forward by Broselow 1995: 184) that vowels cannot spread onto a consonantal slot is mistaken: a C slot is not meant to host consonants exclusively, but non-syllabic segments. If a syllable has one syllabic segment, then a long vowel is hosted by a VC sequence on the skeleton, as Clements and Keyser (1983: 12) argue. In (12d), the empty intervocalic C position is involved in the copying, but, as it is preconsonantal in the reduplicant, it serves as an anchor for the preceding vowel, unlike in the stem, where it is pre-vocalic. This is shown in (13a). Pre-vocalic stems blindly copy the initial empty C position, and so only the first two “real” segments form the reduplicant. (12g) seems to cause a problem now: here the reduplicant appears to be [and-], i.e. VCC, instead of the expected VC. Raimy (1999) suggests an obvious solution: if [nd] is analysed as [ ndd], a geminate pre-nasalized stop, then the situation is identical to that in (12f). The stem-initial empty C must be filled to satisfy onset maximalization: it is impossible to have a coda consonant followed by an empty onset. This is illustrated in (13b). (The reduplicant and the stem are enclosed in brackets for easier identification.) (13)

Reduplication and empty onsets a. [C w

V C] [C

V

i

i

w

C

V] a

b. [C

V

C] [C

V

C

V

C]

o

n

o

n

o

p

In fact, Levin herself suggests the empty-C-slot analysis as an escape hatch for the CV skeleton, but rejects the idea on the grounds that the vowel of the causative prefix [ka-] does not lengthen when prefixed to vowel-initial stems (e.g.

The Skeleton

11

[ka+adanki] > [ka(danki] ‘to name’, [ka+uru(r] > [kauru:r] ‘to be funny’). Vowels do not usually lengthen by filling a pre-vocalic empty C position (cf. Hayes 1989: 281); what is more, it is hard to expect a long vowel or a diphthong to further lengthen. The conclusive test, the prefix [ak-], which is expected to geminate its consonant if prefixed to a vowel-initial stem if there was an empty consonantal slot, “was only found prefixed to C-initial stems” (Levin 1985: 40). We can conclude that the hypothesis that vowel-initial stems carry an empty consonantal position at their left edge is not refuted by Levin’s data.

3.2

Redundancy of C and V

A better argument against CV skeletons is that specifying syllabicity on the skeleton is redundant if the same information can be read off higher prosodic structures, like syllabic constituents, especially the nucleus. Lowenstamm and Kaye (1986) argue that simple syllable trees, like those in (14), adequately define the slots of the skeleton. (14)

Syllable trees a.

b.

c.

N

They suggest that labeling the trees is unnecessary since this information also follows from the configuration. Nevertheless, some minimal labeling (N, i.e. “nucleus”) is necessary to distinguish CVC, (14b), and CVV, (14c), syllables – consider, for example, the Arabic templates of binyans II ([kattab]) and III ([kaatab]) (see (9)). Lowenstamm and Kaye (1986) raise the issue of whether the skeleton is an independent level in phonological representations, or merely a projection of higher prosodic structure, in particular syllable structure. A consequence of this assumption is that the nodes representing syllabic constituents (like onset or nucleus) cannot be distinguished from the skeletal position(s) that they dominate. That is, it is impossible to conceive of skeletal positions not dominated by higher prosodic structure, or of a syllabic constituent that does not dominate a skeletal position. Charette utilizes “pointless onsets” in an analysis of h-aspiré words in French (1991: 90f). She claims that “normal” vowel-initial words begin with an onset that does not dominate any skeletal position, while those which contain h-aspiré – words that phonetically begin with a vowel, but phonologically behave as consonantinitial – begin with a regular “pointful” onset, dominating a skeletal position which is not associated with any melody. The vowel of the definite article is unpronounced before vowel-initial words, but it is pronounced before consonant- and h-aspiréinitial words. (15) illustrates the first part of the two cases using Charette’s examples: l’amie [lami] ‘the girlfriend’ and la hache [la aœ] ‘the axe’. (15)

Two types of empty onset a. O

N

O

N

x

x

x

l

a

a

−

b. O

N

O

N

x

x

x

x

l

a

a

−

12

Péter Szigetvári

According to Charette’s analysis, the vowel of the article is deleted before a pointless onset as a result of the Obligatory Contour Principle, since the two nuclei are “adjacent” if the onset between them lacks a skeletal slot, as in (15a). When such an onset is linked to a skeletal slot, it inhibits the deletion process, as in (15b). This analysis faces difficulties on several counts. On the one hand, the Obligatory Contour Principle controls the appearance of identical melodic elements linked to adjacent skeletal positions. The nodes labeled “nucleus” do not qualify as such. On the other hand, liaison calls for the opposite representation of the two types of vowel-initial words. As mentioned in §1.3, some morphemes that are vowelfinal preconsonantally exhibit a consonant when followed by a vowel-initial word. The plural of the definite article is an example: les amies [lez ami] ‘the girlfriends’ vs. les haches [le aœ] ‘the axes’ (recall that h-aspiré-initial words behave as if they were consonant-initial). The final [z] of the cliticized article is pronounced when there is no skeletal position for it to anchor to, and it is not pronounced when there is one, i.e. without further stipulations, Charette’s analysis predicts just the opposite of the attested liaison facts. The impoverished structures of (14) are also impossible if labels like “onset” and “nucleus” are treated separately from what they label: the skeletal slots. To summarize, there is no compelling reason to distinguish skeletal points and the syllabic constituents containing them. Allowing pointless constituents or constituentless skeletal points makes unnecessary contrasts possible. But then, if prosodic nodes like onset and nucleus are not distinct from skeletal slots, then skeletal slots do carry the basic information of syllabic status: such a skeleton does contain Cs and Vs, irrespective of whether this is penciled on paper as Cs and Vs, Os and Ns, or something else. The two levels must, nevertheless, be kept distinct if more than one skeletal slot can be associated with a single syllabic constituent, i.e. if branching onsets and nuclei are posited. §5 discusses a model where even these are claimed not to exist.

4

Moras

As we have seen in the case of Mokilese reduplication (§3.1), preconsonantal empty C positions are available as targets for the spreading of a preceding vowel; intervocalic ones are not. In many languages, a similar asymmetry characterizes these two consonantal positions. Stress processes, for example, may treat a preconsonantal consonant as being on a par with vowels, but not pre-vocalic consonants.3 Hock (1986) argues that the notion of mora must be (re)introduced into phonological theory. The mora has been around in linguistic discussions for at least two centuries (see Allen 1974: 100); it is its theoretical status that is at issue here (see also chapter 33: syllable-internal structure). Hock’s proposal is to introduce 3

It is common at this point to offer a disclaimer with respect to Everett and Everett (1984) (who claim that Pirahã is different in this respect) or to Davis (1988) (who collects cases where the quality of the onset seems to play a role in stress assignment). However, as Hayes states: “I believe that the ability of Moraic Theory to account for wide-spread patterns of markedness should be given more weight in assessing the evidence than any particular awkwardness in the analysis of individual languages” (1989: 303). This is probably true for any theory. Furthermore, some of the very few onset-sensitive systems have been shown to be re-analyzable so that they are not onset-sensitive (Goedemans 1996; Takahashi 1999).

The Skeleton

13

the mora as an autosegment, rather similar to tones: in his proposal tones are indeed linked to moras. If compensatory lengthening could only lengthen a vowel in compensation for the loss of a tautosyllabic consonant, the “standard” CV or X skeleton would be fully capable of dealing with the process. We have seen, however, that compensatory lengthening also occurs at a distance: the loss of a vowel in the following syllable may lead to lengthening across an intervening onset consonant. Relevant cases are Greek glide loss (e.g. Proto-Greek [odwos] > Ionic [o(dos] ‘threshold’; Steriade 1982: 118) and Middle English schwa apocope (e.g. [talH] > [ta(l] ‘tale’; Minkova 1982). In both cases the melody delinks and the vowel spreading is separated by a consonant that apparently remains linked to the skeleton. (16)

Problematic cases of compensatory lengthening a. V

C

C

V

C

b. C

V

C

V

c. C

V

C

V

o

d

w

o

s

t

a

l

H

t

a

l

H

Actually, as (16a, b) show, the consonant standing in the way of compensatory lengthening is shifted to the right by one slot in both cases. This process, proposed by Steriade (1982: 126–128), is referred to as “double flop” by Hayes (1989: 265 –267). The Greek case in (16a) can be explained by universal principles: the loss of [w] leaves us with an empty onset (provided that the syllabification is [od.wos]). The resulting [od.os] violates the onset maximalization principle, thus resyllabification ensues. But the skeletal position does not resyllabify, since there is an empty onset slot, recently vacated by [w]. It is to this slot that the [d] associates, leaving its original slot empty, triggering the lengthening of the preceding vowel. The lengthening triggered by apocope, exemplified by Middle English [talH] > [ta(l] in (16b), is more problematic for a theory which lacks moras. The mechanism appears to be the same as in (16a), but now the consonant before the disappearing word-final vowel is supposed to flop to a vocalic position, to the nuclear slot of the last syllable. In addition, the position it leaves is not one that should cause lengthening of the preceding vowel. The alternative, whereby the vowel spreads out immediately to the vacated vocalic slot, as in (16c), is even worse, as it violates the axiomatic constraint inhibiting the crossing of association lines.4 In fact, with both CV and X skeletons it is hard to explain why the spreading of a vowel to some consonantal slots should cause lengthening, while in other cases an apparently similar vowel spreading does not. For example, the empty onset in Hungarian pia [pijA] ‘drink’ is filled by the spreading of the melody of the preceding vowel, as in (8).5 Yet the result is not a long vowel, which it is in film ‘film’, for which the pronunciation [fi(m] is possible (Siptár and Törkenczy 2000: 281) (cf. Hayes 1989: 281–283). 4

This problem could be avoided by placing vowels and consonants on separate autosegmental planes (as in (10) and (11)); however, such a modification would loosen the theory beyond desirable limits: we would now find it hard to explain why so many processes deemed possible by the framework never occur. 5 While it may be argued that pia is underlyingly [pijA], the question still holds why the same structure, the melody of [i] doubly linked to a V and a C slot, is [ij] in one case and [i(] in the other.

Péter Szigetvári

14

Hock’s (1986) proposal is to attach a mora ([) to each weight-bearing position, that is, to each vocalic position, as well as to some consonantal positions, notably codas. The two cases are shown in (17). (17) Double flop with moras a.

q

b.

q

q

q

V

C

C

V

C

C

V

C

V

o

d

w

o

s

t

a

l

H

[

[

[

[

[

The moraic analysis of [odwos] > [o(dos] in (17a) is not significantly different from the moraless one, shown in (16a). It nevertheless suggests a reason for the asymmetry between onset and coda consonants: the former do not possess a mora, while the latter do. The advantage of the mora analysis becomes clear in the lengthening of a vowel caused by apocope: [talH] > [ta(l], (17b). The intervening onset consonant is not affected by the process at all, since it is not associated with a mora. Thus, the mora left floating after the final vowel is lost can associate to the stem internal vowel “above the head” (or rather “below the foot”) of the intervening moraless consonant, much like in a vowel harmony process, where intervening consonants not possessing the relevant vocalic feature are transparent. Hayes (1989) rearranges the relationship of the syllable and the mora by making the latter an integral part of prosodic structure, dominated by the syllable node. In a more radical innovation he also gets rid of the skeleton as previously conceived. In his view, the function of the skeleton is taken over by moras, and moraless consonants are either associated directly with the syllable node or share a mora with the moraic segment. Accordingly, the two processes displayed in (16) and (17) would be represented as in (18). (18)

Moras as the skeleton a.

q

b.

q

[

[

o

d

[ w

o

s

t

q

q

[

[

a

l

H

The simple double-flop case of Greek glide deletion in (18a) does not require much comment, as the mechanism is the same as before. For Middle English apocope, however, Hayes needs an extra stipulation, called parasitic delinking: the loss of an overt nucleus in a syllable entails the dissolution of the whole syllable. What is now left of the last syllable, a mora and an [l], is joined to the first one, yielding the correct result. In Hock’s analysis, on the other hand, the [l] remains in place, and does not have to be delinked and relinked, as can be seen in (17b).

The Skeleton

15

Despite this complication, Hayes’s model has definite advantages over Hock’s use of moras. On the theoretical count, it is simpler in that it lacks the CV or X skeleton. On the empirical count, it predicts that compensatory lengthening of a vowel is only caused by the loss of a moraic segment that follows the vowel, never by the loss of one that precedes it. As (19a) shows, Hock’s representations easily allow the latter case, which is not attested according to Hayes. His hypothetical example is [Hla] > [la(]. (19)

Compensatory lengthening triggered by loss of preceding vowel a.

V

C

V

b. q

q

H

l

a

[

[

[

H

[

l

a

In Hayes’s model, (19b), the freed mora of the first syllable cannot be captured by the second mora, because the onset consonant inhibits this. The price to pay for this solution is the stipulative parasitic delinking mentioned above: if the moraic segment of a syllable is delinked, the onset consonant is also delinked, as in (18b). Without this an onset will always block the linking of a heterosyllabic mora. Note that in Hock’s model not only the loss of a vowel, but also the loss of a moraic consonant, could lead to the lengthening of a following vowel (e.g. Proto-Greek [esmi] > hypothetical *[emi(]). Such changes also seem to be unattested, as predicted by Hayes. While theoretically attractive, dispensing with the skeleton has serious repercussions. Recall that linking IPA symbols to elements of higher prosodic structure (slots of the skeleton, moras, syllables) is misleading, since segments are not atomic. In partial trees like those in (20), where the Greek letters a–e stand for (auto)segments, the temporal order of these autosegments is not specified. The string bc is usually referred to as a branching onset; d is a moraless coda, which may occur word finally even in languages with moraic codas, like English. (20)

Autosegmental representations without a timing tier a.

b

c

q

b. q

[

[

a

e

d

Accordingly, the order of two adjacent tautosyllabic or tautomoraic segments must be given by some stipulation. Kaye, for example, provides such a stipulation: “By universal convention the less sonorous of the two elements associated to the same point is produced first in the speech chain” (1985: 289). It remains to be seen if this can be maintained. For syllable-initial consonants, (20a), this is exactly what the sonority sequencing principle dictates. In the domain of single segments, affricates follow this convention, but the existence of prenasalized stops casts some

16

Péter Szigetvári

doubt on its validity. Apart from light diphthongs (like French [wa] in trois [trwa] ‘three’), monomoraic rhymal sequences, as in (20b), obviously cannot be subject to this generalization, since they are invariably ordered in the opposite way, with the more sonorous (vowel) first and the less sonorous (consonant) second. Be that as it may, without some similar (set of) principle(s) an autosegmental representation without a timing tier is uninterpretable. To overcome this difficulty, one might wish to introduce root nodes, a notion familiar from frameworks organizing features into hierarchical structures, “feature geometries” (Clements 1985; Sagey 1986; McCarthy 1988; chapter 27: the organization of features). The root node is the topmost node of such a hierarchy, containing all of the features making up the given segment, that is, the entirety of the segment. If the graphical order of root nodes specified their temporal order as well – as assumed in the contour-segment model of affricates – then “root node” would be just another name for “skeletal slot,” i.e. one would simply reintroduce the skeleton into the representation. The skeleton is apparently indispensable.

5

A return to the CV skeleton

The modern career of the mora was launched by the need to distinguish onset consonants from coda consonants. Only the latter are capable of contributing to the weight of a syllable, that is, of behaving like a vowel; onsets are not (chapter 55: onsets and chapter 57: quantity-sensitivity). A mora is therefore assigned to consonants in the rhyme, but not to those in the onset. Note, however, that the reasoning is circular: codas are equipped with a mora because we observe that they behave differently, and then refer to these moras to explain their difference. But we could just as well imagine an alternative world in which onsets were moraic and codas were not. There is no inherent property of coda consonants that means they are fated to be moraic, as opposed to onset consonants. To make things worse, we will see that it is not exactly true that onset consonants are never moraic, or at least that their loss never entails compensatory lengthening. It turns out to be an oversimplification to tie diverse phenomena like compensatory lengthening, stress assignment algorithms, the assignment of tone-bearing units, etc. to a single property of the representation, i.e. moras (Hayes 1995: 299; Gordon 2004). It is a version of the once rejected CV skeleton that might bring us closer to understanding this asymmetry in the behavior of consonants at the two edges of the syllable. To distinguish it from the McCarthy and Clements and Keyser type of CV skeleton, I will refer to it as the “strict CV skeleton.” In §1.3 and §3.1, we saw why it is useful to suppose that some skeletal positions are empty. So far, we have only seen empty consonantal positions, but there is no particular reason why emptiness, i.e. the state of not being associated to any melodic material, should be limited to consonantal positions. The claim that the host of the vowel (the nucleus) is the head of the syllable, and therefore cannot be missing, is not a very strong one. Syntactic heads, for example the complementizer of a complementizer phrase, may remain empty (e.g. I know [CP [C Ø] she’ll come]).6 But other prosodic units like the foot may also exist without an overt head: in the previous sentence 6

In fact, in English it is by default empty in non-questions, that is, there is an empty complementizer at the beginning of the matrix clause too.

The Skeleton

17

the first headed foot begins with know, and the pronoun I before it forms a headless, degenerate foot. Feet and syllables are similar types of prosodic units, and headless syllables are therefore not inconceivable entities. If nuclei may remain unpronounced, a very restricted syllable structure becomes available.7 Lowenstamm (1996) proposes that underlyingly all languages have the same skeleton, the simplest one available, comprising non-branching onsets and non-branching nuclei in strict alternation. Accordingly, no two consonants and no two vowels are adjacent on the skeleton, as they are always separated by a position of the opposite type. (21) gives the four cluster types of (5) in a strict CV model. (21)

The strict CV representation of vowel and consonant clusters a. V

C

b. V

V

A

A

C

c.

V

C

V t

u

C

d. C n

V

C t

Recall the discussion in §3.2: the CV skeleton contains redundant information that can be read off higher prosodic structures. But this only holds if there is any higher prosodic structure. In fact, strict CV analyses generally do not call for the recognition of such structure, and certainly not of any further syllabic constituency. The two cases of compensatory lengthening in Greek – Proto-Greek [esmi] > Attic [e(mi] and Aeolic [em(i] – are illustrated in (22). (22)

Compensatory lengthening in a strict CV skeleton a. V

C

e

s

V

C

V

b. V

C

m

i

e

s

V

C

V

m

i

With the delinking of the [s], two skeletal slots are opened up for association: both the consonantal slot of the delinked coda and the vocalic slot enclosed within the original [sm] cluster. The choice is apparently controlled by a dialect-specific parameter, just as in any other theory of the skeleton. The mora of moraic theories is an independent entity, which can be assigned to segments as the analyst needs it – it is only empirical considerations that stop them from assigning a mora to onsets. In the strict CV approach, moras are an inevitable consequence of the way the skeleton is built up (Scheer and Szigetvári 2005). A coda consonant is moraic because it is followed by an unpronounced vocalic slot. That is, the moraicness of the coda is only apparent: it is the following vocalic slot that carries weight. In this view, only vocalic slots are moraic. The loss of an intervocalic consonant does not free any “buried” empty vocalic slot, as (23a) shows. The loss of a preconsonantal consonant, on the other hand, makes an otherwise unreachable vocalic slot available as a target for spreading, as in (23b). 7

Note that “empty” and “unpronounced” are not equivalent. In a privative feature framework, empty skeletal positions may be phonetically interpreted, as a sound maximally lacking any contrast, like e.g. [H] or [?]. Some empty skeletal positions may thus be pronounced; others may remain unpronounced if they satisfy certain conditions. See Kaye et al. (1985, 1990), Charette (1991), and Harris (1994) for details.

Péter Szigetvári

18 (23)

The loss of an intervocalic and a preconsonantal consonant a.

V

C

V

b. V

C

a

t

a

a

s

V

C

V

t

a

The weight of closed syllables containing a short vowel is language-specific. For example, in English and Cairene Arabic such syllables count as heavy, while in Khalkha Mongolian and Yidiny they count as light (Zec 1995: 89). This parametric variation is trivially encoded in moraic frameworks: coda consonants are sometimes assigned a mora, and sometimes not. In the strict CV model, the same fact is encoded by parameterizing whether or not an unpronounced vocalic slot is counted by the relevant process. Crucially, however, since the shape of the skeleton is constant – it is always a strict alternation of vocalic and consonantal positions – the uncounted vocalic slot is there even when it is not counted by a certain process (say, stress assignment). One prediction running counter to those of Moraic Theory follows from this fact: compensatory lengthening of a vowel should be possible even if coda consonants are not moraic in a language. Kavitskaya (2002) claims that at least two languages, Piro and Ngajan, are exactly like this. One could claim that the mora associated with the coda in such languages is one which does not contribute to weight but does allow compensatory lengthening (as an anonymous reviewer points out). This then means that there are two types of mora, a “weight mora” and a “compensatory lengthening mora.” The strict CV model predicts exactly this: there are two types of Vs. Pronounced Vs obligatorily contribute to weight, unpronounced ones are parameterizable. In the strict CV framework, when an empty vocalic position enclosed between two consonants is “unearthed,” compensatory lengthening may ensue, irrespective of whether this target of spreading is to the left or to the right of the vowel to lengthen. That is, the loss of an onset consonant may result in the lengthening of the vowel that follows it, as (24) shows. (24)

Onset loss yielding compensatory lengthening C b

V

C

V

c

a

The theory dictates that this option is available only for postconsonantal onsets, not for intervocalic ones (see (23a)). Confirmation of this prediction comes from southwestern dialects of Finnish where gradated [k] is lost, with compensatory lengthening. The data in (25) come from Kiparsky (2008); doubled vowels are long, as in standard Finnish orthography. (25)

Compensatory lengthening in southwestern Finnish dialects input /jalka-t/ /nælæ-n/ /halko-t/

SW dialect jalaat nælææn haloot

standard jalat næljæn halvot

‘legs’ ‘hunger-gen’ ‘logs’

The Skeleton

19

In the Finnish data, the lost consonant is always preceded by another consonant, and is never intervocalic. This is important, because the empty vocalic slot is available between two consonants, but not after a vowel, as (23a) shows. Samothraki Greek exhibits a similar type of compensatory lengthening. In this dialect, pre-vocalic [r] is lost, and is only retained in preconsonantal position – a mirror image of the distribution in non-rhotic dialects of English. The loss of postconsonantal [r] is illustrated in (26a). Intervocalic [r] is lost without trace, as in (26b), as expected. (The data are from Topintzi 2006, who attributes them to Katsanis 1996.) (26)

Loss of /r/ in Samothraki Greek a.

b.

input /’protos/ /’frena/ /’xroma/ /’:rafo/ /’leftirus/ /va’reO/ /’mera/ /’skara/

output [’po(tus] [’fe(na] [’xo(ma] [’:a(fu] [’leftius] [va’eO] [’mia] [’skaa]

‘first’ ‘brakes’ ‘color’ ‘I write’ ‘free’ ‘barrel’ ‘day’ ‘grill’

To provide the missing mora, Hayes (1989: 283) has to hypothesize an epenthesis stage before the loss of the [r]: [’frena] > [fe’rena] > [fe’ena] > [’fe(na]. The strict CV analysis is rather similar, though the only difference is a very important one: the slot of the “epenthetic” vowel is lexically available, since any two consonants are always separated by such an empty slot. The relevance of this difference between the two analyses is that there is no empirical evidence for epenthesis in this case. Furthermore, this assumption creates a paradox in the ordering of the historical events (Kavitskaya 2002: 98), and Hayes’s hypothesis is therefore not plausible. The strict CV skeleton, however, has a vocalic position to which the vowel can spread without any extra process. But even the strict CV model seems to be taken by surprise when it comes to the loss of word-initial [r]: this loss also triggers compensatory lengthening, as the words in (27) show. (27) Loss of /r/ in Samothraki Greek input output /’ruxa/ [’u(xa] /’rema/ [’e(ma]

‘clothes’ ‘stream’

Scheer and Ségéral (2001) introduce the notion of “coda mirror.” The coda is a typical lenition environment, being the position in the word that is not followed by a vowel, i.e. a preconsonantal or word-final position. The coda mirror is the opposite case: it is the position not preceded by a vowel, i.e. a postconsonantal or word-initial position, which is claimed to be the strong position, where lenition is unlikely. Scheer and Ségéral’s theory is built on the strict CV skeleton: for them, “not followed by a vowel” means followed by an unpronounced vowel, and “not preceded by a vowel” means preceded by an unpronounced vowel. It

20

Péter Szigetvári

is this empty vocalic position that causes the lengthening of the vowel in the Finnish and the Greek data discussed here. Not only postconsonantal, but also word-initial consonants are assumed to be preceded by an empty vowel, a proposal first argued for by Lowenstamm (1999). Accordingly, the loss of a word-initial consonant may also cause compensatory lengthening, as shown in (28). (28)

Word-initial consonant loss yielding compensatory lengthening (C)

V

C

V

C

V

r

u

x

a

Since consonant loss is not common in the coda mirror position, compensatory lengthening is also rare in this environment. The peculiarity of Samothraki Greek, then, is that it unexpectedly exhibits [r] loss in the coda mirror position and not in the expected coda position. The ensuing compensatory lengthening is a consequence predicted by the strict CV skeleton.

6

Conclusion

The phonological skeleton evolved as a result of the autosegmental idea taken to its logical conclusion: segments, after autosegmentalizion of all their melodic content, leave behind “traces” that encode their relative temporal order. The debates concerning the phonological skeleton are (i) whether skeletal slots specify any phonetic property (consonantalness vs. vocalicness) or none, i.e. whether the skeleton contains Cs and Vs or uniform Xs; and (ii) whether the mora can replace skeletal slots, with moraless consonants linked directly to the syllable node. This chapter has argued that skeletal slots are Cs and Vs, not merely Xs, but there is no further prosodic constituency (e.g. onsets, nuclei, or syllables). Furthermore, it has been claimed that the mora is not an independent element of the representation, but a consequence of parametrical settings on vocalic skeletal slots: pronounced V slots are universally moraic; unpronounced ones are moraic in some, but not in other languages. Consonants, on the other hand, are never moraic.

ACKNOWLEDGMENTS I acknowledge useful comments and advice from two anonymous reviewers, Marc van Oostendorp, Ádám Nádasdy, Péter Siptár, and László Varga. I thank them all. Remaining errors are mine.

REFERENCES Allen, W. Sidney. 1974. Vox Graeca: A guide to the pronunciation of Classical Greek. 2nd edn. Cambridge: Cambridge University Press. Broselow, Ellen. 1995. Skeletal positions and moras. In John A. Goldsmith (ed.) The handbook of phonological theory, 175 –205. Cambridge, MA & Oxford: Blackwell.

The Skeleton

21

Charette, Monik. 1991. Conditions on phonological government. Cambridge: Cambridge University Press. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clements, G. N. 1985. The geometry of phonological features. Phonology Yearbook 2. 225–252. Clements, G. N. 1999. Affricates as noncontoured stops. In Osamu Fujimura, Brian D. Joseph & Bohumil Palek (eds.) Proceedings of LP ’98: Item order in language and speech, 271–299. Prague: Karolinum Press. Clements, G. N. & Samuel, J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Davis, Stuart. 1988. Syllable onsets as a factor in stress rules. Phonology 5. 1–19. E. Abaffy, Erzsébet. 2003. Az ómagyarkor: Hangtörténet [The Old Hungarian period: Sound changes]. In Jen4 Kiss & Ferenc Pusztai (eds.) Magyar nyelvtörténet [A history of Hungarian], 301–351. Budapest: Osiris. Everett, Daniel L. & Keren Everett. 1984. Syllable onsets and stress placement in Pirahã. Proceedings of the West Coast Conference on Formal Linguistics 3. 105 –116. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5. 253–286. Gimson, A. C. 1989. An introduction to the pronunciation of English. 4th edn., revised by Susan Ramsaran. London: Edward Arnold. Gimson, A. C. 2001. Gimson’s pronunciation of English. 6th edn. London: Edward Arnold. Goedemans, Rob. 1996. An optimality account of onset-sensitive stress in quantityinsensitive languages. The Linguistic Review 13. 33–47. Goldsmith, John A. 1976. Autosegmental phonology. Ph.D. dissertation, MIT. Gordon, Matthew. 2004. Syllable weight. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.) Phonetically based phonology, 277–312. Cambridge: Cambridge University Press. Gussmann, Edmund. 2002. Phonology: Analysis and theory. Cambridge: Cambridge University Press. Halle, Morris & Jean-Roger Vergnaud. 1980. Three-dimensional phonology. Journal of Linguistic Research 1. 83–105. Harris, John. 1994. English sound structure. Oxford: Blackwell. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hoberman, Robert D. 1988. Local and long-distance spreading in Semitic morphology. Natural Language and Linguistic Theory 6. 541–549. Hock, Hans Henrich. 1986. Compensatory lengthening: In defense of the concept “mora.” Folia Linguistica 20. 431–460. Ingria, Robert. 1980. Compensatory lengthening as a metrical phenomenon. Linguistic Inquiry 11. 465 –495. Jakobson, Roman, C. Gunnar M. Fant & Morris Halle. 1952. Preliminaries to speech analysis: The distinctive features and their correlates. Cambridge, MA: MIT Press. Katsanis, Nikolaos. 1996. Sn ckwqqijó idíw[a rgp Ra[nhoájgp [The dialect of Samothraki Greek]. DÑ[np Ra[nhoájgp [Municipality of Samothraki]. Kavitskaya, Darya. 2002. Compensatory lengthening: Phonetics, phonology, diachrony. London & New York: Routledge. Kaye, Jonathan. 1985. On the syllable structure of certain West African languages. In Didier L. Goyvaerts (ed.) African linguistics: Essays in honour of M. W. K. Semikenke, 285–308. Amsterdam: John Benjamins. Kaye, Jonathan. 1989. Phonology: A cognitive view. Hillsdale, NJ: Lawrence Erlbaum. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1985. The internal structure of phonological elements: A theory of charm and government. Phonology Yearbook 2. 305–328.

22

Péter Szigetvári

Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7. 193–231. Kenstowicz, Michael & Jerzy Rubach. 1987. The phonology of syllabic nuclei in Slovak. Language 63. 463–497. Kiparsky, Paul. 2008. Compensatory lengthening. Paper presented at the CUNY Conference on the Syllable. Lahiri, Aditi & B. Elan Dresher. 1999. Open Syllable Lengthening in West Germanic. Language 75. 678–719. Leben, William R. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. Lowenstamm, Jean. 1996. CV as the only syllable type. In Jacques Durand & Bernard Laks (eds.) Current trends in phonology: Models and methods, 419 –441. Salford: ESRI. Lowenstamm, Jean. 1999. The beginning of the word. In Klaus Kühnhammer & John Rennison (eds.) Phonologica 1996, 153–166. The Hague: Holland Academic Graphics. Lowenstamm, Jean & Jonathan Kaye. 1986. Compensatory lengthening in Tiberian Hebrew. In W. Leo Wetzels & Engin Sezer (eds.) Studies in compensatory lengthening, 97–132. Dordrecht: Foris. McCarthy, John J. 1979. On stress and syllabification. Linguistic Inquiry 10. 443–465. McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12. 373–418. McCarthy, John J. 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 17. 207–263. McCarthy, John J. 1988. Feature geometry and dependency: A review. Phonetica 45. 84–108. Minkova, Donka. 1982. The environment for open syllable lengthening in Middle English. Folia Linguistica Historica 3. 29 –58. Moravcsik, Edith A. 1978. Reduplicative constructions. In Joseph H. Greenberg, Charles A. Ferguson & Edith A. Moravcsik (eds.) Universals of human language, vol. 3: Word structure, 297–334. Stanford: Stanford University Press. Odden, David. 1986. On the role of the Obligatory Contour Principle in phonological theory. Language 62. 353–383. Prince, Alan. 1984. Phonology with tiers. In Mark Aronoff & Richard T. Oehrle (eds.) Language sound structure: Studies in phonology presented to Morris Halle by his teacher and students, 234–244. Cambridge, MA: MIT Press. Prunet, Jean-François. 1987. Liaison and nasalization in French. In Carol J. Neidle & Rafael Núñez Cedeño (eds.) Studies in Romance languages, 225 –235. Dordrecht: Foris. Raimy, Eric. 1999. Strong syllable reduplication in Mokilese. In Rebecca Daly & Anastasia Riehl (eds.) Proceedings from ESCOL ’99, 191–202. Ithaca, NY: CLC Publications. Roca, Iggy. 1994. Generative phonology. London & New York: Routledge. Sagey, Elizabeth. 1986. The representation of features and relations in nonlinear phonology. Ph.D. dissertation, MIT. Scheer, Tobias & Philippe Ségéral. 2001. La coda-miroir. Bulletin de la Société de Linguistique de Paris 96. 107–152. Scheer, Tobias & Péter Szigetvári. 2005. Unified representations for the syllable and stress. Phonology 22. 37–75. Siptár, Péter & Miklós Törkenczy. 2000. The phonology of Hungarian. Oxford: Oxford University Press. Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Ph.D. dissertation, MIT. Takahashi, Toyomi. 1999. Constraint interaction in Aranda stress. In S. J. Hannahs & Mike Davenport (eds.) Issues in phonological structure: Papers from an international workshop, 151–181. Amsterdam & Philadelphia: John Benjamins. Topintzi, Nina. 2006. Moraic onsets. Ph.D. dissertation, University College, London. Zec, Draga. 1995. Sonority constraints on syllable structure. Phonology 12. 85 –129.

55

Onsets Nina Topintzi

Onsets are obligatory in the most typical syllable found cross-linguistically, the consonant–vowel (CV) syllable, and as such, are found ubiquitously across languages. This chapter explores various aspects of onsets, covering much of their structural, segmental, and suprasegmental behavior. Using empirical data as a point of departure, various stances and theoretical views will be addressed on a number of issues. These include the presence of the onset in unmarked CV syllables (§1), onset clusters and the role of sonority in their formation (§2), and the structure and representation of the onset within the syllable (§3). The focus will then shift to the onset’s often disregarded role in suprasegmental phonology with reference to several weight-based phenomena (§4). The chapter closes by briefly reviewing approaches that tackle the onset–coda asymmetry (§5).

1

Onsets in unmarked syllables

Most phonologists agree that the most unmarked syllable universally is a CV syllable (Jakobson 1962: 526; chapter 33: syllable-internal structure), i.e. a syllable that consists of a nucleus and a preceding consonant, the onset. When the onset consists of a single segment then it is simplex; when it contains a consonant cluster then it is complex. The present section deals with the former. Evidence for the unmarkedness of CV syllables comes from a variety of sources. First, CV syllables exist in all languages (unlike other syllable types, which only occur in some) and indeed there may be languages whose sole syllable type is CV, e.g. Hua (Blevins 1995) or Senoufo (Zec 2007). While it is the case that every language will have CV syllables, it is not equally true that every syllable in a language will have an onset. Unlike Totonak and Dakota (and of course Hua and Senoufo), where onsets are obligatory, in many other languages they are optional, e.g. Greek, English, and Fijian (Zec 2007). The naturalness of CV syllables is also indicated by the fact that they are the first syllables produced by children during the initial stages of language acquisition (chapter 101: the interpretation of phonological patterns in first language acquisition). The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0055

2 (1)

Nina Topintzi CV outputs by a Dutch child at age 1;5,2 (Levelt et al. 2000) /pus/ /klar/ /oto/ /api/

[pu] [ka] [toto] [tapi]

‘cat’ ‘finished’ ‘car’ ‘monkey’

As Buckley (2003) shows, however, children’s initial productions may also involve VC syllables. Importantly though, these never seem to arise independently, i.e. without CV syllables also being present in the language. The dominance of CV syllables is seemingly contradicted by Arrernte (also known as Aranda; Breen and Pensalfini 1999), Barra Gaelic, and Kunjen – especially its dialect Oykangand – whose syllables are claimed to be of the VC type (with extra codas if need be) and not of the CV type (Blevins 1995 and references therein). These cases are rather weak, however, since for the most part alternative explanations that actually make use of the CV syllable type have been proposed. For instance, Blevins (1995: 230–231) observes that in Kunjen, aspiration only appears prevocalically. In principle, this could be understood as occurring either syllable-initially or syllable-finally, but empirical facts suggest that only the former analysis is viable. If aspiration were to apply syllable-finally, then it should also emerge word-finally, something that never occurs. The facts are thus only compatible with syllabification in the onset. Perhaps the strongest argument in favor of the existence of CV syllables, though, comes from a rule of utterance-initial reduction that deletes initial onsetless syllables, presumably as a means to achieve more well-formed onsetful syllables, as in (2). (2)

Oykangand reduction in utterance-initial position (Sommer 1981: 240) unreduced igigun amamaI uIgul

2

reduced gigun mamaI gul

‘keeps going’ ‘mother (voc)’ ‘there’

deleted material [i] [a] [uI]

Complex onsets

As well as simplex onsets, onsets can also be complex, usually composed of two segments and hence considered maximally binary (Blevins 1995; Morelli 1999; Baertsch 2002; among many others), as in Greek [’tre.xo] ‘I run’, [’pe.tra] ‘stone’, [’vli.ma] ‘missile’, or [’tu.vlo] ‘brick’. Longer sequences such as [str] or [spl] are also commonly allowed, as in English [stre>] stray or [spl>t] split, but usually these are not considered to exceed the binarity maximum, as there is evidence that the [s] here is not part of the onset (see chapter 38: the representation of sc clusters). Yet in some work, the existence of complex clusters is denied altogether. For example, Lowenstamm (1996) and Scheer (2004) claim that all surface syllable types are subsumed under the CV matrix with the addition of empty positions, e.g. English [dØ][ri][mØ] dream. Duanmu (2008) interprets complex onsets such as pl, fr, kl, kr as complex sounds under a single timing slot, on the assumption that such sounds are possible if the articulatory gestures of two sounds can overlap (chapter 54: the skeleton).

Onsets

3

Most phonological models, however, allow complex onsets and provide relevant analyses to account for them. In government phonology (van der Hulst and Ritter 1999; Kaye 2000), for example, binarity is explicitly integrated within the model through the Binarity Theorem (Kaye 1990, 2000), which states that constituents cannot dominate more than two positions, so that onsets may either exhibit single association to a skeletal point (3a) or be maximally binary branching (3b). (3)

Onsets within government phonology (van der Hulst and Ritter 1999) a. O

b. O

x

x

x

More commonly, the binarity of the onset and the combinatorial possibilities among segments within it are attributed to co-occurrence restrictions between adjacent segments (Clements 1990; Zec 2007: 164). In fact, a number of proposals subscribe to the idea that onset syllabification – like the other components of the syllable – is governed by sonority considerations (e.g. Hooper 1976; Steriade 1982; Selkirk 1984; Clements 1990; among others). Briefly, in this approach, more sonorous segments are preferred toward the center of the syllable, whereas less sonorous ones make better syllable margins, i.e. onsets and codas (Clements 1990).1 Despite certain objections to sonority (see below; and also Parker 2002; chapter 49: sonority), its importance for phonological theory is generally acknowledged (Steriade 1982; Selkirk 1984; Clements 1990; Rice 1992; Kenstowicz 1994; Zec 1995). One fairly standard version of the sonority hierarchy is shown below (after Clements 1990). (4)

Sonority scale (> = more sonorous than) vowels > glides > liquids > nasals > obstruents2

One principle that makes use of this scale is the Sonority Sequencing Principle (SSP; Clements 1990), which states that the sonority profile of a syllable must be such that sonority rises sharply toward the peak and gradually lowers after it. Evidence for the SSP comes from various sources. One example is Imdlawn Tashlhiyt Berber (e.g. Dell and Elmedlaoui 1985), known for its long sequences of consonants. Indeed, there may be words that consist of no vowel at all, e.g. [tftkt] ‘you suffered a sprain’. These seemingly highly complicated strings can, however, be easily analyzed if one utilizes the SSP, plus a few other assumptions. Bearing in mind that in Imdlawn Tashlhiyt Berber: (i) any segment can be a syllable nucleus, (ii) onsetless syllables are only allowed word-initially, (iii) codas may appear word-finally, and (iv) complex onsets are banned, the following examples are syllabified in such a way that the nucleus of each syllable comprises a sonority peak.

1

For more detailed discussion on the Sonority Sequencing Principle and the Minimal Sonority Distance, see chapter 49: sonority. 2 For a discussion of other variants see Parker (2002).

4 (5)

Nina Topintzi Imdlawn Tashlhiyt Berber syllabification /ut-x-k/ [u.tUk] /rks-x/ [y.kzx] /t-msx-t/ [tx.sUt]

‘I struck you’ ‘I hid’ ‘you have transformed’

Additional evidence for the SSP comes from onset cluster simplification processes, as in Sanskrit (see Steriade 1988 and chapter 119: reduplication in sanskrit for relevant data) or Attic Greek (Steriade 1982), whereby C1C2 onset strings are reduced to simplex onsets in reduplication. Notably, the surviving C is the least sonorous one, resulting in a more abruptly rising slope toward the nucleus. Similar facts arise in child speech (chapter 101: the interpretation of phonological patterns in first language acquisition), as is evident in the outputs of an English-learning girl aged 2;9 reported on by Gnanadesikan (1995). (6)

Cluster simplification to the least sonorous consonant clean snow friend sky

[kin] [so] [fen] [kaj]3

Not all languages admit the same inventory of complex onsets. It is generally held to be true that the larger the distance in sonority between C1 and C2, the more well-formed the onset cluster. Thus, obstruent (O) + glide (G) clusters are highly favored, followed by O + liquid (L), O + nasal (N), and so on. Onset clusters preferably satisfy a Minimal Sonority Distance restriction in order to be allowed in a language (Vennemann 1972; Hooper 1976; Steriade 1982; Selkirk 1984; Baertsch 2002). In Bulgarian, no distance at all is necessary, thus all of OL, NL, ON, LL, NN, and OO clusters are admitted (Zec 2007); in other languages, different degrees of Minimal Sonority Distance are applicable: in Chuckchee, only OL, NL, and ON clusters are well-formed (Levin 1985); in Spanish, only OL onset clusters (Baertsch 2002); and in Huariapano (Parker 1994), only OG clusters. In a sense, Minimal Sonority Distance generates the expectation that if a language allows C1C2 onset clusters where C2 is of sonority X, then it should also admit onset clusters with a C2 whose sonority is higher than X. But as we have just seen, this is not always the case: e.g. Spanish, which bans *OG clusters. To make things worse, many languages also allow sonority plateaus and even reversals. For example, Greek plateaus like [kt], [fh], and [v:] are tolerated, as in [ktirio] ‘building’, [akti] ‘coast’, [fhiro] ‘impair’, [afhonos] ‘abundant’, [v:azo] ‘remove’, [av:o] ‘egg’. Russian also permits reversals, e.g. [rtut] ‘mercury’ and [lvov] (city name) (Gouskova 2001), which, however, are often considered not to be complex

3

Under the assumption that [sk] is a complex onset, the fricative [s] must be more sonorous than the stop [k] (cf. Dell and Eldmedlaoui 1985; de Lacy 2006; among others). In the sonority hierarchy I have adopted here, this distinction is not made. On the other hand, a difference in sonority of fricatives as opposed to stops would yield incorrect results in other accounts, e.g. Kreitman (2006). If [s], however, is not part of the onset (cf. chapter 38: the representation of sc clusters), this issue does not arise in the first place.

Onsets

5

onsets; rather, the segment(s) violating the SSP can be realized as syllabic, e.g. [y.tut], or even extrasyllabic, attaching to some higher level of prosodic structure, e.g. the foot or prosodic word (see chapter 40: the foot and chapter 38: the representation of sc clusters for more discussion). Such data partly explain why the validity of sonority is sometimes contested. Other objections to sonority include the lack of a clear way to phonetically define and measure it, and its inability to explain the frequent ban on sequences of the type ji, wu, bw, or dl (quite likely an Obligatory Contour Principle effect). Some researchers have therefore gone as far as to discard sonority. For example, Ohala (1990) and Harris (2006) claim that attested sequences in languages can be best captured through the perceptual distance between neighboring sounds in terms of a number of different acoustic properties, including amplitude, periodicity, spectral shape, and fundamental frequency (F0) (Ohala 1990: 334). As Ohala (1990: 334–335) admits, however, this view explains which sequences should be found in languages, but does not explain how and why they are grouped into syllables. This is perhaps why – despite criticism – sonority still remains highly influential in current work on syllabification (cf. Baertsch 2002; Gouskova 2004; Zec 2007; among many others). But there is yet another possibility. Rather than completely endorsing or abandoning sonority, we can accept it, but loosen somewhat the predictions and generalizations it makes. Berent et al. (2007) put forward a proposal along these lines. In particular, they suggest a more flexible version of sonority-based generalizations regarding the profile of onset clusters. They state that: In any given language: (a) The presence of a small sonority rise in the onset implies that of a large one. (b) The presence of a sonority plateau in the onset implies that of some sonority rise. (c) The presence of a sonority fall in the onset implies that of a plateau. (Berent et al. 2007: 594)4

On this view, Spanish is no longer problematic (since OL clusters involve high sonority, there is no reason that there should be OG clusters too), and the plateaus of Greek are expected, given that it also has sonority rises, while Russian has falls only because it also has plateaus. More generally, Berent et al. (2007) test the statements above against the sample of Greenberg (1978) and find that they overwhelmingly hold true typologically. Other typological surveys on onset clusters also tend to employ sonority, usually with some modification or enrichment of the theory. For instance, Morelli (2003) investigates the patterns of obstruent onset clusters and proposes implicational relationships between them, as schematized in (7), where fricative + stop (FT) clusters are the least marked, TT the most marked, and TF somewhere in between. FF clusters merely imply the existence of FT, without further implicational relationship with other clusters. 4

Berent et al. (2007) seem to adopt Greenberg’s (1978) characterization of small and high sonority. High-sonority rises are OL clusters; low-sonority rises are NL and ON; plateaus are OO, and falls are LN and NO clusters.

Nina Topintzi

6 (7)

Implicational relationships between obstruent onset clusters (Morelli 2003)5 TT TF FF

FT

Kreitman (2006) focuses on sonorant (S) and obstruent (O) clusters and proposes the implicational hierarchy SO ⇒ SS ⇒ OO ⇒ OS, with OS clusters being the most unmarked, and SO ones the most marked. These are respectively the most and least favored clusters as far as sonority is concerned. SS and OO clusters involve sonority plateaus, but do not randomly appear in languages as one would expect; instead, the presence of SS systematically implies OO. To account for this fact, Kreitman points to the increased salience of obstruents as opposed to sonorants (cf. Ohala 1983: 193). Since obstruents are considered to carry more information, due to their acoustic form, they are easier to distinguish from non-obstruents. Thus, combinations between obstruents should be perceptually favored over those between sonorants. What all these studies highlight is that removing sonority from the equation is not useful; rather it seems that consideration of other factors, e.g. the role of perceptual salience, may enhance the role of sonority conceptually and improve its empirical coverage.

3

The status of the onset within the syllable

Moving away from the principles that regulate onset syllabification, let us consider the representation of the onset within the syllable. Various models of the syllable have been proposed throughout the years (see Blevins 1995; van der Hulst and Ritter 1999 for overviews; see also chapter 33: syllable-internal structure), which due to lack of space will not be discussed here in detail. Nonetheless, reference will be made to those that are especially relevant to onsets. Broadly speaking, we can identify two major theories: (i) those that distinguish between onsets and rimes (Pike and Pike 1947; KuryÓowicz 1948; Fudge 1969; Selkirk 1982; Levin 1985; Kaye et al. 1990; Blevins 1995), and (ii) moraic models that do away with the rime, i.e. the nucleus + coda string, as a separate constituent (Hyman 1985; Hayes 1989; Morén 2001).

3.1

Onset–rime models

No single version of the onset–rime model is available, and there are significant divergences between models. For instance, Fudge (1969) accepts the syllable as a constituent, whereas Kaye et al. (1990) explicitly do away with it, but nonetheless treat the onset and rime as “an inseparable package” (van der Hulst and Ritter 1999: 23). 5

Inclusion of sC clusters among the FT clusters and their treatment as onset clusters, at least wordinitially, is quite problematic for Morelli, however, in light of evidence showing how sC clusters differ from true branching onsets in various ways (see chapter 38: the representation of sc clusters).

Onsets (8)

7

A typical representation of the onset–rime model (Blevins 1995) q R O N

C

C

C

V

Specific syllable models make different claims about constituenthood. For instance, Blevins (1995) essentially only recognizes the rimal constituent and sees no strong argument for an onset constituent – and for that matter, a coda constituent. For government phonology (van der Hulst and Ritter 1999; Kaye 2000), on the other hand, onsets, nuclei, and rimes are constituents. The basic argument for the rime hinges on the idea that co-occurrence restrictions are always more likely to occur between nuclei and codas, rather than between either onsets and nuclei or onsets and codas. The strongest argument for the rime though comes from weight facts (Blevins 1995; van der Hulst and Ritter 1999: 23). Consider stress, for example. As is well known, in many languages heavy syllables attract stress in contrast to light syllables (e.g. Hopi; Jeanne 1982). Importantly, heaviness implies a binary rime, [VV]R or [VC]R, or both, depending on the language. Since the presence of onsets is disregarded in such an evaluation, it must mean that rimes form a constituent that clearly excludes the onset. Nonetheless, each of the arguments in support of the rime has been challenged. Davis (1985) attacks the reliability of co-occurrence and phonotactic restrictions, given that those are not exclusive to nuclei and codas, but are also found between onsets and nuclei or onsets and codas. For instance, in Korean (Cho 1967), fronted vowels do not appear after labial onsets, while in Yindjibarndi (Wordick 1982), the presence of /r/ in both the onset and a coda of a syllable is banned. Another objection to the onset–rime distinction is found in Yip (2003), who claims that if it were valid, then the boundary between the two constituents should be clear and consistent, and thus segments should uniformly belong to either the onset or the rime, but not to both. English and Mandarin pre-nuclear glides, however, behave sometimes like onsets and sometimes as rimes. As for the weight effects induced by the rime, it is possible to capture them in a different manner without reference to the rime per se. This is what moraic theory does, as we will see in a moment. Before moving on, though, it is notable that the onset–rime debate is also predominant in psycholinguistic studies that explore the onset–rime boundary in terms of implicit and explicit, i.e. non-conscious vs. conscious, phonological awareness. Work by Treiman (1986 and references therein) on various segmentation and substitution tasks in both adults and children suggests that there is a closer connection between VC than CV, thus offering support for the onset–rime boundary. In the same vein, Uhry and Ehri (1999) show that English-speaking kindergarten children preferred to keep VC, rather than CV, intact during segmentation. The opposite result, however, was found by Lewkowicz and Low (1979).

Nina Topintzi

8

More recently, Geudens and Sandra (2003), in a series of four experiments on Dutch-speaking pre-readers and beginning readers, found no support for the onset–rime boundary. Importantly, they applied strict criteria regarding the selection of items under investigation, such that they could control for distributional and sonority effects. In particular, they used items of different sonority equally often and found that syllables with obstruents were easier to perceive and segment than syllables with sonorants (2003: 172); see also chapter 8: sonorants. The influence of sonority may in fact explain some of the findings of previous studies, such as Schreuder and van Bon’s (1989) finding that Dutch first-graders break up a CV string more easily than a VC one. In their study, sonorants were mainly used, but sonorants undergo more vocalization in coda rather than onset position, possibly explaining why children find it harder to break them up in a VC environment rather than a CV one. All in all, psycholinguistic experimentation also reflects contradictory evidence with regard to the onset–rime boundary debate. What this absence of consensus at the very least suggests is that the boundary dispute is well grounded.

3.2

Moraic model

A common response to criticism against the rime has been to dispense with it as a constituent altogether and to replace it with the concept of mora. In moraic theory (Hyman 1985; Hayes 1989), only segments under – what used to be – the rime node may bear moras. Since the latter are needed independently to account for a number of phenomena related to syllable weight, the natural conclusion has been to structurally eliminate the rime from representations. The representation of a CVC syllable in this model is presented next (compare with (8)). Note that the bracket around the mora of the coda indicates that this may be moraic or not on a language-specific basis (cf. Weight-by-Position; Hayes 1989). (9)

Moraic model (Hayes 1989) q [ ([) C

V

C

Within moraic theory, there is no definite agreement as to where exactly the onset associates to. According to Hayes, it directly adjoins to the syllable as in (9). For Hyman (1985), Itô (1989), and Buckley (1992), though, it attaches to the following mora, as in (10). (10)

Onset association (Hyman 1985) q [ ([) C

V

C

Onsets

9

In both these versions of moraic theory, the onset is not recognized as a constituent. This is much more clearly shown in (9), where it directly links to the syllable node, but it is visible even in (10), since the mora is shared between the onset and the nucleus. While Hayes’s representation is the most widely employed, there is nevertheless some evidence for (10). Katada (1990) describes the Japanese chain language game shiritori, in which players say a word that must begin with the final mora of the previous player’s word. If the word ends in a CV syllable, as in [tu.ba.me] ‘swallow’, then the next word can be something like [me.da.ka] ‘killfish’. If the word ends in a long vowel, then the last mora is the second half of the vowel, to the exclusion of the first half, as well as the onset. Thus [bu.doo] ‘grapes’ can be followed by [o.ri.ga.mi] ‘folding paper’ but not by *[doo.bu.tu] ‘animal’. Importantly, a word like [riN.go] ‘apple’ (where N is a moraic nasal) cannot be followed by *[o.ri.ga.mi], but must begin with [go]. This is easily explained if the final mora in [go] also associates to the onset, as claimed by (10), rather than linking directly to the syllable (9). The game ends if the final mora cannot form a proper onset, as happens when it is a moraic nasal, e.g. [ki.riN] ‘giraffe’. Since the moraic model identifies no rime constituent, it bypasses the problems faced by the onset–rime model with regard to the extension of co-occurrence restrictions beyond the rimal node, as well as the absence of a clear boundary between the onset and the rime. Superficially, however, it does equally well as the onset–rime model in accounting for syllable weight, simply by stating or – more accurately, stipulating – that moras are strictly limited to nuclei and codas. But even this assertion has been contested. Work by Hajek and Goedemans (2003), Gordon (2005), and Topintzi (2006, 2010) has shown that there is good evidence for the existence of onset weight. We explore this issue next.

4

The suprasegmental phonology of onsets

Contrary to popular belief, onsets do seem to be prosodically active, albeit in a limited number of languages. Their effects become evident in a range of phenomena, including stress, compensatory lengthening, gemination, word minimality, and tone. This section examines the relevant data and theoretical issues that stem from them.

4.1

Stress

Of all these phenomena, onset-sensitive stress has received the most extensive attention. In brief, three patterns are attested: (i) onset effects due to the presence of an onset, (ii) onset effects due to the quality of an onset, and (iii) patterns (i) and (ii) combined. Starting from (i), we find that in a number of languages onsetful syllables attract stress more than onsetless ones. Languages of this type include Arrernte (Strehlow 1944), Alyawarra (Yallop 1977), and other Australian languages, such as Lamalama, Mbabaram, Umbuygamu, Umbindhamu, Linngithig, Uradhi, KukuThaypan, Kaytetj, and Agwamin (most of them are Cape York and Arandic languages; see Davis 1985, Goedemans 1998, and Blevins 2001 for more details). Beyond Australia, this pattern is attested in unrelated languages of North and

Nina Topintzi

10

South America, Iowa-Oto (Robinson 1975), Banawá (Buller et al. 1993), and Juma (Abrahamson and Abrahamson 1984).6 In Arrernte, C-initial words receive stress on the first syllable (11a), but V-initial ones have stress on the second syllable (11b). One exception is disyllabic words, where stress is word-initial regardless of whether the word begins with a vowel or a consonant (11c). This is probably attributed to Arrernte’s avoidance of final stress or preference for creating binary feet, as the lack of final secondary stress in words like *[a(’ralka)(‘ma)] reveals. (11)

Arrernte stress (Strehlow 1944) a.

consonant-initial ’ra(tama ’kutun‘gula ’lelan‘tinama

words of three or more syllables ‘to emerge’ ‘ceremonial assistant’ ‘to walk along’

b.

vowel-initial words of three or more syllables er’guma ‘to seize’ a’ralkama ‘to yawn’ u’lambu‘lamba ‘water-fowl’

c.

words of two syllables (C- or V- initial) ’ilba ‘ear’ ’a(twa ‘man’ ’kala ‘already’ ’gura ‘bandicoot’

A common denominator is that stress may shift – albeit very locally – to dock on a syllable with an onset. This is not the only possibility, however. In other languages, the stress location remains constant, but if it falls on an onsetless syllable, this acquires an onset. Consider Dutch (Booij 1995: 65). In instances of hiatus where the first vowel is /a/, a glottal stop is inserted before the second vowel, but only if this is stressed by the normal algorithm, e.g. /paelja/ → [pa.’?el.ja] ‘paella’, /aDrta/ → [a.’?Dr.ta] ‘aorta’. Otherwise, no insertion is applicable: /xaDs/ → [’xa(.Ds] ‘chaos’, /farao/ → [’fa(.ra.o(] ‘Pharaoh’. Most analyses view this as a prominence (Smith 2005) or alignment (Goedemans 1998; Topintzi 2010) effect. In yet other languages, the mere presence of an onset is not the issue (see Topintzi 2010: 48 for details on Karo); it is the quality of the onset that matters. This is the case in Karo (Gabas 1999) and possibly Arabela (Payne and Rich 1988). In the former, stress falls on the final syllable, except when the penultimate syllable is a better stress bearer. Better stress bearers are, in order of priority, a syllable with (i) a high tone, (ii) a nasal vowel, or (iii) a voiceless or sonorant onset. When (i) and (ii) are irrelevant, (iii) is taken into consideration and stress falls on the final syllable if the onset is a sonorant (12a) or voiceless (12b) or a voiced obstruent preceded by another voiced obstruent onset (12c). 6

However, the case of Juma should be treated with caution, because only a handful of data are available and because it is possible to re-analyze it. In particular, words like [pe’jikD’pia] ‘bird (sp.)’ may be argued to contain a final diphthong, i.e. [pe.’ji.kD.’pia], rather than a sequence of heterosyllabic vowels, i.e. [pe.’ji.kD.’pi.a], which would lend support to the onset effect. Interestingly, Juma is the sole language where the effect appears at the right edge of the word and not at the left. This may perhaps be an additional indication that it is not truly onset-sensitive.

Onsets (12)

11

Karo final stress and onset voicing (Gabas 1999: 14, 39–41)7 a.

final syllable with sonorant onset kD’jD ‘crab’ ja?’mbD ‘yam (sp.)’ kq7q’wep¬ ‘butterfly’

b.

final syllable with voiceless onset pa’k(D ‘fontanel’ ma?’pe ‘gourd’ ku7u?’cu ‘saliva’

c.

final and prefinal syllables with voiced obstruent onsets ki7i’bDp¬ ‘frog (sp.)’ mq7q’7qj ‘toad (sp.)’

Stress, however, falls on the penult if the final syllable has a voiced obstruent onset and the previous one does not, indicating the stress-attracting nature of the voiceless obstruents and the sonorants in this language. (13)

Karo penult stress and onset voicing (Gabas 1999: 14, 39–41) ’jaba ’we7e ’mHga

‘rodent (sp.)’ ‘frog’ ‘mouse’

’pibe? ‘foot’ ’ka7o ‘macaw’ i?’cDgD ‘quati (sp.)’

Nonetheless, other cases where stress is seemingly sensitive to the onset quality have been shown to be much less robust or even wrong. One example of the latter arises in Mathimathi, where stress is normally word-initial unless attracted by the second syllable when it begins with a coronal onset. Davis (1988) attributes this to genuine onset-sensitivity. Gahl (1996), on the other hand, shows that another account is more plausible, namely one that considers Mathimathi stress to be morphologically based. She claims that stress is located on the last stem syllable of the word (or better, last stem vowel). Stems are generally monosyllabic or bisyllabic. It so happens that apparent stress shift appears on stems of the type C1VC2VC3, where the medial consonant is invariably coronal (Gahl 1996: 329). Evidence for Gahl’s analysis comes from monosyllabic C1VC2 stems, where C2 is again coronal. Addition of a suffix to such stems renders C2 an onset of the second syllable. If Davis were right, then stress here should also be peninitial. However, it is initial, as predicted by Gahl’s morphological account; cf. peninitial stress in bisyllabic stems such as [‘gu.’ra.g+i] ‘sand’ vs. initial stress in monosyllabic stems such as [’wa.Õ+a.{+a] ‘to come’. In both cases, C2 is coronal. Thus, re-examination of the facts in light of morphological considerations may reveal the lack of true onset-sensitive effects (see also Nanni 1977 on the English suffix -ative or Davis et al. 1987 on Italian infinitives). A final pattern that emerges involves the combination of true onset-presence and onset-quality effects. A well-known example is Pirahã (Everett and Everett 1984; Everett 1988), an Amazonian language where codas are banned. Onsetless light syllables (V) do not occur, and stress may only dock on one of the three final 7

Note that [7] in Karo behaves like [d], which is otherwise missing from the inventory (Gabas 1999: 12).

Nina Topintzi

12

syllables of the word. The weight and stress hierarchy the language motivates is: PVV > BVV > VV > PV > BV (P = voiceless; B = voiced). In particular, VV nuclei attract stress more than V ones (14c), and voiceless onsets have the same effect as opposed to voiced ones (14a) and (14d). Crucially, and unlike Karo, Pirahã ‘voiced’ consonants also include sonorants, which appear as allophones of voiced stops, e.g. /b/ may surface as [b], [m], or the bilabial trill [b]. Consequently, in this language, only voiceless obstruents attract stress. Between equally heavy syllables in terms of nucleic weight, onsetful ones attract stress over onsetless (14b). Finally, if there is more than one equal contender for stress, the rightmost one receives it (14e). (14)

Pirahã examples (Everett and Everett 1984; Everett 1988) a.

PVV > BVV ’káo.bá.bai ‘almost fell’ pa.’hai.bií ‘proper name’ b. BVV > VV ’bii.oá.ii ‘tired (lit.: being without blood)’ poo.’gáí.hi.aí ‘banana’ c. VV > PV pia.hao.gi.so.’ai.pi ‘cooking banana’ d. PV > BV ti.’po.gi ‘species of bird’ ’?í.bo.gi ‘milk’ e. rightmost heaviest stress ho.áo.’íi ‘shotgun’ *ho.’áo.íi ti.’po.gi ‘species of bird’ *’ti.po.gi paó.hoa.’hai ‘anaconda’ *paó.’hoa.hai *’paó.hoa.hai

(1988: 239) (1984: 708) (Everett, p.c.) (1984: 709) (1984: 710) (1984: 710) (Everett, p.c.) (1984: 710) (1984: 710) (1984: 707)

What is common to all these examples is that the voiceless obstruent onsets systematically attract stress, contrary to the voiced obstruent ones. Various analyses have been offered to account for the Pirahã facts (and many fewer for Karo). These are examined in Topintzi (2010). Some make use of the increased prominence of onsetful syllables and voiceless onsets over onsetless syllables and voiced onsets respectively (Everett and Everett 1984; Hayes 1995; Goedemans 1998; Smith 2005). Some treat certain onsets as weightful and some as weightless (Topintzi 2006, 2010), and others offer a mixed system that utilizes weight but sees it as a function of prominence (Gordon 2005). Due to space limitations, these proposals will not be reviewed here. However, there is one important empirical argument that favors the onset weight approach, namely the existence of other phenomena beyond stress that are weight-related and influenced by onsets.

4.2

Compensatory lengthening, geminates, and word minimality

An explicit prediction of the onset–rime and the moraic models is that onsets will never participate in weight-related processes. For the former, this is because onsets are excluded from the prosodic hierarchy (van der Hulst and Ritter 1999: 31). For the latter, it is because onsets never bear moras (Hayes 1989). However,

Onsets

13

both assertions are entirely stipulative and subject to modifications given the existence of counterevidence. First, consider compensatory lengthening (chapter 64: compensatory lengthening), a phenomenon widely utilized in support of standard moraic theory. In standard moraic theory (Hayes 1989), it is predicted that onsets will neither induce compensatory lengthening (through deletion) nor undergo it (through lengthening). Yet several cases of both types have been reported. In Samothraki Greek, the onset /r/ deletes and generally leads to lengthening of the following vowel, e.g. /’rema/ > [’e(ma] ‘stream’, /’ruxa/ > [’u(xa] ‘clothes’, /’Ïedru/ > [’Ïedu(] ‘tree’, /kra’to/ > [ka(’to] ‘I hold’ (Katsanis 1996: 50–51). Onondaga (Michelson 1988) is somewhat similar, although /r/-deletion leads to lengthening, whether it is in an onset or a coda originally. Numerous other examples have been reported (Rialland 1993; Beltzung 2007), all of which, however, are highly morphologized. For instance, in Romanesco Italian, the initial /l/ of the definite article and of the object clitic /lo la li le/ optionally deletes (Loporcaro 1991: 280), causing lengthening of the unstressed vowel that follows, e.g. [lo ’stupido] > [o( ’stupido] ‘the stupid (masc)’ or [la ’bru(œo] > [a( ’bru(œo] ‘I burn her’. Beyond this environment, such compensatory lengthening does not appear. Analogous effects are observed in Anuak/Anywa, Lango, Gyore, Turkana, and Ntcham (see Beltzung 2007; Topintzi 2010 and references therein). Nonetheless, one could question the validity of this approach in terms of onset weight structure and instead provide a more phonetic explanation, as done by Kavitskaya (2002). She observes that vowels in CVC syllables are phonetically longer when followed by certain consonants whose transitions can be misheard as part of the vowel (i.e. sonorants, approximants). On deletion of such consonants, the ‘excess’ length of the preceding vowels can be phonologized, so that listeners reinterpret them as phonemically longer. Thus vowels are reinterpreted by listeners as phonemically longer. This approach also extends to compensatory lengthening induced by onsets, but only works when highly sonorous consonants are deleted. In principle, this is appropriate for some of the cases, e.g. Samothraki Greek or Romanesco Italian, but is nevertheless problematic. For instance, it cannot explain why the same phonologization of length has not occurred with regard to the Samothraki coda r, especially since this is the prototypical position for compensatory lengthening. More troublesome, though, is the inability to account for cases like Ntcham, where the onset that is lost is the highly non-sonorous /k/. More strikingly, onsets also can serve as the target of compensatory lengthening. This means that a segment deletes and the preceding onset lengthens, i.e. geminates in order to compensate for its loss.8 For instance, Pattani Malay (Yupho 1989; Topintzi 2008) contrasts singletons and geminates in onsets, but only word-initially (on initial geminates see chapter 47: initial geminates), e.g. [‘bu’wDh] ‘fruit’ vs. [’b(u‘wDh] ‘to bear fruit’, [‘Áa’le] ‘road’ vs. [’Á(a‘le] ‘to walk’ (Yupho 1989: 135). Moreover, it exemplifies a case of compensatory lengthening (Michael Kenstowicz, personal communication). In instances of free variation, one variant involves loss of the word-initial syllable and gemination of the second 8

This characterization is unavoidably linked to a broader discussion of what exactly constitutes a geminate. Briefly, the debate relates to whether geminates are inherently moraic (i.e. heavy) or involve double linking to higher structure (i.e. long). This issue is thoroughly examined in chapter 37: geminates and chapter 47: initial geminates.

Nina Topintzi

14

onset, as in e.g. [buwi] ~ [w(i] ‘give’, [sqdadu] ~ [d(adu] ‘police’, [pqmatD] ~ [m(atD] ‘jewelry’ (Yupho 1989: 130). That these geminates are moraic is supported by another fact of the language, namely stress. Primary stress is word-final, unless the word begins with a geminate, in which case it shifts to the initial syllable (the other syllables receive secondary stress). We can easily understand this effect by claiming that the syllable hosting a geminate is bimoraic and therefore heavy, and as such, attracts stress in preference to monomoraic syllables. (15)

Stress in Pattani Malay (Yupho 1989: 133–135) a.

words lacking geminates ‘a’le ‘road path’ ‘da’le ‘in, deep’ ‘mã‘ke’n ‘food’

b.

words with initial geminates ’m(a‘tD ‘jewelry’ ’Á(a‘le ‘to walk’

As well as Pattani Malay, Trukese provides evidence that onset geminates are moraic (see also chapter 37: geminates for discussion). First, Trukese words are minimally (C)VV, e.g. [maa] ‘behavior’, [oo] ‘omen’, or CiCiV, i.e. a geminate plus a short vowel, e.g. [tto] ‘clam (sp.)’, [ŒŒa] ‘blood’ (Davis and Torretta 1998; Muller 1999).9 CVC and CV words are not allowed (Davis 1999), thus singleton codas contribute no mora (Muller 1999). Presumably, minimality is satisfied by bimoraic words, provided of course that geminates add a mora to their syllable. An additional process of compensatory lengthening following the deletion of the final mora in a word corroborates the moraicity of onset geminates (chapter 37: geminates). Various proposals within the standard moraic theory tradition have been put forward to account for initial moraic geminates (Davis 1999; Curtis 2003), common to which has been the lack of any association between the geminate’s mora and the onset, in line with a major tenet of the theory, namely the ban on onset moraicity. Crucially, to achieve this effect, these approaches link the geminate’s mora to some position other than the onset, which is made possible by the double linking commonly assigned to geminates (see chapter 37: geminates). But this solution is not available in cases of moraic initial consonants that are singletons rather than geminates. Such cases exist. In Bella Coola (Bagemihl 1998) the minimality criterion is fulfilled by VV, VC, and CV words, but crucially not by V words.10 Topintzi (2006, 2010) argues that the easiest way to uniformly understand these data and place them alongside the root-maximality facts of the language – that make reference to mora structure – is by stating a bimoraic word minimum and by allowing onsets to bear moras. 9

Many languages impose a minimum size for words to be well-formed. Commonly, words are required to be at least bimoraic (C)VV as in Ket or Mocha, or (C)VV/(C)VC as in English or Evenki (Gordon 2006), or bisyllabic, e.g. Pitta-Pitta (Hayes 1995: 201). 10 In fact, minimal words with two unsyllabified consonants CC are also allowed. Evidence for the existence of unsyllabified consonants would take us too far afield; see Bagemihl (1998) and Topintzi (2006) for details.

Onsets

15

To capture these facts, Topintzi (2006, 2010) puts forward a flatter syllable structure (reminiscent of Davis 1985), whereby all syllable constituents come in either moraic or non-moraic versions. This is hardly surprising for codas; cf. moraic codas in Latin, Delaware, English, Kiowa, and Turkish vs. non-moraic ones in Wargamay, Lenakel, Eastern Ojibwa, and Khalkha Mongolian (e.g. Hayes 1995; Zec 1995, 2007; Morén 2001). The claim extends to onsets too, e.g. moraic for onset geminates in Trukese or voiceless obstruents in Pirahã vs. non-moraic in a host of other languages. Applying the same distinction to nuclei is also not too far-fetched, as it has been suggested that they can occasionally be weightless, for example in Malagasy (Erwin 1996), Kabardian (Peterson 2007), Alamblak (Mellander 2003), and Chuvash and Mari (Hyman 1985). The following representation illustrates the proposal outlined by Topintzi (2006, 2010).11 (16)

q ([) ([) ([) C

V

C

Even with this modification, though, moraic theory faces problems when it encounters data such as those in Seri and Kikamba (Roberts-Kohno 1995) and Onondaga and Alabama (Broselow 1995 and references therein), and French haspiré (Boersma 2007 and references therein). In Seri (Marlett and Stemberger 1983; Crowhurst 1988; Broselow 1995), the distal prefix [jo-] attaches to either C- or Vinitial stems. In the former, nothing remarkable occurs (17a), but in V-initial stems, things become more complex. In general, when the first vowel of the stem is low back /a/ or low front /æ/ the prefix vowel deletes and compensatory lengthening results (17b). But in some specific stems, no deletion (and consequently no compensatory lengthening) occurs. Instead, a hiatus context is created (17c). (17)

Seri distal forms a.

b.

c.

stem C-initial stems -mækæ ‘be lukewarm’ -pokt ‘be full’

distal jo-meke jo-pokt

general pattern of /a, e/-initial stems -ataø ‘go’ -æmæ ‘be used up’

jo(-taø jo(-me

exceptional pattern of /a, e/-initial stems -amwx ‘be brilliant’ jo-amwx -ænx ‘play stringed instrument’ i-jo-enx

*jo(-mwx *i-jo(-nx

According to Crowhurst (1988), these data support a mixed representation that includes both X slots and moras (chapter 54: the skeleton). The idea is that the stems in (17c) are underlyingly specified with an empty slot in the onset, whose 11

Simultaneous moraicity on all three positions is presumably attested in Karo (see Topintzi 2010: 49).

Nina Topintzi

16

net effect is to block deletion (and compensatory lengthening), because of its intervening position between the two vowels. Effectively, then, (17c) acts as if it were a C-initial stem (17a). Data of this type can also be easily accommodated in government phonology (Kaye et al. 1990), which by its nature allows reference to empty positions. It is, however, not entirely clear that Seri cannot be accommodated by moraic theory alone (especially if onsets may bear moras). Unlinked moras appear in numerous works (cf. van Oostendorp 2005; Topintzi 2007) and are in fact suggested by Crowhurst herself. We could assume then that the input for [jo-amwx] is /jo-Mamwx/, where M indicates a floating mora. If on the surface this mora remains unassociated but anchored at the left edge of the stem, then it can produce the same blocking effect of deletion that Crowhurst achieves by means of an unassociated x-slot. Even if this is feasible, it is unlikely that all similar kinds of facts will be subject to reanalysis. One solution would be to reconsider representations that simultaneously use x-slots and moras, as Crowhurst does. This idea has reappeared in Muller’s (2001) Composite Model with respect to geminates, and in Vaux (2009) as a more complete model of timing. Whether such enrichment of the theory is justified remains to be seen. Alternatively, one could entertain Itô’s suggestion (1989: 255 and references therein) that “the role previously played by lexically empty skeletal slots can be taken over, wholly or in part, by bare melodic root nodes.”

4.3

Tone

Another phenomenon where onsets seem to be involved, albeit rarely, is tone (chapter 45: the representation of tone). Relevant cases reported include Musey (Shryock 1995) and Kpelle (Welmers 1962; Hyman 1985). In Musey, consonants are divided into Type A (or High consonants) and Type B (or Low consonants). Type A consonants include the sonorants and the historically voiceless obstruents. Type B ones correspond to the historically voiced consonants. Both Type A and B obstruents are basically voiceless (Shryock 1995: 68–69), with Type A stops presenting longer positive voice onset time (VOT), less closure voicing, and higher F0 at the onset of the following vowel than the Type B ones. The rightward displacement of lexical L tone when a suffix is added in (18) shows the genuine contrast between the two types of consonants as well as their tonal effects. When the lexical L tone shifts, the vowel that hosted the tone is interpreted as mid or high if the onset is Type A, but as low if the onset is Type B. (18)

Rightward displacement of lexical L tone in Musey a.

b.

cliticization of /-na/ Type A sà → sanà → sanà Type B Âù → Âùnà subjunctive Type A tò ‘sweep’ Type B dò ‘pick’

‘person’ ‘goat’

subjunctive with affixation tdå ‘sweep it’ dòå ‘pick it’

Onsets

17

Thus, at some level of representation, the onset consonants above seem to bear tone – be it by conditioning it or by having it floating in the input – which subsequently surfaces on the neighboring vowel to the right. What is more interesting is that the tone induced depends on the quality of the consonant involved: voiceless obstruents (and sonorants, which I will come back to in a moment) cause M tone, voiced obstruents cause L tone. This fact correlates precisely with data we find in tonogenesis (cf. Vietnamese (Haudricourt 1954) or synchronically in Kammu dialects (Svantesson 1983); see also chapter 97: tonogenesis), where the historical contrast between voiceless and voiced obstruents is neutralized in favor of voiceless obstruents and is reinterpreted by means of tone, as shown below. (19)

Common pattern in tonogenesis voicing contrast; no tone pa > ba >

no voicing contrast; presence of tone pá pà

This pattern is phonetically grounded: in voiceless obstruents, the cricothyroid muscle stretches the vocal folds to obstruct vocal fold vibration resulting in vocal fold tensing, which in turn leads to a higher F0. In voiced obstruents the larynx and hyoid bone are lower and a lowered larynx results in a lower F0 (Yip 2002: 6–7; Honda 2004). In fact, depression of F0 after voiced stops is very likely universal, as Kingston and Solnit (1988b) state (chapter 114: bantu tone). Sonorants, on the other hand, do not automatically perturb the F0 of adjacent vowels, and thus may cause either elevation or depression of the F0 (Kingston and Solnit 1988a: 276). This finding is also in line with the behavior of sonorants in onset-sensitive stress discussed above. Recall that in Karo sonorants act like voiceless obstruents in attracting stress, but in Pirahã like voiced obstruents in avoiding it. Reviewing the vast literature on the phonological effects of the onset/tone interaction phenomenon is well beyond the goals of the present chapter (see Yip 2002; Gordon 2006; van Oostendorp 2006; Tang 2008 for relevant overviews). For our purposes and in light of the data above, it suffices to say that Musey exhibits mixed behavior. On the one hand, it has not entirely lost the voicing contrast between stops (see the discussion on Type A and B consonants) – since it retains phonetic voicing by means of short vs. long VOT – but is moving in that direction, as the facts above reveal; on the other hand, it has introduced tone, which is commonly associated to specific onset quality, but has not (yet?) extended this pattern throughout the system. One thing seems quite clear: onsets in Musey may act as phonological tone bearers. And as expected, voiceless obstruents produce tone raising and voiced ones tone lowering. The more neutral sonorants here pattern with the voiceless obstruents. Along similar lines, we can understand the data in Kpelle. However, unlike Musey, Kpelle onsets act as surface tone-bearing units (TBUs). First, consider minimal pairs such as (20), where a sonorant onset can appear toneless, L-toned, or H-toned. This is hardly surprising, given the capacity of sonorants to bear any type of tone. (20)

mare-kêi åare ké 3are ké

‘a question’ ‘ask him’ ‘ask me’

Nina Topintzi

18

Moreover, the possessive form involves an underlyingly H-toned nasal prefix for the 1st singular or a floating L tone for the 3rd singular (plus the independent processes of voicing assimilation in obstruent-initial stems and total assimilation and nasal simplification in sonorant-initial stems),12 both of which surface on onset positions. (21)

Kpelle onsets as TBUs (Hyman 1985: 44) stem a.

b.

‘my’

‘his/her’

initial obstruent pólù 3bólù túe Udúé kFF ØgFF fíí 3víí

‹ólù ›úé fiFF ‚íí

‘back’ ‘front’ ‘foot, leg’ ‘hard breathing’

initial sonorant lbb Ubb jéé ±éé mXlóI 3XlóI JWI ±WI

Ybb ≤éé åXlóI ≤WI

‘mother’ ‘hand, arm’ ‘misery’ ‘tooth’

These examples show that sonorants and voiced obstruents may appear as surface TBUs, but the same does not hold for voiceless obstruents. This is entirely expected, given that the physical correlate of tone is F0, thus only voiced segments should be able to present it, i.e. vowels, sonorants, and voiced obstruents (Gordon 2006). The Musey data nonetheless have suggested that voiceless onsets should be allowed to be input phonological TBUs (a similar claim for Kpelle appears in Topintzi 2010); if this view is along the right lines, future investigation should focus on how the phonology–phonetics mapping of onset–tone association is accomplished.

5

Onset–coda weight asymmetry

Finishing this chapter, it should by now be obvious that while there is evidence that onsets participate in at least some of the phenomena that codas do, the frequency with which they do so is indisputably much lower and in some cases exceedingly rare. This issue has been mentioned but barely dealt with in the literature; nevertheless, it deserves some brief discussion. Of course, for those who deny any role for onsets in prosody (cf. the standard moraic theory of Hayes 1989), there is not much to explain in the first place. The asymmetry in behavior is the outcome of the more restricted – moraically speaking – structural representation of onsets, compared to that of codas. However, as we have just seen, this approach is too restrictive when it encounters many of the empirical data presented previously. 12

A reviewer points out that the input for the 3rd singular could instead include a low-toned nasal that on the surface fuses with the onset consonant, similarly to what happens in sonorant-initial stems. This is certainly a possibility, but not one Hyman seems to assume. In any case, this issue is orthogonal to the point made here.

Onsets

19

To my knowledge, the first explicit attempt to account for the rarity of onset weight and hence of the onset–coda prosodic asymmetry was offered by Goedemans (1998). Through a set of perception experiments using synthetic stimuli, Goedemans found that Dutch listeners are more attuned to perceive fluctuations in vowel or coda duration rather than onset duration. He next devised an additional experiment to check for the possibility that there is inherently a human bias against perceiving onset duration, but found no evidence in support of this. He therefore concluded that the effect described above must genuinely be due to the weightlessness of onsets. One problem posed by this account is that Goedemans found that listeners recognize duration shifts in onset sonorants better than obstruents. This implies that the former should be preferred as weight bearers to the latter, contra the empirical data, which suggest that in onsets the real difference is between voiced and voiceless obstruents (and that sonorants may pattern with either; cf. Karo vs. Pirahã). More troublesome for this proposal is how to accommodate later work by the same author (cf. Hajek and Goedemans 2003), where onset weight is emphatically argued for, albeit for geminates only (Rob Goedemans, personal communication). Other, more functional accounts of the onset–coda weight asymmetry include Smith (2005) and Gordon (2005), both of which accept onset-sensitivity, but only with regard to stress. To explain why onsets may have a stress-attracting effect, they offer variants of the same idea relating phonological considerations to more general cognitive abilities, such as the sensitivity to auditory stimuli (Viemeister 1980; Delgutte 1982). More specifically, they allude to the evidence of “neural response patterns that the presence of an onset, and specifically a low-sonority onset, does in fact enhance the perceptual response to a syllable” (Smith 2005: 50). Empirically, though, as we know, sonorant onsets may also contribute to weight (or prominence), a fact that both functional accounts fail to capture. Despite this problem, Gordon (2005) claims that in most cases, i.e. most languages, the onset effect is subordinated to the perceptual energy of the rime itself, which is why rimal weight is prioritized over onset weight. Finally, Topintzi (2006, 2010) does not confront this issue in much detail, but nonetheless claims that instead of a single property, it is a constellation of phonological factors, perhaps complemented by the functional accounts above, which may prove enlightening (for details, see Topintzi 2010: §3.3.3, §5.4.1, §6.2.3). For example, the rarity of onset-sensitive tone is attributed to the fact that tone and onset-weight requirements are incompatible with one another. Tone requires the presence of F0, whereas moras that can bear tone in the onset are best assigned to voiceless onsets, which by nature lack F0. In spite of the virtues of each approach, none simultaneously manages to combine accurate empirical coverage with a convincing account that acknowledges the onset–coda asymmetry in its correct perspective and offers a plausible explanation. Future research may fill in the missing pieces of the puzzle.

ACKNOWLEDGMENTS Thanks to Beth Hume, Marc van Oostendorp, and two anonymous reviewers for their instructive comments on various aspects of this chapter. All errors are my own.

20

Nina Topintzi

REFERENCES Abrahamson, Arno & Joyce Abrahamson. 1984. Os fonemas da língua Júma. In Robert A. Dooley (ed.) Estudos sobre línguas tupí do brasil, 157–174. Brasília: Summer Institute of Linguistics. Baertsch, Karen. 2002. An optimality theoretic approach to syllable structure: The split margin hierarchy. Ph.D. dissertation, Indiana University. Bagemihl, Bruce. 1998. Maximality in Bella Coola (Nuxalk). In Ewa Czaykowska-Higgins & M. Dale Kinkade (eds.) Salish languages and linguistics: Theoretical and descriptive perspectives, 71–98. Berlin & New York: Mouton de Gruyter. Beltzung, Jean-Marc. 2007. Allongements compensatoires: Une typologie. Paper presented at the 7th Biennial Meeting of the Association for Linguistic Typology, Paris. Handout available (June 2010) at http://jeanmarc.beltzung.free.fr/?page_id=3. Berent, Iris, Donca Steriade, Tracy Lennertz & Vered Vaknin. 2007. What we know about what we have never heard: Evidence from perceptual illusions. Cognition 104. 591–630. Blevins, Juliette. 1995. The syllable in phonological theory. In Goldsmith (1995), 206–244. Blevins, Juliette. 2001. Where have all the onsets gone? Initial consonant loss in Australian Aboriginal languages. In Jane Simpson, David Nash, Mary Laughren, Peter Austin & Barry Alpher (eds.) Forty years on: Ken Hale and Australian languages, 481–492. (Pacific Linguistics 512.) Canberra: Australian National University. Boersma, Paul. 2007. Some listener-oriented accounts of h-aspiré in French. Lingua 117. 1989–2054. Booij, Geert. 1995. The phonology of Dutch. Oxford: Clarendon Press. Breen, Gavan & Rob Pensalfini. 1999. Arrernte: A language with no syllable onsets. Linguistic Inquiry 30. 1–25. Broselow, Ellen. 1995. Skeletal positions and moras. In Goldsmith (1995), 175–205. Buckley, Eugene. 1992. Kashaya laryngeal increments, contour segments, and the moraic tier. Linguistic Inquiry 23. 487–496. Buckley, Eugene. 2003. Children’s unnatural phonology. Proceedings of the Annual Meeting, Berkeley Linguistics Society 29. 523–534. Buller, Barbara, Ernest Buller & Daniel L. Everett. 1993. Stress placement, syllable structure, and minimality in Banawá. International Journal of American Linguistics 59. 280–293. Cho, Seung-bog. 1967. A phonological study of Korean with a historical analysis. Uppsala: Almqvist & Wiksell. Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 283–333. Cambridge: Cambridge University Press. Crowhurst, Megan J. 1988. Empty consonants and direct prosody. Proceedings of the West Coast Conference on Formal Linguistics 7. 67–79. Curtis, Emily. 2003. Geminate weight: Case studies and formal models. Ph.D. dissertation, University of Washington. Davis, Stuart. 1985. Topics in syllable geometry. Ph.D. dissertation, University of Arizona. Davis, Stuart. 1988. Syllable onsets as a factor in stress rules. Phonology 5. 1–19. Davis, Stuart. 1999. On the representation of initial geminates. Phonology 16. 93–104. Davis, Stuart & Gina Torretta. 1998. An optimality-theoretic account of compensatory lengthening and geminate throwback in Trukese. Papers from the Annual Meeting of the North East Linguistic Society 28. 111–125. Davis, Stuart, Linda Manganaro & Donna Jo Napoli. 1987. Stress on second conjugation infinitives in Italian. Italica 64. 477–498. de Lacy, Paul. 2006. Markedness: Reduction and preservation in phonology. Cambridge: Cambridge University Press.

Onsets

21

Delgutte, Bertrand. 1982. Some correlates of phonetic distinctions at the level of the auditory nerve. In Rolf Carlson & Björn Granström (eds.) The representation of speech in the peripheral auditory system, 131–150. Amsterdam: Elsevier Biomedical. Dell, François & Mohamed Elmedlaoui. 1985. Syllabic consonants and syllabification in Imdlawn Tashlhiyt Berber. Journal of African Languages and Linguistics 7. 105–130. Duanmu, San. 2008. Syllable structure: The limits of variation. Oxford: Oxford University Press. Erwin, Sean. 1996. Quantity and moras: An amicable separation. UCLA Occasional Papers in Linguistics 17. 2–30. Everett, Daniel L. 1988. On metrical constituent structure in Pirahã phonology. Natural Language and Linguistic Theory 6. 207–246. Everett, Daniel L. & Keren Everett. 1984. On the relevance of syllable onsets to stress placement. Linguistic Inquiry 15. 705–711. Fudge, Erik C. 1969. Syllables. Journal of Linguistics 5. 253–286. Gabas, Nilson, Jr. 1999. A grammar of Karo (Tupi). Ph.D. dissertation, University of California, Santa Barbara. Gahl, Susanne. 1996. Syllable onsets as a factor in stress rules: The case of Mathimathi revisited. Phonology 13. 329–344. Geudens, Astrid & Dominiek Sandra. 2003. Beyond implicit phonological knowledge: No support for an onset–rime structure in children’s explicit phonological awareness. Journal of Memory and Language 49. 157–182. Gnanadesikan, Amalia E. 1995. Markedness and faithfulness constraints in child phonology. Unpublished ms., University of Massachusetts, Amherst (ROA-67). Goedemans, Rob. 1998. Weightless segments. Ph.D. dissertation, University of Leiden. Goldsmith, John A. (ed.) 1995. The handbook of phonological theory. Cambridge, MA & Oxford: Blackwell. Gordon, Matthew. 2005. A perceptually-driven account of onset-sensitive stress. Natural Language and Linguistic Theory 23. 595–653. Gordon, Matthew. 2006. Syllable weight: Phonetics, phonology, typology. London: Routledge. Gouskova, Maria. 2001. Falling sonority onsets, loanwords and syllable contact. Papers from the Annual Regional Meeting, Chicago Linguistic Society 37. 175–186. Gouskova, Maria. 2004. Relational hierarchies in Optimality Theory: The case of syllable contact. Phonology 21. 201–250. Greenberg, Joseph H. 1978. Some generalizations concerning initial and final consonant clusters. In Joseph H. Greenberg, Charles A. Ferguson & Edith A. Moravcsik (eds.) Universals of human language, vol. 2: Phonology, 243–279. Stanford: Stanford University Press. Hajek, John & Rob Goedemans. 2003. Word-initial geminates and stress in Pattani Malay. The Linguistic Review 20. 79–94. Harris, John. 2006. The phonology of being understood: Further arguments against sonority. Lingua 116. 1483–1494. Haudricourt, André-George. 1954. De l’origine des tons en vietnamien. Journal Asiatique 242. 69–82. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Honda, Kiyoshi. 2004. Physiological factors causing tonal characteristics of speech: From global to local prosody. In Bernard Bel & Isabelle Marlien (eds.) Speech prosody 2004. Nara, Japan. Available (June 2010) at www.isca-speech.org/archive/sp2004/ sp04_739.pdf. Hooper, Joan B. 1976. An introduction to natural generative phonology. New York: Academic Press.

22

Nina Topintzi

Hulst, Harry van der & Nancy Ritter. 1999. Theories of the syllable. In Harry van der Hulst & Nancy Ritter (eds.) The syllable: Views and facts, 13–52. Berlin & New York: Mouton de Gruyter. Hyman, Larry M. 1985. A theory of phonological weight. Dordrecht: Foris. Itô, Junko. 1989. A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7. 217–259. Jakobson, Roman. 1962. Selected writings, vol. 1: Phonological studies. The Hague: Mouton. Jeanne, LaVerne Masayesva. 1982. Some phonological rules of Hopi. International Journal of American Linguistics 48. 245–270. Katada, Fusa. 1990. On the representation of moras: Evidence from a language game. Linguistic Inquiry 21. 641–646. Katsanis, Nikolaos. 1996. Sn ckwqqijó idíw[a rgp Ra[nhoájgp [The dialect of Samothraki Greek]. DÑ[np Ra[nhoájgp [Municipality of Samothraki]. Kavitskaya, Darya. 2002. Compensatory lengthening: Phonetics, phonology, diachrony. London & New York: Routledge. Kaye, Jonathan. 1990. “Coda” licensing. Phonology 7. 301–330. Kaye, Jonathan. 2000. A users’ guide to Government Phonology. Unpublished ms., University of Ulster. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7. 193–231. Kenstowicz, Michael. 1994. Sonority-driven stress. Unpublished ms., MIT (ROA-33). Kingston, John & David Solnit. 1988a. The inadequacy of underspecification. Papers from the Annual Meeting of the North East Linguistic Society 19. 264–278. Kingston, John & David Solnit. 1988b. The tones of consonants. Unpublished ms., Cornell University & University of Michigan. Kreitman, Rina. 2006. Cluster buster: A typology of onset clusters. Papers from the Annual Regional Meeting, Chicago Linguistic Society 42. 163–179. KuryÓowicz, Jerzy. 1948. Contribution à la théorie de la syllabe. Bulletin de la Société Polonaise de Linguistique 8. 80–113. Levelt, Clara C., Niels O. Schiller & Willem J. Levelt. 2000. The acquisition of syllable types. Language Acquisition 8. 237–264. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. Lewkowicz, Nancy K. & Leone Y. Low. 1979. Effects of visual aids and word structure on phonemic segmentation. Contemporary Educational Psychology 4. 238–252. Loporcaro, Michele. 1991. Compensatory lengthening in Romanesco. In Pier Marco Bertinetto, Michael Kenstowicz & Michele Loporcaro (eds.) Certamen phonologicum II: Papers from the 1990 Cortona Phonology Meeting, 279–307. Turin: Rosenberg & Sellier. Lowenstamm, Jean. 1996. CV as the only syllable type. In Jacques Durand & Bernard Laks (eds.) Current trends in phonology: Models and methods, 419–441. Salford: ESRI. Marlett, Stephen A. & Joseph Paul Stemberger. 1983. Empty consonants in Seri. Linguistic Inquiry 14. 617–639. Mellander, Evan W. 2003. Weightless syllables in Alamblak and the case against Head Dependence. Poster presented at the CASTL Kick-Off Conference, University of Tromsø. Michelson, Karin. 1988. A comparative study of Lake-Iroquoian accent. Dordrecht: Kluwer. Morelli, Frida. 1999. The phonotactics and phonology of obstruent clusters in Optimality Theory. Ph.D. dissertation, University of Maryland at College Park. Morelli, Frida. 2003. The relative harmony of /s+stop/ onsets: Obstruent clusters and the Sonority Sequencing Principle. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 356–371. Cambridge: Cambridge University Press. Morén, Bruce. 2001. Distinctiveness, coercion and sonority: A unified theory of weight. New York & London: Routledge.

Onsets

23

Muller, Jennifer. 1999. A unified mora account of Chuukese. Proceedings of the West Coast Conference on Formal Linguistics 18. 393–405. Muller, Jennifer. 2001. The phonology and phonetics of word-initial geminates. Ph.D. dissertation, Ohio State University. Nanni, Debbie. 1977. Stressing words in -Ative. Linguistic Inquiry 8. 752–763. Ohala, John J. 1983. The origin of sound patterns in vocal tract constraints. In Peter F. MacNeilage (ed.) The production of speech, 189–216. New York: Springer. Ohala, John J. 1990. Alternatives to the sonority hierarchy for explaining segmental sequential constraints. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26(2). 319–338. Oostendorp, Marc van. 2005. Expressing inflection tonally. Catalan Journal of Linguistics 4. 107–126. Oostendorp, Marc van. 2006. Tone and segmental structure. Handout for a course given at the GLOW/DGfS Linguistics Summer School on Micro- and Macrovariation, Stuttgart. Available (June 2010) at www.vanoostendorp.nl/pdf/tone.pdf. Parker, Steve. 1994. Coda epenthesis in Huariapano. International Journal of American Linguistics 60. 95–119. Parker, Steve. 2002. Quantifying the sonority hierarchy. Ph.D. dissertation, University of Massachusetts, Amherst. Payne, David & Furne Rich. 1988. Sensitivity to onset in Arabela stress. Unpublished ms., Instituto Lingüístico de Verano. Peterson, Tyler. 2007. Issues of homophony and the minimal word in the Adyghan languages. Paper presented at the Conference on the Languages of the Caucasus, Max Planck Institute for Evolutionary Anthropology, Leipzig. Pike, Kenneth L. & Eunice V. Pike. 1947. Immediate constituents of Mazateco syllables. International Journal of American Linguistics 13. 78–91. Rialland, Annie. 1993. L’allongement compensatoire: Nature et modèles. In Bernard Laks & Annie Rialland (eds.) L’architecture et la géometrie des représentations phonologiques, 59–92. Paris: CNRS. Rice, Keren. 1992. On deriving sonority: A structural account of sonority relationships. Phonology 9. 61–99. Roberts-Kohno, R. Ruth. 1995. Vowel coalescence and hiatus in Kikamba. In Akinbiyi Akinlabi (ed.) Theoretical approaches to African linguistics, 314–328. Trenton, NJ: Africa World Press. Robinson, Lila W. 1975. Some phonological rules of Iowa-Oto (Siouan). Proceedings of the Mid-America Linguistics Conference, 439–448. Lawrence: University of Kansas. Scheer, Tobias. 2004. A lateral theory of phonology, vol. 1: What is CVCV, and why should it be? Berlin & New York: Mouton de Gruyter. Schreuder, Robert & Wim H. J. van Bon. 1989. Phonemic analysis: Effects of word properties. Journal of Research in Reading 12. 59–78. Selkirk, Elisabeth. 1982. The syllable. In Harry van der Hulst & Norval Smith (eds.) The structure of phonological representations. Part II, 337–383. Dordrecht: Foris. Selkirk, Elisabeth. 1984. On the major class features and syllable theory. In Mark Aronoff & Richard T. Oehrle (eds.) Language sound structure, 107–136. Cambridge, MA: MIT Press. Shryock, Aaron M. 1995. Investigating laryngeal contrasts: An acoustic study of the consonants of Musey. UCLA Working Papers in Phonetics 89. Available at http:// escholarship.org/uc/item/7jb1r066. Smith, Jennifer L. 2005. Phonological augmentation in prominent positions. New York & London: Routledge. Sommer, Bruce. 1981. The shape of Kunjen syllables. In Didier L. Goyvaerts (ed.) Phonology in the 1980s, 231–244. Ghent: Story-Scientia.

24

Nina Topintzi

Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Ph.D. dissertation, MIT. Steriade, Donca. 1988. Reduplication and syllable transfer in Sanskrit and elsewhere. Phonology 5. 73–155. Strehlow, Theodor G. H. 1944. Aranda phonetics and grammar. Sydney: University Press. Svantesson, Jan-Olof. 1983. Kammu phonology and morphology. Malmö: CWK Gleerup. Tang, Katrina Elizabeth. 2008. The phonology and phonetics of consonant–tone interaction. Ph.D. dissertation, University of California, Los Angeles. Topintzi, Nina. 2006. Moraic onsets. Ph.D. dissertation, University College London. Topintzi, Nina. 2007. Weight polarity in Ancient Greek and other languages. In Mary Baltazani, George K. Giannakis, Tasos Tsangalidis & George J. Xydopoulos (eds.) Proceedings of the 8th International Conference on Greek Linguistics, 503–517. Ioannina: Department of Linguistics, University of Ioannina. Topintzi, Nina. 2008. On the existence of moraic onsets. Natural Language and Linguistic Theory 26. 147–184. Topintzi, Nina. 2010. Onsets: Suprasegmental and prosodic behaviour. Cambridge: Cambridge University Press. Treiman, Rebecca. 1986. The division between onsets and rimes in English syllables. Journal of Memory and Language 25. 476–491. Uhry, Joanna K. & Linnea C. Ehri. 1999. Ease of segmenting two- and three-phoneme words in kindergarten: Rime cohesion or vowel salience. Journal of Educational Psychology 91. 594–603. Vaux, Bert. 2009. On the phonological representation of timing and prosody. Paper presented at the CUNY Conference on the Foot, City University of New York. Handout available (June 2010) at www.cunyphonologyforum.net/FOOTPAPERS/ vauxhandout.pdf. Vennemann, Theo. 1972. On the theory of syllabic phonology. Linguistische Berichte 18. 1–18. Viemeister, Neal. 1980. Adaptation of masking. In G. van den Brink & F. A. Bilsen (eds.) Psychophysical, physiological and behavioural studies in hearing, 190–198. Delft: Delft University Press. Welmers, William E. 1962. The phonology of Kpelle. Journal of African Languages 1. 69–93. Wordick, Frank. 1982. The Yindjibarndi language. (Pacific Linguistics C71.) Canberra: Australian National University. Yallop, Colin. 1977. Alyawarra: An Aboriginal language of Central Australia. Canberra: Australian Institute of Aboriginal Studies. Yip, Moira. 2002. Tone. Cambridge: Cambridge University Press. Yip, Moira. 2003. Some real and not-so-real consequences of comparative markedness. Theoretical Linguistics 29. 53–64. Yupho, Nawanit. 1989. Consonant clusters and stress rules in Pattani Malay. Mon-Khmer Studies 15. 125–137. Zec, Draga. 1995. Sonority constraints on syllable structure. Phonology 12. 85–129. Zec, Draga. 2007. The syllable. In Paul de Lacy (ed.) The Cambridge handbook of phonology, 161–194. Cambridge: Cambridge University Press.

56 Sign Syllables Ronnie Wilbur

1

Introduction

This chapter focuses on the notion of syllable in sign languages. Although there is now a consensus on the defining feature of a syllable in sign languages, i.e. that there must be a movement, the initial idea of having “syllables” in sign languages met with considerable resistance on its introduction in the early 1980s, in large part because sign languages are fundamentally monosyllabic languages (see Coulter 1982 on American Sign Language (ASL); similar evidence has been provided for Finnish Sign Language by Jantunen 2007). There was also a strong undercurrent that using concepts borrowed from spoken language linguistics for sign language phenomena would be problematic, if not inappropriate. However, I had an opportunity to ask my colleague Ray Kent, a speech researcher who was then editor of the Journal of Speech and Hearing Research, what he would require if I were to send his journal a paper on syllables in sign languages. He said he would look for evidence that the concept was linguistically meaningful and could be reliably measured, and his response will begin our tour on this topic. Actually, proving to spoken language researchers that there are syllables in sign languages on those two criteria was remarkably easy (Wilbur and Nolen 1986; Wilbur and Allen 1991). In contrast, the phonological representation of sub-syllabic structure has been under constant and lively debate. Accordingly, the presentation of evidence for the existence of syllables will be brief, and the bulk of the chapter will focus on describing the issues of disagreement and providing evidence for an answer to the question of the representation of syllables in sign languages.

2

Historical background

Early sign language research treated the sign as the unit of analysis. This is best observed in Stokoe (1960), where each sign was treated as a unit. Then, when this unit was analyzed further, it was observed that a sign was composed of a “simultaneous” bundle of aspects/primes/parameters, including the “big four”: handshape, location, movement, and orientation (Stokoe 1960; Friedman 1974, 1977; Battison 1978; Siple 1978; Klima and Bellugi 1979; Wilbur 1979; see also The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0056

2

Ronnie Wilbur

Figure 56.1

ASL ARRIVE is a monosyllabic form (arrow shows direction of movement)

Figure 56.2 ASL BABY has two syllables, i.e. two movements (arrows show direction of movement)

chapter 9: handshape in sign language phonology; chapter 10: the other hand in sign language phonology; chapter 24: the phonology of movement in sign language). Several studies suggested that there are syllables in ASL and that these syllables have internal “sequential” organization (Kegl and Wilbur 1976; Chinchor 1978, 1981; Newkirk 1979, 1980, 1981; Liddell 1984). Before addressing the question of internal organization of sign syllables, some clarification of the notion “syllable” in sign languages is necessary. To better appreciate the status of a syllable in sign languages, we must consider the difference between a syllable and a sign. In many cases, the sign is a single syllable and the boundaries of the two, i.e. a syllable or a sign, coincide (Coulter 1982, 1990). Figure 56.1 is an example of a single-syllable sign. In other signs, such as BABY in Figure 56.2, there are two syllables in a single sign. The sign BABY is a larger unit than ARRIVE (Figure 56.1). In still other signs, such as MOTHER (Figure 56.3), there are specifications for handshapes, locations, and hand orientations, but, critically, these signs do not have their own movement specifications. Since every syllable must have a movement, the phonological specification for these lexical items is smaller than a syllable; it has been proposed that the epenthetic/transition movement to, or away from, the target location provides the prosodic feature needed to produce a full syllable (Wilbur 1985; Brentari 1990b, 1998; Geraci 2009). We have so far sketched sign syllables by providing three groups of signs with different movement specifications, i.e. with single movement, with two movements,

Sign Syllables

3

Figure 56.3 ASL MOTHER is a sign without a movement specification. By permission of Dr. Bill Vicars

and without lexical movement. We now consider the difference between a syllable and a morpheme in order to distinguish their functions. A morpheme is defined as the smallest possible unit of meaning. In Figures 56.1–3, each sign is also a single morpheme. In Figure 56.1, the morpheme and the syllable are the same size. In Figure 56.2, the morpheme is larger than the syllable (it has two syllables), and in Figure 56.3, the morpheme is smaller than the syllable (it is missing movement). A morpheme may also be as small as the feature specification for a single handshape, as in classifier constructions (see chapter 9: handshape in sign language phonology), or location, as in verb agreement. It should be noted that two-syllable lexical items are highly constrained with respect to their movement specifications: the movement in the second syllable is either the exact opposite (180° rotation) of the movement in the first syllable (as in BABY, Figure 56.2) or it is 90° rotated (Figure 56.4; note transitional movement inserted when the end of the first movement is not the starting position of the second movement) (Wilbur and Petersen 1997). Clearly movement is central to syllable structure (see chapter 24: the phonology of movement in sign language). The first attempt to break sign movement into smaller sequential pieces was Newkirk (1979, 1980, 1981). Considering rhythmic features of movement, he analyzed them into [onset] [movement] [offset]. Subsequently, a number of sequential and simultaneous proposals were offered. This brings us to the ongoing debate – what is the internal structure of a syllable with respect to sequentiality and simultaneity? In the following section, I provide an overview of what everyone does agree upon, i.e. that syllables in sign languages exist. In the subsequent section, I provide evidence of the behavior of syllables with respect to higher phonological organization. Finally, we dive inside the sign syllable and consider the evidence for the two theoretical options – sequential and simultaneous organization of syllable structure.

4

Ronnie Wilbur

3

1

2

4

Figure 56.4 ASL CANCEL/CORRECT/CRITICIZE and schematic of movement showing second syllable (3 to 4) perpendicular to first (1 to 2)

3

Sign syllables exist

Syllables have been reliably measured, and in conversational contexts they have roughly the same duration as spoken language syllables (Wilbur and Nolen 1986). As will become clear below, there has been no shortage of linguistic uses for the syllable in sign language (morpho-)phonology, hence it is linguistically meaningful. It is fair to say that sign language phonologists now take the notion of sign syllable as a given, and that movement is its nucleus (the carrier of its perceptual salience; Jantunen and Takkinen 2010).

3.1

Syllable measurements

Investigators have measured sign duration, including signs that are clearly single syllables (Bellugi and Fischer 1972; Friedman 1976; Liddell 1978, 1984). For those signs which are monosyllabic, the measured duration means range from 233 to 835 msecs as a function of context. Liddell (1978) reported the effects of sentence position and syntactic function on duration of the monosyllabic signs DOG and CAT. His measurements show phrase-final lengthening, as the durations were longest in sentence-final position. Duration in sentence-initial position was next longest, and medial position in relative clauses had the shortest duration. His measurements also show a syntactic effect: objects were shorter than subjects or heads of relative clauses. My investigation of syllable duration began in 1984, when videotape was “reel to reel,” which meant that the tapes could be moved by hand, forward and backward, and measurement was in “fields” – 60 per second. To measure syllables, movements, and holds, we had to provide our own mechanical guidelines and demonstrate that they could be reliably used (Wilbur and Nolen 1986; Wilbur and

Sign Syllables

5

Schick 1987). We started with the cues identified by Green (1984) for beginnings and ends of signs, which worked well for signs that were perceptually monosyllabic. These cues included points of contact and changes in facial expressions and eye gaze. However, Green’s procedures were not sufficient for determining syllable boundaries when the sign and the syllable are not coterminous, i.e. when we have more than one syllable in a sign. To capture the behavior of multisyllabic signs such as bidirectional signs (Figure 56.2 above), reduplicated forms, and compounds, additional cues were needed. With the aid of native signers, we determined that a change in the direction of movement marked a boundary between two adjacent syllables. For elliptical movements, we accepted Newkirk’s (1979, 1980, 1981) argument that they were segmentable into two parts, and then we used the change in direction of movement as the boundary between the two parts. By contrast, circular movements, which show no internal structure, were treated as one syllable per circle. For holds, we established a procedure in which the end of a hold would be marked by one or more of the following cues: start of the next movement, loss of tension in the signing hand(s), change of eye gaze, initiation of signing by the other signer, or change of eye gaze by the other signer (Wilbur and Nolen 1986). Syllables were measured by two people at a time; if they could not agree, a third person was consulted. Over three thousand syllables were measured in four situations – natural conversation, elicited paragraphs, lists of signs, and phrases and compounds. In conversations, 889 syllables from three signers had a mean of 248 msecs, comparable to the estimated 250 msecs for spoken English (Adams 1979; Hoequist 1983). This similarity may be a reflection of an underlying timing mechanism for motion that may surface not only in speech and signing, but also in non-linguistic motor behaviors. For example, in baseball, a bat swing takes about 200 msecs (Schmidt and Lee 2005). For the lists, mean syllable duration was 299 msecs for the first production, and 417 msecs for the second. Thus, signers can have different durations at different times. For paragraphs, 14 signers produced paragraphs with either a stressed or unstressed target sign. The stressed target mean was 317 msecs, and the unstressed target mean 299 msecs. There were more syllables in the stressed condition, i.e. repeated syllables and/or resyllabification (similar to English please pronounced as puh-leeze). In the last condition, compounds may have two syllables or can be reduced to one (Coulter 1982; Liddell 1984). Eighteen sets of compounds and their two component signs were provided by Ursula Bellugi. In each set, the two signs appeared in a phrase in isolation (e.g. FACE CLEAN) and in context (HE HAS FACE CLEAN ‘He has a clean face’). The same morphemes also appeared in a compound (FACE-CLEAN ‘handsome’) in isolation and in context (HE FACE-CLEAN ‘He is handsome’). The compounds had significantly more syllables per sign than simple lexical items (Wilbur and Nolen 1986). Also, the signs in isolation (whether simple lexical items or compounds) had significantly more syllables than in context, reflecting prosodic effects. Thus signers manipulate both syllable duration and number of syllables in their sign productions. The evidence so far lends support to one of the two criteria that started our discussion, i.e. “Can syllables be reliably measured in sign languages?” We turn now to the other criterion, i.e. “Are syllables linguistically useful?”

6

4

Ronnie Wilbur

Phonological applications identified for “syllable”

In this section, I briefly review the arguments that have been offered to show that syllables contribute to the statement of phonological processes and to our understanding of why some processes behave the way they do. The notion of syllables has proven to be useful in the statement of a variety of historical changes (Battison 1978; Frishberg 1978), synchronic morphological processes (Chinchor 1981), and phonological processes (Coulter 1982; Wilbur 1990b, 1993). Blevins (1993), Padden (1993) and the work of Brentari (1990a, 1990b, 1993, 1996, 1998) provide further arguments in favor of the role of syllable structure in ASL phonology. I review only a few, and refer the reader to the original authors for further evidence and discussions.

4.1

Fingerspelling and fingerspelled loan signs

Battison’s (1978) discussion of the creation of new signs from fingerspelled words provides data to support the notion of syllables. Theoretically, each fingerspelled letter consists of a handshape and, when produced in slow sequences, a transition movement (handshape change) to reach each handshape. Thus, there could potentially be as many syllables as there are letters in the word being fingerspelled, because there could be one movement to make each handshape and each could therefore be a syllable. In actuality, fluent fingerspelling is performed with a phrasal rhythm that smoothes the transition handshape changes and reduces the prominence of certain handshapes while increasing the prominence of others (Akamatsu 1982, 1985; Wilcox 1992). In the process of becoming a lexicalized fingerspelled loan sign, some letters in a word are dropped, and remaining handshapes are associated to syllabic nodes, reducing the number of syllables produced. Fingerspelling the word “sick” (Figure 56.5) involves handshape changes from each letter to the following letter. Since fingerspelling is based on English spelling, each English word will have a different set of handshapes and a corresponding different set of transitional handshape changes. In contrast, in the fingerspelled loan sign #SICK (where # denotes a fingerspelled loan sign (Figure 56.6), the middle letters have been dropped, and the handshape change from S to K has created the movement nucleus of the syllable, to which a slight directional path movement has been added (the arrow does not do justice to this – the middle finger can appear to flick forward from the fist while the index finger straightens up). At the syllable-internal level, the features for

Figure 56.5

Handshapes S, I, C, K, with three transition movements between them

Sign Syllables

Figure 56.6

7

Fingerspelled loan sign #SICK, with one lexical movement

the handshapes S and K are associated with the same syllabic node, i.e. the only syllabic node. The handshape change from S to K is not permitted in core lexical items as reflected in Brentari’s model, thus this form is clearly identified as of foreign origin from English. Brentari (1994) used the fingerspelled loan signs from published ASL lectures (Valli and Lucas 1992) to determine the processes involved in lexicalization. She found that long fingerspelled words with as many as eight or more letters reduced to fewer handshapes and just two movements. The result is that the newly lexicalized forms fit the phonotactics of ASL, having a maximum of two syllables.

4.2

Evidence for a sonority hierarchy

Several researchers have suggested a relationship between visibility and syllable sonority (Corina 1990; Perlmutter 1992; Sandler 1993; Brentari 1998). The sonority hierarchy treats movements made with joints closer/proximal to the body/trunk, such as elbows and shoulders, as more sonorous, because of their visibility in motion when compared to those lower down and more distal, such as hands and fingers, which are considered less visible and hence less sonorous (1) (see chapter 49: sonority for an overview of issues surrounding the sonority hierarchy in spoken language): (1)

Sonority hierarchy with respect to the relevant joints (from Brentari 1998) Most sonorous shoulder > elbow > wrist joint joint joint

>

base finger joints

Least sonorous > non-base finger joints

Movements made with the wrist joint have a higher sonority value, i.e. are more visible due to larger movements, than movements made with the use of finger joints. With respect to the development of fingerspelled loan signs above, the addition of the slight path movement in the loan sign #SICK may be viewed as adding a higher sonority value to the less sonorous handshape change from S to K, thereby producing a more sonorous syllable and hence more acceptable lexical item. Thus what we observe is that the proposed prosodic hierarchy provides a rationale for why finger positions are dropped and wrist or elbow movements (depending on how the small path is made) are retained in fingerspelled loan words (chapter 95: loanword phonology).

8

Ronnie Wilbur

A

Figure 56.7

4.3

B

Metathesis for the ASL sign DEAF. (a) A-to-B. (b) B-to-A

Contact metathesis

Another situation where the notion of syllable is phonologically relevant is found with two-location contacting signs, which in certain circumstances can undergo a change that causes the two locations to switch their order (i.e. metathesis; Kegl and Wilbur 1976; Johnson 1986; Sandler 1986; Wilbur 1987; chapter 59: metathesis). For example, the location points in signs such as DEAF, PARENTS, and FLOWER may switch from A-to-B to B-to-A (Figure 56.7), depending on preceding phonological context: The process of metathesis is limited to signs that are both single morphemes and single syllables. Some were originally compounds and may be articulated in a way which reflects their origins. For example, the sign PARENTS might be deliberately made to emphasize its origin from FATHER + MOTHER: movement to contact at the forehead (FATHER) and then movement to contact at the chin (MOTHER). This form would be two syllables, and not subject to metathesis. In the monosyllabic version of PARENTS, the forehead contact is preceded by a transition movement, and the lexical syllable consists of the change in location from the forehead contact to the chin contact. This latter form can undergo metathesis, i.e. the hand can touch the chin first and the forehead second. The statement of metathesis in terms of syllables greatly simplifies the formalization of the rule. We can see that the statement in terms of syllables is the correct one, as opposed to morphemes or signs. PARENTS can undergo metathesis in one form of production but not in another, so the rule about which signs can undergo metathesis could not refer to “contacts in the same sign” or “contacts in the same morpheme,” but only to “contacts in the same syllable.”

4.4

Handshape change (change in aperture)

Brentari and Poizner (1994) show that handshape change timing is different within syllables than it is between syllables (chapter 9: handshape in sign language phonology). Within syllables, if there is path movement and handshape change, the handshape change coordinates with the beginning and end of the path. However, when two signs in sequence have different handshapes, there must be a transitional handshape between the end of the first sign and the beginning of the second (as discussed for fingerspelling, above). In such conditions, the

Sign Syllables

9

change in handshape is not coordinated with the path movement in the same way as within signs, that is, the timing of the change does not distribute evenly over the transitional path movement.

4.5

Consistency in movement

Likewise, Tyrone et al. (2007) compare monosyllabic sign movements toward the body to the same location (e.g. forehead) in two conditions: (i) the movement is part of the sign (THINK), and (ii) the movement is transitional prior to the sign (SMART). They report that within-syllable within-sign movements show typical bell-shaped velocity curves for targeted movement, whereas transitional movement between signs is less regular. These findings, along with the handshape change findings above, converge on the necessity of separating phonological syllable movement from phonetic epenthetic/transitional movement.

4.6

Minimal word

Simply put, the syllable is the smallest possible well-formed sign/word. Furthermore, two syllables is the maximum for well-formed core lexical signs (Perlmutter 1992; Sandler 1993; Brentari 1998; Jantunen 2007). Alternative formulations without mention of syllables would necessarily be more complex.

4.7

Prosodic constraints

Miller (1997) argued, on the basis of Quebec Sign Language (LSQ), that Phonological Phrases require a disyllabic foot. Similarly, van der Kooij and Crasborn (2008) suggest that in Sign Language of the Netherlands (NGT) the phonological constraint on the addition of sentence-final pointing must be stated in syllabic terms: sentence-final pointing is permitted only if the outcome is a disyllabic foot. Wilbur (1999a) observes that ASL pronouns in sentence-final position tend to be extrametrical with respect to stress assignment at the phrase level, supporting prosodic constraints proposed in Halle and Vergnaud (1987) for spoken languages.

5

Syllables and prosody

The last two arguments in favor of syllables also provide evidence for the prosodic hierarchy in sign languages. That is, metrical structure (lexical, phrasal, and clausal stress assignment), rhythmic structure, and intonational phrasing are dependent to some degree on the syllable level. (2) shows the prosodic hierarchy we adopt for further discussions (see chapter 33: syllable-internal structure; chapter 40: the foot; chapter 51: the phonological word; chapter 57: quantity-sensitivity; chapter 84: clitics for more discussion of the prosodic hierarchy). Prosodic words will be discussed in this chapter; for discussion of Intonational Phrases see Wilbur (1994), Sandler and Lillo-Martin (2006), and Weast (2008). (2)

Prosodic hierarchy syllable < prosodic word < prosodic phrase < intonational phrase

Ronnie Wilbur

10

GIVE-A-GIFT1

repeat2

repeat3

] repeat4

repeat5

WORK(ERS)

Figure 56.8 One prosodic word, composed of GIVE + distributive aspect, repeated five times, accompanied by one Posture-NM, followed by the next prosodic word, containing WORK

5.1

Prosodic words

As indicated, the minimal prosodic word is at least one syllable, and the prosodic constraint on well-formed lexical items is a maximum of two syllables. Brentari and Crossley (2002) demonstrated that changes in lower face tension (mouth and cheeks) mark the end of a prosodic word (PW), which is above the syllable in the prosodic hierarchy. Figure 56.8 shows a single lower face position, i.e. closed mouth with lip corners slightly down, referred to as posture non-manuals (P-NM), which stretches over one long PW, followed by the sign WORK, which has a round mouth and is in a different prosodic word. The context was “every year at Christmas time, the boss gives each of the employees a gift.” Note that the single PW contains five syllables (five repetitions of the lexical item GIVE-A-GIFT). (3) represents the marking of the relevant prosodic words through sign language glossing conventions. The tier above the glosses represents non-manual marking, and the line indicates the spread of the non-manual marker: (3)

Prosodic grouping for Figure 56.8 P-NM P-NM !GIVE-A-GIFT [Repeat × 5]# !WORK(ERS)# @PW $ @PW $

In contrast, Figure 56.9 shows “one car hits another car three times.” The signer produces three mouth changes (Transition-NMs), once for each repetition. These changes result in three PWs, as represented in (4): (4)

Prosodic grouping for Figure 56.9 T-NM T-NM T-NM !HIT-CAR # !HIT-CAR # !HIT-CAR # @PW $ @PW $ @PW $

Sign Syllables

] CAR-HIT1 ]PW

] CAR-HIT2 ]PW

11

] CAR-HIT3 ]PW

Figure 56.9 Sequence of three CAR-HITs, involving three syllables, three mouth changes, and three PWs

5.2

Stress assignment

Stress assignment is a prosodic process, and may occur on lexical items, compounds, and phrases. Early research on ASL stress focused on marking stress on lexical items (Covington 1973; Friedman 1976; Wilbur and Schick 1987; Coulter 1990). Stressed signs can be set off from unstressed signs by several cues: (i) faster/shorter transition movement than between unstressed signs, breaking the rhythmic pattern; (ii) higher in the signing space compared to their unstressed counterparts; (iii) increased repetitions compared to their unstressed counterparts, changing the duration; (iv) increased speed (higher peak instantaneous velocity) compared to their unstressed counterparts; (v) increased muscle tension compared to their unstressed counterparts; and (vi) stressed signs have a following pause (Wilbur and Schick 1987; Wilbur 1990a, 1990b, 1999b, 2009; Allen et al. 1991).

5.2.1 Lexical items So far, no sign language has been shown to have distinctive lexical stress, comparable to English ’permit and per’mit (Jantunen and Takkinen 2010). The predominance of monosyllabic lexical items is partly responsible for this absence. Another reason is that polysyllabic signs are restricted to three possibilities: Lexicalization of repetition: A sign may have more than one syllable if it is formed as a result of lexicalization of a repeated form (e.g. ASL FINGERSPELL; Brentari 1998: 169). The result is a two-movement sign with a Return transition in the middle: A-Return-A. In these forms, only the first syllable is prominent/ stressed (Supalla and Newport 1978; Coulter 1990). Lexical disyllables: A sign may have two syllables if it is a lexical disyllable, i.e. if the morpheme itself requires two syllables. There are two types of disyllables, both of which are subject to constraints on the nature of the movements in each syllable (Wilbur 1990b). In the first type, the movement of the second syllable must be rotated in direction 180° from that of the first, returning the hands to their original location (BABY; Figure 56.2). In the second type, the movement of the second syllable is rotated 90° from the first (creating a crossing movement) (CANCEL; Figure 56.4). Supalla and Newport (1978) discuss the first type, and note that prominence is equal on both syllables. It is also the case that prominence is equal on both syllables in the second type. Thus all lexical disyllables have equal stress on both syllables. Similarly, van der Kooij and Crasborn (2008) show that NGT has both trochaic and iambic stress patterns for disyllabic signs, the

12

Ronnie Wilbur

type being predictable on the basis of the phonotactics of the rest of the sign. Thus, as for ASL, stress is not distinctive in NGT. Lexical items have stress on the first, and perhaps only, syllable. Lexical disyllables are exceptional in being specified at the morphemic level for two syllables and in requiring equal prominence on both syllables. Compounds and phrases: A sign may have two syllables if it is a compound, with the first weaker than the second. Unlike lexical items, the assignment of stress to ASL compounds and syntactic phrases follows a very general pattern. In a compound or phrase, a single stress is assigned to the most prominent syllable of the rightmost lexical item.

6 6.1

The internal structure of syllables The debate

Historically, models of the internal structure of syllables (chapter 33: syllableinternal structure) have taken one of two views. The first view is that sign syllables, like spoken syllables, are composed of sequences of segments (chapter 54: the skeleton). These segments are of two types (like consonant and vowel in spoken language), namely Movement (M) (the hands are in motion) and a contrasting type with the hands not in motion. Distinctive features are distributed among these phonological segments parallel to spoken language C and V. Liddell (1984) argued for two types of segments, movements (M) and holds (H). The remaining information – handshape, contact, orientation, location, and facial expression – is represented as features occurring simultaneously with each segment, thus there is a sequence of feature matrices within each sign. Signed syllables could then be of several types, e.g. M, MH, HM, HMH. Sandler (1986, 1989, 2008) proposed a different model, in which the segment opposition is between movement (M) and location (L), with handshape configuration on a separate autosegmental tier. The presence or absence of holds would be characterized by a binary feature in the location feature matrix; rather than having holds underlyingly, there will be some phonetic holds (list rhythm), some phonological holds (at utterance boundary), some morphological holds (ASL aspectual inflections may include final hold as part of their pattern), and some pragmatic holds (end of conversational turn, waiting for back-channel nod). For these models (e.g. Liddell and Johnson 1989; Sandler 1989, 2008; Sandler and Lillo-Martin 2006), the segments are at the top of the phonological trees containing the distinctive features, i.e. the mother nodes in a feature geometry model. That is, the syllable is composed of segments, which are characterized by relevant phonological features. In the other view, supported in Brentari (1998) for ASL and van der Kooij (2002) for NGT, movements are dynamic prosodic units with similar autosegmental status as tones in contrastive tonal languages (e.g. Mandarin, Cantonese; chapter 45: the representation of tone; chapter 107: chinese tone sandhi). An important step leading to this alternative view was van der Hulst’s (1993) Head–Dependency Model, in which features that did not change during the sign were considered to be heads, with changing features treated as dependents (for detailed discussion see chapter 24: the phonology of movement in sign language). Head features

Sign Syllables

13

could be location, orientation, the active (selected) fingers, or their configuration. Movement itself was dependent on change of location (path movement) or hand configuration or orientation (local movement) (see similar arguments in Wilbur 1987). Brentari (1998) provides arguments against the notion of dependent/emerging movement, and instead identifies those features that do not change within the syllable as Inherent Features (IF) and those that do change as Prosodic Features (PF). From this perspective, ASL syllables contain distinctive features which may be accessed by phonological rules only in terms of their tiers and syllabic positions (e.g. syllable-initial, syllable-final), without further subdivision or organization. The segments are abstract timing slots at the bottom of the tree, onto which the phonological features are mapped, i.e. the terminal nodes in a feature geometry approach. Thus the question arises of how these two models should be distinguished.

6.2

Evidence related to syllable structure

Jantunen and Takkinen (2010) observe that there is no “direct phonetic evidence” to support the sequential segmental models. In fact, evidence against the segmental arrangement of internal syllable structure comes from a variety of experimental sources: tapping, slips of the hand, and backwards signing.

6.2.1 Tapping Spoken syllables have a rhythmic focus at the onset of the nuclear vowel (Allen 1972). That is, native English speakers who tap in time to speech cluster their taps at the stressed vowel onset. In a comparable study of ASL, native Deaf signers, native hearing signers and sign-naive hearing subjects were asked to “tap the rhythm” of five different three-sentence signed stories. Each story was presented 30 times. One story was repeated as the sixth condition (30 repetitions) for reliability; these conditions represent “tap the rhythm” (Allen et al. 1991). Finally, another one of the stories was repeated (30 repetitions) with new instructions to “tap the syllables.” Analysis of the tap responses in this condition showed that for all groups, taps are evenly distributed within syllables and do not differ from chance distribution. That is, no syllable-internal rhythmic focus is apparent (Wilbur and Allen 1991). This result is very crucial, and can only be predicted if the sign syllable is composed of constantly changing movement (smoothly changing muscular activity), meaning that there is no single point in time which attracts perceptual attention in the way that the onset of a spoken stressed vowel does, with large changes in muscular and acoustic energy (Allen 1972). The absence of such peaks is consistent with the proposal in the Prosodic Model that there is no further segmentation inside the sign syllable.

6.2.2 Slips of the hand Additional arguments against segmental models come from sign errors (Meier 1993; Wilbur 1993). English slips of the tongue tend to involve all the features of the segments involved (Fromkin 1971, 1973). If sign phonological features are distributed across segments, as suggested in segmental models, all features associated to each segment should be able to behave as a group. Therefore, parallel to speech, we might expect that the initial segments of two signs could switch with everything else remaining the same. In the corpus of 131 slips of the hand (Klima and Bellugi 1979), the predicted segmental switch did not occur. Instead,

Ronnie Wilbur

14

observed slips involved handshape, location, orientation, or handedness (one vs. two hands) features, with handshape involvement being the most common. In one slip involving BLACK and WHITE, the handshape sequence in WHITE (open fingers and thumb changing to closed fingers and thumb touching at tips) is anticipated in BLACK, with its regular handshape completely replaced by the handshape change from WHITE, whereas its location (at the forehead) and movement direction (brushing across) remained unaffected (Klima and Bellugi 1979: 139). What did not happen was a complete replacement of the initial handshape, location, and orientation of BLACK with those of WHITE, which would have created a form that started at the chest, not at the forehead. In none of their examples did the features act together as a group, as would be predicted from segmental models.

6.2.3 Backwards signing Backwards signing demonstrates that signers have access both to syllable sequences and to individual features within syllables which can be exchanged in temporal sequence, but not to units corresponding to segments as defined by the segmental models. This contrasts with the evidence from spoken language games and backwards speaking (Sherzer 1970; Cowan and Leavitt 1981, 1990; Cowan et al. 1982; Treiman 1983; Cowan et al. 1985; Cowan et al. 1987; Cowan 1989). Cowan et al. (1985: 680) report that fluent backwards talkers segment speech into “phonemic or syllabic” units, and then reverse their order. Their subjects fall into two groups, using either orthography or phonology as the basis for reversal. For example, for ‘terrace’, orthographic reversers would say /ekaret/, including the final “silent e” and adjusting the pronunciation of the letter “c” followed by back vowels to /k/. Phonological reversers would say /s>ret/, simply reversing phonological segment order. Data from backwards signing (Wilbur and Petersen 1997) provide evidence that signers treat monosyllabic signs in ways that are not compatible with segmental models. For example, Liddell’s (1984) representation for THINK in (5) consists of two segments MH, i.e. M-[approach](AP) followed by H with contact. The exchange of these segments should yield HM, i.e. H with contact followed by M-[approach]. Signers actually produce “contact” followed by “move away,” a result which would be predicted if the starting and ending locations of the movement are exchanged. That is, if movement is represented as a sequence of features, say [−contact] [+contact] (5b) (or [neutral] [forehead] rather than [approach]), then exchanging those features within-syllable to [+contact] [−contact] will result in the correct prediction of movement away from the forehead. Note that in Liddell’s model in (5a), there is a sequence of [−contact] [+contact], but these features cannot switch independently of the segments AP and H to which they are assigned. This is what we mean by having the segments at the top of the model. (5)

a.

Liddell’s model of THINK THINK Segment AP H Handshape 1 1 Orientation TI TI Location FH FH Contact – + Non-manual markings – –

Sign Syllables b.

15

Feature representation of change seen in backwards signing of THINK q

q [−contact] [+contact]

→

[+contact] [−contact]

Incorrect predictions from segmental models are more obvious with the sign FLY, represented as a single M segment. The predicted backwards version should be the same as the original, because there is nothing available to exchange. Backwards signing shows that the direction of movement of FLY is reversed (Figure 56.10), comparable to the movement reversal in THINK. This remains an inexplicable fact in segmental models which treat movement as a single M segment with its own feature matrix. The only recourse is to change the representation to H1MH2 and then reverse the two Hs, but evidence for the presence of those H segments would need to be provided. In any case, the lack of analogy with spoken segment sequences can be seen: the backwards form of cat /kæt/ is /tæk/, with the vowel unchanged. But, clearly, in the backwards form of FLY the movement has changed. Any segmental model containing M segments will have the same problem, because it is the phonological features associated with the movement that must be available to signers to be exchanged. In backwards signing, movements are consistently reversed by exchanging end specification with start specification, as though initial and final features are exchanged on their own tiers: end location with beginning location; end handshape with beginning handshape; end orientation with beginning orientation. Wilbur and Petersen (1997) argue that movement is not inside the syllable, but rather that movement is the syllable, a conception of “syllable” that takes movement as a dynamic gesture with only starting and ending specifications for the movement trajectory and no further linguistically meaningful internal specifications (see current arguments from gestural phonology approaches, e.g. Mauk and Tyrone 2008; Tyrone and Mauk 2008). For speech, Bagemihl (1989) and Pierrehumbert and Nair (1995) argue that sub-syllabic constituents, such as onset and rime or coda, do not participate in language game behavior, with Pierrehumbert and Nair extending this observation to phonological theory in general, claiming that these sub-syllabic constituents do not exist and that a flat syllable model, such as that proposed by Clements and Keyser (1983), is adequate to account for the facts. Bagemihl (1989: 485f.) notes that language

(a) normal Figure 56.10 signing

(b) backwards

The sign FLY, made (a) with normal movement and (b) in backwards

16

Ronnie Wilbur

games in cultures lacking phonemic alphabet writing systems do not use segments, only syllables. Furthermore, children do not use segmental language games until they are exposed to those writing systems. He suggests that alphabetic writing systems may be necessary for the development of metalinguistic awareness of segments as opposed to syllables. Brentari (1998) capitalizes on these and other observations by separating the specifications that do not change during a syllable (Inherent Features) from those that do (Prosodic Features). Each syllable has two timing slots, one after the other, representing sequentiality, and the Prosodic Features are associated accordingly. The Inherent Features spread across both slots. Thus, her timing slots (sequentiality) are at the bottom of the tree, whereas for Sandler and Liddell, the sequentiality is at the top of their models. Jantunen and Takkinen (2010) review the sign language studies, and note that there is no evidence for internally structured sequential segmental syllables of the kind found in spoken languages (such as an onset–nucleus–coda distinction). Hence there is no justification for positing an intermediate level between segment (referring here to the timing slots in Brentari’s model) and syllable, or more than two segments/slots per syllable. Finally, another benefit of the Prosodic Model for sign languages is that it provides seamless access to the prosodic hierarchy above the syllable.

7

What does simultaneity in syllable structure buy us?

It is time to turn our attention to the benefits of the notion of simultaneity in sign syllables, that is, what it accounts for that the other approaches do not. There are two important concepts that come from this model of syllable representation, namely the notion of syllable weight and the analogue to spoken language sonority. In addition, aspectual reduplication can be seen to operate prosodically on verb roots, whether one or two syllables.

7.1

Syllable weight

Consider the difference in speech between syllables of different structures CV, CVC, CVCC, CCVC, CCVCC. It is easy to identify an increase in syllable weight as more consonants are added to these syllables, even without knowing what those consonants are or the type of vowel in the nucleus or whether there is a distinction between short and long vowels or open syllable vowel lengthening (chapter 57: quantity-sensitivity). If sign syllables do not have the same internal structure as spoken language syllables, then is there a syllable weight distinction in sign languages, and, if so, how does it manifest itself? Brentari (1998) argues that there is a weight distinction in ASL, based on the number of simultaneous movements specified for the syllable. Syllables with one movement are light and those with two are heavy; more technically, a weight unit is constructed for every prosodic foot. With this analysis, she can explain the pattern of verbs that can and cannot take reduplication to form nouns, i.e. respectively light and heavy verbs. For example, the sign FLY in Figure 56.10a above is able to form a repeated nominalization for AIRPLANE because it is a light (one-movement) syllable. In contrast, syllables

Sign Syllables

17

with complex movements (for example, a path movement combined with a handshape change) cannot undergo reduplicated nominalization, even if the verb qualifies semantically. Similarly, activity verbs that form activity nouns with the addition of the feature [trilled movement] must have light syllable structure to start with (Brentari 1998: 242–243). Brentari also shows the correlation between verb heaviness and preference for sentence-final position, that is, the word order is sensitive to the weight of the verb, which is determined by the number of movements in the syllable (and if reduplicated, the number of syllables). In Brentari’s analysis, the maximum number of weight units per syllable is two. Using data from Finnish Sign Language (FinSL), Jantunen (2005; see also Jantunen and Takkinen 2010) argues that an extended system is necessary for FinSL because more than two weight units per syllable are possible if one takes the non-manual movements into account – in fact, three or four may be possible. For Jantunen, a movement is complex (not simple) if more than one articulator is involved. He is then able to make a weight distinction between two monosyllabic signs, MUSTA ‘black’ which has only path movement, hence one weight unit, and UJO ‘shy’, which has local movement accompanied by a head movement, and hence has two weight units. Thus, even though both are monosyllabic, the difference in weight results from the non-manual head specification. Another benefit of this line of reasoning is that in FinSL there are lexical items which are made entirely with non-manual articulation (there are a few in ASL also) – these would be assigned a single weight unit, and that is the desired result. One additional generalization can be stated: both FinSL and ASL prefer syllables with simple movements over complex movements (Brentari 1998; Jantunen and Takkinen 2010). This generalization would be lost from a segmental perspective on syllable structure.

7.2

Sonority

Sonority is not built on syllable weight, as a mora-based generalization might suggest (Perlmutter 1992). Brentari suggests that sonority be approached as multidimensional salience. She suggests that sonority is correlated articulatorily with closeness of the articulator to the body’s midline, and that articulation closer to the midline has greater visual salience than articulation further away. Thus, strengthening of visual salience (Enhancement Theory; Stevens and Keyser 1989) by choice of articulator higher up on the hierarchy (and likewise, reduction by choice lower down) is captured directly by the Prosodic Model in a way that segmental models cannot (Brentari 1998: 135). She suggests the following hierarchy, repeated here from (1) above: (6)

Brentari hierarchy shoulder joint

> elbow > joint

wrist > joint

base finger joints

> non-base finger joints

This distinction can be observed in cases where a movement can be articulated by different articulators, that is, if a wrist movement and an elbow movement

18

Ronnie Wilbur

can convey the same phonology. In such cases, if the movement is made by an articulator that is up the hierarchy, say an elbow joint replacing a wrist movement, then “proximalization” is said to occur, whereas if the articulator used is down the hierarchy, from elbow to wrist, then “distalization” is said to happen (see Mirus et al. 2001 for an empirical test of the factors involved). Crasborn (2001) likewise provided evidence in NGT for some of the factors relating to proximalization/ distalization. An important methodological aspect of Crasborn’s study is that it looked at fluent L1 signers, whereas Mirus et al. looked at L2 acquisition, for which the presence of sonority effects might be obscured by developmental performance factors. Looking at data from British Sign Language and the echoing of manual movements by the signer’s mouth, Woll (2001) suggests than non-manual articulations can also provide insight into the sonority hierarchy. Based on detailed investigation of these aspects of Finnish Sign Language, Jantunen (2005, 2006, 2007) suggests that non-manual movement should be included in the hierarchy, as in (7), from Jantunen (2005: 56): (7)

7.3

upper body and head > hands (including Brentari’s hierarchy) > mouth

Another perspective on reduplication: Templatic vs. prosodic

Klima and Bellugi (1979) treat reduplication as part of a templatic approach to aspectual modification (chapter 100: reduplication).1 They refer to formational terms such as Planar locus (horizontal, vertical), Cyclicity (repetition), Direction (e.g. upward, downward), Geometric array (line, arc, circle, other arrangement), Quality (small, large), and Manner (continuous, hold, restrained). Thus, each morphological function (e.g. iterative, durative) involves a template composed of some of these formational features. But the choices of feature combinations in each template are not explained. Similarly, Sandler (chapter 24: the phonology of movement in sign language) argues for a templatic approach to reduplication, using additional M (movement) or L (location) segments to account for differences in movement type or final holding (what Klima and Bellugi refer to as “end marking”). Further discussion of reduplication with Klima and Bellugi led us to an interesting separation of function for spatial and temporal formational properties, with the spatial properties providing information about the arguments of the verb and the temporal/rhythmic properties providing information about aspect on the verb (Wilbur et al. 1983). We speculated that reduplication could be analyzed the same way as in spoken languages. It took over 20 years to work it out, but a standard Base–Copy reduplication approach can be applied to sign languages using the Prosodic Model (Wilbur 2005, 2009). In Brentari’s model (8), the node dominating syllables and associated features is the root (chapter 24: the phonology of movement in sign language). 1

It is important to distinguish repetition from reduplication. Here, repetition is viewed as prosodically driven, to fill the needs of a prosodic foot. Lexicalized repetition creates nouns from verbs, with only two formations of the lexically meaningful movement required. Reduplication is aspectually driven.

Sign Syllables (8)

19

Brentari’s Prosodic Model of syllable structure root inherent features articulator non-manual

place of articulation

prosodic features setting ∆

manual H2

H1

path orientation ∆ aperture ∆ x

x q

Whether one syllable or two in a root, the Base for reduplication is the root, and the entire Base is copied. A simple example with Base and Copies is illustrated in Figure 56.11; planar difference indicates argument differences. There are, however, two modality differences between standard Base–Copy and what occurs in sign languages. First, multiple copies are common (Figure 56.11); indeed, a single copy implies “dual,” so aspectual reduplication typically has two or more copies of the Base. A second difference is that for many aspectual reduplications the hand must return to its initial position in the Base before it can articulate the Copy, thus the sequence is Base–Return–Copy. Aspectual reduplication is a combinatorial system of Base event (verb root), followed by Return to initial position, which reflects the time between the end of the Base event and the onset of the repeated event in aspects involving iteration (Wilbur 2005, 2009). Different aspects determine the size of the Return (smaller than, equal to or greater than the Base) (Table 56.1). Whether the shape includes a stopping point or is smooth (circular, elliptical) is dependent entirely on whether the verb is telic (contains a stop) or atelic (cannot stop) (Wilbur 2008).

Figure 56.11 Base–Copy reduplication schemata for apportionative external and internal. By permission of Ursula Bellugi, The Salk Institute for Biological Studies

20

Ronnie Wilbur Table 56.1 Combinations of Return options and Base event type yield aspectual inflections time between events

telic event root

atelic event root

habitual incessant iterative

durative n/a continuative

[return = root] [return < root] [return > root]

For telic events (Figure 56.12), when Return and Base are equal in size, there is the appearance of equal prominence on both (habitual); when the Return is smaller, there is a tendency for the Base to reduce as well (incessant). When the Return is larger than the Base, an arc is added (the morpheme EXTRA; Wilbur 2008). Thus, featurally, incessant aspect has [repeat] [return] [less than], habitual [repeat] [return] [equal] and iterative [repeat] [return] [greater than]. For atelic events (Figure 56.13), only two of the three options are possible, and the Base must be curved. Both of these requirements result from the absence of stops in the formation of atelic roots. Durative has [repeat] [return] [equal] and continuative has [repeat] [return] [greater than]. The atelic equivalent of the incessant, [repeat] [return] [less than], is not possible, because shortened movements would be perceptually equivalent either to stops, creating confusion with telics, or to trilled movement, which has a different interpretation (stative, not repeated). The difference between the modifications shown in Figures 56.12 and 56.13 is the Base root, which reflects the event structure in the semantics of the verb.

LOOK-AT

LOOK-ATincessant

LOOK-AThabitual

LOOK-ATiterative

Figure 56.12 LOOK-AT and three inflections. By permission of Ursula Bellugi, The Salk Institute for Biological Studies

LOOK

LOOK-durative

LOOK-continuative

Figure 56.13 LOOK-stative and durative and continuative aspects. By permission of Ursula Bellugi, The Salk Institute for Biological Studies

Sign Syllables

21

These observations are not captured by a templatic approach to reduplication. Sandler’s (1989, 2008) evidence for templatic analysis (for signs like BLOW-TOP, FAINT, and SNOOP) are all compatible with the generalization that aspectual reduplication copies the verb root. She argues that for the sign BLOW-TOP, which is a two-syllable compound created from HEAD and EXPLODE-OFF, EXPLODEOFF is copied but not HEAD. Her explanation is that the rightmost M is copied in reduplication, whereas in lexicalized monosyllabic (single M) signs like FAINT (originally from MIND+DROP), the whole form is repeated. But in the alternative analysis presented here it is expected that EXPLODE-OFF will be copied. Similarly, Sandle cites SNOOP (from NOSE+STICK-IN) as an exception, with the syllable associated with STICK-IN reduplicating, even though the entire form is monosyllabic (single M). An explanation for this is that the initial movement to the nose for NOSE is purely epenthetic (as for MOTHER (Figure 56.3), and one of the two versions of PARENTS in §4.3). That is, SNOOP starts its STICK-IN movement at the nose but does not return there for subsequent repetitions. If so, it might be appropriate to consider the initial location at the nose to be something akin to a prefixal location adjoined/cliticized onto the beginning location of STICK-IN, resulting from the compound-to-lexical-item reduction process, leaving STICK-IN to be copied by reduplication with its original location, as shows up in subsequent repetitions. This discussion highlights the kind of phonological level analysis that can be conducted with respect to syllables and their contents. Thus, factors other than phonology, such as verbal telicity and type of aspectual morphology, affect the final form of reduplicated signs (Wilbur 2005, 2009).

8

Conclusion

In this chapter, we started with two basic criteria for discussion of syllables in sign languages: linguistic meaningfulness and reliable measurement. To address the requirements of these criteria, we introduced the reader to relevant historical discussions of the status of syllables in sign languages. The key is that movement is the defining feature of a phonological syllable in sign languages. Then, we provided evidence that syllables in sign languages can be reliably measured. As for the criteria of linguistic meaningfulness, we reviewed several phenomena for which one has to make reference to the syllable for a reasonable account. Among those discussed were the phonologization of fingerspelled loan words, contact metathesis, handshape changes within and between signs, and some prosodic constraints. We then turned to the debate concerning the formal representation of sign syllables. We reviewed the sequential and simultaneous models of syllable representation. Data presented – tapping, slips of the hand, and backwards signing – strongly favor the Prosodic Model proposed by Brentari (1998), resulting in a syllable model that is internally different from those that exist in speech. We then considered the implications of this conclusion, especially since it goes against expectations of similarity between signed and spoken languages. Issues of sonority, syllable weight, and Base–Copy reduplication indicate that the syllable performs similarly in sign language and spoken language phonologies despite the internal differences in organization.

22

Ronnie Wilbur

To get to the point where we can use syllables for explaining other phenomena, we have needed a consistent and well-developed syllable model that makes empirically testable claims. This model has been tested in a variety of ways reviewed in this chapter, and there is more that has been omitted for reasons of space (variability of non-manual marking; judgments of well-formed syllables; Brentari and Wilbur 2008). The critical feature of the model is that it is a prosodic model, and the lowest level of the prosodic hierarchy is the syllable, for sign languages as well as spoken languages.

ACKNOWLEDGMENTS The preparation of this chapter was funded in part by NIH-NIDCD: “Prosodic features of American Sign Language” (1993–98); NIH-NIDCD: “Modeling the nonmanuals of American Sign Language” (2004–10); NSF: “A basic grammar of Croatian Sign Language” (2004–10); NSF: “Syllables in American Sign Language” (1984–87). A special acknowledgment to Diane Brentari for discussion in the process of developing this chapter, and to Kadir Gökgöz for suggestions and revisions.

REFERENCES Adams, Corinne. 1979. English speech rhythm and the foreign learner. The Hague: Mouton. Akamatsu, Carol. 1982. The acquisition of fingerspelling in pre-school children. Ph.D. dissertation, University of Rochester. Akamatsu, Carol. 1985. Fingerspelling formulae: A word is more or less than the sum of its letters. In Stokoe & Volterra (1985), 126–132. Allen, George. 1972. The location of rhythmic stress beats in English speech, Parts I & II. Language and Speech 15. 72–100, 179–195. Allen, George, Ronnie Wilbur & Brenda Schick. 1991. Aspects of rhythm in American Sign Language. Sign Language Studies 72. 297–320. Bagemihl, Bruce. 1989. The crossing constraint and “backwards languages.” Natural Language and Linguistic Theory 7. 481–549. Battison, Robbin. 1978. Lexical borrowing in American Sign Language. Silver Spring, MD: Linstok Press. Bellugi, Ursula & Susan Fischer. 1972. A comparison of sign language and spoken language: Rate and grammatical mechanisms. Cognition 1. 173–200. Blevins, Juliette. 1993. The nature of constraints on the nondominant hand in ASL. In Coulter (1993), 43–61. Brentari, Diane. 1990a. Licensing in ASL handshape change. In Lucas (1990), 57–68. Brentari, Diane. 1990b. Underspecification in American Sign Language phonology. Proceedings of the Annual Meeting, Berkeley Linguistics Society 16. 46–56. Brentari, Diane. 1993. Establishing a sonority hierarchy in American Sign Language: The use of simultaneous structure in phonology. Phonology 10. 281–306. Brentari, Diane. 1994. Prosodic constraints in American Sign Language. In Helen Bos & Trude Schermer (eds.) Sign Language Research 1994, 39–51. Hamburg: Signum Press. Brentari, Diane. 1996. Trilled movement: Formal representation and phonetic representation. Lingua 98. 43–71. Brentari, Diane. 1998. A prosodic model of sign language phonology. Cambridge, MA: MIT Press. Brentari, Diane & Laurinda Crossley. 2002. Prosody on the hands and face: Evidence from American Sign Language. Sign Language and Linguistics 5. 105–130.

Sign Syllables

23

Brentari, Diane & Howard Poizner. 1994. A phonological analysis of a Deaf Parkinsonian signer. Language and Cognitive Processes 9. 69–100. Brentari, Diane & Ronnie Wilbur. 2008. A cross-linguistic study of word segmentation in three sign languages. In Ronice Müller de Quadros (ed.) Sign languages: Spinning and unraveling the past, present and future – Theoretical Issues in Sign Language Research 9, 48–63. Petrópolis, Brazil: Editora Arara Azul. Chinchor, Nancy. 1978. The syllable in ASL. Paper presented at the MIT Sign Language Symposium. Chinchor, Nancy. 1981. Numeral incorporation in American Sign Language. Ph.D. dissertation, Brown University. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Corina, David. 1990. Handshape assimilation in hierarchical phonological representation. In Lucas (1990), 27–49. Coulter, Geoffrey. 1982. On the nature of ASL as a monosyllabic language. Paper presented at the 56th Annual Meeting of the Linguistic Society of America, San Diego. Coulter, Geoffrey. 1990. Emphatic stress in ASL. In Fischer & Siple (1990), 109–126. Coulter, Geoffrey (ed.) 1993. Current issues in ASL phonology. New York: Academic Press. Covington, Virginia. 1973. Features of stress in American Sign Language. Sign Language Studies 2. 39–58. Cowan, Nelson. 1989. Acquisition of Pig Latin: A case study. Journal of Child Language 16. 365–386. Cowan, Nelson & Lewis Leavitt. 1981. Talking backward: Exceptional speech play in late childhood. Journal of Child Language 9. 481–491. Cowan, Nelson & Lewis Leavitt. 1990. Speakers’ access to the phonological structure of the syllable in word games. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26. 45–59. Cowan, Nelson, Lewis Leavitt, Dominic Massaro & Raymond Kent. 1982. A fluent backward talker. Journal of Speech and Hearing Research 25. 48–53. Cowan, Nelson, Martin Braine & Lewis Leavitt. 1985. The phonological and metaphonological representation of speech: Evidence from fluent backward talkers. Journal of Memory and Language 24. 679–698. Cowan, Nelson, Cristy Cartwright, Carrie Winterowd & Molly Sherk. 1987. An adult model of preschool children’s speech memory. Memory and Cognition 15. 511–517. Crasborn, Onno. 2001. Phonetic implementation of phonological categories in sign language of the Netherlands. Ph.D. dissertation, University of Leiden. Fischer, Susan D. & Patricia Siple (eds.) 1990. Theoretical issues in sign language research, vol. 1: Linguistics. Chicago: University of Chicago Press. Friedman, Lynn A. 1974. On the physical manifestation of stress in the American Sign Language Unpublished ms., University of California, Berkeley. Friedman, Lynn A. 1976. Phonology of a soundless language: Phonological structure of American Sign Language. Ph.D. dissertation, University of California, Berkeley. Friedman, Lynn A. (ed.) 1977. On the other hand: New perspectives on American Sign Language. New York: Academic Press. Frishberg, Nancy. 1978. The case of the missing length. Communication and Cognition 11. 57–67. Fromkin, Victoria A. 1971. The non-anomalous nature of anomalous utterances. Language 47. 27–52. Fromkin, Victoria A. 1973. Slips of the tongue. Scientific American 229. 109–117. Geraci, Carlo. 2009. Epenthesis in Italian Sign Language. Sign Language and Linguistics 12. 3–51. Green, Kerry. 1984. Sign boundaries in American Sign Language. Sign Language Studies 42. 65–91.

24

Ronnie Wilbur

Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: MIT Press. Hoequist, Charles, Jr. 1983. Syllable duration in stress-, syllable- and mora-timed languages. Phonetica 40. 203–237. Hulst, Harry van der. 1993. Units in the analysis of signs. Phonology 10. 209–241. Jantunen, Tommi. 2005. Mistä on pienet tavut tehty? Analyysi suomalaisen viittomakielen tavusta prosodisen mallin viitekehyksessä [What are the little syllables made of? A Prosodic Model account of the Finnish Sign Language syllable]. Licentiate thesis, University of Jyväskylä, Finland. Jantunen, Tommi. 2006. The complexity of lexical movements in FinSL. SKY Journal of Linguistics 19. 335–344. Jantunen, Tommi. 2007. Tavu suomalaisessa viittomakielessä [The syllable in Finnish Sign Language] (with English abstract). Puhe ja kieli 27. 109–126. Jantunen, Tommi & Ritva Takkinen. 2010. Syllable structure in sign language phonology. In Diane Brentari (ed.) Sign languages, 312–331. Cambridge: Cambridge University Press. Johnson, Robert. 1986. Metathesis in American Sign Language. Paper presented at the Conference on Theoretical Issues in Sign Language Research, Rochester, NY. Kegl, Judy & Ronnie Wilbur. 1976. When does structure stop and style begin? Syntax, morphology and phonology vs. stylistic variation in American Sign Language. Papers from the Annual Regional Meeting, Chicago Linguistic Society 12. 376–396. Klima, Edward S. & Ursula Bellugi. 1979. The signs of language. Cambridge, MA: Harvard University Press. Kooij, Els van der. 2002. Phonological categories in Sign Language of the Netherlands: The role of phonetic implementation and iconicity. Ph.D. dissertation, University of Leiden. Kooij, Els van der & Onno Crasborn. 2008. Syllables and the word-prosodic system in Sign Language of the Netherlands. Lingua 118. 1307–1327. Liddell, Scott K. 1978. Non-manual signals and relative clauses in ASL. In Siple (1978), 59–90. Liddell, Scott K. 1980. American sign language syntax. The Hague: Mouton. Liddell, Scott K. 1984. THINK and BELIEVE: Sequentiality in American Sign Language. Language 60. 372–399. Liddell, Scott K. & Robert E. Johnson. 1989. American Sign Language: The phonological base. Sign Language Studies 64. 197–277. Lucas, Ceil (ed.) 1990. Sign language research: Theoretical issues. Washington, DC: Gallaudet University Press. Mauk, Claude & Martha Tyrone. 2008. Sign lowering as phonetic reduction in American Sign Language. Proceedings of the 8th International Speech Production Seminar, 185–188. Available (September 2010) at http://issp2008.loria.fr/proceedings.html. Strasbourg. Meier, Richard. 1993. A psycholinguistic perspective on phonological segmentation in sign and speech. In Coulter (1993), 169–188. Miller, Christopher. 1997. Phonologie de la langue des signes québecoise: Structure simultanée et axe temporel. Ph.D. dissertation, Université de Québec à Montréal. Mirus, Gene, Christian Rathmann & Richard Meier. 2001. Proximalization and distalization of sign movement in adult learners. In Valery L. Dively, Melanie Metzger, Sarah Taub & Anne Marie Baer (eds.) Signed languages: Discoveries from international research, 103–119. Washington, DC: Gallaudet University. Newkirk, Donald. 1979. The form of the continuative aspect on ASL verbs. Unpublished ms., Salk Institute for Biological Studies, La Jolla, CA. Reprinted 1998 in Sign Language and Linguistics 1. 75–80. Newkirk, Donald. 1980. Rhythmic features of inflection in American Sign Language. Unpublished ms., Salk Institute for Biological Studies, La Jolla, CA. Reprinted 1998 in Sign Language and Linguistics 1. 81–100.

Sign Syllables

25

Newkirk, Donald. 1981. On the temporal segmentation of movement in American Sign Language. Unpublished ms., Salk Institute for Biological Studies, La Jolla, CA. Reprinted 1998 in Sign Language and Linguistics 1. 59–97. Padden, Carol. 1993. Response to Sandler’s “Linearization of phonological tiers in ASL.” In Coulter (1993), 131–134. Perlmutter, David. 1992. Sonority and syllable structure in American Sign Language. Linguistic Inquiry 23. 407–442. Pierrehumbert, Janet B. & Rami Nair. 1995. Word games and syllable structure. Language and Speech 38. 77–114. Sandler, Wendy. 1986. The spreading hand autosegment of American Sign Language. Sign Language Studies 50. 1–28. Sandler, Wendy. 1989. Phonological representation of the sign: Linearity and non-linearity in American Sign Language. Dordrecht: Foris. Sandler, Wendy. 1993. A sonority cycle in American Sign Language. Phonology 10. 243–279. Sandler, Wendy. 2008. The syllable in sign language: Considering the other natural modality. In Barbara Davis & Kristine Zajdo (eds.) The syllable in speech production, 379–408. New York: Taylor Francis. Sandler, Wendy & Diane Lillo-Martin. 2006. Sign language and linguistic universals. Cambridge: Cambridge University Press. Schmidt, Richard & Tim Lee. 2005. Motor control and learning: A behavioral emphasis. Urbana, IL: Human Kinetics. Sherzer, Joel. 1970. Talking backwards in Cuna: The sociological reality of phonological descriptions. Southwestern Journal of Anthropology 26. 343–353. Siple, Patricia (ed.) 1978. Understanding language through sign language research. New York: Academic Press. Stevens, Kenneth N. & Samuel J. Keyser. 1989. Primary features and their enhancement in consonants. Language 65. 81–106. Stokoe, William C. 1960. Sign language structure: An outline of the visual communication systems of the American deaf. Silver Spring, MD: Linstok Press. Stokoe, William C. & Virginia Volterra (eds.) 1985. SLR ‘83: Proceedings of the 3rd International Symposium on Sign Language Research. Rome: Consiglio Nazionale delle Ricerche. Supalla, Ted & Elissa Newport. 1978. How many seats in a chair? The derivation of nouns and verbs in American Sign Language. In Siple (1978), 91–132. Treiman, Rebecca. 1983. The structure of spoken syllables: Evidence from novel word games. Cognition 15. 49–74. Tyrone, Martha & Claude Mauk. 2008. Sign lowering in ASL: The phonetics of wonder. Paper presented at the SignTyp Conference, Storrs, CT. Tyrone, Martha, Louis Goldstein & Gaurav Mathur. 2007. Movement kinematics and prosody in American Sign Language. Poster presented at the 20th Annual CUNY Conference on Human Sentence Processing, La Jolla, CA. Valli, Clayton & Ceil Lucas. 1992. Linguistics of American Sign Language. Washington, DC: Gallaudet University. Weast, Traci. 2008. Questions in American Sign Language: A quantitative analysis of raised and lowered eyebrows. Ph.D. dissertation, University of Texas, Arlington. Wilbur, Ronnie. 1979. American Sign Language and sign systems: Research and applications. Baltimore: University Park. Wilbur, Ronnie. 1985. Towards a theory of “syllable” in signed languages: Evidence from the numbers of Italian Sign Language. In Stokoe & Volterra (1985), 160–174. Wilbur, Ronnie. 1987. American Sign Language: Linguistic and applied dimensions. Boston: College Hill Press. Wilbur, Ronnie. 1990a. An experimental investigation of stressed sign production. International Journal of Sign Language 1. 41–60.

26

Ronnie Wilbur

Wilbur, Ronnie. 1990b. Why syllables? What the notion means for ASL research. In Fischer & Siple (1990), 81–108. Wilbur, Ronnie. 1993. Segments and syllables in ASL phonology. In Coulter (1993), 135–168. Wilbur, Ronnie. 1994. Eyeblinks and ASL phrase structure. Sign Language Studies 84. 221–240. Wilbur, Ronnie. 1999a. Metrical structure, morphological gaps, and possible grammaticalization in ASL. Sign Language and Linguistics 2. 217–244. Wilbur, Ronnie. 1999b. Stress in ASL: Empirical evidence and linguistic issues. Language and Speech 42. 229–250. Wilbur, Ronnie. 2005. A reanalysis of reduplication in American Sign Language. In Bernhard Hurch (ed.) Studies in reduplication, 593–620. Berlin & New York: Mouton de Gruyter. Wilbur, Ronnie. 2008. Complex predicates involving events, time and aspect: Is this why sign languages look so similar? In Josep Quer (ed.) Signs of the time: Selected papers from TISLR 2004, 217–250. Hamburg: Signum. Wilbur, Ronnie. 2009. Effects of varying rate of signing on ASL manual signs and nonmanual markers. Language and Speech 52. 245–285. Wilbur, Ronnie & George Allen. 1991. Perceptual evidence against internal structure in ASL syllables. Language and Speech 34. 27–46. Wilbur, Ronnie & Susan Nolen. 1986. Duration of syllables in American Sign Language. Language and Speech 29. 263–280. Wilbur, Ronnie & Lesa Petersen. 1997. Backwards signing and ASL syllable structure. Language and Speech 40. 63–90. Wilbur, Ronnie & Brenda Schick. 1987. The effects of linguistic stress on ASL signs. Language and Speech 30. 301–323. Wilbur, Ronnie, Edward Klima & Ursula Bellugi. 1983. Roots: On the search for the origins of signs in ASL. Papers from the Annual Regional Meeting, Chicago Linguistic Society 19. 314–336. Wilcox, Sherman. 1992. The phonetics of fingerspelling. Amsterdam & Philadelphia: John Benjamins. Woll, Bencie. 2001. The sign that dares to speak its name: Echo phonology in British Sign Language. In Penny Boyes Braem & Rachel Sutton-Spence (eds.) The hands are the head of the mouth: The mouth as articulator in sign languages, 87–98. Hamburg: Signum.

57

Quantity-sensitivity Draga Zec

1

Introduction

Quantity-sensitivity is an important property of prosodic constituents, which are subclassified along this dimension as either light or heavy. In a typical hierarchical organization of prosodic units, as in (1) (Selkirk 1978, 1980; Nespor and Vogel 1986), each of the prosodic levels may be instantiated by constituents that vary in length, segment quality, or structural complexity (see chapter 33: syllable-internal structure; chapter 40: the foot; chapter 51: the phonological word; chapter 84: clitics; chapter 50: tonal alignment). (1)

Prosodic hierarchy Prosodic phrase Prosodic word Foot Syllable

This variation, in at least some of its aspects, introduces distinctions in quantity among constituents at the same level of the hierarchy, evidenced by distinctions in phonological behavior. While quantity-sensitivity is most clearly manifested at the level of the syllable, other prosodic levels exhibit this property as well. Quantity-sensitivity characterizes a wide range of phonological phenomena, including stress, tone, poetic meter, and various prosodic effects on morphosyntax. Moreover, quantity-sensitivity can be manifested either as a binary or as a scalar property. For these and other reasons to be addressed in this chapter, prosodic quantity needs to have its place in the formal representation of prosody, and is a central issue in any discussion of phonological representations. The chapter is organized as follows: §2 addresses crucial aspects of weightsensitivity in the syllable, providing a typology of weight patterns supported by a wide range of attested cases. §3 focuses on formal representations of the syllable and its weight, while §4 addresses the relevance of vowel length for quantity-sensitivity. §5 shows that weight distinctions could be binary in some languages, and multivalued in others. §6 documents inconsistencies in weight The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0057

Draga Zec

2

patterns, both with respect to phonological processes and phonological contexts. §7 addresses quantity-sensitivity in feet, focusing on binary patterns, and §8 focuses on scalar patterns of quantity. §9 touches upon quantity-sensitivity at the higher levels of the prosodic hierarchy, in prosodic words and prosodic phrases. §10 offers some remarks on markedness, and §11 concludes.

2

The syllable and quantity-sensitivity

Phonological quantity is primarily associated with the syllable. One of the traditional classifications of syllables is into those that are light and those that are heavy. This distinction is motivated on empirical grounds and is brought to relief by a number of quantity-sensitive phonological phenomena, including stress, tone, and poetic meter. We focus here only on those languages that do exhibit quantitysensitivity at the syllable level, as this is not a universal prosodic property. In §2.1 we present a paradigm case of syllable weight, and then, in §2.2, turn to the typology of syllable-based quantity-sensitivity, supported by a wide range of phonological phenomena.

2.1

A paradigm case of quantity-sensitivity

Quantity-sensitivity has figured prominently in studies of classical languages and their prosody (Allen 1973). Latin provides a paradigm case of quantity-sensitivity, already known to the early grammarians such as Quintilian. Latin stress (chapter 39: stress: phonotactic and phonetic evidence) is quantity-sensitive, as illustrated in (2). If the penultimate syllable is heavy it is stressed, as in (2a), but if it is light the antepenultimate syllable is stressed, as in (2b). (2)

Latin stress a.

If the penultimate syllable is heavy, it is stressed. for’tena ‘fortune (nom sg)’ ’gaudbns ‘rejoicing (nom sg)’ gau’dentem ‘rejoicing (acc sg)’

b.

If the penultimate syllable is light, the antepenultimate is stressed. ’anima ‘soul (nom sg)’

Syllables that function as light are of the CV type, containing a short vowel, as the penultimate syllable in (2b). Syllables that function as heavy are more diverse, as shown in (2a). They are either of the CVV type, containing a long vowel or a diphthong, or of the CVC type, with a short vowel followed by a consonant. Significantly, syllables that are functionally equivalent may differ in their segmental content (chapter 54: the skeleton). Within the set of heavy syllables, differences in segmental content are found not only across CVV and CVC syllables but also within the class of CVV syllables, which may contain either a long vowel or a diphthong. In this case, as in many others, onset consonants are excluded from the computation of weight (chapter 55: onsets). This same pattern of syllable weight also figures in Latin poetic meter. In one of the meters of Horace’s Odes, known as the First Asclepiad (borrowed from

Quantity-sensitivity

3

Greek), a line of verse contains a sequence of metrical positions that admit either heavy or light syllables (marked as – and ∪, respectively, with „ marking the caesura), as in (3). In (3a), the first three metrical positions, all heavy, are filled with CVV syllables that contain either a diphthong or a long vowel, while in (3b) these same metrical positions are filled with CVC syllables. The light metrical positions, such as the next two, are filled in both lines with CV syllables. (The remaining metrical positions are filled in the same fashion, with the exception of the last vowel of tollere in (3b) which is elided, and therefore not scanned.) (3)

Latin poetic meter: Horace, First Asclepiad a.

– – – ∪∪ – „ – ∪∪ – ∪ x Maecbnas atavcs bdite rbgibus ‘O Maecenas, born from kingly ancestors!’ (Odes 1.1.1)

b.

– – – ∪ ∪– „ – ∪ ∪ –∪ x Certat tergemincs tollere honoribus ‘vies to lift [him] with triple magistracies.’ (Odes 1.1.8)

Thus, in Latin, both stress and poetic meter are sensitive to distinctions in quantity, with weight characterized identically in the two phonological subsystems.

2.2

Patterns of quantity-sensitivity

The system of quantity in Latin exemplified in (2) and (3) was taken in much relevant work to be the standard mode of computing quantity, with broad empirical support. This is how syllable weight was characterized in KuryÓowicz (1948) and later in Newman (1972), among others. Newman (1972), in particular, identifies quantity-sensitivity in a number of languages, all exemplifying the pattern of quantity with light CV and heavy CVV and CVC syllables. In addition to Latin, the list includes Classical Greek, Finnish, Estonian, Classical Arabic, and Gothic, as well as three Chadic languages, Bolanci, Kanakuru, and Hausa. In fact, a number of researchers stated important generalizations about quantitysensitivity solely in terms of the Latin pattern of weight (e.g. Kiparsky 1979, 1981; Halle and Vergnaud 1980; Clements and Keyser 1983). The Latin pattern, however, is not the only empirically attested mode of computing quantity, as shown in Hyman’s (1977) broad survey of stress systems and in much later work. In what follows we present the range of quantity patterns that have been empirically attested, and a typology of weight distinctions.

2.2.1 Weight patterns: A typology McCarthy (1979) made the crucial theoretical statement that quantity-sensitivity can be instantiated in more than one way. In addition to the Latin weight pattern, with light CV and heavy CVV and CVC syllables as in (4a), henceforth type 1, there is a further weight pattern, one in which only CVV syllables are heavy and both CV and CVC syllables are light, as in (4b), henceforth type 2. A number of languages were identified to belong to this weight type: for example, Huasteco Mayan in Hyman (1977) and Yidiny and Tiberian Hebrew in McCarthy (1979). Many more such cases figure in Hayes (1980, 1995) and Gordon (2006).

Draga Zec

4 (4)

Possible weight patterns (first approximation) a. b.

Type 1: heavy CVV, CVC vs. light CV Type 2: heavy CVV vs. light CV, CVC

Thus the weight of CVC syllable is “parameterized”: while in (4a) such syllables form a natural class with CVV, in (4b) they form a natural class with CV. This is crucially due to the status of the final consonant in a CVC syllable, which contributes to weight in (4a), but not in (4b) types of languages. McCarthy (1979) further identifies an important implicational relation: a language with heavy CVC syllables also has heavy CVV syllables. This supports the prediction about the following impossible weight pattern: no language can have heavy CVC but light CVV syllables. But computation of quantity can be even more fine-grained than in (4), and in order to show this we invoke the sonority of segments. In particular, vowels are more sonorous than consonants, and within the class of consonants, sonorants (chapter 8: sonorants) are more sonorous than obstruents. (For a general discussion of sonority, see chapter 49: sonority.) In addition to the two weight systems in (4), one in which all consonants contribute to weight, and one in which no consonants contribute to weight, there is also a type 3 system, in which only some consonants contribute to weight (see Prince 1983; Zec 1988, 1995). In such split systems, the subset of consonants contributing to weight is generally sonorants. Such a case is exemplified by Kwakw’ala, to be discussed in §2.2.2, in which heavy syllables are CVV and CVR (sonorant), while light syllables are CV and CVO (obstruent). This yields the implicational relation that if CVO syllables are heavy, so are CVR syllables; and excludes the impossible system, with CVV and CVO syllables being heavy, and CV and CVR syllables being light (Zec 1995). Furthermore, while other splits in the hierarchy should in principle be possible, say, with liquids being weight-bearing to the exclusion of obstruents and nasals, such systems have not been attested. Only major splits within the sonority hierarchy appear to be exploited for distinctions in quantity, those in particular that correspond to splits imposed by the major class features (Chomsky and Halle 1968). To summarize, a basic typology of weight patterns is given in (5). In type 1 languages, all segments contribute to weight, so that both CVV and CVC syllables are heavy; in type 2 languages only vowels are weight-bearing, which makes CVC syllables light; and in type 3 languages vowels and sonorant consonants are weight-bearing, to the exclusion of obstruents. The set of weight-bearing segments follows the sonority hierarchy: if a less sonorous segment contributes to weight, so does a more sonorous segment. (5)

Typology of weight patterns Type 1 Type 2 Type 3

Heavy CVV, CVC CVV CVV, CVR

Light CV CV, CVC CV, CVO

A special case of type 2 languages is those that lack CVC syllables. In a syllable inventory including only CV and CVV syllables, the former are light

Quantity-sensitivity

5

and the latter are heavy, as in Fijian (Hayes 1995; among others). Likewise, a special case of type 3 languages consists of those that lack CVO syllables, with an inventory that includes light CV and heavy CVV and CVR syllables, as in Tiv (Zec 1995); or Manam, in which the set of heavy syllables includes CVV and CVN (nasal), but excludes syllables closed with liquids (Lichtenberk 1983; Buckley 1998). Crucially, the onset is excluded from the computation of weight: the number of segments in the onset does not affect the weight status of a syllable. This empirically grounded property of onsets will need to be captured in the representation of the syllable, an issue to be addressed in §3. But although broadly attested, this is not a universal property of onsets; see chapter 55: onsets for cases of weight-sensitive onsets, which constitute counterexamples to this claim. To conclude, it has been shown that there is a measure of language-specificity, with different modes of quantity computation employed in different languages. It has also been shown that there is an implicational relation across occurring weight patterns, or, more specifically, across the sets of heavy syllables in different languages, as in (6): (6)

Implicational relations among heavy syllables a. b.

If a language has heavy CVC syllables, it also has heavy CVV syllables. If a language has heavy CVO syllables, it also has heavy CVR syllables.

2.2.2 Weight patterns: Case studies The three patterns of quantity in (5) are documented below with two types of quantity-sensitive phonological phenomena: stress and tone. We begin with stress, which provides the most striking cases of quantity-sensitivity. It should be noted though that only some stress systems are quantity-sensitive. According to Gordon’s (2006: 20–21) extensive survey, based on 408 languages, 310 languages have culminative accent systems. Out of those, 136 (43.9 percent) exhibit quantitysensitivity, and 86 belong to one of the three weight systems we exemplify here. Languages with quantity-sensitive stress show a clear preference for placing stress on heavy syllables (cf. Hyman 1977; Hayes 1980, 1995; Halle and Vergnaud 1987; Halle and Idsardi 1995). Simply stated, heavy syllables attract stress (Prince 1990). Moreover, languages with quantity-sensitive stress systems are of either type 1 or type 2, and rarely of type 3. The Latin stress pattern illustrated in §2.1, which belongs to type 1, is found in a number of languages. Out of 86 languages with quantity-sensitive stress in Gordon’s survey, 42 languages are of type 1. It is found, for example, in Modern Classical Arabic (as described in Ryding 2005), where stress falls on the penultimate heavy syllable, CVV or CVC, otherwise on the antepenultimate syllable. Note, however, the pattern in Classical Arabic, where stress falls on the rightmost (non-final) heavy syllable, as in (7a), otherwise on the first syllable, as in (7b) (McCarthy 1979, and the references therein).1 1

Final syllables have a special status, in at least two respects. Stress does not fall on final CVC syllables, but CVVC and CVCC syllables, which are only found word-finally, do bear stress. The special behavior of final elements is a more general issue, to be addressed in §6.2.

Draga Zec

6 (7)

Type 1: Classical Arabic a.

b.

ki’taabun manaa’diilu ju’œaariku ’mamlakatun ’kataba ’balapatun

‘book (nom sg)’ ‘kerchiefs (nom)’ ‘he participates’ ‘kingdom (nom sg)’ ‘he wrote’ ‘date (nom sg)’

The type 1 weight pattern is also noted in English, although the overall stress system is rife with idiosyncrasies. A small portion of the English lexicon, the set of underived nouns, has a relatively regular stress pattern: stress falls on the heavy penult, either CVV as in e’litist, ma’rina, and Ari’zona, or CVC as in a’genda, a’malgam, and co’nundrum; otherwise on the antepenult, as in ’discipline, ’labyrinth, and A’merica (Hayes 1982). This stress pattern is again reminiscent of Latin. Many more type 1 stress systems are documented in Hayes (1995) and Gordon (2006). Quantity-sensitive stress systems of type 2 are evidenced in a wide range of languages, just like type 1 (Hayes 1995; Gordon 2006; among others): 40 out of 86 quantity-sensitive stress systems in Gordon’s survey. It is found, for example, in the Mongolian language Buriat (Poppe 1960; Walker 1996), illustrated in (8): stress falls on the initial syllable in words with no long vowels, as in (8a), and on the rightmost non-final heavy syllable in words with more than one long vowel or diphthong, as in (8b). If a word has only one CVV syllable, stress falls on that syllable even if it is final, as in (8c). Note that CVC syllables figure in the language, yet do not attract stress, for example, the third syllable in /ta’ruulagdaxa/, in (8b). (8)

Type 2: Buriat a. b.

c.

’xada mo’rjooroo dalai’gaaraa ta’ruulagdaxa xa’daar

‘mountain’ ‘by means of his own horse’ ‘by one’s own sea’ ‘to be adapted to’ ‘through the mountain’

Another type 2 system is Huasteco Mayan (Larsen and Pike 1949; Hyman 1977; Hayes 1995: 296): stress falls on the rightmost CVV syllable, otherwise on the initial CV syllable. Again, CVC syllables pattern with CV rather than CVV syllables. And, in Aguacatec Mayan (McArthur and McArthur 1956; Hayes 1980, 1995), stress falls on a CVV syllable regardless of its position within a word, as in (9a); stress is final in words with no long vowels, as in (9b).2 (9)

Type 2: Aguacatec Mayan stress a.

2

Forms with CVV syllables ?in’ta( ‘my father’ ’Œi:bah ‘meat’ ’?e:q’um ‘carrier’

All examples are from McArthur and McArthur (1956), who list no cases of final CV syllables. Also, they claim that stress falls on the rightmost CVV syllable, yet no words with more than one CVV syllable are found in this source.

Quantity-sensitivity b.

7

Forms with no CVV syllables wu’qan ‘my foot’ ?al’k’om ‘thief’ tpil’ta? ‘courthouse’

Quantity-sensitive stress systems are very rarely of type 3. In Gordon’s (2006) survey, only four out of 86 languages are of type 3. Here we illustrate the distribution of stress in Kwakw’ala, in which CVV and CVR syllables pattern as heavy, while CV and CVO pattern as light (Boas 1947; also Zec 1988 and references therein). Stress falls on the leftmost heavy syllable, either CVV, in (10a), or CVR, in (10b). In words that contain only light syllables, CV or CVO, stress is final, as in (10c) and (d). (10)

Type 3 language: Kwakw’ala ’qa(sa ’n’a(la ’ts’e(kwa t’H’li(dzu b. ’m’Hnsa ’dHlxa ’dzHmbHtHls mH’xHnxHnd c. nH’pa bH’ha m’HkwH’la ts’HxH’laH d. ts’Ht’xa tHO’ts’a kw’Hs’xa

a.

‘to walk’ ‘day’ ‘bird’ ‘large board on which fish are cut’ ‘to measure’ ‘damp’ ‘to bury in hole in ground’ ‘to strike edge’ ‘to throw a round thing’ ‘to cut’ ‘moon’ ‘to be sick’ ‘to squirt’ ‘to warm oneself’ ‘to splash’

The three weight patterns in (5) can be further exemplified with tonal phenomena, those provided by languages with lexical, i.e. contrastive tone, in simpler systems commonly High, or High and Low (chapter 45: the representation of tone). Quantity-sensitive tonal phenomena differ substantially from quantity-sensitive stress.3 Crucial evidence for quantity-sensitivity comes from the so-called contour tones. If no more than one tone is sponsored by a light syllable and no more than two by a heavy one, we can say that multiple tones, standardly referred to as contour tones, may occur on heavy, but not on light, syllables. In other words, we focus on those languages in which a light syllable has one tone-bearing unit and a heavy syllable has two (see Zhang 2002 for different characterizations of

3

It is not typical for tone to be attracted to a heavy syllable, although some cases have been interpreted in this light. Thus Hopi, as described in Jeanne (1982), has been interpreted as a quantity-sensitive stress system (Hayes 1995): stress occurs on initial heavy syllables, either CVV or CVC, otherwise on non-final peninitial syllables; stress is initial in all disyllables. However, because stress is realized as tonal prominence, this system has also been interpreted as a tonal system in which High tone is attracted to the initial heavy syllable, otherwise to the second syllable, if non-final (Yip 2002: 245). This stress-like behavior of tone, if indeed correctly interpreted, is truly atypical.

Draga Zec

8

contour tones).4 We further focus on those languages in which the mapping between tones and tone-bearing units is fairly straightforward: a tone-bearing unit may be associated with at most one tone. With this background, we turn to the evidence for the three weight patterns coming from the tonal domain. We again rely on Gordon’s (2006: 32–33) survey: out of 408 languages in his survey, 111 use contrastive tone and, of those, 61 use tone in a quantity-sensitive mode. Type 2 and type 3 weight patterns are widely exploited by weight-sensitive tonal phenomena, while type 1 is rarely associated with quantity-sensitive tone. Type 2 pattern is found in 28 languages (four without CVC syllables), and type 3 is found in 30 languages. In type 2 languages, contour tones occur on CVV syllables, but are absent from both CV and CVC syllables. In Navajo, contour tones occur only on CVV syllables as in (11a), while simple tones occur on all syllable types; (11b) exemplifies the absence of contour tones on CV and CVC syllables (Zhang 2002, based on Young and Morgan 1987). (11)

Type 2: Navajo contour tone a.

b.

sáànìì hákòónèè? tèíl?á háá?ált’è? pìkh ìn

‘old woman’ ‘let’s go’ ‘they extend’ ‘exhumation’ ‘his house’

Another type 2 language is Ju|’hoansi, in which, as reported in Miller-Ockhuizen (1998; also Zhang 2002), contour tones are found only on long vowels and diphthongs, but not on CV syllables or syllables closed with nasals (the only type of closed syllable in the language). Type 3 weight pattern is exemplified by a number of languages, including Nama (Khoisan), Lithuanian (Indo-European), and Tiv (Niger-Congo). Lithuanian has a pitch accent system, in which contour tones appear on heavy, but not on light, syllables. In particular, a Low High tonal contour, the so called circumflex accent, occurs on heavy syllables: CVV, as in /víinas/ ‘wine’, /zúikas/ ‘rabbit’, and CVR, as in /gársas/ ‘sound’, /bálsas/ ‘voice’, and /lánkas/ ‘rainbow’. Syllables that pattern as light are CV and CVC, and those that pattern as heavy are CVV and CVR (Zec 1995 and references therein). We now turn to the tonal evidence for the type 1 weight pattern. Contour tones are rare, and phonetically difficult to realize, on syllables closed with an obstruent. In his broad survey of quantity-sensitive tone, Gordon (2006) documents only three such cases: Hausa, Luganda, and Musey. Zhang (2002: 51) also lists Ngizim, and Yip (2002: 141–142) mentions the Nilo-Saharan language Kunama (Eritrea). Here we present evidence from Hausa, based on Gordon’s (2006) experimental data. Hausa has three tones, two level tones, High and Low, and a contour High Low tone. As shown in (12), on the targeted initial syllables, the two level tones occur on all syllable types, while the contour tone occurs on CVV,

4

Note, however, that tone languages vary as to what constitutes a tone-bearing unit. What we described here is one of several modes of selecting a tone-bearing unit. On tone and tone-bearing units, see chapter 45: the representation of tone.

Quantity-sensitivity

9

CVR and CVO, but not on CV syllables. That is, the contour tone occurs on heavy, but not on light syllables. (12)

Type 1: Hausa contour tone CV CVV CVR CVO

L fàsá( mà(má( ràndá( fàskí(

H sáfú( rá(ná( mándá( máskó(

HL – lâ(lá( mântá( râssá(

Of interest here is the fact that while sonorants, both vowels and consonants, are capable of phonetically realizing pitch, obstruents are not. As shown by Gordon (2006: 92), although the phonological weight of the CVO syllable provides the two tone-bearing units required for the realization of contour tone, the contour is phonetically realized on the vowel, which in this case has greater duration. No comparable increase in duration is evidenced in CVV and CVR syllables with contour tones. Thus, Hausa presents an interesting case of a mismatch between phonology and phonetics. Other phonological phenomena that provide evidence for quantity-sensitivity include vowel shortening in closed syllables, to be addressed in §6, as well as compensatory lengthening (see chapter 64: compensatory lengthening) and poetic meter, which in §2.1 served as evidence for Latin. Onsets may on occasion exhibit quantity-sensitivity; for such cases, see chapter 55: onsets and chapter 47: initial geminates.

3

Representation of syllable quantity

The relevance of quantity-sensitivity, as well as its representation, was clearly recognized in early theoretical approaches to phonology. Both Jakobson (1931) and Trubetzkoy (1939) document weight distinctions among syllables, and cast them in terms of the unit of weight traditionally referred to as the mora: a light syllable contains one mora, and a heavy syllable contains two moras. Quantity-sensitivity was also recognized by KuryÓowicz (1948), who pursued the characterization of quantity in configurational terms, that is, in terms of a subconstituent of the syllable, the rhyme, whose structure is branching for heavy, and non-branching for light syllables. These two theoretical approaches to quantity-sensitivity, one in terms of constituency and the other in terms of arboreal configuration, emerged again in the 1970s and 80s, as competing representations of syllable weight, as well as the weight of other constituents in the prosodic hierarchy. These two approaches both express an important intuition: that quantity formally corresponds to a binary structure. This will emerge as highly relevant in the representation of the syllable and its internal structure. This is also relevant for the representation of feet, as will be shown in §7. The questions to be addressed in this section are: (i) how is weight computed from the representation of the syllable?; and (ii) how are different weight patterns represented? (For a general discussion of syllable structure and its representation, see chapter 33: syllable-internal structure.)

10

3.1

Draga Zec

Quantity represented in configurational terms

We begin with the configurational approach to syllable weight. In the representation in (13), the syllable branches into an onset and a rhyme, with the latter obligatorily dominating the nucleus and, optionally, the coda. The sub-syllabic constituent which is taken to be the domain of weight is the rhyme: if the rhyme branches, the syllable is heavy (13b); otherwise it is light (13a). An alternative assumption has been that a branching nucleus, as in (13c), has its role in the computation of quantity. (13)

b. Heavy: Branching rhyme q

a. Light q O

R

O

N C

V

C

R

c. Heavy: Branching nucleus q O

N

Co

V

C/V

R N

C

V

V

While this constituency was motivated on other grounds as well, capturing syllable quantity has been one of its important rationales. It was generally assumed that encoding weight distinctions is a crucial role of syllable structure. This representation was advocated, in this or somewhat modified form, by Kiparsky (1979, 1981), McCarthy (1979), Halle and Vergnaud (1980), Hayes (1980), Steriade (1982, 1988), and Levin (1985), among others. In all these approaches, the weight domain, provided by the rhyme subconstituent, crucially excludes the onset consonants, which do not participate in any of the weight patterns characterized in §2.2. How does this representation capture the three weight patterns presented in (5)? In some proposals that primarily focus on type 2 languages (e.g. Halle and Vergnaud 1980), both CVV and CVC syllables are represented in terms of a branching rhyme, that is, as (13b). Capturing both type 1 and type 2 languages called for modifications. In one modification, CVV syllables are represented in terms of a branching nucleus, as in (13c), and CVC syllables in terms of a branching rhyme, as in (13b) (e.g. Hayes 1980). In another modification, different configurations are posited for type 1 and type 2 languages (e.g. McCarthy 1979). Type 3 language posed a special challenge: heavy CVR syllables in this language type were represented in terms of a branching nucleus (13c), with the weight-bearing sonorants residing in the nucleus together with vowels (Steriade 1990).

3.2

Quantity represented by constituency

Another way of capturing syllable weight is in terms of constituency. By positing the mora as a sub-syllabic constituent, syllable weight is represented in terms of the number of moras that the syllable dominates. A syllable with one mora is light, and a syllable with more than one mora is heavy.

Quantity-sensitivity (14)

a. Light q [

11

b. Heavy q [

[

While the mora as a unit of syllable weight goes back at least as far as the study of classical languages, it was introduced to theoretical phonology by Jakobson (1931) and Trubetzkoy (1939). Arguments for representing the mora as a sub-syllabic entity are primarily due to Hyman (1985), McCarthy and Prince (1986), Hayes (1989), and Zec (1988). Crucially, moras do not uniquely map to the level of segments. Moraic representations in (14) are thus sufficiently flexible to capture all three systems of syllable quantity. What needs to be stated is the set of segments that can be dominated by the second mora of the syllable: all segments, as in the type 1 weight pattern, only vowels, as in type 2, and vowels and sonorant consonants, as in type 3. How this is to be implemented varies with specific phonological models, which may rely either on rules or on constraints. Thus Hayes (1989) posits a weight-by-position rule, Zec (1988, 1995) posits language-specific sets of moraic segments that act as constraints on the second mora of a syllable, and Morén (1999) proposes optimality-theoretic constraints on moraic segments that parallel Prince and Smolensky’s (1993) constraints on syllable nuclei.

4

Quantity-sensitivity and vowel length

Quantity-sensitivity is not a necessary property of the syllable. A number of languages, some listed in Hayes (1995), do not exhibit quantitative distinctions at the level of the syllable, for example, Bulgarian (Indo-European), Piro (Arawakan), Garawa (Karawic), and Modern Greek (Indo-European). Significantly, all these languages also lack vowel length (chapter 20: the representation of vowel length). This strongly suggests that the basic weight contrast is in fact that between short and long vowels, and raises the question of possible implicational relations between syllable weight and vowel length, either phonemic or non-phonemic. A strong claim on the relation of CVV and CVC syllables, proposed by KuryÓowicz (1948) and Newman (1972), among others, is that a language with heavy CVC syllables also has phonemic vowel length. While true in a number of specific cases, including Latin, Classical Arabic, and Fijian, this claim is too strong. A weaker claim is that the CVV syllable type is available in languages with heavy CVC syllables even if a language does not have phonemic vowel length (cf. Hayes 1989; Zec 1988, 1995). In such languages, vowel length could arise due to phonological processes such as compensatory lengthening, as in Ilokano (Hayes 1989), or iambic lengthening, as in Hixkaryana (Hayes 1995: 205 and the references therein). This claim rests crucially on a representation already available in a language (see §3), rather than on its phonemic distinctions.

5

Are weight distinctions binary or multivalued?

Cases of quantity-sensitivity presented thus far are characterized by two degrees of weight: a syllable is either light or heavy. The representations of syllable weight

Draga Zec

12

in §3 characterize quantity-sensitivity as a binary opposition, with two degrees of weight. However, a further question to be explored is whether there are cases of more than two degrees of weight, that is, whether quantity distinctions can be construed as scalar in nature. Weight patterns with weight-bearing consonants, types 1 and 3, present an obvious point of departure. In a language with light CV and heavy CVV and CVC syllables, what is the status of CVVC and CVCC syllables? Are such syllable shapes allowed? And, if allowed, are they superheavy? That is, do they call for syllable structures that are either ternary branching or trimoraic? Likewise, what is the status of CVVR (and the less likely CVRR) syllables in type 3 languages? Starting with type 1 languages, we find the following two cases. First, a language may have a syllable inventory that includes CVVC and CVCC syllables. In Hindi, such syllables give rise to three degrees of weight, as in (15a). Evidence for this ternary weight pattern comes from quantity-sensitivity in the stress system. Stress falls on a superheavy syllable if there is one, otherwise on a heavy syllable, otherwise on a light syllable (glossing over the complexities of this system, for details and examples, see §8). By contrast, Latin also has CVVC and CVCC syllables in addition to the standard type 1 inventory, yet exhibits only two degrees of weight, as in (15b). In this case, CVVC and CVCC syllables are functionally non-distinct from heavy syllables, CVV and CVC. This functional identity is supported by both stress and poetic meter. (15)

a.

b.

Hindi light heavy superheavy

CV CVV, CVC CVVC, CVCC

Latin light heavy

CV CVV, CVC, CVVC, CVCC

Newman (1972) claims that all weight distinctions are binary, pointing to languages like Latin. However, languages like Hindi clearly show that ternary weight distinctions are an attested reality. Second, a language may have a restricted syllable inventory, with only CV, CVV, and CVC, excluding both CVVC and CVCC syllable shapes. Such languages impose binarity as an upper limit to syllable complexity both in terms of weight, or mora count, and in terms of the number of consonants that may occur at the right margin of the syllable. This situation is clearly illustrated by Turkish (Clements and Keyser 1983). The syllable inventory of Turkish, a type 1 language, includes light CV and heavy CVV and CVC syllables, and systematically lacks CVVC and CVCC syllables. If the prohibited syllable types arise by virtue of morpheme concatenation, they are eliminated by phonological processes. In (16a), the underlying long vowel is shortened in a closed syllable, (nominative and ablative), but not in the open syllable (accusative). And in (16b), the two post-vocalic consonants in the underlying form are split by an epenthetic vowel (chapter 67: vowel epenthesis), in order to avoid a CVCC syllable (nominative and ablative).

Quantity-sensitivity (16)

13

Turkish a.

CVVC → CVC /zAmA(n-/ ‘time’ /ispA(t-/ ‘proof’

accusative zAmA(nQ ispA(tQ

nominative zAmAn ispAt

ablative zAmAndAn ispAttAn

b.

CVCC → CVCVC /kArn-/ ‘abdomen’

kArnQ

kArQn

kArQndAn

Type 3 languages, or at least the known cases, do not provide evidence for ternary weight distinctions. Lithuanian, for example, has the following syllable shapes in its inventory, classified in terms of weight: (17)

Lithuanian Light Heavy

CV, CVO, CVOO CVV, CVR, CVVO, CVRO

This weight pattern, as we saw in §2, is supported by the system of Lithuanian pitch accents (chapter 42: pitch accent systems): only heavy syllables, that is, CVV and CVR, can have contour tones. Lithuanian also provides evidence for strict binarity. This is evidenced by the process known as ablaut which applies in verbal morphology, with the effect of lengthening the root vowels in the preterite and infinitive, but not in the present form (Zec 1995). Vowel lengthening due to ablaut takes effect in all preterite forms: the root vowel occurs in an open syllable, due to the vowel-initial ending -ee, and is free to lengthen. In the infinitive forms, the root vowel is in a closed syllable, due to the consonantinitial ending -ti. Lengthening takes place in (18a), i.e. in roots that end in an obstruent, but not in roots that end in a sonorant (18b). (18)

Lithuanian: Ablaut in verbal forms a.

CVO

b.

CVR

root tupdrebvirmir-

present tupia drebia viria miria

preterite tuupee dreebee viiree miiree

infinitive tuupti dreebti virti mirti

‘perch’ ‘splash’ ‘boil’ ‘die’

That is, ablaut may not create a superheavy CVVR syllable, and is therefore prevented from taking effect in the infinitives of the roots in (18b). While type 2 languages may tolerate CVCC and CVVC syllables in their syllable inventories, such syllables do not form a natural class: the former has the weight of CV, and the latter has the weight of CVV syllables. The extended syllable inventories we document in this section call for representations richer than those discussed in §3. This was directly addressed in moraic representations of the syllable and its weight: a constraint restricts the number of moras per syllable to no more than two; and this constraint can be violated in some languages, giving rise to trimoraic syllables, as in Hindi. The syllable inventory in Latin is accommodated by allowing some non-moraic consonants at the syllable’s right margin (for a detailed discussion, see Sherer 1994).

Draga Zec

14

6

Inconsistencies in weight patterns

The representations in §3, despite some conceptual differences, make the strong prediction that quantity distinctions in a language will be of the same type across the board, that is, in all relevant phonological processes, and in all contexts. However, a challenge to this strong position comes from many known cases of weight inconsistencies.

6.1

Weight inconsistencies with respect to phonological process

In §2.1 we saw that Latin belongs to the type 1 weight pattern both in its stress system and in the system of poetic meter. The phenomenon of compensatory lengthening (chapter 64: compensatory lengthening) conforms to this same weight pattern, as in /kasnus/ → [ka(nus]. While not uncommon, weight consistency across different phonological processes, as evidenced in Latin, is not the general case. Weight inconsistencies are encountered in a number of languages, as noted by Steriade (1990), as well as Hayes (1995) and Gordon (2006). One such case is Kiowa (Watkins 1984). As shown in (19), vowels are shortened in syllables closed by sonorants (19a), as well as those closed by obstruents (19b) and (19c), suggesting a type 1 weight system that obeys strict binarity. (19)

Kiowa short vowels in closed syllables: Type 1 a.

b.

c.

gú(lê( gûl gúltG( cá(dG( cát cátpé t h F( t h F(dèkh ì( t h Fp

‘write-imperf-fut’ ‘write-imp’ ‘write-fut’ ‘from the doorway’ ‘entrance, doorway’ ‘against the doorway’ ‘beyond’ ‘next day’ ‘away beyond’

However, the distribution of contour tones, shown in (20), clearly points to a weight system of type 3. Contour tones occur on CVV syllables and syllables closed by a sonorant, as in (20a), but not on either CV syllables or syllables closed with an obstruent, as exemplified in (20b). (20)

Kiowa contour tones: Type 3 a.

b.

pá(lê( sân khûl sà(né sép

‘weak’ ‘child’ ‘pull off’ ‘snake’ ‘rain’

Another case is Lhasa Tibetan (Gordon 2006, based on Dawson 1980), in which the stress system treats only CVV syllables as heavy, while the system of tone treats

Quantity-sensitivity

15

as heavy both CVV and CVR syllables (Gordon 2006 and references therein). In other words, Lhasa Tibetan is a type 2 language in its stress system, and a type 3 language in its tonal system. According to Steriade (1990), variability in weight is also found in Classical Greek, in which CVV syllables are heavy for the purposes of tone, yet all syllables are heavy for the purposes of stress distribution. Thus, stress falls on the penultimate syllable if the final syllable is heavy, either CVC(C) or CVV(C), otherwise it is antepenultimate. However, only CVV syllables can sustain tonal contours, either HL or LH. Cases of weight variability in different phonological subsystems within a single language present an important challenge to formal representations, and call for fresh perspectives on the syllable and its quantity.

6.2

Weight inconsistencies with respect to phonological context

It has been noted in much work on stress that the weight of a syllable may be computed differently in word-internal and word-final positions. Thus in Classical Arabic, as shown in §2, stress falls on the rightmost CVV or CVC syllable, yet never on the final CVC. That is, CVC syllables are computed as heavy wordinternally and as light word-finally. A further fact, not mentioned in §2, is that final CVCC syllables are always stressed, i.e. they are computed as heavy (CVCC do not occur word-internally). In other words, word-final consonants do not contribute to weight. Such cases of variable weight were subsumed in Hayes (1980) under the more general rubric of extrametricality (chapter 43: extrametricality and non-finality), according to which certain phonological entities, segments as well as higher constituents, are “invisible” to phonological processes at word edges. There have been proposals, however, to treat contextual differences in weight as representational differences (Davis 1987; Kager 1989; Rice 1995; Rosenthall and van der Hulst 1999; see also chapter 36: final consonants). Under this view, the CVC sequence in Classical Arabic would be parsed as a heavy syllable wordinternally, and as a light syllable word-finally. It has been shown, however, that contextual weight differences are not restricted to word edges. Several cases of this type have been reported in Hayes (1994, 1995), among them Cahuilla and Eastern Ojibwa, as well as Central Alaskan and Pacific Yupik. In the Pacific Yupik dialect of Chugach, CVV syllables are heavy in all positions, while CVC syllables are heavy only initially, and light elsewhere. The distribution of stress in Chugach is fairly complex, and there can be more than one stress per word (for details, see Leer 1985; Kager 1993; Hayes 1995). We focus here on the evidence for the variable weight of CVC syllables. While initial CVV and CVC syllables are stressed, as in /’ta(ta’qa/ ‘my father’ and /’anciku’kut/ ‘we’ll go out’, initial CV syllables are not, as in /mu’lu’ku(t/ ‘if you take a long time’. But in medial position, CVC syllables pattern with CV rather than CVV syllables. Note that the second syllable in /’kal’ma(nuq/ ‘pocket’, a CVV syllable, is stressed. Neither CV nor CVC syllables are stressed in this same environment, as in the forms /’anku’taøtu’a/ ‘I’m going to go out’ and /’atmax’Œiqu’a/ ‘I will backpack’. Another relevant case is Goroa (Hayes 1980; Rosenthall and van der Hulst 1999, and references therein), in which stress falls on the leftmost CVV syllable, as in

Draga Zec

16

(21a), or on the final CVC syllable, as in (21b); or on the penultimate syllable, as in (21c). Crucially, CVC syllables in positions other than final are not heavy: the second CVC syllable in /giram’bo(da/ does not win over the following CVV syllable, nor do the CVC syllables in /axe’mis/ and /idir’dana/ attract stress. (21)

Goroa stress: Variable weight of CVC syllables a.

Leftmost CVV du(gnuno( giram’bo(da heni’nau

stressed ‘thumb’ ‘snuff’ ‘young’

b.

Final CVC stressed a’dux ‘heavy’ axe’mis ‘hear’

c.

Penultimate syllable stressed oro’mila ‘because’ am’rami ‘ivory arm ring’ idir’dana ‘sweet’

Contextually conditioned variation in syllable quantity affects CVC syllables, those that cross-linguistically could be either light or heavy. Thus the variability of the weight of CVC syllables found across languages has also been evidenced within individual languages. The phenomenon of contextually conditioned weight inconsistency of CVC syllables has been addressed, with a fair amount of success, in the Optimality Theory framework, most notably in Rosenthall and van der Hulst (1999).

7

Quantity-sensitivity of the foot

Syllables are grouped into feet, which belong to the next higher level of the prosodic constituency in (1) (see chapter 40: the foot; chapter 41: the representation of word stress). Quantity-sensitivity of the syllable is directly reflected at the level of the foot, as noted in Hayes (1980, 1995), McCarthy and Prince (1986), and Prince (1990), among others. Feet play an important role in the characterization of stress and in prosodic morphology, and our examples will come from both domains. As shown in a vast body of literature, feet tend to be binary. That is, feet are prosodic constituents resulting from grouping at most two constituents at the next lower level (Hayes 1995; McCarthy and Prince 1986; Prince 1990; among others). How this proceeds depends crucially on whether a language has a quantity-sensitive or a quantity-insensitive foot system (Hayes 1980). In quantity-insensitive systems, pairs of syllables are incorporated into feet regardless of their weight. Relevant for our discussion is foot formation in quantity-sensitive systems, in which syllable weight plays a crucial role. An important property of such systems is the commensurability of a heavy syllable with two lights. There are two types of quantitysensitive feet, trochaic and iambic (chapter 44: the iambic–trochaic law). In quantity-sensitive trochaic systems a foot corresponds to either one heavy syllable, as in (22a), or two light syllables, as in (22b); feet are left-headed, that is, have initial prominence, shown in (22b) by underlining.

Quantity-sensitivity (22)

17

Trochaic foot inventory a. b.

qH qL qL

This receives a straightforward interpretation in moraic theory of syllable structure: a foot contains two moras, a condition met either by one heavy syllable, as in (22a), or by two lights, as in (22b). A heavy syllable has a dual status: it counts not only as a syllable but also as a foot. This foot inventory is active in the stress system of Fijian, a type 2 language (Hayes 1995, and references therein). In words with only a light syllable, pairs of syllables are incorporated into feet, computing from right to left, and foot-initial syllables are assigned stress. As a result, stress falls on every second syllable, computed from the right edge, as shown in (23). Parsing of syllables into feet obeys strict binarity, but is not necessarily exhaustive. In words with an odd number of syllables, as in (23c) and (23e), a syllable at the left edge is not footed. (The rightmost stressed syllable bears primary stress; others bear secondary stress.) (23)

Fijian stress: Light syllables only a. b. c. d. e.

(24)

(’lako) (’talo) ßi (’naka) (‘ndiko) (’nesi) pe (‘resi) (’tendi)

‘go’ ‘pour’ ‘good’ ‘deaconess’ ‘president’

Fijian stress: Light and heavy syllables a. e. c. d. e.

ki (’la() (‘mbe() (’leti) (‘mbele) (‘mbo() (’tomu) pa(‘ro() ka (’ramu) (‘mi() (‘sini) (’Igani)

‘know’ ‘belt’ ‘bellbottoms’ ‘program’ ‘machine-gun’

In words with both light and heavy syllables, each heavy syllable corresponds to a foot, and is stressed. Right-to-left footing is thus disrupted by heavy syllables, and has to work around them. In the disyllabic form with a heavy final syllable, in (24a), the initial syllable is left unfooted. And the form in (24d), which has five syllables, two light syllables, the first and the third, are left unfooted. All syllables are footed in the remaining forms in (24). The inventory of feet in (22) captures the distribution of stress in a number of trochaic quantity-sensitive systems, including some of the cases presented in §2. In particular, stress in Latin follows the same pattern as in Fijian, with one notable difference: The final syllable is ignored for the purposes of stress (another case of so-called extrametricality, see §6.2). As a result, trisyllabic forms with only light syllables have initial stress, as in (’ani)ma ‘soul (nom sg)’. Likewise, final heavy syllables are not stressed: in (’gau)dbns ‘rejoicing (nom sg)’ the penultimate heavy, but not the final heavy, is footed, and stressed (for a detailed analysis, see Mester 1994; Hayes 1995). We also present a case of prosodic morphology that employs the foot inventory in (22). In the system of Japanese hypocoristic formation, as characterized in

Draga Zec

18

Poser’s (1990) detailed study, hypocoristics are formed by adding the suffix -tjan to proper names, either to their full or modified form. As shown by Poser, what is considered to be modification is really a case of template satisfaction. Crucially, the template corresponds to a trochaic foot: either to two light syllables or one heavy syllable. Japanese, a type 1 language, has light CV and heavy CVV and CVC syllables. As shown by the truncated versions of the proper name Hanako, the suffix is added to two light syllables, as in (25a) or one heavy, as in (25b) and (25c). The truncated form cannot be smaller than a foot, corresponding to a single light syllable, as in the ill-formed (25d). Nor can the truncated form be greater than a foot, corresponding to three light syllables, as shown by the ill-formed (26b). Proper names corresponding to a light syllable are converted to a heavy syllable, that is, to a foot; in (27a) this is accomplished by vowel lengthening. Note that (27b) is also available, as -tjan can be added to any proper name in its full form regardless of its size. (25)

Hypocoristic forms for Hanako a. b. c. d.

(26)

Hypocoristic forms for Takatugu a. b.

(27)

hanatjan haatjan hattjan *hatjan

takatjan *takatutjan

Hypocoristic forms for Ti a. b.

tiitjan titjan

Thus, in trochaic prosodic morphology, just as in trochaic stress systems, a heavy syllable is functionally equivalent with two light syllables. Quantity-sensitive iambic feet differ somewhat in shape from the trochaic set, as shown by the inventory in (28). Iambic feet are right-headed, indicated by the underlining. (28)

Iambic foot inventory a. b. c.

qH qL q L qL q H

In this case, as well, syllable quantity plays a central role: for a foot to be wellformed, it needs to contain syllables of the correct weight. The iambic system of quantitative feet captures the distribution of stress in Asheninca (Hayes 1995; Payne 1990). Asheninca has a type 2 weight system, with only CVV heavy syllables. The forms in (29a) contain only light syllables: binary right-headed feet are computed from right to left. The final syllable is regularly left unfooted, which yields initial stress in disyllables, as in /’haka/ ‘here’. Crucial are the forms in (29b), which contain both light and heavy syllables, and can therefore exemplify all members of the foot inventory.

Quantity-sensitivity (29)

19

Asheninca stress system a.

b.

(pa.’me).(na.’ko).(wen.’ta).ke.ro (ha.’ma).(nan.’ta).(ke.’ne).ro (no.’ko).(wa.’we).ta.ka (no.’ton).(ka.’men).to (ka.’man).ta.ke (no.’ma).(ko.’rjaa).(’wai).(ta.’paa).ke (pi.’Jaa).(’paa).ke (i.’kjaa).(’piin).ti (’poo).(ka.’na).ke.ro (’paa).(ti.’ka).ke.ri

‘take care of her’ ‘he bought it for her’ ‘I wanted (it) in vain’ ‘my gun’ ‘he/she said’ ‘I rested a while’ ‘you saw on arrival’ ‘he always enters’ ‘you threw it out’ ‘you stepped on him’

Quantity-sensitive iambic feet also figure in prosodic morphology. In Ulwa, which has a type 1 weight system, the suffix /-ka/ is attached to the leftmost iambic foot, as in (30). It occurs at the right edge of a stem only when the entire stem corresponds exactly to a foot, as in (30a). In (30b), the only way for /-ka/ to be attached to an iambic foot is to occur stem-internally. (30)

Ulwa construct state (from McCarthy and Prince 1990: 228) a.

b.

base al bas kii sana amak sapaa suulu kuhbil baskarna siwanak anaalaaka karasmak

possessed al-ka bas-ka kii-ka sana-ka amak-ka sapaa-ka suu-ka-lu kuh-ka-bil bas-ka-karna siwa-ka-nak anaa-ka-laaka karas-ka-mak

‘man’ ‘hair’ ‘stone’ ‘deer’ ‘bee’ ‘forehead’ ‘dog’ ‘knife’ ‘comb’ ‘root’ ‘chin’ ‘knee’

Trochaic and iambic systems differ with regard to the role of quantity, as noted in Hayes (1985) and Prince (1990) as well as chapter 44: the iambic–trochaic law. The preferred type of trochaic disyllabic feet includes two light syllables, while iambic feet optimally correspond to a sequence of a light and heavy syllable. Thus, disyllabic trochaic feet are preferably even, while iambic feet are preferably of uneven quantity. Evidence for even trochaic quantity comes from the so-called trochaic shortening, which makes an uneven trochee even by vowel shortening, as exemplified by Fijian. The form in (31a), with an underlying long vowel, undergoes shortening when integrated into a disyllabic foot, as in (31b). (31)

Fijian: Trochaic shortening a. b.

’ta( ’ta-ja

‘chop’ ‘chop-trans-3sg obj’

By contrast, uneven quantity is an important feature of iambic systems. A number of iambic stress systems are characterized by iambic lengthening, including

Draga Zec

20

Menomini, Hixkaryana, and Kashaya (Hayes 1995). In Hixkaryana (Caribian), prominent CV syllables undergo vowel lengthening, as in (32a) and (32b), and thus become heavy. Note that prominent CVC syllables, which are already heavy, are not subject to lengthening, as in the second foot in (32c), and the initial syllables in (32a) and (32b). (32)

Hixkaryana: Iambic lengthening a. b.

owtohona tohkurjehonahaœaka

c.

mqhananqhno

(’ow)(to’ho()na (’toh)(ku’rje()(ho’na() (ha’œa()ka (mq’ha()(na’nqh)no

‘to the village’ ‘finally in Tohkurye’ ‘you taught him’

Generalizations about the quantity of trochaic and iambic groupings are stated in Hayes (1995) as the Iambic–Trochaic Law (see chapter 44: the iambic–trochaic law): (33)

The Iambic–Trochaic Law a. b.

8

Elements contrasting in intensity naturally form groupings with initial prominence. Elements contrasting in duration naturally form groupings with final prominence.

Scalar quantity systems

While binary quantity systems are based primarily on grouping syllables into feet, scalar quantity systems are based on prominence, defined along some dimension (Prince and Smolensky 1993; Hayes 1995). A central prominence dimension is syllable weight, although other dimensions, such as tone and vowel height, have been evidenced as well. We present two cases with syllable weight as the prominence dimension. One is Kashmiri, with examples given in (34) (Kenstowicz 1993; Rosenthall and van der Hulst 1999). In Kashmiri, CVV syllables are heavier than CVC, which in turn are heavier than CV. Thus, in words with only CV and CVV syllables, stress falls on the leftmost CVV, as in (34a). In words with only CV and CVC, stress falls on the leftmost CVC, as in (34b). In words with both CVC and CVV syllables, stress falls on the CVV syllable, as in (34c). Finally, with only CV syllables present, stress is initial, as in (34d). The final syllable is excluded from scansion. (None of the sources supply glosses for Kashmiri forms.) (34)

Kashmiri stress: CVV > CVC > CV a. b.

mu’si(bah a’jo(gjH ta( ba’gandarladin juni’varsiti

c. d.

am’ri(ka mas’ra(wun ’tsaripop ’paharadari(

Quantity-sensitivity

21

Languages in which stress is assigned on the basis of scalar syllable prominence may have several degrees of syllable weight. Thus Hindi (for the dialect described in Kelkar 1968) has three degrees of syllable weight: superheavy syllables CVVC and CVCC are more prominent than heavy syllables, CVV and CVC, which in turn are more prominent than CV syllables, as stated in (35). (35)

CVVC, CVCC > CVV, CVC > CV

Excluding the final syllable from scansion, stress is assigned to the heaviest available syllable, as in (36). In both forms stress falls on a CVVC syllable, which in (36b) wins over a CVV syllable, and in (36a) over both a CV and CVV syllable. (36)

a. b.

’œo(x–aba(ni( ’re(zga(ri(

‘talkative’ ‘small change’

If there is a tie, stress is assigned to the rightmost (non-final) syllable: to a CV syllable in (37a), and a CVV syllable in (37b) and (37c). (37)

a. b. c.

sa’miti ro(’za(na( ka(’ri(gari(

‘committee’ ‘daily’ ‘craftsmanship’

Interestingly, when the final syllable is the heaviest in the word, it is not excluded from scansion, as in (38): (38)

ki’dhar ru’pia as’ba(b

‘which way’ ‘rupee’ ‘goods’

Quantity in Hindi is thus computed along a scale of syllable weight, with the superheavy syllable being most prominent, followed by the heavy syllable and then by the least prominent light syllable. This case is analyzed in precisely these terms in Hayes (1995) and Prince and Smolensky (1993), although in different frameworks: in rule-based metrical theory and in Optimality Theory, respectively. An interesting mode of computing prominence is found in Pirahã, a Mura language of Brazil (Everett 1988; Hayes 1995). The Pirahã prominence scale combines syllable weight and onset quality (on onsets, see chapter 55: onsets). While CVV syllables are more prominent than CV syllables, voiceless onsets are more prominent than voiced onsets, and presence of onset is more prominent than its absence, yielding the scale in (39). (39)

KVV > GVV > VV > KV > GV

[K = voiceless, G = voiced]

Stress falls on one of the last three syllables of the word that is highest on this scale, as in (40a). In the event of ties, the rightmost syllable wins, as in (40b). (40)

Pirahã prominence-based stress a.

’ka(gai ?apa’ba(si ’?ibogi

b.

ko’po ?aba’pa paohoa’hai

22

Draga Zec

Further dimensions for computing prominence are in no obvious way related to patterns of syllable quantity we surveyed here. Yet, because of their scalar nature, they are highly reminiscent of quantity-based systems of prominence. One such dimension is vowel quality: given the sonority scale, stress falls on the most sonorous vowel. Prominence systems of this type have been analyzed in Kenstowicz (1997) and de Lacy (2004). In Mordwin, for example, non-high vowels are more prominent than high vowels (chapter 21: vowel height). In words with only non-high vowels, or with only high vowels, stress falls on the leftmost syllable. However, in words that contain both high and non-high vowels, stress falls on a non-high vowel, regardless of its position in the word. Another dimension is tonal prominence: syllables associated with High tones are more prominent than syllables associated with Low tones, and thus more likely to be associated with stress. Prominence systems of this type are described in Hayes (1995) and de Lacy (2002); for a somewhat different perspective, see Zec (2003). Of particular interest is the complex case of Nanti, a Kampa language of Peru: its stress system, which is of the iambic type, is also governed by several types of prominence, including syllable quantity and vowel quality (Crowhurst and Michael 2005).

9

Quantity-sensitivity at the higher levels of the prosodic hierarchy

When focusing on higher levels of the prosodic hierarchy, the prosodic word and the prosodic phrase, we are in fact dealing with morphosyntax/prosody interface. Quantity-sensitivity is a specifically prosodic phenomenon and is not known to play any role in other modules of the grammar. Any effects of quantitysensitivity in either morphology or syntax are therefore to be attributed to prosody. We addressed the interfaces with morphology in §8, with two cases of affixes that select not only a morphological class, but also a prosodic type of the stem; both in this case select for the foot. Many more such cases are found in the literature (McCarthy and Prince 1986, 1990, 1995; among others). The word level of the prosodic hierarchy is constituted by a grouping of feet (chapter 41: the representation of word stress; chapter 51: the phonological word). In practice, however, one foot is sufficient for a prosodic word to achieve the desired quantity, as documented by numerous cases of minimal word size. Moreover, in a number of languages, no minimal size beyond a single syllable is imposed on prosodic words. This is broadly documented in Hayes (1995) and Downing (2006), among others. In sum, the prosodic word provides no evidence for quantity-sensitivity of the sort found at the level of the syllable and the foot: its binary structure is not distinct from that of a foot. There are no known cases of a prosodic word minimally branching into two feet, yet this would be expected, based on the situation at the lower end of the prosodic hierarchy. However, quantity-sensitivity has been evidenced at the higher end of the hierarchy, that is, at the level of the prosodic phrase. The distribution of a syntactic constituent should not be affected by its length or internal complexity. When such effects arise, they are generally attributed to prosody. We focus here

Quantity-sensitivity

23

on cases of branching in prosodic phrases, typical cases of apparent quantitysensitivity of syntactic constituents. Cases of binary branching prosodic phrases were reported by Nespor and Vogel (1986), with evidence from Italian, French, and English. In Italian, for example, a prosodic phrase preferably contains more than one prosodic word, as shown by the following cases (Nespor and Vogel 1986): (41)

Prosodic phrase formation in Italian a.

Av’ra trovato (il pescecane)t ‘He will have found the shark.’

b.

(I cari’bu nanni)t sono estinti ‘Dwarf caribous are extinct.’

c.

Hanno dei (car’ibú)t (molto piccoli)t ‘They have very small caribous.’

While complements that correspond to single prosodic words, as in (41a), form one-word prosodic phrases, multiple word complements, as in (41b) and (41c), correspond to branching prosodic phrases. The prosodic phrasing in (41c) further shows that complements with three prosodic words do not correspond to a single prosodic phrase, as branching prosodic phrases contain at most two prosodic words. By contrast, Serbo-Croatian sentence-initial topics have to include at least two prosodic words (Zec and Inkelas 1990), and thus exemplify obligatory branching in prosodic phrases. This line of research has been further continued by Ghini (1993), Selkirk (2000), and Sandalo and Truckenbrodt (2002), among others.

10 Remarks on markedness It is important to note that the markedness (chapter 4: markedness) of light and heavy constituents is not identical across prosodic levels: heavy constituents are marked at the level of the syllable, while light constituents are marked at the level of the foot. This is directly encoded in Optimality Theory. Constraints listed in (42) assign marked status to heavy syllables: to CVV syllables, as in (42a), and to syllables with coda consonants, as in (42b) and (42c). While (42b) targets any coda consonant, (42c) targets any weight-bearing segment. (42)

a.

NoLongVowel (Rosenthall 1994) A vowel should not be long, i.e. linked to more than one mora.

b.

NoCoda (Prince and Smolensky 1993) Syllables must not have a coda.

c.

*Mora[seg] (Morén 1999) Do not associate a mora with a particular type of segment.

These constraints, which belong to the markedness family, penalize binary structures at the syllable level, thus favoring a simple CV syllable, which is light. Thus, light syllables emerge as the unmarked case: all languages have light syllables, and some may also have heavy syllables. Superheavy, i.e. trimoraic, syllables are,

24

Draga Zec

of course, also marked, and are penalized as such by a constraint against trimoraic syllables proposed by Sherer (1994). By contrast, heavy feet are preferred over light ones. Binary constituents are highly desirable at the level of the foot, both in trochaic and iambic systems. Non-binary, or light, feet are permitted in some languages under very special conditions and banned in others. The unmarked condition for feet is thus to be binary, that is, heavy, either under moraic or syllabic analysis, and this is codified in Optimality Theory by a corresponding constraint: (43)

FootBinarity (McCarthy and Prince 1993) Feet must be binary under a syllabic or moraic analysis.

At the higher prosodic levels, constituent size is largely determined by morphosyntax, as is the distribution of light and heavy constituents. However, where permitted by morphosyntax, heavy, i.e. branching, constituents are preferred over light ones (see §9).

11

Conclusion

Quantity-sensitivity is an important property of prosodic structure, evidenced at each of its levels. As we have seen, constituents at any level of the prosodic hierarchy can be classified into those that are light and those that are heavy. While quantity-sensitivity is typically associated with the syllable and the foot, all prosodic levels exhibit this property. Whether a syllable is light or heavy crucially depends on its segmental setup; quantity at the level of the foot relies on, and is largely characterized in terms of syllable quantity; quantity-sensitivity of the prosodic word is non-distinct from that of the foot; and quantity-sensitivity at the higher prosodic levels is heavily influenced by morphosyntax. While characterization of quantity largely depends on level-specific criteria, a general property of heavy constituents is their greater size and complexity, and often their binary structure. It is interesting, however, that preference, or dispreference, for heavy constituents varies across prosodic levels. The unmarked condition for syllables is to be light while the unmarked condition for feet is to be heavy. The latter condition persists through the higher levels of the prosodic hierarchy. Thus, while light syllables are preferred over heavy ones, feet and prosodic words are preferably heavy. Heavy prosodic phrases are preferred as well, although in a very weak sense.

ACKNOWLEDGMENTS I am most grateful to Adam Cooper, Michael Weiss, and Kimberly Will, to an anonymous referee, and to the editors, Marc van Oostendorp and Keren Rice, for their invaluable comments and suggestions, which have led to major improvements to this chapter.

Quantity-sensitivity

25

REFERENCES Allen, W. Sidney. 1973. Accent and rhythm. Cambridge: Cambridge University Press. Boas, Franz. 1947. Kwakiutl grammar with a glossary of the suffixes. Transactions of the American Philosophical Society: New Series 37. 201–377. Buckley, Eugene. 1998. Alignment in Manam stress. Linguistic Inquiry 29. 475–496. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Crowhurst, Megan J. & Lev Michael. 2005. Iterative footing and prominence-driven stress in Nanti (Kampa). Language 81. 47–95. Davis, Stuart. 1987. Coda extrametricality in English nouns. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23. 66–75. Dawson, Willa. 1980. Tibetan phonology. Ph.D. dissertation, University of Washington. de Lacy, Paul. 2002. The interaction of tone and stress in Optimality Theory. Phonology 19. 1–32. de Lacy, Paul. 2004. Markedness conflation in Optimality Theory. Phonology 21. 145–199. Downing, Laura J. 2006. Canonical forms in prosodic morphology. Oxford: Oxford University Press. Everett, Daniel L. 1988. On metrical constituent structure in Pirahã phonology. Natural Language and Linguistic Theory 6. 207–246. Ghini, Mirco. 1993. X-formation in Italian: A new proposal. Toronto Working Papers in Linguistics 12. 41–78. Gordon, Matthew. 2006. Syllable weight: Phonetics, phonology, typology. London: Routledge. Halle, Morris & William J. Idsardi. 1995. General properties of stress and metrical structure. In John A. Goldsmith (ed.) The handbook of phonological theory, 403–443. Cambridge, MA & Oxford: Blackwell. Halle, Morris & Jean-Roger Vergnaud. 1980. Three-dimensional phonology. Journal of Linguistic Research 1. 83–105. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. Cambridge, MA: MIT Press. Hayes, Bruce. 1980. A metrical theory of stress rules. Ph.D. dissertation, MIT. Hayes, Bruce. 1982. Extrametricality and English stress. Linguistic Inquiry 13. 227–276. Hayes, Bruce. 1985. Iambic and trochaic rhythm in stress rules. Proceedings of the Annual Meeting, Berkeley Linguistics Society 11. 429–446. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce. 1994. Weight of CVC can be determined by context. In Jennifer Cole & Charles W. Kisseberth (eds.) Perspectives in phonology, 61–79. Stanford: CSLI. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hyman, Larry M. 1977. On the nature of linguistic stress. In Larry M. Hyman (ed.) Studies in stress and accent, 37–82. Los Angeles: Department of Linguistics, University of Southern California. Hyman, Larry M. 1985. A theory of phonological weight. Dordrecht: Foris. Jakobson, Roman. 1931. Die Betonung und ihre Rolle in der Wort- und Syntagmaphonologie. Travaux du Cercle Linguistique de Prague 2. Reprinted (1962) in Selected writings, vol. 1: Phonological studies, 117–136. The Hague: Mouton. Jeanne, LaVerne Masayesva. 1982. Some phonological rules of Hopi. International Journal of American Linguistics 48. 245–270. Kager, René. 1989. A metrical theory of stress and destressing in English and Dutch. Dordrecht: Foris.

26

Draga Zec

Kager, René. 1993. Alternatives to the iambic-trochaic law. Natural Language and Linguistic Theory 11. 381–432. Kelkar, Ashok R. 1968. Studies in Hindi-Urdu I: Introduction and word phonology. Poona: Deccan College. Kenstowicz, Michael. 1993. Peak prominence stress systems and Optimality Theory. Proceedings of the 1st International Conference of Linguistics and Chosun University, 7–22. Gwangju: Foreign Culture Research Institute, Chosun University, Korea. Kenstowicz, Michael. 1997. Quality-sensitive stress. Rivista di Linguistica 9. 157–187. Kiparsky, Paul. 1979. Metrical structure assignment is cyclic. Linguistic Inquiry 10. 421–441. Kiparsky, Paul. 1981. Remarks on the metrical structure of the syllable. In Wolfgang U. Dressler, Oskar E. Pfeiffer & John R. Rennison (eds.) Phonologica 1980, 245–256. Innsbruck: Innsbrucker Beiträge zur Sprachwissenschaft. KuryÓowicz, Jerzy. 1948. Contribution à la théorie de la syllabe. Bulletin de la Société Polonaise de Linguistique 8. 80–113. Larsen, Raymond S. & Eunice V. Pike. 1949. Huasteco intonations and phonemes. Language 25. 268–277. Leer, Jeff. 1985. Toward a metrical interpretation of Yupik prosody. In Michael Krauss (ed.) Yupik Eskimo prosody systems: Descriptive and comparative studies, 159–172. Fairbanks: Alaska Native Language Center Research Papers. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. Lichtenberk, Frantisek. 1983. A grammar of Manam. Honolulu: University of Hawaii Press. McArthur, Henry & Lucille McArthur. 1956. Aguacatec (Mayan) phonemes within the stress group. International Journal of American Linguistics 22. 72–76. McCarthy, John J. 1979. On stress and syllabification. Linguistic Inquiry 10. 443–465. McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. McCarthy, John J. & Alan Prince. 1990. Foot and word in prosodic morphology: The Arabic broken plural. Natural Language and Linguistic Theory 8. 209–283. McCarthy, John J. & Alan Prince. 1993. Generalized alignment. Yearbook of Morphology 1993. 79–153. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 249–384. Amherst: GLSA. Mester, Armin. 1994. The quantitative trochee in Latin. Natural Language and Linguistic Theory 12. 1–61. Miller-Ockhuizen, Amanda. 1998. Towards a unified decompositional analysis of Khoisan lexical tone. In Mathias Schladt (ed.) Language, identity, and conceptualization among the Khoisan, 217–243. Cologne: Rüdiger Köppe Verlag. Morén, Bruce. 1999. Distinctiveness, coercion and sonority: A unified theory of weight. Ph.D. dissertation, University of Maryland at College Park. Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Newman, Paul. 1972. Syllable weight as a phonological variable. Studies in African Linguistics 3. 301–323. Payne, Judith. 1990. Asheninca stress patterns. In Doris L. Payne (ed.) Amazonian linguistics: Studies in lowland South American languages, 185–209. Austin: University of Texas Press. Poppe, Nicholas N. 1960. Buriat grammar. Bloomington: Indiana University Publications. Poser, William J. 1990. Evidence for foot structure in Japanese. Language 66. 78–105. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14. 19–100. Prince, Alan. 1990. Quantitative consequences of rhythmic organization. Papers from the Annual Regional Meeting, Chicago Linguistic Society 26(2). 355–398. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder.

Quantity-sensitivity

27

Rice, Curt. 1995. An Optimality Theoretic analysis of weight variability and variation in primary stress in English, Dutch, and Norwegian. Unpublished ms., University of Tromsø. Rosenthall, Sam. 1994. Vowel/glide alternation in a theory of constraint interaction. Ph.D. dissertation, University of Massachusetts, Amherst. Rosenthall, Sam & Harry van der Hulst. 1999. Weight-by-position by position. Natural Language and Linguistic Theory 17. 499–540. Ryding, Karin C. 2005. A reference grammar of Modern Standard Arabic. Cambridge: Cambridge University Press. Sandalo, Filomena & Hubert Truckenbrodt. 2002. Some notes on phonological phrasing in Brazilian Portuguese. Delta 19. 1–30. Selkirk, Elisabeth. 1978. On prosodic structure and its relation to syntactic structure. In Thorstein Fretheim (ed.) Nordic prosody II, 111–140. Trondheim: Tapir. Selkirk, Elisabeth. 1980. Prosodic domains in phonology: Sanskrit revisited. In Mark Aronoff & Mary-Louise Kean (eds.) Juncture, 107–129. Saratoga: Anma Libri. Selkirk, Elisabeth. 2000. The interaction of constraints on prosodic phrasing. In Merle Horne (ed.) Prosody: Theory and experiment, 231–261. Dordrecht: Kluwer. Sherer, Timothy. 1994. Prosodic phonotactics. Ph.D. dissertation, University of Massachusetts, Amherst. Steriade, Donca. 1982. Greek prosodies and the nature of syllabification. Ph.D. dissertation, MIT. Steriade, Donca. 1988. Review of Clements & Keyser (1983). Language 64. 118–129. Steriade, Donca. 1990. Moras and other slots. Proceedings of the Formal Linguistics Society of Midamerica 1. 254–280. Trubetzkoy, Nikolai S. 1939. Grundzüge der Phonologie. Göttingen: van der Hoeck & Ruprecht. Walker, Rachel. 1996. A third parameter for unbounded stress. Papers from the Annual Meeting of the North East Linguistic Society 26. 441–455. Watkins, Laurel J. 1984. A grammar of Kiowa. Lincoln & London: University of Nebraska Press. Yip, Moira. 2002. Tone. Cambridge: Cambridge University Press. Young, Robert & William Morgan. 1987. The Navajo language: A grammar and dictionary. Albuquerque: University of New Mexico Press. Zec, Draga. 1988. Sonority constraints on prosodic structure. Ph.D. dissertation, Stanford University. Zec, Draga. 1995. Sonority constraints on syllable structure. Phonology 12. 85–129. Zec, Draga. 2003. Prosodic weight. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 123–143. Cambridge: Cambridge University Press. Zec, Draga & Sharon Inkelas. 1990. Prosodically constrained syntax. In Sharon Inkelas & Draga Zec (eds.) The phonology–syntax connection, 365–378. Chicago: University of Chicago Press. Zhang, Jie. 2002. The effects of duration and sonority on contour tone distribution: Typological survey and formal analysis. New York & London: Routledge.

58

The Emergence of the Unmarked Michael Becker Kathryn Flack Potts

1

Introduction

The term “The Emergence of the Unmarked” (TETU), originally coined by McCarthy and Prince (1994), refers to situations where some marked structure is generally allowed in a language, but banned in particular contexts; the complementary unmarked structure thus “emerges.” In Nuu-chah-nulth (Wakashan, referred to by McCarthy and Prince by its former name, Nootka), for example, syllables can generally have codas; reduplicants, however, are exceptional in that codas are banned. This results in words like [Œi-Œim.s’i(p] ‘hunting bear’ and [wa(-wa(s.Œix] ‘naming where’, in which unmarked (codaless) syllables emerge in reduplicants despite the presence of marked codas in bases. TETU effects came to prominence in phonological theory with the advent of Optimality Theory (OT; Prince and Smolensky 1993). In OT terms, these effects typically follow from rankings like (1), where a markedness constraint M is dominated by a faithfulness constraint F1, which blocks M’s activity in some, though crucially not all, contexts. M is free to become active in contexts where F1 isn’t relevant; here, M can motivate violation of still lower-ranked faithfulness constraints (F2). (1)

F1 >> M >> F2

The Nuu-chah-nulth pattern described above results from a ranking of this type, as shown in (2) and (3). The markedness constraint NoCoda is dominated by the anti-deletion constraint IO-Max; this ranking protects underlying codas from deletion, eliminating the unmarked, codaless candidate (2b). Since reduplicants are assumed not to stand in correspondence with inputs, however (chapter 100: reduplication), high-ranking IO-Max is irrelevant in their evaluation.1 Because NoCoda dominates BR-Max, the emergence of unmarked CV syllables is permitted 1

Correspondence between input and output candidates is evaluated by input–output (IO) faithfulness constraints. Reduplicants stand in correspondence relationships with the output forms of their bases, and are evaluated by base–reduplicant (BR) faithfulness constraints (McCarthy and Prince 1999). Faithfulness constraints in this chapter assess IO correspondence, unless otherwise noted. The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0058

2

Michael Becker & Kathryn Flack Potts

in reduplicants. Concretely, candidates (3a) and (3b) are identical, except that the reduplicant in (3b) contains a copy of the coda of the root-initial syllable, while the reduplicant in (3a) doesn’t. Because NoCoda dominates BR-Max, the additional NoCoda violation in (3b) rules out this candidate in favor of the less marked (3a). (2)

/Œims-’i(p/ ☞ a. Œim.s’i(p b. Œi.s’i(

(3)

IO-Max NoCoda BR-Max ** **!

/red-Œims-’i(p/ IO-Max NoCoda BR-Max ☞ a. Œi.Œim.s’i(p b. Œim.Œim.s’i(p

** ***!

**** ***

Increasing attention to TETU effects was a natural result of inquiry into Optimality Theory. As McCarthy and Prince note, TETU is a direct result of two fundamental properties of OT. First, OT is a theory of ranked, violable constraints. Constraints are frequently active in a language even if they are not always satisfied; this is at the heart of TETU effects, which occur when a markedness constraint is dominated but still active. They observe that this “sharply differentiates OT from approaches to linguistic structure and interlinguistic variation based on parameters, rules, or other devices that see linguistic principles in globally all-or-nothing terms” (1994: 363–364).2 Second, distinctions between marked and unmarked structures are fundamental to OT, allowing the existence and emergence of unmarkedness to be formally defined. As McCarthy and Prince explain, “OT (Prince and Smolensky 1993) offers an approach to linguistic theory that aims to combine an empirically adequate theory of markedness with a precise formal sense of what it means to be ‘unmarked’ ” (1994: 333). At the heart of OT are two basic constraint types: those demanding identity, typically between inputs and outputs (faithfulness), and those penalizing particular output structures (markedness) (chapter 63: markedness and faithfulness constraints; see also chapter 4: markedness). Marked structures are defined as exactly those structures which violate a markedness constraint.3 “Emergence” can be defined with similar precision, again by reference to basic properties of OT: an unmarked structure can be said to emerge in a language if the markedness constraint violated by that structure is dominated by some (typically faithfulness) constraint which blocks its activity in some, but not all, contexts in that language. §2 elaborates on this basic understanding of TETU as “activity despite domination,” surveying three types of cases in which a dominating constraint is inactive in a particular evaluation, allowing a lower-ranked markedness constraint 2

This view is elaborated in McCarthy (2002: 129–134), where it is noted that theories with ordered rules can mimic some TETU effects with the application of default rules. 3 More precisely, structures which violate some markedness constraint M1 are marked with respect to M1; if these structures do not violate some other markedness constraint M2, no conflict arises in saying that they are also unmarked with respect to M2. In OT, markedness is multidimensional, assessed by each markedness constraint individually.

The Emergence of the Unmarked

3

to emerge. §3 then describes gradient TETU effects found in languages where the emergent markedness constraint is never categorically active. Finally, §4 compares true TETU effects with situations where faithfulness, rather than markedness, constraints are active despite domination and thus emergent.

2

TETU typology

The typical TETU ranking is F1 >> M >> F2, with M emerging in evaluations where F1 is not decisive. This section will discuss three subclasses of TETU rankings, following from three different contexts in which high-ranking F1 may be rendered inactive. §2.1 looks at output segments and structures which have no input correspondents and so are invisible to IO-faithfulness constraints; these include reduplicants, epenthetic segments, and syllable boundaries. §2.2 considers evaluations in which multiple candidates tie on a particular high-ranking constraint, and §2.3 surveys faithfulness constraints which evaluate only some positions or aspects of outputs while ignoring others. In each of these situations, a high-ranking constraint is inactive and a dominated markedness constraint becomes active, choosing the winning output.

2.1

Output segments and structures without input correspondents

TETU is commonly observed in output structures which lack input correspondents and thus cannot be evaluated by IO-faithfulness. Recall the Nuu-chah-nulth ranking in (2) and (3), of the form IO-F >> M >> BR-F. Because reduplicants have no input correspondents in this theory, they cannot be evaluated by IO-faithfulness, allowing the effects of M (NoCoda in Nuu-chah-nulth) to emerge. This section describes similar TETU patterns found in two other structures which are present in outputs but not inputs: epenthetic segments and syllable boundaries.

2.1.1 Epenthesis Kager (1999) observes that markedness constraints which are generally freely violated in a language often determine the quality of epenthetic segments (chapter 67: vowel epenthesis). These segments are typically featurally unmarked; epenthetic vowels like [i], [q], and [H], and consonants like [?], [h], and glides, are cross-linguistically common, while marked segments like [f] and [æ] are rarely epenthesized.4 This is due to TETU rankings like IO-Ident >> M, where M is a featural markedness constraint. When a constraint demanding identity between input and output features outranks markedness (here, IO-Ident >> M), the latter has little power to ban marked features in the language as a whole. While the presence of an epenthetic segment violates the anti-epenthesis constraint Dep, its lack of an input correspondent means that it is invisible to high-ranked IO-Ident; thus, epenthetic segments are subject to markedness constraints which require them to have unmarked feature values. 4

See Vaux (2002, 2008) for a survey of epenthetic segments and a diachronic perspective. See also Steriade (2001, 2009) for the view that epenthetic segments are chosen by faithfulness constraints minimizing the perceptual distance between representations with and without the epenthetic segment.

4

Michael Becker & Kathryn Flack Potts

2.1.2 Syllable structure Not every aspect of linguistic outputs is evaluated by faithfulness constraints; some output properties, like prosodic structure above the mora level, are generally taken to be governed by markedness constraints only (chapter 33: syllable-internal structure). In a language like Timugon Murut (Austronesian), where Dep >> Onset as in (4) and (5) (McCarthy and Prince 1994), the dominated markedness constraint Onset emerges to make decisions in cases where Dep cannot distinguish between candidates. Dep’s high ranking results in a language where epenthesis never occurs in order to avoid onsetless syllables, thus allowing words like [ambi’luo] ‘soul’ in (4). Dep (and similarly Max, Ident, etc.) cannot, however, distinguish between the candidates in (5), which differ only in syllabification. Because faithfulness constraints cannot see these differences, the decision is handed down to the emergent markedness constraint Onset. (4)

/ambi’luo/ ☞ a. am.bi.’lu.o b. ?am.bi.’lu.?o

(5)

/ambi’luo/ ☞ a. am.bi.’lu.o b. am.bil.’u.o

Dep

Onset **

**! Dep

Onset ** ***!

Cross-linguistically, the markedness constraint Onset commonly triggers epenthesis, deletion, and other changes to prevent onsetless syllables. But its effects can also emerge even in languages like Timugon Murut, where Onset is crucially dominated and so cannot require all syllables to have onsets; here, Onset nonetheless requires syllabification of available consonants as onsets rather than codas. This contrasts with a parameter-based view of phonology, where onsetless syllables are present only when the Onset parameter is “off,” and thus cannot affect syllabification in any way.

2.2

Output candidates not distinguished by dominating constraints

Unmarkedness can also emerge when multiple output candidates are evaluated identically by all constraints dominating the emergent markedness constraint. This section discusses allomorph selection, which has been traditionally analyzed as a TETU effect of this sort within OT, as well as a similar example from the syntax–phonology interface.

2.2.1 Allomorphy Mascaró (2004) observes that when a morpheme has multiple underlying forms, Gen supplies candidates that vary in the forms they correspond to (chapter 99: phonologically conditioned allomorph selection). In cases like English a/an, where the indefinite article has two lexically listed allomorphs, some members of the candidate set stand in correspondence with underlying a, while others stand

The Emergence of the Unmarked

5

in correspondence with underlying an. For this reason, the two output candidates shown in (6), a wug and *an wug, tie on all high-ranked IO-faithfulness constraints. While the ranking of faithfulness constraints (here, simply Faith) above NoCoda generally permits codas throughout English, NoCoda nevertheless emerges as decisive here, ruling out *an wug in this unique case where multiple possible outputs are equally faithful to their respective inputs.5 {a, an} /wZg/ Faith NoCoda

(6)

☞ a. H.wZg b. Hn.wZg

* **!

A more complex TETU analysis of lexically specific allomorph selection is offered in Becker’s (2009) discussion of the Turkish aorist (Lees 1961; Napiko-lu and Ketrez 2006). The aorist suffix has two allomorphs: /-Ir/, with a high vowel, is used after all polysyllabic roots; /-Er/, with a non-high vowel, is used after all monosyllabic obstruent-final roots (the backness and height of these vowels are determined by vowel harmony; chapter 118: turkish vowel harmony). Monosyllabic sonorant-final roots allow lexical exceptions: some take /-Ir/, while others take /-Er/. (7)

shape of stem polysyllabic obstruent-final monosyllabic {r l n}-final monosyllabic

affix -Ir -Er -Ir -Er

[gere’k-ir] [sA’t-Ar] [kA’l-qr] [dA’l-Ar]

‘need’ ‘sell’ ‘stay’ ‘dive’

[ŒAlq’œ-qr] [œ’p-er] [gœ’r-yr] [œ’r-er]

‘work’ ‘kiss’ ‘see’ ‘knit’

Turkish vowels are typically faithful to their underlying height specification, both in roots and in affixes; for example, the affix /-E/ (dative) (e.g. [je’re] ‘to the place’) contrasts with the affix /-I/ (3sg poss) (e.g. [je’ri] ‘his/her place’). This indicates that Ident[high] outranks both of the markedness constraints in (8). When two allomorphs are available to choose from, however, as in these aorist examples, Ident[high] is satisfied regardless of the choice of allomorph; the markedness constraints can thus emerge as decisive. (9) illustrates how *’q/high consistently selects the /-Er/ allomorph in monosyllabic obstruent-final roots. (8)

a. b.

(9)

*’q/high No stressed high vowels. *RER No non-high vowels between sonorants. /sAt-{-Er, -Ir}/ Ident[high] *’ /high *RER

☞ a. sA’t-Ar b. sA’t-qr 5

*!

The ranking Faith >> Onset similarly chooses [Hn.Zg] over *[H.Zg].

6

Michael Becker & Kathryn Flack Potts

The situation is more complex in the monosyllabic sonorant-final roots, which don’t behave uniformly. Some of these occur with /-Er/, violating *RER, as shown in (10), while others occur with /-Ir/, violating *’q/high, as in (11). Becker argues that sonorant-final monosyllabic roots are linked to lexically specified constraint rankings: for /-Er/-selecting roots like /dAl/, *’q/high >> *RER, while the opposite ranking holds for /-Ir/-selecting roots like /kAl/. The overall pattern is one where each markedness constraint is emergent for a particular class of roots. See Becker (2009) for further details of the analysis, including the treatment of polysyllables and mechanisms for learning both affix URs and lexically specific rankings. (10)

/dAl-{-Er, -Ir}/ Ident[high] *’ /high *RER ☞ a. dA’l-Ar b. dA’l-qr

(11)

* *!

/kAl-{-Er, -Ir}/ Ident[high] *RER a. kA’l-Ar ☞ b. kA’l-qr

*’ /high

*! *

This TETU analysis of the Turkish aorist accounts for the fact that the lexically specific distribution of this affix is limited to sonorant-final roots. Since *RER is ranked below Ident[high], its effect can only be observed when a root contributes one sonorant and one of two lexically listed allomorphs (here, of the aorist affix) contributes the other. This contrasts with a diacritic-based approach to exceptionality; since such an approach isn’t based on markedness constraints, it runs the risk of missing phonological restrictions on the distribution of exceptions. See Gouskova (2010) for further arguments in favor of a grammar-based approach to exceptionality (also chapter 106: exceptionality). Rankings in each of these allomorphy examples take the form F >> M, where F cannot distinguish between candidates containing different allomorphs. Because multiple candidates are equally faithful, satisfying M does not require violating a lower-ranked F2, as is required in the prototypical TETU cases discussed in §1. The Turkish example shows that satisfying an emergent markedness constraint can also require violating a lower-ranked markedness constraint, in a ranking like F >> M1 >> M2. This occurs because markedness constraints can conflict with each other, as well as with faithfulness constraints. The following discussion of phonological phrasing and the syntax–phonology interface carries this observation further, demonstrating that markedness constraints can also emerge in contexts where a higher-ranked, conflicting markedness constraint is inactive.

2.2.2 Phonological phrasing Because faithfulness constraints do not evaluate prosodic structure, analyses of phonological phrasing are generally based on rankings of conflicting markedness constraints. While most familiar TETU rankings involve domination by a conflicting faithfulness constraint, the dominating constraint may also be a second markedness constraint. That is, dominated M2 may also emerge in a ranking like (12).

The Emergence of the Unmarked (12)

7

M1 >> M2

Truckenbrodt (1999) proposes an analysis of phonological phrasing based only on markedness constraints; in contexts where high-ranking markedness constraints are rendered inactive, lower-ranking markedness constraints emerge. In Chewa (Bantu, referred to by Truckenbrodt as Chicheºa), a complex VP like [V NP NP]VP is produced as single phonological phrase, (V NP NP)PhP, rather than *(V NP)PhP (NP)PhP, as in (13). The large phrase satisfies Wrap-XP, a constraint that penalizes any syntactic phrase whose elements are parsed into smaller phonological phrases. Other dominated markedness constraints express conflicting preferences for smaller phonological phrases: Align-XP demands alignment of the right edge of each syntactic phrase with the right edge of a corresponding phonological phrase. The winner in (13) incurs a violation due to the first NP, which has no phrase break at its right edge. The ranking Wrap-XP >> Align-XP generally thwarts Align-XP’s desire for additional phonological phrases. Align-XP’s effects emerge, however, under focus. Align-Foc requires focused verbs to fall at the end of phonological phrases; the ranking Align-Foc >> Wrap-XP rules out candidate (14a). Candidates (14b) and (14c) both satisfy Align-Foc, and both violate Wrap-XP, rendering Wrap-XP inactive as well in selecting the optimal output. Because these candidates tie on high-ranking constraints, we again see a TETU effect: Align-XP emerges, selecting the unmarked candidate, (14c). (13)

/[V NP NP]VP/

Align-Foc

☞ a. (V NP NP)PhP b. (V NP)PhP (NP)PhP (14)

Wrap-XP

Align-XP *

*!

/[VFOC NP NP]VP/

Align-Foc

a. (VFOC NP NP)PhP b. (VFOC)PhP (NP NP)PhP

*!

☞ c. (VFOC)PhP (NP)PhP (NP)PhP

Wrap-XP

Align-XP

*

* *!

*

Truckenbrodt notes that this analysis of Chewa is particularly interesting, due to the non-local nature of the TETU effect: the appearance of an (unmarked) prosodic break after the focused verb causes another break to appear after a subsequent non-focused noun phrase.

2.3

Output segments not evaluated by specific faithfulness

This final subsection discusses situations where general IO-faithfulness is lowranked, and the emerging markedness constraint is instead dominated by a different type of faithfulness constraint. In other words, these are TETU rankings of the type Special-F >> M >> General-F. We discuss three kinds of faithfulness that can outrank general IO-faithfulness: positional faithfulness, which protects strong positions inside a candidate; output–output faithfulness, which protects the base in a morphologically complex form; and UseListed, which protects correspondents of existing forms in a speaker’s lexicon.

8

Michael Becker & Kathryn Flack Potts

2.3.1 Positional faithfulness Beckman (1999) examines patterns where contrasts are licensed only in strong positions like initial syllables, stressed syllables, and onsets. She analyzes these using positional faithfulness constraints, which assess correspondence only for segments in particular output positions (here, onsets). Catalan (Romance) allows contrastive voicing in onsets, but bans voiced obstruents in codas (chapter 69: final devoicing and final laryngeal neutralization). Beckman accounts for this with the ranking shown in (15)–(17). Underlyingly voiced coda obstruents are devoiced in surface forms, due to the ranking *VoiObs >> Ident[voice], as in (15). In onsets, however, underlying voicing surfaces faithfully, due to the high-ranking positional faithfulness constraint Ident[voice]/Onset, as in (16)–(17). Here the markedness constraint *VoiObs is dominated, yet active in non-onset contexts, making this a TETU effect. (15)

/griz/ ‘gray (masc)’

Ident[voice]/Onset *VoiObs Ident[voice]

a. ’griz ☞ b. ’gris (16)

/gos-a/ ‘dog (fem)’ a. ’go.zH ☞ b. ’go.sH

(17)

/griz-a/ ‘gray (fem)’ ☞ a. ’gri.zH b. ’gri.sH

**! *

*

Ident[voice]/Onset *VoiObs Ident[voice] *!

** *

*

Ident[voice]/Onset *VoiObs Ident[voice] *!

** *

*

In addition to protecting phonologically strong positions (initial syllable, stressed syllables, onsets), positional faithfulness may also protect morphologically strong contexts such as roots (McCarthy and Prince 1995) and nouns (Smith 1999, 2001; see also chapter 102: category-specific effects). Smith notes that in Spanish, stress is lexically marked in nouns but predictable in verbs. She analyzes a range of such patterns using noun-specific faithfulness constraints (F/Noun) in the ranking schema F/Noun >> M >> F. Here, nouns may be faithful to lexically specified stress thanks to high-ranking F/Noun; in verbs, however, stress is instead governed by emergent markedness constraints. The activity of markedness constraints in all of these rankings, despite their domination by a conflicting (here position-specific) faithfulness constraint, identifies these as TETU effects. There is a significant difference, however, between many positional faithfulness patterns and most other TETU patterns. In the TETU ranking schemata discussed in previous sections, general IO-faithfulness constraints dominate emergent markedness constraints. Here, though, markedness dominates (general, though not position-specific) IO-faithfulness. This results in different surface distributions of the emergent unmarked structures. Typically, when IO-F >> M, marked structures are licensed in most contexts throughout the language; unmarkedness emerges in specific, less frequent contexts like reduplicants, epenthetic segments, or allomorphs. When TETU results

The Emergence of the Unmarked

9

from high-ranking positional faithfulness, however, the reversed ranking M >> IO-F can result in a language which is largely unmarked; in these languages, marked structures are restricted to the specific set of contexts protected by the positional faithfulness constraint. When this set of markedness licensing contexts is small, as for Faith/q1 or Faith/’q (faithfulness to word-initial and stressed syllables, respectively), unmarked structures are required in the majority of contexts: the set of positions in which markedness may occur is atypically smaller than those where unmarkedness is required.6 This distributional pattern will be discussed further in §4.

2.3.2 Output–output faithfulness Another family of constraints which evaluates only some outputs and so gives rise to TETU effects is output–output (OO) faithfulness (Benua 1997). OO-faithfulness constraints evaluate correspondence between the base of a morphologically complex word and that base’s stand-alone surface form. Harris (1990) discusses examples of Aitken’s Law in dialects of the Central Scottish Lowlands. Here, stressed vowels in roots have predictable length: when followed by any consonant other than /r v Ï z/, vowels are short; otherwise, they are long. /> Z / are exceptions, remaining short in all positions. (See also chapter 20: the representation of vowel length.) For example, stop-final feed has a short vowel, while the open syllable key has a long vowel. The past tense keyed, however, keeps the long vowel which is present in its base key, despite its final stop coda. This can be attributed to protection from high-ranking OO-Faith, as described below (Benua 1997; McCarthy 2002). Because OO-faithfulness constraints target only a subset of a language’s output forms – those which are morphologically complex – they can give rise to TETU effects. A ranking like OO-F >> M >> IO-F operates much like the positional faithfulness TETU ranking discussed above. The markedness constraint *V(C] (“no long vowels in syllables closed by any consonant other than /r v Ï z/”) dominates IO-faithfulness; tableau (18) shows that this results in a language which is typically unmarked: long vowels are absent from closed syllables. Long vowels appear in open syllables, as in (19); because OO-Ident(length) >> *V(C], long vowels also appear in closed syllables in morphologically complex forms derived from roots with long vowels, as in keyed [ki(d] (cf. key [ki(]) in (20). (18)

/fi(d/ OO-Id(length) V(C] IO-Id(length) ☞ a. fid b. fi(d

(19)

/ki(/ a. ki ☞ b. ki(

6

* *! OO-Id(length) V(C] IO-Id(length) *!

Positional faithfulness TETU rankings can also result in languages where, as is more typical of TETU, unmarkedness is the less frequent pattern; this occurs when the positional faithfulness constraint targets a broad set of positions, e.g. Faith/Root.

10 (20)

Michael Becker & Kathryn Flack Potts /ki(-d/ OO-Id(length) V(C] IO-Id(length) a. kid ☞ b. ki(d

*!

* *

Here, again, a markedness constraint is active in the language despite its domination by (here, OO) faithfulness. Similar TETU effects are possible in other theories that use faithfulness relations between members of a paradigm, such as McCarthy’s (2005) Optimal Paradigms.

2.3.3

UseListed

Zuraw (2000) proposes an additional novel kind of faithfulness constraint, UseListed, which protects items that are listed in a speaker’s lexicon. Listed items include all roots and all morphologically complex forms that a particular speaker has heard, with more frequent items assumed to be more strongly listed. In producing a previously heard, morphologically complex form, the speaker has two options: they could use either the lexically listed forms of the root and affixes as inputs to the grammar, or they could instead use the single lexically listed complex form (again, as input to the grammar). Zuraw proposes that these two possible input structures compete in a single evaluation, with UseListed penalizing outputs derived from productive combinations of morphemes. For novel roots and novel complex forms (i.e. novel combinations of roots and affixes, even if a speaker is familiar with each morpheme in other contexts), however, no lexical listing is available. Thus outputs based on any of these forms will violate UseListed equally. Markedness constraints ranked below UseListed can therefore emerge in evaluations of unfamiliar items, as in Hayes and Londe’s (2006) analysis of Hungarian vowel harmony. The Hungarian dative appears with a back vowel when the root’s final syllable has a back vowel ([glyko(z-nDk] ‘glucose-dat’), and it appears with a front vowel when the root’s final syllable has a front rounded vowel ([œofø(r-nek] ‘chauffeur-dat’) (chapter 123: hungarian vowel harmony). When the root’s final syllable has a front unrounded vowel, some items take a front suffix ([tsi(m-nek] ‘address-dat’) and others take a back suffix ([hi(d-nDk] ‘bridge-dat’). Taking a back suffix is especially likely when the final front unrounded vowel is preceded by a back vowel ([a(tse(l-nDk] ‘steel-dat’). Here, the relevant markedness constraints will be Local[e(], which penalizes back vowels in the syllable immediately following an [e(], and Distal[back], which penalizes front vowels in any syllable following a back vowel. Hungarian speakers agree on the dative forms of familiar (lexically listed) items such as [a(tse(l-nDk]. UseListed is decisive in these cases, preferring the listed form over productive combinations of the root and the suffix, and thus rendering lower-ranked markedness constraints on vowel harmony inactive. The two candidates (21a) and (21b) are generated from the listed form [a(tse(l-nDk], and thus satisfy UseListed (despite the unfaithful surface form of this input in (21b)). The second two candidates are generated productively by combining the root /a(tse(l/ with the dative suffix, and are thus ruled out by UseListed.

The Emergence of the Unmarked (21)

/a(tse(l-{nek, nDk}/, listed: [a(tse(l-nDk]

11

Use Ident Local Distal Listed [back] [e(] [back]

☞ a. /a(tse(l-nDk/ → a(tse(l-nDk b. /a(tse(l-nDk/ → a(tse(l-nek

*

* **

*

*

*!

c. /a(tse(l-{nek, nDk}/ → a(tse(l-nDk

*!

d. /a(tse(l-{nek, nDk}/ → a(tse(l-nek

*!

**

But when Hungarian speakers hear a novel root containing a back vowel followed by a front unrounded vowel, e.g. [ha(de(l], the suffix vowel in the dative forms can agree with either root vowel: some speakers prefer [ha(de(l-nDk], as in (22a), while others prefer [ha(de(l-nek], as in (22b). Both candidates in (22) violate UseListed, since no lexical listing exists for this novel item, and thus the dative form must be derived productively by combining the root /ha(de(l/ with the dative suffix. (22)

/ha(de(l-{n k, nDk}/, listed: [ ] ☞ a. /ha(de(l-{n k, nDk}/ → ha(de(l-nDk ☞ b. /ha(de(l-{n k, nDk}/ → ha(de(l-n k

Use Ident Local Distal Listed [back] [e(] [back] * *

*

* **

Hayes and Londe argue that a particular speaker’s actual output depends on a stochastic ranking between the two competing markedness constraints on vowel harmony, Local[e(] and Distal[back]. Crucially, as in the Turkish example in §2.2.1, one of these two low-ranked markedness constraints emerges; here, this occurs when dominating UseListed cannot distinguish between candidate outputs for a novel input.

3

Gradient TETU

The previous sections have surveyed various ways in which high-ranking constraints can be rendered irrelevant in particular evaluations, allowing lower-ranked markedness constraints to emerge. Of course, not all markedness constraints which are dominated in a particular language emerge; many are ranked too low to ever be active in choosing a winning surface form. Recent work suggests, however, that subtle TETU effects can be identified even for markedness constraints which never distinguish between grammatical and ungrammatical forms. Consonants in Arabic roots are subject to various co-occurrence restrictions (chapter 86: morpheme structure constraints). Among grammatical consonant combinations, preferences for particular combinations are found: some are much more frequent than others in the lexicon, and novel words conforming to the more frequent patterns are judged as more well-formed in rating tasks (chapter 90: frequency effects). Coetzee and Pater’s (2008) analysis of these preferences among grammatical forms casts them as gradient TETU effects, where markedness constraints which are never categorically obeyed nevertheless exert subtle preferences for unmarked forms (chapter 89: gradience and categoricality in

12

Michael Becker & Kathryn Flack Potts

phonological theory). Rankings giving rise to these gradient effects are illustrated in (23) and (24). Roots including both coronal stops and fricatives (e.g. /dasar/ ‘to push’, represented here as TS) and those containing coronal stops and sonorants (e.g. /dalaq/ ‘to spill’, represented here as TL) both surface faithfully in Arabic, although these combinations are underrepresented, i.e. they are attested less often than expected, given the overall frequency of each type of consonant. In the lexicon, however, TS roots are more severely underrepresented than TL roots, suggesting that TL roots are in some sense more easily tolerated. Coetzee and Pater argue that both combinations violate a constraint against roots with two coronals (*TT), while only the dispreferred TS roots violate an additional constraint against roots with two coronals of similar sonority (*TT[son]). (23)

/TS/ ☞ a. TS b. PS

(24)

/TL/ ☞ a. TL b. PL

Ident(place) *TT[son] *TT *

*

*! Ident(place) *TT[son] *TT * *!

Here, no markedness constraint is ranked highly enough to ban TS or TL outputs. These consonant combinations are protected by faithfulness, and are thus attested and grammatical, but they are not judged by speakers to be quite as well formed as roots that lack OCP violations. TS’s additional violation of *TT[son] contributes to its decreased acceptability relative to TL, as observed in the results of word-likeness tasks and similar psycholinguistic experiments. In other words, the markedness constraint *TT[son] is active in Arabic even though it is crucially dominated by Ident(place). This activity is evidenced by the underattestation of actual TS roots and the decreased acceptability of novel TS roots, even though it doesn’t force unfaithful mappings. The incorporation of gradient generalizations into the grammar can also be used to identify relative rankings of undominated markedness constraints, i.e. the opposite of gradient TETU. If neither of two markedness constraints is ever crucially dominated by a conflicting constraint in some language, the relative ranking of these constraints cannot be determined from either categorical phonotactics or paradigmatic information. This approach, however, allows evidence for their relative ranking to come from gradient phonotactics and psycholinguistic data. Coetzee (2009) compares the grammaticality of English homorganic stops after [s], noting that coronals are attested, as in state, but labials and dorsals are not, as in *skake or *spape (see also Davis 1984, 1991; Frisch 1996; Frisch et al. 2004). Coetzee’s psycholinguistic experiments show that speakers rate *spape as less acceptable than *skake, and both are less acceptable than state. He uses this result to propose that while *spVp and *skVk are both undominated in English, the constraint penalizing *spVp is more highly ranked than the constraint penalizing *skVk. This view is also supported by the existence of words that come close to

The Emergence of the Unmarked

13

violating *skVk, such as skag, skulk, or squeak, compared with the non-existence of *spab, *spulp, or *spweep.

4

The emergence of the faithful

As mentioned in §2.3.1, there is an important distinction between the formal definition of TETU and the most intuitive surface-oriented descriptions of these patterns. McCarthy and Prince (1994: 334) define TETU as follows: Even in languages where [some markedness constraint] C is crucially dominated and therefore violated, the effects of C can still be observed under conditions where the dominating constraint is not relevant. . . . this [is] “emergence of the unmarked.”

The same passage describes the typical surface pattern that results from constraint activity despite domination: in the language as a whole, C may be roundly violated, but in a particular domain it is obeyed exactly. In that particular domain, the structure unmarked with respect to C emerges.

Patterns where high-ranking positional faithfulness constraints allow unmarkedness to emerge (e.g. Ident[voice]/Onset >> *VoiObs >> Ident[voice]) demonstrate that activity-despite-domination rankings can also give rise to a converse pattern: a markedness constraint may in fact be obeyed in the language as a whole, but violated in a particular domain. In these cases, the structure unmarked with respect to the markedness constraint emerges in the language as a whole, despite its ungrammaticality in a particular domain. The lack of a necessary connection between a markedness constraint’s activity despite domination and the relative rarity of the resulting unmarkedness is also illustrated in patterns following from the activity of positional markedness constraints. Like positional faithfulness constraints, these are versions of regularly attested markedness constraints which evaluate only structures in particular output positions, e.g. Onset/q1, a constraint that penalizes onsetlessness in the initial syllable only (chapter 55: onsets). An example of this pattern comes from Arapaho (Algonquian, Smith 2002: 127, from Salzman 1956: 53–54). In this language, onsetless syllables occur in non-initial syllables (e.g. the onsetless third syllable in [wo.’?o.u(.so(] ‘kitten’), as shown in (25). Word-initial vowels are, however, banned (e.g. *[o.to?]), as shown in (26). These patterns follow from the ranking Onset/q1 >> Dep >> Onset, and are identical in character to the typical TETU surface pattern: marked structures are licensed in most of the language, but a small pocket of enforced unmarkedness is found in initial syllables. Many other languages of this type are discussed by Smith (2002) and Flack (2009). (25)

/wo’?ou(so(/ Onset/ 1 Dep Onset ☞ a. wo.’?o.u(.so( b. wo.’?o.?u(.so(

* *!

14 (26)

Michael Becker & Kathryn Flack Potts /oto?/ Onset/ 1 Dep Onset ☞ a. ho.to? b. o.to?

* *!

*

Despite the surface similarities between positional markedness patterns and classic TETU patterns, the formal structure of these rankings distinguishes them from TETU rankings. These follow the schematic form M1 >> F >> M2, rather than the TETU form F1 >> M >> F2. These patterns thus might be dubbed “The Emergence of the Faithful”: F emerges (i.e. is active despite domination) in cases where dominating Onset/q1 is inactive. Albright (2004) discusses patterns of this sort, using the term “The Emergence of the Marked” to describe their surface pattern. In Lakota (Siouan), codas are banned in roots but licensed elsewhere (e.g. affixes, reduplicants, function words). This pattern results from the positional markedness ranking NoCoda-root >> F >> NoCoda; as Albright explains, this pattern is a “mirror image of the TETU configuration: here, greater faithfulness emerges outside roots, when a higher-ranked markedness constraint (NoCoda-root) is inapplicable” (2004: 7). Here, markedness “emerges” in distributionally rare root-external contexts. To be clear, the distributional sense of “emerge” used here is different from the formal sense used by McCarthy and Prince: formally, effects of the constraint which is dominated yet active emerges; distributionally, whichever pattern is not generally permitted (markedness vs. unmarkedness) “emerges” in specific, restricted contexts.

5

Conclusion

TETU is a property of theories with violable constraints, and sets these theories apart from those with parameters or inviolable constraints. In TETU rankings, a markedness constraint is shown to be dominated in a language, yet active in situations where the dominating constraints are irrelevant. Three types of such situations are surveyed in §2. Active-yet-dominated markedness constraints have also been used in the analysis of gradient patterns, as discussed in §3. Finally, patterns mirroring TETU which result from active-yet-dominated faithfulness constraints are discussed in §4. TETU effects, which demonstrate the violability of OT constraints, set OT apart from theories with inviolable constraints, also known as parameters in Principles and Parameters Theory (Chomsky 1981, 1986). In Principles and Parameters Theory, the learner starts with parameters set to their default, or unmarked, position; parameters can be switched off given evidence from the ambient language. The NoCoda parameter, for instance, will remain on for a speaker of Hawaiian, as this language doesn’t allow codas. However, speakers of English or Nuu-chah-nulth will switch the NoCoda parameter off, as codas are generally allowed in these languages. Once off, however, the NoCoda parameter can no longer be used to account for the contexts in which these two languages prefer codaless syllables (see §1 and §2.2.1 above), causing a loss of generality in the analysis of these languages (McCarthy 2002: 131–132). Interest in TETU effects initially brought attention to a variety of cases where constraints were shown to be active even in languages where they were roundly

The Emergence of the Unmarked

15

violated, e.g. NoCoda in English. This lent support to the view that there is a single, universal constraint set for all languages, which in turn led to fruitful research on how language-specific rankings of these universal constraints could be learned (see e.g. Tesar 1995; Tesar and Smolensky 1998; and much work since). Early work in OT typically assumed that this universal constraint set was innate; assumptions of both innateness and constraint universality have begun to lose favor in recent years with the advent of proposals that some or all constraints are induced by learners (Flack 2007; Hayes and Wilson 2008; Moreton 2010). TETU effects were a major focus of interest in the early days of Optimality Theory, when the concept of violable constraints was new to the linguistic community. With the increased acceptance of violable constraints in theoretical work, cases of TETU no longer attract special attention, even as interest turns to other theories that incorporate violable constraints, including OT-CC (McCarthy 2007), Harmonic Grammar (Legendre et al. 1990; Pater 2009), and MaxEnt (Goldwater and Johnson 2003; Hayes and Wilson 2008).

REFERENCES Albright, Adam. 2004. The emergence of the marked: Root-domain markedness in Lakhota. Handout of paper presented at the 78th Annual Meeting of the Linguistic Society of America, Boston. Becker, Michael. 2009. Phonological trends in the lexicon: The role of constraints. Ph.D. dissertation, University of Massachusetts, Amherst. Beckman, Jill N. 1999. Positional faithfulness: An optimality theoretic treatment of phonological asymmetries. New York: Garland. Benua, Laura. 1997. Transderivational identity: Phonological relations between words. Ph.D. dissertation, University of Massachusetts, Amherst. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, Noam. 1986. Knowledge of language: Its nature, origin, and use. New York: Praeger. Coetzee, Andries W. 2009. Grammar is both categorical and gradient. In Steve Parker (ed.) Phonological argumentation, 9–42. London: Equinox. Coetzee, Andries W. & Joe Pater. 2008. Weighted constraints and gradient restrictions on place co-occurrence in Muna and Arabic. Natural Language and Linguistic Theory 26. 289–337. Davis, Stuart. 1984. Some implications of onset–coda constraints for syllable phonology. Papers from the Annual Regional Meeting, Chicago Linguistic Society 20. 46–51. Davis, Stuart. 1991. Coronals and the phonotactics of nonadjacent consonants in English. In Carole Paradis & Jean-François Prunet (eds.) The special status of coronals: Internal and external evidence, 49–60. San Diego: Academic Press. Flack, Kathryn. 2007. The sources of phonological markedness. Ph.D. dissertation, University of Massachusetts, Amherst. Flack, Kathryn. 2009. Constraints on onsets and codas of words and phrases. Phonology 26. 269–302. Frisch, Stefan A. 1996. Frequency and similarity in phonology. Ph.D. dissertation, Northwestern University. Frisch, Stefan A., Janet B. Pierrehumbert & Michael B. Broe. 2004. Similarity avoidance and the OCP. Natural Language and Linguistic Theory 22. 179–228. Goldwater, Sharon & Mark Johnson. 2003. Learning OT constraint rankings using a maximum entropy model. In Jennifer Spenader, Anders Eriksson & Osten Dahl (eds.) Proceedings of the Stockholm Workshop on Variation within Optimality Theory, 111–120. Stockholm: Stockholm University.

16

Michael Becker & Kathryn Flack Potts

Gouskova, Maria. 2010. Unexceptional segments. Unpublished ms., New York University. Harris, John. 1990. Derived phonological contrasts. In Susan Ramsaran (ed.) Studies in the pronunciation of English: A commemorative volume in honour of A. C. Gimson, 87–105. London: Routledge. Hayes, Bruce & Zsuzsa Cziráky Londe. 2006. Stochastic phonological knowledge: The case of Hungarian vowel harmony. Phonology 23. 59–104. Hayes, Bruce & Colin Wilson. 2008. A maximum entropy model of phonotactics and phonotactic learning. Linguistic Inquiry 39. 379–440. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Lees, Robert B. 1961. The phonology of Modern Standard Turkish. Bloomington: Indiana University. Legendre, Géraldine, Yoshiro Miyata & Paul Smolensky. 1990. Harmonic Grammar: A formal multi-level connectionist theory of linguistic well-formedness – an application. In Proceedings of the 12th Annual Conference of the Cognitive Science Society, 884–891. Mahwah, NJ: Lawrence Erlbaum. Mascaró, Joan. 2004. External allomorphy as emergence of the unmarked. In John J. McCarthy (ed.) Optimality Theory in phonology: A reader, 513–522. Malden, MA & Oxford: Blackwell. McCarthy, John J. 2002. A thematic guide to Optimality Theory. Cambridge: Cambridge University Press. McCarthy, John J. 2005. Optimal paradigms. In Laura J. Downing, T. A. Hall & Renate Raffelsiefen (eds.) Paradigms in phonological theory, 170–210. Oxford: Oxford University Press. McCarthy, John J. 2007. Hidden generalizations: Phonological opacity in Optimality Theory. London: Equinox. McCarthy, John J. & Alan Prince. 1994. The emergence of the unmarked: Optimality in prosodic morphology. Papers from the Annual Meeting of the North East Linguistic Society 24. 333–379. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 249–384. Amherst: GLSA. McCarthy, John J. & Alan Prince. 1999. Faithfulness and identity in prosodic morphology. In René Kager, Harry van der Hulst & Wim Zonneveld (eds.) The prosody–morphology interface, 218–309. Cambridge: Cambridge University Press. Moreton, Elliott. 2010. Connecting paradigmatic and syntagmatic simplicity bias in phonotactic learning. Paper presented at MIT, April 2010. Napiko-lu, Mine & Nihan Ketrez. 2006. Children’s overregularizations and irregularizations of the Turkish aorist. In David Bamman, Tatiana Magnitskaia & Colleen Zaller (eds.) Proceedings of the 30th Annual Boston University Conference on Language Development, vol. 2, 399–410. Somerville, MA: Cascadilla Press. Pater, Joe. 2009. Weighted constraints in generative linguistics. Cognitive Science 33. 999–1035. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Salzman, Zdenek. 1956. Arapaho I: Phonology. International Journal of American Linguistics 22. 49–56. Smith, Jennifer L. 1999. Noun faithfulness and accent in Fukuoka Japanese. Proceedings of the West Coast Conference on Formal Linguistics 18. 519–531. Smith, Jennifer L. 2001. Lexical category and phonological contrast. In Robert Kirchner, Joe Pater & Wolf Wikely (eds.) Papers in experimental and theoretical linguistics 6: Workshop on the Lexicon in Phonetics and Phonology, 61–72. Edmonton: University of Alberta.

The Emergence of the Unmarked

17

Smith, Jennifer L. 2002. Phonological augmentation in prominent positions. Ph.D. dissertation, University of Massachusetts, Amherst. Steriade, Donca. 2001. Directional asymmetries in place assimilation: A perceptual account. In Elizabeth Hume & Keith Johnson (eds.) The role of speech perception in phonology, 219–250. San Diego: Academic Press. Steriade, Donca. 2009. The phonology of perceptibility effects: The P-map and its consequences for constraint organization. In Kristin Hanson & Sharon Inkelas (eds.) The nature of the word: Studies in honor of Paul Kiparsky, 151–179. Cambridge, MA: MIT Press. Tesar, Bruce. 1995. Computational Optimality Theory. Ph.D. dissertation, University of Colorado, Boulder. Tesar, Bruce & Paul Smolensky. 1998. Learnability in Optimality Theory. Linguistic Inquiry 29. 229–268. Truckenbrodt, Hubert. 1999. On the relation between syntactic phrases and phonological phrases. Linguistic Inquiry 30. 219–255. Vaux, Bert. 2002. Consonant epenthesis and the problem of unnatural phonology. Handout from paper presented at Yale University. Vaux, Bert. 2008. Why the phonological component must be serial and rule-based. In Bert Vaux & Andrew Nevins (eds.) Rules, constraints, and phonological phenomena, 20–60. Oxford: Oxford University Press. Zuraw, Kie. 2000. Patterned exceptions in phonology. Ph.D. dissertation, University of California, Los Angeles.

59

Metathesis Eugene Buckley

The term metathesis – Greek for ‘transposition’ – refers to a reordering of segments. This chapter outlines the range of phenomena that fall under this description, and theoretical perspectives on their insightful analysis. Other cross-linguistic surveys of this topic include Webb (1974), Ultan (1978), Hock (1985), Wanner (1989), Blevins and Garrett (1998, 2004), Becker (2000), and Hume (2001, 2004). The term has traditionally been best known for the description of historical sound changes (chapter 93: sound change), often described as sporadic. For example, Osthoff and Brugmann (1878: xiv, n. 1) cite metathesis, along with dissimilation (chapter 60: dissimilation), as lacking the “mechanical” character of regular sound change. Hock (1985), however, argues that diachronic metathesis is regular when it serves to enforce a structural constraint. For example, in early attestations of Persian, as well as in reconstructed forms, clusters of an obstruent or nasal plus a liquid can be found before a final vowel. Loss of that vowel leads to a final cluster with a rising sonority profile (chapter 49: sonority; chapter 46: positional effects in consonant clusters); this configuration is repaired by metathesis of the two consonants, so that the more sonorous liquid is closer to the vowel. (The segments involved are underlined in (1).) (1)

Persian liquid metathesis (Hock 1985: 534) suxra vafra asru *namra

> > > >

surx barf ars narm

‘red’ ‘snow, ice’ ‘tear’ ‘soft’

Although much of the literature discusses historical metathesis – where copious examples can be found – this chapter focuses on instances of metathesis that are active synchronically. By this I mean alternations in the ordering of segments that appear to be part of a speaker’s productive grammatical knowledge, and therefore must be accounted for in theories of linguistic competence. There is of course an intimate connection between diachronic metathesis and the synchronic alternations that may persist in the grammar as a result, but I will take care to distinguish examples for which only diachronic change is well attested, The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0059

Metathesis

2

and where the results of the change appear to be new underlying forms rather than a new phonological alternation. Similarly, although the emphasis is on phonologically defined patterns, some types of metathesis require reference to morphological context, even if the specific change is expressed in terms of phonological categories. For most of the twentieth century, metathesis was described either in prose or, as formalisms became more sophisticated, as reorderings of indexed objects in a string. Chomsky and Halle (1968: 361) describe metathesis as “a perfectly common phonological process,” and permit transformations that effect permutation. In their notation, /skt/ → [kst] metathesis in Faroese, shown below in (3), could be expressed as follows. (2)

Metathesis as a transformation Structural description

s

k

t

Structural change

1 2

3

→ 2

1

3

The need for indexation distinguishes metathesis from most other processes, such as insertion (chapter 67: vowel epenthesis), deletion (chapter 68: deletion), and featural assimilation (chapter 81: local assimilation). In those sorts of changes, whatever elements of the representation remain after the change maintain their relative ordering on their tier. A true featural equivalent to segmental metathesis would be a swap in feature values on the same tier (or at some non-root node), such as a change from LH to HL tone in a context where underspecification of the L with simple shift of H is not a plausible analysis. As noted in §1.4, there is limited evidence for tonal metathesis of this sort. Following the most common modern usage, in this chapter I apply the term metathesis to permutations of segments regardless of intervening material. §1 deals with local metathesis, including the sequences CC, CV, and VV, followed by brief consideration of other types. §2 considers the long-distance metathesis of non-adjacent segments, as well as the displacement of a segment that is not exchanged with another. §3 considers the relation of metathesis to other phenomena with which it shares some formal properties, such as infixation.

1

Local metathesis

In local metathesis, two adjacent segments are swapped, without any necessary change in their features, although in some cases other processes may affect the outcome. These can be classified formally according to the segments involved in the reversal: two consonants, a consonant and a vowel (in either order), or two vowels.

1.1

CC metathesis

To organize this presentation, I group the processes according to the features of the segments involved. These include the special role of sibilants, place of articulation, and manner of articulation.

3

Eugene Buckley

1.1.1 Sibilants Sibilant consonants are often observed to reverse order with an adjacent stop consonant (Silva 1973; Hume 2001: 12–14; Seo and Hume 2001; Steriade 2001: 234f.; Blevins and Garrett 2004: 139f.; Hume and Seo 2004: 36–39). An example is found in Faroese, where /sk/ followed by /t/ is reversed (Lockwood 1955: 23f.). (3)

Faroese metathesis of /sk/ (Lockwood 1955: 24) masculine fesk-Ár rask-Ár dansk-Ár

neuter feks-t raks-t daIks-t

‘fresh’ ‘energetic’ ‘Danish’

As noted above, metathesis has typically been considered a sporadic or irregular process, unlike phenomena such as assimilation that can often be described in very general and regular terms (see Hume 2001: 1f. for representative quotations). But the Faroese reversal illustrates that a process of metathesis can be fully regular while also quite restricted in scope, simply because the necessary configuration does not often arise. Thus the neuter noun suffix /t/ provides the environment for reversal of stem-final /sk/; but a similar environment in verbs also triggers the changes, as can be seen in /ÁõIks-ti/ ‘wish (past sg)’ compared to the present singular /Áõnsk-ir/ with the underlying ordering thanks to the following vowel (Hume 1999: 294). It was traditionally claimed that metathesis yields sequences that are in some way better formed than the input ordering, usually in the sense of “ease of articulation” or satisfying a language’s phonotactic constraints (Wechssler 1900: 497; Grammont 1933: 239; Ultan 1978: 390). More recent work has placed greater emphasis on the role of perception (chapter 98: speech perception and phonology), and on historical explanations for how metathesis arises (chapter 93: sound change). Faroese can be seen as auditory metathesis – the temporal decoupling of the noise of a fricative, especially a sibilant, from the surrounding signal, which can lead to a sibilant and an adjacent stop being reinterpreted as occurring in the opposite of the original order (Blevins and Garrett 2004: 120). A segment often moves to a position in which it is more easily perceptible, especially due to the formant transitions in an adjacent vowel (Hume 1999: 295f.; Seo and Hume 2001: 215–217). Thus Faroese metathesis places the stop /k/ in a more perceptible position, adjacent to the preceding vowel, while the sibilant remains perceptible without an adjacent vowel. This directionality suggests that confusibility in the ordering of the segments is not the sole factor, since symmetrical confusion predicts random reordering according to the two possible interpretations of an ambiguous auditory signal (Steriade 2001: 233–235); but see Blevins and Garrett (2004: 119f.) for a defense of the misperception account. The outcome in particular languages may depend on prosody, such as the location of stress, and phonetic detail, such as the release of final stops; such differences may explain the symmetrically opposite changes in Late West Saxon (/frosk/ → [froks] ‘frog’) and a certain variety of colloquial French (/fiks/ → [fisk] ‘fixed’) (Blevins and Garrett 2004: 139f.). A transformational rule that reverses the order of segments does not make reference to the apparent motivations of the reordering, such as an improvement in markedness (chapter 4: markedness). But like other phonological processes,

Metathesis

4

metathesis may operate in order to satisfy the phonotactic restrictions of a language. That is, just as the place assimilation in anba → amba satisfies a condition that nasal codas must agree in place with a following stop, so a metathesis such as inma → imna in (6) satisfies a condition on the sequencing of coronal and labial consonants. Recent approaches have attempted to capture this insight and to treat metathesis more on a par with other processes. In the surface orientation of Optimality Theory (Prince and Smolensky 2004), the expected Faroese sequence [skt] can be penalized by a constraint against a stop that occurs between two other consonants (Hume 1999: 298), whether it is defined directly in terms of perceptibility or as a more abstract configuration. This pressure must dominate the correspondence constraint Linearity, which otherwise prevents reorderings of segments, and obviously plays a central role in the analysis of metathesis (Hume 1998: 149, 68f.; McCarthy and Prince 1995: 371f.; McCarthy 2000: 173). Metathesis occurs only when Linearity is ranked below faithfulness constraints such as Max and Dep; these prevent deletion or insertion of material that otherwise might serve to remedy the surface constraint that metathesis addresses. /raskt/ *Stop/C__C Max Dep Linearity

(4)

☞ a. raskt b. rast

*! *!

c. rask

*!

d. raskit

*!

e. rakst

*

Naturally, no violation of Linearity is required in a form such as [raskÁr], where the stop /k/ is adjacent to a vowel, and the sequence surfaces intact. Another relatively restricted case of stop–sibilant metathesis is the Tiberian Hebrew hitpa‘el verb form, where the /t/ of the prefix reverses with a steminitial sibilant (Malone 1993: 52f.; Coetzee 1999: 106; see Malone 1971 for similar facts in other Semitic languages; see also chapter 108: semitic templates). The examples in (5a) show the lack of metathesis with non-sibilants. (5)

Tiberian Hebrew metathesis (Coetzee 1999: 106) a. b.

hit-pallel hit-qaddeœ hit-sappex hit-œammer hit-Wakker hit-zakker hit-s#addeq

→ → → → → → →

hitpallel hitqaddeœ histappex hiœtammer hiWtakker hizdakker his#t#addeq

‘he ‘he ‘he ‘he ‘he ‘he ‘he

prayed’ sanctified himself’ felt attached to’ protected himself’ gave himself into service’ remembered’ considered himself righteous’

For Coetzee (1999: 122f.), the motivation for metathesis in exactly this context, when a /t/ would otherwise precede a sibilant, is that a [t] + sibilant sequence would be subject to reinterpretation as an affricate, a type of segment disfavored in Tiberian Hebrew. He proposes a constraint *t+Sibilant against that sequence,

5

Eugene Buckley

again with relatively low-ranked Linearity. Hume (2004: 222f.), discussing the equivalent metathesis in Modern Hebrew, argues that the poor attestation of [t] + sibilant sequences in Hebrew sets the stage for a reinterpretation with the sibilant in first position: an ambiguous acoustic signal is likely to be interpreted sequentially according to the most commonly attested ordering of those segments, dependent not necessarily on universal principles, but on the lexicon and grammar of the language in question.

1.1.2 Place of articulation Many instances of CC metathesis depend on place of articulation (chapter 22: consonantal place of articulation), with certain orderings of place favored over others. Permutation of this type is found in a range of Malayo-Polynesian languages (Blevins and Garrett 2004: 136). In Cebuano, for instance, a coronal stop or nasal followed by a labial or velar consonant is reversed, optionally in some cases (Blust 1979: 110).1 The consonants come to be adjacent as the result of vowel syncope after a vowel-initial suffix is added. (6)

Cebuano metathesis of coronal + non-coronal clusters (Blust 1979: 110) stem lutuk gitik atup inum

suffixed form lukt-un gitk-anun ~ gikt-anun atp-an ~ apt-an imn-a

‘put the finger in’ ‘ticklish’ ‘roof’ ‘drink’

Stems such as /lakat/ ‘walk’ that already have the preferred ordering maintain it ([lakt-un]), showing that the process is not simply an across-the-board reversal in consonant clusters. In this case, the favored ordering places the coronal in second position. The two changes in Cebuano – deletion of the vowel and reversal in the resulting cluster – were likely ordered historical events, and this history can be modeled easily by ordered synchronic rules. But the same facts are also consistent with simultaneous satisfaction of two surface constraints in OT. The candidates *[lutukun] and *[lutkun] both violate one of these constraints – by the lack of syncope, or the disfavored consonant ordering – whereas [luktun] satisfies both, and wins under low ranking of Linearity and Max-V. (7)

/lutuk-un/ *VCVCV *TK Linearity Max-V a. lutukun b. lutkun ☞ c. luktun

*! *!

* *

*

A phonetic explanation for this type of reordering is co-articulatory metathesis, which results from the overlap in adjacent consonant gestures (Blevins and Garrett 2004: 1

Similar patterns are found in a number of other Philippine languages (Blust 1971: 85f.; 1979: 104f.; Crowhurst 1998: 597), including Tagalog, where, however, it is poorly attested and classified with irregular verbs (Schachter and Otanes 1972: 375–380).

Metathesis

6

136–138); for example, overlapping coronal (T) and non-coronal closures (K) are perceived as the non-coronal, which leads to reversals such as TK → KT in Cebuano (6). A general preference for apicals to follow non-apicals has been cited with regard to metathesis in other languages such as Greek, which may be related to the tendency for coronal codas to assimilate to following non-coronals (Bailey 1970: 348). One abstract phonological approach formalizes the licensing properties of different places of articulation (Rubin 2001: 194–199); unmarked Coronal is a natural head and licenses the place features of a preceding non-coronal, favoring KT over TK. See also Blust (1979: 102f.) and Winters (2001) on the general preference for coronals to occur second in a cluster. Some reorderings have considerably more complex origins. In the Kondh branch of Dravidian, sequences of a velar /k g/ plus a labial /p b/ are reversed. The Pengo examples below illustrate two allomorphs /-pa/ and /-ba/ of the intensivefrequentative or plural action suffix, both of which also occur in contexts without metathesis as seen in (8a). For similar Kui examples, see Hume (2001: 8). (8)

Pengo velar + labial metathesis (Burrow and Bhattacharya 1970: 82f., 201) a. b.

gru(t-pahuz-baÕrik-paku(k-paÈa(k-batog-ba-

→ → → → → →

gru(t-pahuz-baÕripkaku(pkaÈa(bgatobga-

‘fell’ ‘roast’ ‘break’ ‘call’ ‘sacrifice’ ‘be split’

According to Garrett and Blevins (2009: 538ff.), this metathesis pattern arose by re-analysis of complex allomorphy deep in the history of Dravidian. Briefly, causative /p/ could replace the last consonant of the stem, as in Kolami [melg-] ‘grow (intr)’ and derived [mel-p-] ‘rear’. This was interpreted as a rule deleting the velar before the labial, which was extended to other labial-initial suffixes, including the plural action containing [-p-]. This would yield an alternation between simple */ku(k-/ ‘call’ and plural action */ku(k-p/→ *[ku(-p-]. But there is another basic allomorph of the plural action suffix containing /-k-/; if this were added to the existing plural action in order to make the exponence of that category clearer, the result is the pair *[ku(k-] and *[ku(-p-k-], which then gives the appearance of metathesis of the suffixal /p/ and the stem-final /k/. Whatever the historical origin of Pengo metathesis, it became part of the grammar thanks to learners treating it as an active synchronic process. The constraint encoding the Pengo alternation must penalize a velar + labial sequence; call it *KP. In addition to Max and Dep, it is especially relevant here to include the constraint Ident to prevent changes to the features targeted by the phonotactic constraint. (9)

/togba/ *KP Max Dep Ident Linearity a. togba b. toba c. togiba d. tomba ☞ e. tobga

*! *! *! *! *

7

Eugene Buckley

Perceptual factors may contribute to such re-analyses. Placing the labial first in the cluster, at least in related Kui, puts it in the stressed syllable, which may enhance its perceptibility; the weak bursts of labial stops reduce the benefit of being located in the onset (Hume 1999: 296). Experimental evidence indicates that labial place is more perceptible in codas than is velar, and that velars benefit more in perceptibility from being in the onset than do labials (Winters 2001: 238–241). This means that the ordering PK is overall more likely to be heard correctly than KP. Emphasizing historical origin, however, Blevins and Garrett (2004: 136) consider the perceptually based prediction to be PK → KP, as attested in other languages such as Mokilese /apkas/ → [akpas] ‘now’, and claim that the Dravidian pattern favoring PK could arise only by such means as re-analysis of a morphological pattern. Segments undergoing metathesis may originate in different morphemes, as in Pengo, but may also occur inside a single morpheme. Across a morpheme boundary, the offending cluster is created by concatenation; within a morpheme, the context for metathesis may be created directly by a syncope rule that brings the consonants into contact (as in Cebuano (6)), or a triggering context introduced by concatenation but affecting two consonants that are underlyingly adjacent (as in Faroese (3)).

1.1.3 Manner of articulation Classes of consonants defined by manner, such as liquids or sonorants (see chapter 13: the stricture features), are often targeted specifically by metathesis. (One might also include here the sibilants discussed in §1.1.1.) Metathesis involving the class of liquids is found in a number of languages (Blevins and Garrett 2004: 128f.); a historical example from Persian was cited in (1). In Rendille (Cushitic, Kenya), an /r/ and a preceding obstruent or nasal reverse in order after they become adjacent upon deletion of the intervening vowel (Sim 1981: 7, 9f.; Hume 1998: 178; Blevins and Garrett 2004: 129). (10)

Rendille metathesis of /r/ in clusters (Sim 1981: 7, 9) feminine údur-te ágar-te pámar-te

masculine úrd-e árg-e párm-e

‘s/he slept’ ‘s/he saw’ ‘s/he shivered’

In the framework of Blevins and Garrett (2004: 121–125), this perceptual metathesis arises when the cues for a sequence of sounds are perceived by the listener as reordered relative to the speaker’s intention, which is possible when some feature is realized over a relatively long duration and therefore contains ambiguity of analysis. This is true for consonant clusters as well as vowel–consonant sequences (see §1.2). Besides liquids (chapter 30: the representation of rhotics; chapter 31: lateral consonants), other segment types with elongated cues include pharyngeals (chapter 25: pharyngeals), secondary labialization (chapter 29: secondary and double articulation), palatalization (chapter 71: palatalization), and glottalization or aspiration (Blevins and Garrett 2004: 123). Hume (2004: 220–227) argues similarly that ambiguity or indeterminacy in the auditory signal sets the stage for a reinterpretation of linear order, but places a

Metathesis

8

special emphasis on the role of the specific attested sequences in the language, as discussed for Hebrew above. In Kambata (East Cushitic, Ethiopia), a suffix-initial nasal transposes with a preceding obstruent, and is also subject to place assimilation in this position, as illustrated by [Ik] and [mb] resulting from an /n/-initial suffix (Hudson 1980: 105); similarly the related language Sidamo (Vennemann 1988: 55) and several other East Cushitic languages (Garrett and Blevins 2009: 532f.). (11)

Kambata metathesis of obstruent + nasal clusters (Hudson 1980: 105) it-ne(mmi t’u(d-na(mmi oros-na(mmi sok-ne(mmi hab-no(mmi

→ → → → →

inte(mmi t’u(nda(mmi oronsa(mmi soIke(mmi hambo(mmi

‘we ‘we ‘we ‘we ‘we

have eaten’ will see’ will take’ have sent’ forgot’

This metathesis is part of a conspiracy (chapter 70: conspiracies) of changes (including complete assimilation and vowel epenthesis) that avoid ill-formed consonant clusters, in this case obstruent + sonorant; see Hume (1999: 300– 302) for a perceptual-optimization account. In a novel strategy that anticipates Optimality Theory, Hudson (1980: 109) proposes that affixation generates two outputs with alternate orderings of juxtaposed consonants (such as [itne(mmi], [inte(mmi]), where the choice between the outputs is made according to conditions on phonotactics. This technique would not, however, generalize to examples such as Faroese and Cebuano, in which the relevant consonants are not juxtaposed across a morpheme boundary. The obstruent + nasal reversal in East Cushitic has been cited as an example of a metathesis that does not result from a conventional source such as a misperception of the ordering of the cues (Garrett and Blevins 2009: 532–537). Rather, it appears to reflect the pressure of other consonant interactions; I illustrate with the facts of Kambata, but follow the argument of Garrett and Blevins, who use data from related Bayso. In some clusters that occur at stem boundaries, we find regressive assimilation, as in /rn/ → [nn] and /mt/ → [nt]; but in others, there is apparent progressive assimilation to create a geminate, as in /bt/ → [bb]. (12)

Kambata assimilation in clusters (Hudson 1980: 105) a.

b.

im-to(?i ful-na(mmi kam-no(mmi mar-ni ub-to(?i dag-tonti t’u(d-tenti oros-ta(nti

→ → → → → → → →

into(?i funna(mmi kanno(mmi manni ubbo(?i daggonti t’u(ddenti orossa(nti

‘she dug’ ‘we will go out’ ‘we forbade’ ‘we, going’ ‘she fell’ ‘you knew’ ‘you have seen’ ‘you will go’

If the learner seeks to generalize to a single process of regressive assimilation, then an intermediate step of metathesis is necessary to create the right outcome: /bt/ → tb → [bb]. Extending this to instances of obstruent + nasal, such as /bn/

9

Eugene Buckley

→ nb → [mb], also has the effect of preserving the features of the root-final consonant and yielding a nasal + obstruent sequence of the sort that is common in other concatenations. Interestingly, whereas Kambata assimilation and metathesis apply to all places of articulation, in Bayso both are restricted to coronals, which reinforces the connection; this correlation is found across the East Cushitic languages (Garrett and Blevins 2009: 536f.). From the perspective of synchronic phonology, an unavoidable conclusion is that metathesis processes are available to the learner, whether the pattern results from a misperception of the phonetic signal or a generalization of an existing pattern.

1.2

CV metathesis

Ordering reversals of a consonant and vowel involve many of the same principles of explanation and analysis as CC reversals – for example, the historical reinterpretation of an ambiguous signal, and a synchronic constraint that dominates Linearity. From the examples in the literature, however, synchronic CV metathesis appears to be strongly associated with specific morphological contexts, and the reordering may be the main exponence of a grammatical category, something that is not typical of CC metathesis. But before considering such cases, we examine a few more strictly phonological examples.

1.2.1 Phonological reorderings A well-known case that has been treated as metathesis is Cayuga (Iroquoian), in which a laryngeal consonant /h ?/ transposes with a preceding vowel when it occurs in an odd-numbered, non-final syllable (Foster 1982: 69f.; Blevins and Garrett 1998: 509–512). The necessary prosodic context can be analyzed as the weak branch of an iambic foot (Hayes 1995: 222f.; see also chapter 44: the iambic–trochaic law). (13)

Cayuga laryngeal metathesis (Foster 1982: 69f.) kahwista?eks ko?nikõha? akekaha? ahanohae?

→ → → →

kh+’wisd?aes g?o’nikhwa? a’gekhaa? a’hanhwae?

‘it strikes, chimes’ ‘her mind’ ‘my eye’ ‘he washed it’

To some degree, however, it is uncertain whether this process is truly a reversal of segment order or instead a spreading of features across the vowel, resulting in overlap rather than reordering (Foster 1982: 70). Somewhat similar metathesis of vowel + /h/ occurs in Cherokee when a stop consonant precedes the vowel; the result is that the laryngeal is realized as aspiration on the stop (Flemming 1996; Blevins and Garrett 1998: 520f.). In the framework of Blevins and Garrett (1998: 509f., 2004: 121–125), Cayuga and Cherokee show the results of perceptual metathesis. Just as with the CC metathesis involving liquids and other segment types, the spread of laryngealization or devoicing through the vowel leads to the possibility of reinterpretation. Diachronic instances of the same phenomenon include liquid metathesis in Slavic, as in *orbota ‘work’ > Polish /robota/; and reordering of /r/ with schwa in Le Havre French, such as [bHrbi] ‘ewe’ compared to standard [brHbi] (Blevins and Garrett

Metathesis

10

1998: 513, 16f.). The Slavic example is somewhat unusual, in that it involves an initial sequence undergoing metathesis; reordering is cross-linguistically disfavored for root-initial segments, since a disruption in that position interferes with effective word recognition more than metathesis of other segments (Hume and Mielke 2001). A clearer example of synchronic CV metathesis is found in the Austronesian language Leti, discussed in detail by Hume (1998) and Blevins and Garrett (1998: 541–547). Alternating stem forms in Leti are phonologically conditioned according to the following context; in particular, morpheme-final VC reverses to CV to avoid an illicit consonant cluster within a phrase. (14)

Leti VC metathesis (Hume 1998: 153) a. ukar lavan b. ukar ppalu ukar muani

→ → →

ukarlavan ukrappallu ukramwani

‘thumb, big toe’ ‘index finger’ ‘middle finger’

There is, additionally, the same reversal in phrase-final position, so that ‘finger’ appears as [ukra]; here, rather than serving general phonotactics, the metathesized form appears to mark the word as phrase-final (Bonthuis 2001: 37f.). Because metathesis in Leti affects all consonant types – compare /ulit/ → [ulti] ‘skin’, /metam/ → [metma] ‘black’ – it cannot be attributed to the elongated phonetic realization of a class such as laryngeals, and is not perceptual metathesis. Instead, Blevins and Garrett (1998: 539–547) identify it as pseudo-metathesis. By this they mean an alternation in ordering that did not arise historically as a direct reinterpretation of segment order. In the case of Leti, two main steps are posited, with evidence from other patterns within Leti and in related languages. First, an epenthetic vowel was inserted after final consonants, /ulit/ → [ulit”]. Although the inserted vowel was not a copy of the preceding vowel, it nevertheless would have been subject to co-articulatory effects of the more palatal or labial quality of a preceding /i/ or /u/, as in [ulit”]. Second, syncope of medial vowels led to loss of that schwa preceding another word beginning CV ([ulit”] → [ulit]), but loss of the original medial vowel in other contexts ([ulit”] > [ult”]); here, however, the vowel quality of the deleted vowel is preserved in the final schwa due to the coarticulation ([ult”] > [ulti]). Words containing the low vowel, such as /ukar/, do not show palatal or labial co-articulation, but result from the fact that schwa more generally became /a/ in the history of Leti (thus /ukarH/ > /ukrH/ > /ukra/). From the point of view of synchronic phonology, the crucial point is that alternations such as [ulit] ~ [ulti] were successfully integrated into the grammar as learners re-analyzed the historical patterns.

1.2.2 Morphological context As noted above, CV metathesis often appears to occur in the presence of a particular morphological trigger, even if the reordering that occurs can be defined phonologically. A famous example, also from Austronesian, is found in Rotuman. In this language, words appear in two different “phases,” called complete and incomplete (Churchward 1940). The incomplete form is derived from the complete by a variety of means, but the default strategy is metathesis of the final CV to VC, often forming a short diphthong with the preceding vowel.

11 (15)

Eugene Buckley Metathesis in Rotuman phase alternations (Churchward 1940: 14) complete ho.sa i.?a pu.re ti.ko se.se.va

incomplete hoas ia? puer tiok se.seav

‘flower’ ‘fish’ ‘to rule, decide’ ‘flesh’ ‘erroneous’

The process applies to loanwords as well, such as /pe.pa/ → [peap] ‘paper’. Consistent with many other languages, Rotuman short diphthongs must consist of two vowels with rising sonority (i.e. movement from a higher to a lower vowel). Where this condition is not met, the incomplete phase is realized in other ways: by dropping a final vowel, as in /to.ki.ri/ → [to.kir] ‘to roll’; by fusing two vowels brought together by metathesis, as in /mo.se/ → [møs] ‘to sleep’; or by directly changing a vowel sequence to a long diphthong, as in /ke.u/ → [keu] ‘to push’. Blevins and Garrett (1998: 527–529) categorize the Rotuman alternation as compensatory metathesis. Historically, this entails an anticipation or perseveration of vowel features across an intervening consonant toward the stressed vowel, leading to “extreme vowel-to-vowel coarticulation.” In Rotuman there would have been anticipation of the final vowel in the direction of the preceding stressed syllable, followed by loss of the final vowel, essentially /hosa/ > /hoasa/ > /hoas/. In some vowel sequences, further changes occurred, such as /mose/ > /moese/ > /moes/ > /møs/. Metathesis of similar origin is also pervasive in the related language Kwara’ae, where final CV changes in most communicative contexts to VC to mark phrasal boundaries (Sohn 1980: 311f.); the details of Kwara’ae vowel realization lend particular support to the proposed historical origin as compensatory metathesis (Blevins and Garrett 1998: 530f.). The phases of Rotuman were originally described by Churchward in complex syntactic and semantic terms, but some recent work has argued that their specific realization depends on prosody, and therefore that they are basically phonologically determined rather than triggered by a morphological or other grammatical context (Hale and Kissock 1998: 120–123). For example, it has been proposed that the desired outcome achieved by metathesis as well as the other processes is a word-final heavy syllable (Blevins and Garrett 1998: 531–534; McCarthy 2000: 159, 73f.). From this point of view, the reversal in order is just one way of satisfying the heavy-syllable constraint; there is no specific rule demanding metathesis. This analysis relies on the claim of Hale and Kissock (1998) that the complete phase occurs before monomoraic morphemes such as /-me/ ‘hither’ in [ho?a-me] ‘to bring’, and the incomplete before bimoraic morphemes such as transitive /-kia/ in [hoa?-kia] ‘to take (trans)’. The essential idea is that right-aligned prosodic structure in [ho(?a-me)] requires the stem-final CV syllable to be grouped with the CV suffix in a proper bimoraic foot, and the pressure for a stem-final heavy syllable is thwarted. But in [(hoa?)-(kia)], the stem and the suffix are footed independently, and the stem undergoes metathesis to ensure a stem-final heavy syllable. The same result is predicted in the absence of a suffix. Kurisu (2001: 187) cites, in addition to certain exceptional suffixes, minimal pairs from Churchward (1940: 15) showing that the two phases can occur in an

Metathesis

12

identical phonological context, such as complete [?epa la hoa?] ‘the mats will be taken’ and incomplete [?eap la hoa?] ‘some mats will be taken’. This overlap indicates that the phase changes must in some way be triggered by the presence of a morphosyntactic category, the Incomplete Phase. For Kurisu, the high-ranked constraint RealizeMorpheme forces the incomplete to be phonologically distinct from the complete phase (chapter 103: phonological sensitivity to morphological structure); the relative ranking of phonological constraints, including Linearity, determines exactly how the base form is modified. The metathesis outcome is favored by the constraint ranking, although particular configurations (such as vowel sequences with falling sonority) lead to other outcomes, including fusion of the vowel features. Notably, in this more morphologically oriented approach, there is still no specific morphological demand for metathesis; rather, the drive for distinctness of word forms interacts with phonological constraints to produce metathesis, among other results. The examples presented so far involve underlying CV changing to VC, especially stem-finally. The converse change at the left edge, where initial VC changes to CV, is attested in some Northern Paman languages such as Ngko}, following the historical loss of initial consonants (Hale 1976: 17f., 23–28; Blevins and Garrett 1998: 537f., 2004: 135f.). (16)

Ngko} initial VC metathesis (Hale 1976: 23–28) *njipul*nji(na*kulan*puIa*Iali*kami-

> > > > > >

*ipul*ina*ulan*uIa*ali*ami-

> > > > > >

pjulnjalwanIwalajmaj-

‘you (non-sg)’ ‘to sit’ ‘possum’ ‘son’ ‘we (dual incl)’ ‘mother’s mother’

Unlike in Rotuman and a number of other Austronesian languages, however, this metathesis appears to be diachronic only. A somewhat similar pattern is found synchronically for a number of verbs in the Nilo-Saharan language Fur (Jakobi 1990: 57f., 64–74; Hume and Mielke 2001: 141f.). These verbs, when preceded by a monoconsonantal person-marking prefix, undergo reversal of the initial CV. (17)

Fur CV metathesis under prefixation (Jakobi 1990: 57f., 64–74) k-bak-teerk-lat-

→ → →

kabketerkald-

‘we drink’ ‘we forge’ ‘we beat, hit’

Some alternations are quite irregular, such as /li-/ → [al-] ‘wash’ and /tii-/ → [ei-] ‘catch’, so that a plausible alternative is that the allomorphs are lexically listed. This account would also address the formal problems in alternations such as /bul-/ → [ulb-] (→ [ulm-]) ‘find’, which involve two apparent metatheses (Hume 2001: 18f.); see §2.3 below. Even if the allomorphs are listed, however, metathesis was a crucial historical source.

Eugene Buckley

13

1.2.3 Metathesis in templates Languages with templatic morphology express certain inflectional or derivational categories by changes to the syllable structure of the stem (see chapter 105: tier segregation; chapter 108: semitic templates). If a particular paradigm includes different orderings of C and V elements, then the result is a form of metathesis. Templatically created metathesis generally does not derive from general phonological properties of a language, but rather from potentially arbitrary exponence of a morphological category. For example, a relatively productive metathesis applies to derived Classical Arabic nominal stems two syllables in length; initial /Ca/ is reversed to [aC] and a glottal stop onset is inserted (McCarthy and Prince 1990: 213f., 279f.). Examples include /kabar/ → [?akbar] ‘greater, greatest’ and /–anib-at/ → [?a–nib-at] ‘wings’; compare the underived forms [kabi(r] ‘great’ and [–ana(b-at] ‘wing’, without metathesis. The change cannot be treated as a general phonological process, because it is limited to certain morphological categories, and does not occur in verbs such as /katab/ ‘he wrote’ (*[?aktab]). In Mutsun, a Costanoan language of northern California, templatic alternations include the reversal of a stem-final sequence of vowel and consonant; the primary stem is consonant-final and the derived stem is vowel-final (Okrand 1979: 126f.). The choice between alternate verb stem forms depends on what suffix is added; in other cases (18b) the primary stem is a noun, and the derived stem is a verb with related meaning. The derived stem has the uniform shape CVCCV, despite considerable variation in the primary stem shape. (18)

Mutsun stem alternations (Okrand 1979: 126f.) a.

b.

primary pasikliŒ(ejmat(allul(upto(herla(lakposo(l-

derived paskiliŒjematlalulputohrelalkaposlo-

‘to visit’ ‘to stand’ ‘to be face down’ ‘flute / to play the flute’ ‘a cough / to cough’ ‘goose / to get geese’ ‘posole (stew) / to make posole’

Okrand observes that while the vowel-final derived stem is the form used with all suffixes that would create an illicit consonant cluster if added to the primary stem ([liŒje-hte] ‘standing’, *[liŒ(ej-hte]), it also occurs with some suffixes that would be phonotactically well-formed with a preceding consonant ([matla-nu] ‘put (someone) face down’, alongside [mattal-pu] ‘put oneself face down’). Therefore, this morphologically defined reordering does not merely repair phonological violations, even if it sometimes conspires to avoid phonotactically problematic concatenations. It has been pointed out that for similar alternations in related Sierra Miwok, a representation with V/C segregation makes a specific metathesis rule unnecessary (Smith 1985: 366f.; Goldsmith 1990: 91; Stonham 1994: 157f.); more on this below. In Tunisian Arabic, a stem-internal alternation is a cleaner example of metathesis than what we find in Classical Arabic (Kilani-Schoch and Dressler 1986; Becker 2000: 579f.). Historical changes to vowels within stems have led to minimal differences

Metathesis

14

defined by ordering, such as Classical Arabic /malak-a, milk-u/ > Tunisian /mlHk, mHlk/; this pattern is now productive in relating triliteral surface forms. (19)

Tunisian Arabic stem alternations (Kilani-Schoch and Dressler 1986: 62, 65f.) mlHk fhHm prHm kfDr

‘he ‘he ‘he ‘he

possessed’ understood’ forbade’ blasphemed’

mHlk fHhm parm kDfr

‘property’ ‘understanding’ ‘prohibition’ ‘blasphemy’

A similar alternation is found in Alsea (Buckley 2007). In this coastal Oregon language, stems generally show at least two forms; the full stem contains a root vowel, while the short stem lacks it. For stems with a medial sonorant consonant, an additional distinction is found: the full stem occurs in two varieties, light and heavy, according to whether the root vowel follows or precedes the sonorant. The stem choice depends on the presence of particular suffixes as well as an aspectual distinction. (20)

Alsea stems with a medial sonorant (Buckley 2007: 8f.) light (CV) stlaktwihtmus-

heavy (VC) stalktiwhtums-

short (no V) stlktwhtms-

‘slide’ ‘pour’ ‘close’

In the analysis of Buckley (2007: 15–18), the light stem is the underlying form; the short stem is created by deletion of the root vowel, and the heavy stem results from VC metathesis. Since only sonorants undergo this potential reordering, they alone are treated as weight-bearing in the coda, and therefore only in that case can metathesis yield satisfaction of the heavy template requirement. The same approach might be applied to Tunisian Arabic, with the difference that all consonant classes are moraic, and therefore metathesis applies to stems regardless of the medial consonant. The larger point is that the requirement for a heavy syllable is morphologically determined, but the effect is generated phonologically. Similar is the stem alternation found in Klallam and other Straits Salish languages, in two forms called the actual and non-actual aspect (Thompson and Thompson 1969: 215–217; Demers 1974: 17f.; Montler 1989: 96f.). (21)

Klallam stem alternations (Thompson and Thompson 1969: 216) non-actual Œkwuqq’iIq’HmtHqw-

actual Œukwqiq’IHq’mHtqw-

‘shoot’ ‘restrain’ ‘swallow’ ‘put in water’

Anderson (2005: 9–11) argues that synchronically, Klallam and similar languages require a processual rule of metathesis to express this morphological category. Montler (1989: 93), however, expresses this effect for Saanich as a CVCC template

15

Eugene Buckley

that causes metathesis in a form such as /’søHt/ ‘push it’ → [’sHøt] ‘pushing it’. In roots where the CVCC template cannot be satisfied by metathesis, other strategies are available, such as glottal stop insertion after the vowel in /’weqHs/ ‘yawn’ → [’we?qHs] or reduplication to achieve this templatic result, as in /’qen’/ ‘steal’ → [’qeqn’] (→ [’qeqHn’] by epenthesis). The same additional strategies occur in Klallam as well. In a prosodic version of the template approach, Stonham (1994: 173f.) proposes insertion of a mora in Klallam and Saanich that forces CCV to surface as CVC, a heavy syllable, and also causes the related effects of coda insertion and reduplication. The templatic and moraic approaches treat metathesis as one possible means of satisfying the morphologically determined, but phonologically expressed, restriction on shape. As with Rotuman, metathesis is one change among several, and not necessarily the direct goal of the morphological category. In a more strictly phonological approach for closely related Lummi, Demers (1974: 16) proposes a rule that deletes unstressed schwa between obstruents. In this view, the actual and non-actual forms are both based on a CHCH root, but have different stress placement. Schwa deletion yields apparent metathesis in pairs with the surface shapes seen in/’CHCH/ → [’CHC] and /CH’CH/ → [’CCH]. Although the synchronic evidence for exactly this derivation is missing in Klallam, the Lummi pattern suggests the likely diachronic origin of metathesis as a reinterpretation of vowel deletion. Such a historical origin can explain why these templatic changes normally involve reorderings of consonants and vowels, but not of consonants. For example, suppose that (similar to Lummi) the Alsea stem ‘to close’ that alternates between [tmus] and [tums] derives from original *tumus, with deletion of the unstressed vowel in forms with distinct stress patterns due to different suffixation: *’tumus-a > /’tums-a/ ‘door’ and *tu’mus-ø > /’tmus-ø/ ‘is closed’ (Buckley 2007: 22f.). The alternate forms that preserve different vowels are subject to reinterpretation as a stem with a single underlying vowel that is reordered with the adjacent consonant in different suffixal contexts; but vowel deletion by itself will not result in the reordering of consonants. Given the frequency of vowel harmony and syncope, patterns like this can be expected to arise rather often. Despite the crucial role of morphological context in conditioning these reorderings, phonological techniques can often be used to generate the necessary effects. One important tool has been the segregation of vowels and consonants onto different tiers (see chapter 105: tier segregation), so that they have no underlying ordering and no actual metathesis occurs in the derivation (McCarthy 1989: 5, 22f.). The advent of Optimality Theory, with its emphasis on output constraints rather than restricted input representations, makes V/C segregation “superfluous” (McCarthy 2000: 180f.). Even in an approach that does not treat the consonantal root as a morpheme listed independent of any vowels, derivational and inflectional morphemes often consist of vowels that overwrite the underlying vowels of the stem (Ussishkin 2005). Apparent VC metathesis among surface forms is merely the result of different overwriting patterns, as when the elements /h i i/ are imposed on Modern Hebrew /gadal/ ‘grow’ to form [h-igdil] ‘enlarge’. Constraints on the realization of affixal material in the stem lead to particular overwriting patterns, but the vowels of the affixes still have no underlying ordering relation to the consonants of the input word.

Metathesis

16

It is less clear how a vowel-overwriting approach for Semitic can extend to language families such as Miwok-Costanoan and Yokuts, where the vowels and consonants can be reordered, but the vowels do not have the status of separate morphemes (McCarthy 1989: 74, 78). Thus in Mutsun, the verb ‘to visit’ is lexically specified with not only the consonants /psk/ but also the vowels /ai/, combined in different ways, including [paski-] and [pasik-] (18). The overwriting operation would have to be available for subparts of one lexical entry, rather than independent morphemes, in order to account for languages like Mutsun.

1.3

VV metathesis

Although CV and CC metathesis are robustly attested, there is weak evidence for VV metathesis. Webb (1974: 8) states that “[e]ven as a sporadic change metathesis of vowels appears to be quite uncommon.” Kiparsky and O’Neil (1976: 531, n. 7) believe “there are few if any rules that metathesize contiguous syllabic segments in any language.” McCarthy (2000: 176) observes that the few synchronic analyses that posit VV reversals “involve very abstract analyses, in which the underlying representations and/or the consequences of metathesis are by no means apparent.” The rarity of such reversals may be related to the much longer typical duration of vowel gestures compared to consonants, so that a considerable temporal shift would be required for re-analysis of the ordering of two vowels (Steriade 1990: 390f.; McCarthy 2000: 176). A classic example is VV reversal in Kasem, to which Chomsky and Halle (1968: 361) first applied the transformational rule format for metathesis. In particular, the vowel sequence /ia/ is reversed to [ai] when followed by the plural suffix /i/; but the first /i/ deletes and then the remaining vowels coalesce, as in /pia-i/, which surfaces as [pe] ‘sheep (pl)’. Needless to say, on first inspection /piai/ → [pe] is not an obvious example of metathesis. Phelps (1975: 303f., 10f., 25, 1979: 56f.) argues against the Chomsky and Halle VV metathesis rule, but in favor of an entirely different CV metathesis, in derivations such as /boa(l-u/ → [bola(-u] (→ [bolo] ‘valley’). This derivation is again complex, although with different assumptions about underlying forms. Both of Phelps’s general conclusions regarding Kasem metathesis – CV is transposed but not VV – are endorsed, in a more modern framework, by Haas (1988: 241–253, 45f.); see also Burton (1989: 29f.) for an analysis of vowel coalescence without an intermediate reordering. Similar re-analyses have been proposed for other languages with apparent VV metathesis. Keyser (1975: 404) posits a rule for Old English that reverses vowels in order to feed a vowel elision rule, as in /lufa-i/ → lufia → [lufa] ‘love!’; Kiparsky and O’Neil (1976: 535f.) argue that a revised formulation of vowel elision makes metathesis unnecessary. A rule of VV metathesis has been claimed to play “a central role” in Latvian phonology; it reverses the order of elements in the diphthongs /ai au æi æu/, although under restricted conditions (Halle and Zeps 1966: 108). In a more recent treatment of Latvian vowels, although not focusing on metathesis, Anderson and Durand (1988: 34, n. 7) reject some of the synchronic abstractness assumed by Halle and Zeps; instead they assume raising of a monophthongal vowel that then undergoes breaking to form a diphthong, where no metathesis is required.

17

Eugene Buckley

A few diachronic examples of VV metathesis can be cited, especially if we include vowel/glide reversals in this category, since the same set of segmental features may serve as a glide or vowel before and after the metathesis (Ultan 1978: 375f.). Two examples from Portuguese are /genukulum/ > /geoOo/ > /ÚoeOo/ ‘knee’ and /dehonesta(re/ > /deostar/ > /doestar/ ‘to insult’ (Williams 1962: 111); these reversals may have occurred “on the analogy of the more familiar sequence oe” (Ultan 1978: 376).

1.4

Other types of metathesis

Permutations involving something larger than a segment may fall under broader definitions of metathesis (Ultan 1978: 370). These include syllable reversals in language games or ludlings, such as Chasu /i.ku.mi/ → [i.mi.ku] ‘ten’ (Bagemihl 1995: 704). Metathesis has also been proposed for elements such as location in sign language phonology (Sandler 1993: 246). Hyman and Leben (2000: 590) state that there are “sporadic reports of tonal metathesis in the literature”; some examples include Bamileke-Dschang (Pulleyblank 1986: 41, 50), Mixtec (Goldsmith 1990: 25), and Dangme (Holscher et al. 1992: 126). These processes typically involve the movement of a floating tone, originating at the edge of a word or other domain, past a single linked tone. But – like VV metathesis in §1.3 – they are also embedded in complex derivations, and depend on multiple assumptions about how the pieces of the analysis fit together. Under other assumptions, metathesis may not be required. For example, Pulleyblank (1986: 41) proposes that in Bamileke-Dschang a floating L tone moves leftward across a H, and remains floating to represent downstep; Hyman (1985: 71, 73), on the other hand, links the L directly to the H on a second tonal tier as a direct representation of a downstepped H. In essence, the new rule is a merger rather than a reordering, similar to Zoque palatalization in §3 below. It should be kept in mind that “metathesis” of syllabicity, as in French /oj/ > /we/ > modern /wa/ and Proto-Slavic *ew > /ju/ (Ultan 1978: 376), does not involve transposition of segmental features but rather a shift in affiliation relative to the head of the syllable. Thus in French /oj/ > /we/, the round vocoid continues to precede the front vocoid; in Slavic *ew > /ju/, the round element is second, but remains there. The same observation can be made for English /iw/ > /ju/, found in words such as few (Jespersen 1949: 101), which is quite similar to the Proto-Slavic change. None of these represents segmental metathesis.

2

Non-local effects

Often grouped with local metathesis is the exchange of segments that are not adjacent, called long-distance or non-contiguous metathesis (Ultan 1978: 380–383). In fact, metathesis or its equivalent in another language has been used, especially by earlier writers, specifically for such long-distance effects (Blevins and Garrett 1998: 525). Grammont (1933: 239ff., 339ff.) devotes separate chapters to longdistance métathèse and local interversion, a terminology found more recently, for example, in Pierret (1994: 61); but Wechssler (1900: 496) already uses Metathese for both local and long-distance transpositions. As noted at the beginning of this chapter, metathesis here refers to either type of reordering.

Metathesis

2.1

18

Diachronic

A famous diachronic example of transposition over intervening segments is the Spanish metathesis of r . . . l > l . . . r, observable in a few modern words (Ultan 1978: 381; Penny 2002: 36). (22)

Spanish liquid metathesis (Penny 2002: 36) Latin mi(ra(kulum peri(kulum parabola

> > >

Spanish milagro peligro palabra

‘miracle’ ‘danger’ ‘word’

These pronunciations were probably influenced by the greater frequency of consonant + r in the lexicon; various sound changes had eliminated many inherited instances of consonant + l (Ultan 1978: 391; Penny 2002: 70–72); the change can also be viewed as two steps, first the change of /l/ to /r/ in a cluster, and then the well-attested dissimilation of identical liquids (Wanner 1989: 444f.). Comparison of cognate words in the Yuman family of the American Southwest reveals a variety of historical metathesis processes, including root consonants in Walapai /’pil/ ‘burns’ ~ Cocopa /’lip/ ‘flames up’, or Havasupai /ka’to/ ~ Walapai /ta’ko/ ‘chin’ (Langdon 1976). There are also variant forms within languages, such as Ipai Diegueño /mHxH’tun/ ~ /xHmH’tun/ ‘knee’. These alternations are widespread, but remain lexically specific. Swapping of the consonants in largely CVC roots is also common in the Salish family, as seen in apparent cognate pairs such as Shuswap /xwej/ ~ Twana /jHxw/ ‘disappear’, and Klallam /ts’Hq’w/ ~ Upper Chehalis /q’wHts’/ ‘dirty’ (Noonan 1997: 482). The pervasiveness of this pattern in Salish is unusual, and is possibly best explained by historical processes of reduplication and consonant deletion rather than direct metathesis (Hume and Mielke 2001: 143, n. 4; Noonan 1997: 513). Prunet (2006: 57–61) discusses examples of consonant metathesis within Semitic roots. These are said to be particularly common in the Hebrew lexicon, as in synonymous variants such as [keveW] ~ [keWev] ‘lamb’ and related meanings such as [?a(raz] ‘tie packages’ ~ [?a(zar] ‘bind, girdle’ (Horowitz 1960: 228–234). More dramatic examples of non-contiguous consonant metatheses are found in language games in Bedouin Hijazi and Moroccan Arabic, in which the root consonants are scrambled (Prunet et al. 2000: 623f.); in Hijazi, /kattab/ ‘caused to write’ can be realized as [battak], [takkab], [kabbat], [tabbak], and [bakkat]. Although this radical permutation is not part of the basic grammar, such language games show an impressive computational capacity for synchronically active metathesis (Bagemihl 1995: 703f.; Anderson 2005: 11f.).

2.2

Synchronic

Typological surveys have claimed that permutation of non-adjacent segments does not occur as a regular synchronic process (Webb 1974: 5; Wanner 1989: 445; Hume and Mielke 2001: 145f.). Certainly, the permutation of non-adjacent segments is common in speech errors, such as classic Spoonerisms, but such errors

19

Eugene Buckley

also involve strings of segments such as complex onsets and rhymes (Fromkin 1971: 31f.). (23)

Metathesis in speech errors (Fromkin 1971: 31f.) intended kip H te>p fAr mDr pe> ske>l swe7Ì draj>I hip Hv –ZIk

→ → → → →

error tip H ke>p mAr fDr ske> pe>l dre7Ì swaj>I hZIk Hv –ip

‘keep a tape’ ‘far more’ ‘pay scale’ ‘sweater drying’ ‘heap of junk’

These form part of a larger phenomenon of anticipation, perseveration, deletion, and so forth. Speech errors may, however, be a source of sporadic metathesis in historical change (Wanner 1989: 445). The outputs of speech errors, like more systematic metathesis, overwhelmingly respect the existing phonotactics of the language (Wells 1951: 26; Fromkin 1971: 40–42; Dell 1995: 200); but transposition of adjacent consonants, so common in regular metathesis, is “exceptionally rare” as a speech error, such as whipser for whisper (Berg 1987: 9). Elements that transpose by error usually occupy parallel syllable positions, which is not the case for adjacent consonants; instead, as discussed above, metathesis of such segments normally arises historically by misperception rather than production or planning errors. An interesting comparison is an optional metathesis reported for a few words in Turkana (Dimmendaal 1983: 48f.; Hume and Mielke 2001: 139f.; Hume 2004: 218). Here two consonants with the same value of [sonorant] that serve as onsets to successive syllables, and are adjacent to identical vowels, are optionally transposed in fast speech. (24)

Turkana onset metathesis (Dimmendaal 1983: 48f.) preferred Ia-kèmèr-a Ii-kwaIGrDmDk-à e-s>k>n-a`

alternate Ia-kèrèm-a ‘mole’ Ii-kwaIGmDrDk-à ‘a kind of tree’ e-k>s>n-a` ‘breast’

These alternations have the appearance of a common speech error that has become somewhat conventionalized. In particular, it has been widely observed that exchange (and other) errors are more likely when the sounds in question are found in similar phonological environments, so that for example left hemisphere → heft lemisphere, where the initial consonants are both followed by /e/, is more likely than the parallel error in right hemisphere, where the vowels are different (MacKay 1970: 325–328; Dell 1984: 222). Morphologically restricted metathesis (§1.2.2) can apply synchronically to surface non-adjacent segments. For example, Akkadian has a /t/ infixed in reciprocal verbs; it surfaces there in most cases, exemplified by [pitrus-] (25a), which motivates the stem-internal position as basic. But this stop is transposed to word-initial position when the root has an initial coronal obstruent as in (25b) (Caldwell et al. 1977: 118; McCarthy 1981: 381; Buccellati 1996: 233f.; Huehnergard 2005: 390, 531, 611).

Metathesis (25)

20

Akkadian long-distance metathesis (Caldwell et al. 1977; Huehnergard 2005) a.

b.

root /prs/ /rkb/ /kmr/ /s#bt/ /snq/ /zkr/ /dkœ/

infinitive para(sraka(bkama(rs#aba(tsana(qzaka(rdaka(œ-

reciprocal pitrusritkubkitmurtis#buttisnuqtizkurtidkuœ-

*s#itbut*sitnuq*zitkur*ditkuœ-

‘divide’ ‘mount, ride’ ‘heap up’ ‘seize’ ‘be near’ ‘declare’ ‘swell’

The same metathesis occurs in iterative stems: [pitarrus-] but [tis#abbut-] (*[s#itabbut-]). In the analysis of Ìubowicz (2009), Akkadian metathesis serves to move the /t/ outside the stem domain, where it would cause a violation of the Obligatory Contour Principle (OCP) penalizing two tier-adjacent coronal consonants; thus [tizkur] does not violate this constraint, whereas *[zitkur] would, because the /t/ is located within the stem. This approach gives a relatively prominent role to phonology (the OCP constraint on coronals) while maintaining a crucial morphological component, due to the role of the stem domain.

2.3

Displacement

A related phenomenon, which is also called metathesis by many authors (Grammont 1933: 339; Ultan 1978: 372), involves the shift or displacement of a segment over more than one intervening segment. A famous example comes from the Occitan dialect of Bagnères-de-Luchon in southwestern France (Grammont 1905–6: 74, 85, 1933: 341; Blevins and Garrett 1998: 526). Among other processes, a liquid following a stop shifts leftward to form a cluster in the preceding syllable. (26)

Bagnères-de-Luchon long-distance shift of liquids (Grammont 1905–6: 74, 85, 1933: 341) *’kabra *’bespras *’pawpro *’tendro *’kambra *kum’pra *e’spingla

> > > > > > >

’krabo ’brespes ’prawbe ’trende ’krambo krum’pa e’splingo

‘goat’ ‘vespers’ ‘poor’ ‘tender’ ‘room’ ‘to buy’ ‘pin’

A shift like this is formally identical to metathesis when just one segment is skipped. However, with intervening material that includes non-constituent strings such as /esp/, it must be movement of /r/ rather than exchange. A similar shift of /r/ to the initial syllable is attested in South Italian Greek (Rohlfs 1930; Blevins and Garrett 2004: 130f., 34f.). If it is to be classified with long-distance metathesis, displacement might be expected to be absent from synchronic grammars. But synchronically active long-distance displacement is attested at least for laryngeal and pharyngeal features (Blevins and Garrett 2004: 132–134; see also chapter 25: pharyngeals). For example, in the Interior Salish language Colville (Nxilxcín), the pharyngeal

21

Eugene Buckley

consonant of a root is displaced to a stressed suffix, where it lowers the adjacent vowel to [a] (Mattina 1979). (27)

Colville pharyngeal displacement (Mattina 1979: 17f.) ’q’w#aj ’q’w#aj-’us ’q’w#aj-’lstsut

→ → →

’q’w#aj q’wHj’#as q’wHjlsts’#at

‘black’ ‘black man’ ‘his clothes are dirty’

Both pharyngealization and laryngealization can be seen as suprasegmental features in Salish (Mattina 1979: 19f.). Displacement of these features appears to reflect the spread of features over multiple syllables that may then be localized to a salient position (Blevins and Garrett 2004: 122f.). Such displacement resembles the mobility of a tone that shifts from the morpheme with which it is underlyingly affiliated to some phonologically defined position such as the penultimate syllable (Yip 2002: 65f., 89f., 132). A reasonable synchronic analysis is a [pharyngeal] feature affiliated with the root, which is attracted to the stressed syllable and possibly realized there as a segment. By the same token, the displacement of /r/ to the initial syllable in South Italian Greek reflects the salience and perceptual prominence of such syllables due to their location in the word (Blevins and Garrett 2004: 134); in Luchonnais, both stress and initial position favor the first syllable. Patterns of this sort have similar historical origins to simple exchanges of segments – in particular, phonetic cues that are relatively long in the temporal dimension and therefore subject to re-analysis, as discussed above for perceptual metathesis. But since they are displacements rather than exchanges, they are not formally equivalent to true metathesis as the exchange of positions. In particular, if local metathesis is seen as a minimal displacement (across a single segment), then long-distance metathesis would have to involve two simultaneous displacements, one leftward and one rightward, as in /abXcdYef/ → [abYcdXef]. This extra formal complexity may account for the rarity of synchronic non-local metathesis, which seems to be restricted to limited examples such as the optional reversal in Turkana (24) and the morphologically defined environment in Akkadian (25). An ordering alternation in the form of two suffixes is reported for several Costanoan languages, including Mutsun (Okrand 1979; Hume 1998: 170f., 1999: 300f., 2004: 223f.). Both suffixes have the shape CCV after a vowel-final stem, and CVC after a consonant-final stem, which makes phonotactic sense insofar as a three-consonant cluster at the stem boundary would be ill-formed. (28)

Mutsun suffix alternations (Okrand 1979: 127f., n. 17) CCV pire-tka rukka-kma

‘on the ground’ ‘houses’

CVC ?urkan-tak wimmah-mak

‘in the mortar’ ‘wings’

Although the locative [tka] ~ [tak] can be treated as local metathesis, in the plural [kma] ~ [mak], the [k] appears to move across two other segments. If Linearity is gradiently violable, with one violation for each segment over which another is displaced, the minimal change (one ordering reversal) is generally optimal (Hume 1998: 168–171, 2001: 17–19). Gradient violation does still permit Mutsun /mak/ → [kma] when other constraints force multiple violations of

Metathesis

22

Linearity, and the alternation can be seen as part of the synchronic grammar; but such cases seem to be quite rare and limited in scope. In the more recent version of OT that incorporates candidate chains (OT-CC), changes to the representation occur by minimal steps, and non-contiguous metathesis is subject to the requirement that each change in linear order increase well-formedness (McCarthy 2007: 87f.). In a suffix alternation requiring the synchronic derivation /mak/ → mka → [kma], the first step might be motivated by a preference for sonorant codas, but that change does not appear to be motivated more generally in Mutsun; in fact, Hume (1998: 170f.) specifically gives the constraint *m]coda. It might therefore be that the architecture of OT-CC forces the Mutsun alternation to be treated as listed allomorphy (chapter 99: phonologically conditioned allomorph selection).

3

Related processes

In some cases, a pattern that was originally considered to be metathesis was later seen as non-metathesis – and occasionally vice versa. A good example is the Zoque 3rd person singular prefix /j-/, which never surfaces as a strict prefix, but has been described as permuting with the stem-initial consonant, as in /j-pata/ → [p jata] ‘his mat’ (Wonderly 1951: 117f.; Dell 1973: 110). Sagey (1986: 105–111) argues that the glide /j/ actually merges featurally with the following consonant to produce a palatalized segment (chapter 71: palatalization), which may be realized with an offglide, as implied by transcriptions such as [p j]. This position is supported by the independent need for a non-metathesis source of palatalized segments found at compound boundaries; cases such as /kuj-tZm/ → [kujt jZm] ‘avocado’ show that the glide spreads, rather than reversing in order. A similar pattern is found in several languages of Nigeria and Cameroon. Prefixes that can be reconstructed as the high vowels *i and *u are realized as a glide – or a secondary articulation on the consonant – after the stem-initial consonant, as in Noni /k-w-en/ ‘firewood’ from the base /ken/ (Blevins and Garrett 1998: 514–516). See also the cases in Webb (1974: 12f.). Another phenomenon that has a certain affinity to metathesis is infixation, since it likewise requires a reordering from the expected position. In particular, “infixation and metathesis commonly show the potential mobility of full segments” rather than just subsegments such as features or nodes (Zoll 2001: 51). The closest analogy can be found in the infixation of a single consonant across one other consonant, as in the active neutral infix /-m-/ of Atayal /t-m-apeh/ ‘beckon’ (Egerod 1965: 265f.); this is formally similar to the metathesis of adjacent consonants. But infixation encompasses a broader set of phenomena that can include multiple segments in the item that undergoes reordering, as well as multiple segments in the span over which the infix is displaced; both are illustrated by the Tagalog actor focus /-um-/ that (optionally) moves over complex onsets in borrowed words such as /gr-um-adwet/ ~ /g-um-radwet/ ‘graduate’ (Orgun and Sprouse 1999: 204). On the other hand, Halle (2001) argues that the apparent Tagalog infixes appearing as /-um-/ and /-in-/ are actually CV underlyingly, with non-local metathesis of the two leftmost onsets, as /mu-tawag/ → [tu-mawag]. Theoretical and empirical problems with this approach are discussed by Klein (2005: 989–991), who advocates an infixation analysis within Optimality Theory.

23

Eugene Buckley

Another phenomenon that might be seen as involving either metathesis or infixation (or even other possibilities) is imbrication in Bantu languages such as Cibemba (Hyman 1995). In this process, the perfective suffix /-il/ combines with a polysyllabic stem, such that the /l/ of the suffix disappears and the /i/ combines with the rightmost vowel in the stem according to the usual coalescence rules of the language, as in /sákat-il-e/ → [sákeete] ‘seize’. The striking fact is that the suffixal vowel appears to skip over the stem-final consonant; in principle, this could be handled a variety of ways, including either CV metathesis, /sákat-il-e/ → sákaitle → [sákeete]; or infixation of the suffix inside the final consonant, /sáka-il-t-/ → sákailt-e → [sákeete]. These approaches assume subsequent simplification of the consonant cluster, as well as vowel coalescence. Hyman (1995: 11–16) argues in favor of infixation, which he relates to the positioning of the perfective (and the applicative) before the passive and causative suffixes. Diachronically, metathesis is the origin of some instances of infixation (Ultan 1975: 178f.; Yu 2007: 139–148). Another point of comparison is found in Horwood (2002: 170, 2004: 11), who uses Linearity to control the displacement of prefixes and suffixes to infixed positions. A crucial difference is that infixation of this sort (that is, excluding infixation tied to prosodically prominent constituents) is inherently edge-oriented; the infixed material remains as close to the left or right edge of the stem as possible, subject to the phonotactic constraints or other pressures that force deviation from simple prefixation or suffixation (McCarthy and Prince 1993; Prince and Smolensky 2004: 40–43; Yu 2007: 67–71). Metathesis, on the other hand, often occurs at stem edges as the result of morpheme concatenation, but in principle can occur anywhere in a word – recall the stem-medial cases in Cebuano and Rendille (§1.1). In addition, the infix has the status of a morpheme, which may happen to consist of a single segment; but in metathesis the single-segment status is fundamental, and not necessarily correlated with a particular morpheme. It can be noted finally that metathesis as a phenomenon is important evidence in favor of the category segment, however it may be formalized (chapter 54: the skeleton). Whether one considers the category of segment to be innate in the language faculty or something that emerges from the coordination of phonological gestures (Bybee 2001: 85f.), it is impossible to describe reorderings coherently in terms of disparate features or phonetic cues: the essential property of metathesis is that it moves all features associated with a segment, and the cues that instantiate these features are affected as a group. Indeed, the features may be implemented by rather different cues in the new position. For example, the Alsea alternation [stlak] ~ [stalk] affects just two of the five segments in this root. Even if /la/ were to be described as a core syllable, which is then reversed in some sense, the notion of “reversal” makes covert reference to the segments within the CV syllable. Otherwise there must be a claim that the prevocalic /l/ has the same phonetic realization as when it occurs in the coda, and that the release of the /t/ into the /a/ is no different from that into the /l/ in the nonmetathesized form. The need to refer to discrete segments even to characterize metathesis, and even more so to provide a theoretical analysis, presents particularly good evidence against suggestions that segments have no psychological reality, and are a mere artifact of an alphabetic writing system (Ladefoged 2005: 191; Silverman 2006: 6, 203).

Metathesis

24

REFERENCES Anderson, John M. & Jacques Durand. 1988. Underspecification and dependency phonology. In Pier Marco Bertinetto & Michele Loporcaro (eds.) Certamen phonologicum: Papers from the 1987 Cortona Phonology Meeting, 3–36. Turin: Rosenberg & Sellier. Anderson, Stephen R. 2005. Morphological universals and diachrony. Yearbook of Morphology 2004. 1–17. Bagemihl, Bruce. 1995. Language games and related areas. In John A. Goldsmith (ed.) The handbook of phonological theory, 697–712. Cambridge, MA & Oxford: Blackwell. Bailey, Charles-James N. 1970. Toward specifying constraints on phonological metathesis. Linguistic Inquiry 1. 347–349. Becker, Thomas. 2000. Metathesis. In Booij et al. (2000), 576–581. Berg, Thomas. 1987. A cross-linguistic comparison of slips of the tongue. Bloomington: Indiana University Linguistics Club. Blevins, Juliette & Andrew Garrett. 1998. The origins of consonant–vowel metathesis. Language 74. 508–556. Blevins, Juliette & Andrew Garrett. 2004. The evolution of metathesis. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.) Phonetically based phonology, 117–156. Cambridge: Cambridge University Press. Blust, Robert. 1971. A Tagalog consonant cluster conspiracy. Philippine Journal of Linguistics 2. 85–91. Blust, Robert. 1979. Coronal–noncoronal consonant clusters: New evidence for markedness. Lingua 47. 101–117. Bonthuis, Fiorieneke. 2001. Metathesis in Leti. In Hume et al. (2001), 26–52. Booij, Geert, Christian Lehmann & Joachim Mugdan (eds.) 2000. Morphologie: Ein internationales Handbuch zur Flexion und Wortbildung. Berlin & New York: Mouton de Gruyter. Buccellati, Giorgio. 1996. A structural grammar of Babylonian. Wiesbaden: Harrassowitz. Buckley, Eugene. 2007. Vowel–sonorant metathesis in Alsea. International Journal of American Linguistics 73. 1–39. Burrow, Thomas & Sudhibhushan Bhattacharya. 1970. The Pengo language: Grammar, texts, and vocabulary. Oxford: Clarendon Press. Burton, Strang. 1989. Kasem coalescence and metathesis: A particle analysis. Toronto Working Papers in Linguistics 10. 21–32. Bybee, Joan. 2001. Phonology and language use. Cambridge: Cambridge University Press. Caldwell, Thomas A., John N. Oswalt & John F. X. Sheehan. 1977. An Akkadian grammar: A translation of Riemschneiders Lehrbuch des Akkadischen. 3rd edn. Milwaukee: Marquette University Press. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Churchward, C. Maxwell. 1940. Rotuman grammar and dictionary, comprising Rotuman phonetics and grammar and a Rotuman–English dictionary. Sydney: Australasian Medical Publishing Company. Coetzee, Andries W. 1999. Metathesis in Tiberian Hebrew: A perspective from Optimality Theory. Theoretical Linguistics 25. 99–131. Crowhurst, Megan J. 1998. Um infixation and prefixation in Toba Batak. Language 74. 590–604. Dell, François. 1973. Les règles et les sons. Paris: Hermann. Dell, Gary S. 1984. The representation of serial order in speech: Evidence from the repeated phoneme effect in speech errors. Journal of Experimental Psychology: Learning, Memory and Cognition 10. 222–233. Dell, Gary S. 1995. Speaking and misspeaking. In Lila R. Gleitman & Mark Liberman (eds.) An invitation to cognitive science, vol. 1: Language, 183–208. 2nd edn. Cambridge, MA: MIT Press.

25

Eugene Buckley

Demers, Richard. 1974. Alternating roots in Lummi. International Journal of American Linguistics 40. 15–21. Dimmendaal, Gerrit Jan. 1983. The Turkana language. Dordrecht: Foris. Egerod, Sren. 1965. Verb inflexion in Atayal. Lingua 15. 251–282. Flemming, Edward. 1996. Laryngeal metathesis and vowel deletion in Cherokee. UCLA Occasional Papers in Linguistics 16. 23–44. Foster, Michael. 1982. Alternating weak and strong syllables in Cayuga words. International Journal of American Linguistics 48. 59–72. Fromkin, Victoria A. 1971. The non-anomalous nature of anomalous utterances. Language 47. 27–52. Garrett, Andrew & Juliette Blevins. 2009. Analogical morphophonology. In Kristin Hanson & Sharon Inkelas (eds.) The nature of the word: Essays in honor of Paul Kiparsky, 527–545. Cambridge, MA: MIT Press. Goldsmith, John A. 1990. Autosegmental and metrical phonology. Oxford & Cambridge, MA: Blackwell. Grammont, Maurice. 1905–6. La métathèse dans le parler de Bagnères-de-Luchon. Mémoires de la societé de linguistique de Paris 13. 73–90. Grammont, Maurice. 1933. Traité de phonétique. Paris: Delagrave. Haas, Wim G. de. 1988. Phonological implications of skeleton and feature underspecification in Kasem. Phonology 5. 237–254. Hale, Kenneth. 1976. Phonological developments in particular Northern Paman languages. In Peter Sutton (ed.) Languages of Cape York, 7–40. Canberra: Australian Institute of Aboriginal Studies. Hale, Mark & Madelyn Kissock. 1998. The phonology–syntax interface in Rotuman. In Matthew Pearson (ed.) Recent papers in Austronesian linguistics, 115–128. Los Angeles: Department of Linguistics, University of California, Los Angeles. Halle, Morris. 2001. Infixation versus onset metathesis in Tagalog, Chamorro, and Toba Batak. In Michael Kenstowicz (ed.) Ken Hale: A life in language, 153–168. Cambridge, MA: MIT Press. Halle, Morris & Valdis Zeps. 1966. A survey of Latvian morphophonemics. MIT Research Laboratory of Electronics Quarterly Progress Report 83. 104–113. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Hock, Hans Henrich. 1985. Regular metathesis. Linguistics 23. 529–546. Holscher, Dan, Monica Macaulay & Marnie Jo Petray. 1992. Tone metathesis in the Dangme imperative. Proceedings of the Annual Meeting, Berkeley Linguistics Society 17. 20–33. Horowitz, Edward. 1960. How the Hebrew language grew. New York: Jewish Educational Committee Press. Horwood, Graham. 2002. Precedence faithfulness governs morpheme position. Proceedings of the West Coast Conference on Formal Linguistics 21. 166–179. Horwood, Graham. 2004. Order without chaos: Relational faithfulness and position of exponence in Optimality Theory. Ph.D. dissertation, Rutgers University. Hudson, Grover. 1980. Automatic alternations in nontransformational phonology. Language 56. 94–125. Huehnergard, John. 2005. A grammar of Akkadian. 2nd edn. Winona Lake: Eisenbrauns. Hume, Elizabeth. 1998. Metathesis in phonological theory: The case of Leti. Lingua 104. 147–186. Hume, Elizabeth. 1999. The role of perceptibility in consonant/consonant metathesis. Proceedings of the West Coast Conference on Formal Linguistics 17. 293–307. Hume, Elizabeth. 2001. Metathesis: Formal and functional considerations. In Hume et al. (2001), 1–25.

Metathesis

26

Hume, Elizabeth. 2004. The indeterminacy/attestation model of metathesis. Language 80. 203–237. Hume, Elizabeth & Jeff Mielke. 2001. Consequences of word recognition for metathesis. In Hume et al. (2001), 135–158. Hume, Elizabeth & Misun Seo. 2004. Metathesis in Faroese and Lithuanian: From speech perception to Optimality Theory. Nordic Journal of Linguistics 27. 35–60. Hume, Elizabeth, Norval Smith & Jeroen van de Weijer (eds.) 2001. Surface syllable structure and segment sequencing. Leiden: Holland Institute of Generative Linguistics. Hyman, Larry M. 1985. Word domains and downstep in Bamileke-Dschang. Phonology Yearbook 2. 47–83. Hyman, Larry M. 1995. Minimality and the prosodic morphology of Cibemba imbrication. Journal of African Languages and Linguistics 16. 3–39. Hyman, Larry M. & William R. Leben. 2000. Suprasegmental processes. In Booij et al. (2000), 587–594. Jakobi, Angelika. 1990. A Fur grammar: Phonology, morphophonology and morphology. Hamburg: Buske. Jespersen, Otto. 1949. A Modern English grammar on historical principles. Part I: Sounds and spellings. London: George Allen & Unwin. Keyser, Samuel J. 1975. Metathesis and Old English phonology. Linguistic Inquiry 6. 377–411. Kilani-Schoch, Marianne & Wolfgang Dressler. 1986. Métathèse et conversion morphologiques en arabe tunisien. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 39. 61–75. Kiparsky, Paul & Wayne O’Neil. 1976. The phonology of Old English inflections. Linguistic Inquiry 7. 527–557. Klein, Thomas B. 2005. Infixation and segmental constraint effects: UM and IN in Tagalog, Chamorro, and Toba Batak. Lingua 115. 959–995. Kurisu, Kazutaka. 2001. The phonology of morpheme realization. Ph.D. dissertation, University of California, Santa Cruz. Ladefoged, Peter. 2005. Vowels and consonants: An introduction to the sounds of languages. 2nd edn. Malden, MA & Oxford: Blackwell. Langdon, Margaret. 1976. Metathesis in Yuman languages. Language 52. 866–883. Lockwood, W. B. 1955. An introduction to modern Faroese. Copenhagen: Munksgaard. Ìubowicz, Anna. 2009. Infixation as morpheme absorption. In Steve Parker (ed.) Phonological argumentation: Essays on evidence and motivation, 261–284. London: Equinox. MacKay, Donald G. 1970. Spoonerisms: The structure of errors in the serial order of speech. Neuropsychologia 8. 323–350. Malone, Joseph L. 1971. Systematic metathesis in Mandaic. Language 47. 394–415. Malone, Joseph L. 1993. Tiberian Hebrew phonology. Winona Lake, IN: Eisenbrauns. Mattina, Anthony. 1979. Pharyngeal movement in Colville and related phenomena in the interior Salishan languages. International Journal of American Linguistics 45. 17– 24. McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12. 373–418. McCarthy, John J. 1989. Linear order in phonological representation. Linguistic Inquiry 20. 71–99. McCarthy, John J. 2000. The prosody of phase in Rotuman. Natural Language and Linguistic Theory 18. 147–197. McCarthy, John J. 2007. Hidden generalizations: Phonological opacity in Optimality Theory. London: Equinox. McCarthy, John J. & Alan Prince. 1990. Foot and word in prosodic morphology: The Arabic broken plural. Natural Language and Linguistic Theory 8. 209–283.

27

Eugene Buckley

McCarthy, John J. & Alan Prince. 1993. Generalized alignment. Yearbook of Morphology 1993. 79–153. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.) Papers in Optimality Theory, 249–384. Amherst: GLSA. Montler, Timothy. 1989. Infixation, reduplication, and metathesis in the Saanich actual aspect. Southwest Journal of Linguistics 9. 92–107. Noonan, Michael. 1997. Inverted roots in Salish. International Journal of American Linguistics 63. 475–515. Okrand, Marc. 1979. Metathesis in Costanoan grammar. International Journal of American Linguistics 45. 123–130. Orgun, Cemil Orhan & Ronald L. Sprouse. 1999. From MParse to Control: Deriving ungrammaticality. Phonology 16. 191–224. Osthoff, Hermann & Karl Brugmann. 1878. Morphologische Untersuchungen auf dem Gebiete der indogermanischen Sprachen. Leipzig: Hirzel. Penny, Ralph J. 2002. A history of the Spanish language. 2nd edn. Cambridge: Cambridge University Press. Phelps, Elaine. 1975. Simplicity criteria in generative phonology: Kasem nominals. Linguistic Analysis 4. 297–332. Phelps, Elaine. 1979. Abstractness and rule ordering in Kasem: A refutation of Halle’s maximizing principle. Linguistic Analysis 5. 29–69. Pierret, Jean-Marie. 1994. Phonétique historique du français et notions de phonétique générale. Louvain-la-Neuve: Peeters. Prince, Alan & Paul Smolensky. 2004. Optimality Theory: Constraint interaction in generative grammar. Malden, MA & Oxford: Blackwell. Prunet, Jean-François. 2006. External evidence and the Semitic root. Morphology 16. 41–67. Prunet, Jean-François, Renée Béland & Ali Idrissi. 2000. The mental representation of Semitic words. Linguistic Inquiry 31. 609–648. Pulleyblank, Douglas. 1986. Tone in Lexical Phonology. Dordrecht: Reidel. Rohlfs, Gerhard. 1930. Etymologisches Wörterbuch der unteritalienischen Gräzität. Halle, Saale: Niemeyer. Rubin, Dominic. 2001. Place-licensing: Opposite orderings are cognitive, not phonetic. In Hume et al. (2001), 189–209. Sagey, Elizabeth. 1986. The representation of features and relations in nonlinear phonology. Ph.D. dissertation, MIT. Sandler, Wendy. 1993. A sonority cycle in American Sign Language. Phonology 10. 243–279. Schachter, Paul & Fe Otanes. 1972. Tagalog reference grammar. Berkeley: University of California Press. Seo, Misun & Elizabeth Hume. 2001. A comparative account of metathesis in Faroese and Lithuanian. In Hume et al. (2001), 210–229. Silva, Clare M. 1973. Metathesis of obstruent clusters. OSU Working Papers in Linguistics 14. 77–84. Silverman, Daniel. 2006. A critical introduction to phonology: Of sound, mind, and body. London & New York: Continuum. Sim, Ronald J. 1981. Morphophonemics of the verb in Rendille. Afroasiatic Linguistics 8. 1–33. Smith, Norval. 1985. Spreading, reduplication and the default option in Miwok nonconcatenative morphology. In Harry van der Hulst & Norval Smith (eds.) Advances in nonlinear phonology, 363–380. Dordrecht: Foris. Sohn, Ho-min. 1980. Metathesis in Kwaraae. Lingua 52. 305–323. Steriade, Donca. 1990. Gestures and autosegments: Comments on Browman and Goldstein’s paper. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 382–397. Cambridge: Cambridge University Press.

Metathesis

28

Steriade, Donca. 2001. Directional asymmetries in place assimilation: A perceptual account. In Elizabeth Hume & Keith Johnson (eds.) The role of speech perception in phonology, 219–250. San Diego: Academic Press. Stonham, John T. 1994. Combinatorial morphology. Amsterdam & Philadelphia: John Benjamins. Thompson, Laurence C. & M. Terry Thompson. 1969. Metathesis as a grammatical device. International Journal of American Linguistics 35. 213–219. Ultan, Russell. 1975. Infixes and their origins. In Hansjakob Seiler (ed.) Linguistic workshop III, 157–205. Munich: Fink. Ultan, Russell. 1978. A typological view of metathesis. In Joseph H. Greenberg, Charles A. Ferguson & Edith A. Moravcsik (eds.) Universals of human language, vol. 2: Phonology, 367–402. Stanford: Stanford University Press. Ussishkin, Adam. 2005. A fixed prosodic theory of nonconcatenative templatic morphology. Natural Language and Linguistic Theory 23. 169–218. Vennemann, Theo. 1988. Preference laws for syllable structure and the explanation of sound change: With special reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. Wanner, Dieter. 1989. On metathesis in diachrony. Papers from the Annual Regional Meeting, Chicago Linguistic Society 25. 434–450. Webb, Charlotte. 1974. Metathesis. Ph.D. dissertation, University of Texas, Austin. Wechssler, E. 1900. Giebt es Lautgesetze? Forschungen zur romanischen Philologie: Festgabe für Hermann Suchier, 349–538. Halle: Niemeyer. Wells, Rulon. 1951. Predicting slips of the tongue. Yale Scientific Magazine, December, 9–30. Williams, Edwin B. 1962. From Latin to Portuguese. 2nd edn. Philadelphia: University of Pennsylvania Press. Winters, Steve. 2001. VCCV perception: Putting place in its place. In Hume et al. (2001), 230–247. Wonderly, William L. 1951. Zoque II: Phonemes and morphophonemes. International Journal of American Linguistics 17. 105–123. Yip, Moira. 2002. Tone. Cambridge: Cambridge University Press. Yu, Alan C. L. 2007. A natural history of infixation. Oxford: Oxford University Press. Zoll, Cheryl. 2001. Constraint and representations in subsegmental phonology. In Linda Lombardi (ed.) Segmental phonology in Optimality Theory: Constraints and representations, 46–78. Cambridge: Cambridge University Press.

60

Dissimilation Patrik Bye

1

Introduction

Dissimilation prototypically refers to a situation in which a segment becomes less similar to a nearby segment with respect to a given feature. As a synchronic alternation, it can be exemplified by liquid dissimilation in Georgian, where the ethnonym-forming suffix {-uri} becomes [uli] when an /r/ precedes it anywhere within the word (Fallon 1993; Odden 1994). The resulting pattern of alternation is shown in (1). (1)

Georgian r-dissimilation a. b. c.

p’olon-uri somø-uri sur-uli p’rusi-uli avst’ral-uri kartl-uri

‘Polish’ ‘Armenian’ ‘Assyrian’ ‘Prussian’ ‘Australian’ ‘Kartvelian’

In (1a), the suffix surfaces in its basic (non-dissimilated) form. The forms in (1b) illustrate the result of unbounded dissimilation within the word. In the word meaning ‘Prussian’ it takes place despite the presence of the intervening consonant. If a lateral /l/ intervenes between the two rhotics, however, dissimilation does not apply. This is shown in (1c). A very similar example of liquid dissimilation comes from Latin (e.g. Kent 1936, 1945; Steriade 1987), where the alternation is reversed. The adjectival suffix -alis, as in navalis ‘naval’, dissimilates to -aris whenever another /l/ precedes it in the word, e.g. lenaris ‘lunar’. Dissimilation is similarly blocked whenever /r/ intervenes between the trigger and the target, e.g. fldralis ‘floral’, *fldraris. As a diachronic change, dissimilation is most often sporadic, applying to random lexical items (Posner 1961). The historical development of Latin and the Romance languages furnish several examples of sporadic liquid dissimilation, e.g. Latin arbor > Spanish arbol ‘tree’, peregrcnus > Late Latin pelegrcnus ‘pilgrim’. Regular synchronic alternations involving dissimilatory processes are far more rare and, The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0060

Dissimilation

2

as a result, dissimilation has been afforded somewhat less systematic attention than other more common segmental patterns like assimilation. Nevertheless, the study of dissimilation phenomena offers a valuable source of insights into the fundamental questions phonologists ask. These questions include (a) the nature of rules and representations, and the relation between the two, (b) the division of labor between the grammar and the lexicon, and (c) whether phonological patterns reflect possibly innate cognitive biases or extralinguistic factors operating during acquisition. The organization of the remainder of this chapter is as follows. §2 sets out the major parameters of dissimilation, explaining which features participate, along with restrictions on the interaction of context and focus determined by locality and domain of application. §3 addresses the contribution that the study of dissimilation phenomena has made to phonological theory, assessing how it has shaped our understanding of both representations and rules. §4 provides an overview of the motivations for dissimilatory patterns proposed in the literature. Conclusions and questions for future research are given in §5.

2

Dissimilatory patterns and their parameters

2.1 Participating features Suzuki (1998) presents a comprehensive survey of cross-linguistic dissimilatory patterns. His survey includes 39 dissimilatory alternations. Table 60.1 provides a somewhat revised summary with a total of 46 alternations, adducing a few additional cases not covered in Suzuki’s original survey, and suppressing cases that on closer inspection turn out not to be true dissimilation.1 The second column of the table specifies the locality condition on the process (for illustration

1

For example, Suzuki’s cases 54 –57 are grouped under “polarity”, but on closer inspection they appear to have little in common. The reasons for reclassifying or not including these cases here are, briefly, as follows. In Russian jakan′e (54) a pre-tonic non-high vowel reduces to [i] or [a], depending on the quality of the following stressed vowel. The high or low quality of the reduced vowel gives the impression of maximizing the contrast in vowel height, e.g. /s j e’mj ju/ → [s j a’m j ju] ‘seven (inst)’, but /dj i’sjatka/ → /d j i’sjatka/ ‘tenfold’. Crosswhite (1999: 79–83), however, argues that the dissimilatory effect is only epiphenomenal, and actually has nothing to do with dissimilation or enhancement of vowel height contrast. Based on work by Alderete (1995), she argues that what is at issue is actually a difference in foot structure. Syllables with prominent nuclei may constitute feet on their own. In [dji(’s j a[[t)ka], the stressed syllable forms a foot on its own, whereas in [(s j a’m j ju)] the pre-tonic syllable must also be incorporated because the stressed nucleus is not sufficiently prominent. The choice of raising or lowering thus comes to depend on whether the focus is parsed into a foot or not. Dinka (55) represents a morphological exchange rule, which is highly controversial in linguistic theory. See Wolf (2007) for an alternative analysis of exchange rules in terms of featural affix allomorphs. Huamelultec Chontal (56) appears to be a case of [spread glottis] dissimilation, not polarity. Margi (57) represents a case of allomorphy. Also not included is Thurneysen’s Law in Gothic, which recent research by Woodhouse (1998) shows to be a case of analogical relexicalization rather than a phonological rule. Suzuki also includes Finnish consonant gradation (Keyser and Kiparsky 1984; Alderete 1997), but it is excluded here, since it is neither synchronically properly phonological nor obviously a dissimilation rule. One language cited as evincing low vowel dissimilation is the Chadic language Kera (Ebert 1974; Kenstowicz and Kisseberth 1979), where /a/ is claimed to dissimilate to [H] preceding another /a/. Recent work by Pearce (2008), however, shows that the effect is due to reduction in unstressed syllables.

3

Patrik Bye

Table 60.1 Dissimilatory alternations (Rt = root adjacency; q = syllable adjacency; unbounded (within the word); P = progressive; R = regressive) Feature

Locality

Example

labial

Rt

Tashlhiyt Berber (P: Elmedlaoui 1985, 1995; Jebbour 1985; Boukous 1987; Larsi 1991; Odden 1994) Cantonese (P: Yip 1982, 1988), Palauan (R: Josephs 1975, 1990; Finer 1986) Akkadian (R: Soden 1969; McCarthy 1979; Yip 1988; Hume 1992; Odden 1994), Tashlhiyt Berber (2×R; see references above), Palauan (R; see references above)

q unbounded

coronal

Rt

Dakota (R: Shaw 1976, 1985)

lateral

unbounded

Kuman (R: Trefry 1969; Walsh Dickey 1997), Latin (P: Kent 1936, 1945; Posner 1961; Johnson 1973; Steriade 1987, 1995; Odden 1994; Walsh Dickey 1997), YidiJ (R: Dixon 1977; Steriade 1995; Walsh Dickey 1997), Yimas (P: Foley 1991; Odden 1994; Walsh Dickey 1997)

rhotic

Rt unbounded

Ainu (R: Maddieson 1984; Shibatani 1990) Georgian (P: Fallon 1993; Odden 1994), Modern Greek (R: Newton 1971; Walsh Dickey 1997), Sundanese (R: Cohn 1992; Holton 1995), Yindjibarndi (P: Wordick 1982)

voice

q

Bantu (R: Bennett 1967; Davy & Nurse 1982; Odden 1994; Lombardi 1995), Japanese (Itô & Mester 1986; Alderete 1997)

spread glottis

q

Ancient Greek (R: Grassmann 1863), Huamelultec Chontal (P: Waterhouse 1949, 1962; Kenstowicz & Kisseberth 1979), Inari Saami (P: Itkonen 1986–91)

constricted glottis

q

Seri (R: Marlett & Stemberger 1983; Yip 1988), Cuzco Quechua (R: Parker 1997)

nasal NC

Rt q unbounded

Chukchi (R: Odden 1987) Gooniyandi (P: McGregor 1990; Odden 1994; Evans 1995) Gunindji (P: McConvell 1988; Odden 1994; Evans 1995), Yindjibarndi (P: Wordick 1982; Odden 1994)

continuant

Rt

Modern Greek (P: Kaisse 1988), Northern Greek (R: Newton 1971), Tsou (P: Szakos 1994), (P: Quintero 2004), North Central Spanish (R: González 2008)

high

q Rt

Guere (R: Paradis & Prunet 1989) Arusa (R: Levergood 1987), Wintu (Pitkin 1984)

low

q

Marshallese (R: Bender 1968, 1969; Kenstowicz & Kisseberth 1977), Woleaian (R: Sohn 1975; Sohn & Tawerilmang 1976; Poser 1982)

length

q

Gidabal (P: Geytenbeek & Geytenbeek 1971), Latin (R: Leumann 1977; Sihler 1995), Oromo (P: Gragg 1976; Lloret 1988; Alderete 1997), Slovak (P: Kenstowicz & Kisseberth 1977, 1979; Rubach 1993)

H

Rt q unbounded

Bantu (P: Goldsmith 1984; Odden 1994) Bantu (P: Goldsmith 1984; Odden 1994) Arusa (P: Levergood 1987; Odden 1994)

L

unbounded

Peñoles Mixtec (P: Daly 1993; Odden 1994)

Dissimilation

4

of dissimilations with different locality conditions, see §2.2). Also indicated is the direction of dissimilation and, in case the language has more than one, the number of dissimilative patterns. The numbers of progressive and regressive dissimilations are more or less evenly split, with 21 and 24 cases respectively. Major class features such as [consonantal], [sonorant], and [approximant] do not appear to participate in dissimilation. The existence of some of these features is indeed contested in the literature. While there is something of a consensus that [sonorant] is necessary,2 Hume and Odden (1996) propose that [consonantal] may be dispensed with. There is also a widespread assumption that the feature [approximant] is not contrastive in language, although see Levi (2008) for evidence to the contrary. Beyond the major class features, all classes of feature may be involved in dissimilation, including place of articulation, laryngeal state, manner (continuancy, liquid, nasality), vowel height, and suprasegmental properties such as length and tone. We shall illustrate some of these in this section; other alternations that raise particular theoretical issues will be illustrated in the relevant sections. Thus, see §3 for examples of nasal dissimilation, and §4.3 for continuant dissimilation. Of the place of articulation features, only [labial] dissimilation is common. Labial dissimilation is illustrated in (2) with data from Tashlhiyt Berber (Odden 1994). The labial nasal /m/ in a prefix dissimilates to [n] if the stem contains a labial consonant anywhere within it. (2)

Labial dissimilation in Tashlhiyt Berber a. b.

las agur # rmi bur # azum

‘shear’ ‘remain’ ‘be tired’ ‘remain celibate’ ‘fast’

am-las am-agur an-#rmi an-bur an-#azum

‘shearer’ ‘abandoned’ ‘tired person’ ‘bachelor’ ‘faster’

Dissimilation of [coronal] is only attested in a single case, Dakota (Shaw 1976, 1985). Underlying coronal non-continuants /t Πn d/ are all neutralized to [k] (or, with regressive voicing assimilation, [g]) before another coronal consonant. The examples in (3) are from Shaw (1985: 184); see also Shaw (1976: 337). (3) Coronal dissimilation in Dakota reduplication a.

b.

2

/Œek/ /Œap/ /t’hs/ /khuœ/ /sut/ /Úat/ /t heŒ/ /Œ heŒ/ /nin/

Œek-’Œeka Œap-’Œapa t’hs-’thza khuœ-’khuÚa suk-’suta Úag-’Úata t hek-’t heŒa Œ hek-’Œ heŒa nig-’nina

‘stagger’ ‘trot’ ‘draw tight’ ‘lazy’ ‘strong’ ‘curved’ ‘be new’ ‘to look like’ ‘very’

This consensus naturally does not extend to those representational theories like Government Phonology, where the elements of representation must have autonomous interpretations (see especially Kaye et al. 1985). Obviously, [sonorant] has no phonetic interpretation independent of the place and manner features with which it is associated.

Patrik Bye

5

There are apparently no attested examples of dissimilation involving the feature [dorsal]. Alternations involving laterals and rhotics are relatively common. We shall provide examples of liquid dissimilation in §2.2 in connection with the discussion of locality parameters. Several Australian languages show dissimilation of prenasalized stops or nasal + stop clusters (NC). In Gurindji (Pama-Nyungan, Northern Territory; McConvell 1988; Odden 1994), this process is unbounded, e.g. /lutcu-Ika/ ‘ridge’, /pinka-Ika/ → [pinka-ka] ‘river-loc’, /kankula-mpa/ → [kankula-pa] ‘high ground-loc’. Several languages have restrictions on consecutive heavy nuclei that do not appear reducible to prosodic structure. In Slovak, for example, a long nucleus becomes short following a long nucleus, according to a rule known as the Rhythmic Law (Rubach 1993: 172–175). Thus, the suffixes {-a(} (fem sg) and {-e(mu} (dat sg) shorten their first vowel following a long vowel, e.g. [ma’l-a] ‘small-fem sg’ vs. [’ml(kv-a] ‘silent-fem sg’; [ma’l-emu] ‘small-masc dat sg’ vs. [’ml(kv-emu] ‘silent-masc dat sg’. The alternation is apparently unrelated to stress (Rubach 1993: 41– 42). At least in Western Slovak, the main stress falls on the initial syllable of the word, and sources report a binary stress pattern, some with the possibility of ternary alternation. The Rhythmic Law nevertheless applies in odd-numbered syllables, where we would expect resumption of secondary stress on a binary alternating pattern. This is shown by derivations with the agentive suffix {-ni(k} and the diminutive {-ik}, e.g. [hutni(k] ‘steelworker’ vs. [Œalu(nnik] ‘wallpaperer’, [xlebi(k] ‘bread’ vs. [’–ba(nik] ‘pot’. Several languages of Vanuatu have productive Low Vowel Dissimilation (Lynch 2003). In Maskelynes (Malayo-Polynesian), the nominalizer prefix is realized as [nH-] when the following vowel is low /e a o/, and [na-] following a high vowel /i H u/. (4)

a.

b.

na-vis na-xHmar na-xut nH-matu nH-gor

‘banana’ ‘men’s house’ ‘louse’ ‘right (hand)’ ‘green coconut’

Dissimilation is occasionally also used to refer to the deletion of one of a pair of similar neighboring sounds. Hall (2009), for example, describes this phenomenon with reference to /r/ in American English, in principle giving alternations like [fAçm] farm vs. [fAmÌ] farmer, and [istÌn] eastern vs. [istHnÌ] easterner. All the cases we have looked at so far involve the elimination of sequences of similar sounds. Preventive dissimilation is when the creation of new sequences of similar sounds is blocked. One example is provided by Inari Saami (Itkonen 1986 –91), which has a morphologically conditioned process of consonant gradation. An overlong obstruent in the “strong” grade generally alternates with the corresponding singleton in the “weak” grade, as shown in (5). In each of the examples below, the strong grade form on the left represents the nominative singular, the weak grade form on the right the accusative-genitive singular. Examples have been adapted from Finno-Ugric transcription into IPA, according to the conventions set out in Bye et al. (2009) ([Z] is a somewhat low central vowel; [∆] is “ultrashort”).

Dissimilation

6

(5) Inari Saami consonant gradation (obstruents) tsuop(ph∆ fAt(t∆ Jeœ(œi

tsuophZ fAAtZ Jeeœi

‘meat of fish’ ‘yard’ ‘mud, slush’

Under normal circumstances, the overlong aspirated velar stop /k(kh/ alternates with /h/, as shown in (6). This may be taken to reflect a general process that debuccalizes /kh/, leaving bare [h]. (6) Inari Saami consonant gradation: Debuccalization of aspirated velar stop kak(khu Œok(kh i kAl(kkh∆

kaahu, *kaak h u Œohii, *Œokh ii kAAlhZ, *kAAlk hZ

‘unleavened rye-bread’ ‘peak, summit’ ‘chalk’

However, there is one situation where debuccalization to [h] fails to take place. As (7) shows, this is when the onset of the preceding syllable is also /h/. (7) Inari Saami consonant gradation: Debuccalization blocked hik(k h i hiik h i, *hiihi ‘hay-basket’ h h hAk(k ∆ hAAk Z, *hAAhZ ‘canon’ h h hul(kk ∆ huulk Z, *huulhZ ‘knife-sheath’

2.2 Locality and domains Dissimilation may be associated with one of three locality conditions listed in (8). This parameter was first addressed in detail by Odden (1994). Suzuki’s (1998) survey largely confirms this picture. (8) Locality conditions a. Root adjacency b. Syllable adjacency c. Unbounded Liquid dissimilation may be used to exemplify all three locality conditions. Ainu, a language isolate of Japan, illustrates the root-adjacency condition (Shibatani 1990: 13). Given an underlying cluster /rr/, the first /r/ dissimilates to [n], as shown in (9). (9) Ainu r-dissimilation kukor kur kor mat

‘my husband’ ‘his wife’

kukon rusuj kon rametok

‘I want to have (something)’ ‘his bravery’

Yimas (Foley 1991: 54), a Sepik-Ramu language of Papua New Guinea, illustrates liquid dissimilation operating under syllable adjacency. An /r/ dissimilates to [t] if there is an /r/ in the immediately preceding syllable. The examples in (10) show variation in the shape of the inchoative suffix {-ara} (1991: 290).

Patrik Bye

7 (10)

Yimas r-dissimilation tuak-ara‘break open’ kamprak-ara- ‘snap’ apr-ata‘open, spread’

Dissimilation may also be unbounded within the word, as we have already seen in the Georgian example in (1) with which we opened this chapter.

3

Dissimilation in the grammar

One of the fundamental issues in generative phonology has always been whether linguistically significant generalizations should be assigned to particular designated levels of representation, such as the underlying or surface level, or to the rules that map one representation onto another (McCarthy 2007; Bye, forthcoming). In earlier approaches in the style of SPE (Chomsky and Halle 1968), dissimilations were described in terms of feature-changing rules of the general form shown in (11). (11)

X → [−F] / __ [+F]

Rules of this kind were criticized because the pairing of structural change and environment was arbitrary. Because of this, they were unable to distinguish between natural assimilations like (12a) and arbitrary rules like (12b) (example from Odden 1987). (12)

a. b.

[+consonantal] → [+voice] / __ [+voice] [+consonantal] → [+voice] / __ [+continuant]

Concerns about the generative power of feature-changing rules thus motivated the development in the mid-1970s and 1980s of non-linear approaches to phonological representation (Goldsmith 1976) that permitted greater elegance and simplicity in the statement of natural rules. Assimilation rules were remodeled as feature-filling spreading (see chapter 81: local assimilation and chapter 77: long-distance assimilation of consonants). In non-linear terms, dissimilation is simply the deletion or delinking of a feature and, in accounts that retain a view of features as binary, independently motivated insertion of a default value (Odden 1987, 1994; chapter 27: the organization of features). For example, Chukchi has a process changing underlying /I/ to [:] before another nasal, shown in (13). (13)

taraI-Hk inawrHI-Hk pit?iI

‘build a dwelling’ ‘to give as a gift’ ‘cold’

nH-tara:-mori inawrH:-nin pit?i:-Iinqij

‘we built a dwelling’ ‘he gave it’ ‘boy with a cold’

Odden (1987: 242) provides an analysis of this alternation as delinking of [+nasal] before another [+nasal], as shown in (14). Subsequently, redundancy rules fill in the feature [−nasal] by default.

Dissimilation (14)

denasalization

default

→

→

inawrHI-Hn [+N][+N]

inawr H K- H n [+N]

8

inaw r H : -H n [−N][+N]

Accompanying the development of non-linear representations was a return to the idea that at least certain phonological generalizations are best stated as constraints on surface forms. This conception made it feasible to explain how it was possible that certain rules seemed to share the same functional teleology (the “conspiracy” problem; see Kisseberth 1970 and chapter 70: conspiracies). Such constraints could trigger the application of repairs, such as the deletion of the first of the two [+nasal] features, or block the application of rules that would otherwise apply (see examples of preventive dissimilation in §2). The first such constraint on output representations was the Obligatory Contour Principle (Leben 1973; Goldsmith 1976; McCarthy 1979, 1981, 1986, 1988; Odden 1988; Yip 1988, 1989), one formulation of which is provided in (15). (15)

Obligatory Contour Principle (OCP) (McCarthy 1986, 1988) At the melodic level, adjacent identical elements are prohibited.

The OCP was originally used in accounting for tonal phenomena, especially adjacency of high tones, but it was subsequently extended to include other features. The OCP specifies a negative output target, and dissimilation only represents one strategy for satisfying it. Other repair strategies include merger of adjacent identical nodes, blocking of syncope (McCarthy 1986), and the insertion of epenthetic segments (Yip 1988). The OCP was incorporated into work couched in the framework of Optimality Theory (OT: McCarthy and Prince 1993; Prince and Smolensky 1993), where it became a violable constraint. Alderete (1997) and Itô and Mester (2003) propose that the OCP may represent a local self-conjunction of more primitive markedness constraints (Smolensky 1995, 1997). OCP[F] is violated precisely when *[F] is violated more than once within some local domain. Another major theoretical concern during the 1980s especially was locality conditions on the application of rules, and dissimilation played a major part in this debate. The autosegmentalization of representations into tiers permitted the elimination of many kinds of apparent long-distance effects. Sounds that are nonadjacent on the level of the segmental root may nevertheless dissimilate, provided that the relevant features are adjacent on the same autosegmental tier. Steriade (1987) argues that the Latin facts mentioned at the beginning of this chapter may be accounted for by a version of the OCP with jurisdiction over the [lateral] tier, over which interactions between tier-adjacent liquids may be described. The diagrams in (16) show how the liquids in the words lenaris and fldralis are projected onto separate tier-specifying values of [lateral]. This allows us to explain the ungrammaticality of the counterfactual form *lenalis as a result of the two occurrences of [+lateral] being adjacent on the [lateral] tier, in violation of OCP[lateral]. In fldralis, on the other hand, there is an intervening [−lateral] between the two occurrences of [+lateral], so the OCP is not violated.

9 (16)

Patrik Bye Lateral dissimilation OCP

*OCP

OCP

[+lat] [−lat]

[+lat] [−lat]

[+lat] [−lat] [+lat]

lun-aris

*lun-alis

f l o r - a l i s

Even with the possibility of factoring the representation into tiers, though, there is still an empirical residue that poses a problem for a strict interpretation of locality. In some theories, vocalic place (V-Place) and consonantal place (C-Place) are represented on separate planes (Clements 1991; Clements and Hume 1995; Morén 2003; see also chapter 19: vowel place and chapter 22: consonantal place of articulation). This organization implies that non-adjacent consonants and vowels should not display any interaction, but this expectation is not borne out. Akkadian (Soden 1969: 64ff.), for example, has a nominalizer prefix {ma-} that dissimilates to {na-} if followed by a labial consonant in the stem, e.g. /ma-W?al-t-u/ ‘question’ but /ma-rkab-t/ → [na-rkab-t] ‘chariot’. If a labial vowel or glide intervenes between the trigger and the target, however, dissimilation is blocked, e.g. /ma-wmii-t-um/ → *[ma-amii-t-um} ‘oath’. Odden (1994: 319) argues for an additional adjacency parameter, transplanar locality, to cover these cases, but it is unclear how this is to be formalized.

4

Motivations for dissimilation

There are a number of theories of what causes dissimilation. The purpose of this section is to review the major proposals as well as some others of more limited applicability. Our point of departure will be Ohala’s Co-articulationHypercorrection Theory (CHT; Ohala 1981, 1993, 2003), which is presented in §4.1. According to the CHT, dissimilation results when the listener reverses a perceived co-articulation. The central prediction of the CHT is that dissimilation should only occur with features that have cues that are significantly extended in time. Other theories assume a processing motivation. Frisch et al. (2004) argue that similarity avoidance effects are due to the difficulties associated with processing the sequencing of similar segments. This bias is reflected in the statistical structure of the lexicon and is described in §4.2. Following on from this, §4.3 considers the possibility that dissimilation in manner between pairs of adjacent fricatives or stops may be understood in terms of the enhancement of place cues. In §4.4 we look at dissimilation-like phenomena in certain kinds of reduplication (“echo” reduplication) and language games, which exploit non-identity for aesthetic, ludic, or secret purposes.

4.1 Dissimilation as listener reversal of co-articulation The phonetic realization of certain features may extend over long temporal domains. Long-domain features are interesting because they create an ambiguity for the listener faced with the task of reconstructing the feature’s place in phonological structure. This ambiguity creates conditions favorable to reanalysis,

Dissimilation

10

which – in the case of temporally extended features – may take one of three forms: assimilation, metathesis, or dissimilation (Blevins 2004). In dissimilation, when one instance of a distinctive feature occurs within the phonetic domain of another instance of the same distinctive feature, there is an ambiguity as to whether the phonetic effects should be ascribed to the first or second instance of the feature, or both. In a series of highly influential publications, Ohala (1981, 1993, 2003) argues that dissimilation as a sound change is the result of reversal by the listener of perceived co-articulation (see chapter 98: speech perception and phonology). The driving force of the change on this view is the overzealous application of reconstructive rules, with the result that long-domain effects that are actually intended by the speaker become reversed. This is known as hypercorrection. The mechanism of Ohala’s Co-articulation-Hypercorrection Theory is schematized in (17). (17) Dissimilation as sound change by the listener (after Ohala 1981: 187) Speaker

Listener /yt/

produced as

/ut/ reconstructed as

[yt]

Listener-turned-Speaker

heard as

produced as [yt]

[ut]

In this example, the speaker intends to say [yt], which is also the form that the listener actually hears. However, the listener is in possession of tacit phonetic knowledge that coronal consonants raise the value of F2 on neighboring vowels. Drawing on this knowledge, he concludes that the intended quality of the vowel has been distorted due to its proximity to the coronal consonant. The perceived distortion is then eliminated by reconstructing the intended form as /ut/: the /y/ dissimilates. There are three important entailments of the CHT. The first follows from the assumption that dissimilation involves co-articulation. Segments are only expected to dissimilate to the extent that they entail overlapping articulations. Many dissimilations involve segments that are not phonologically adjacent on the level of the segmental root node. A well-known example is Grassmann’s Law in Indo-European (Grassmann 1863). In Ancient Greek, which provides one instantiate of the sound law, there cannot be more than one aspirated stop in a pair of adjacent syllables (Smyth 1956 [1920]: 31). Thus, earlier /t h rikh-os/ ‘hair (gen sg)’ became [trikhos] rÙíunp (cf. [t h riks] yÙím (nom sg)). Grassmann’s Law apparently represents an interaction between two non-adjacent consonants. Once we take into account the co-articulatory effect of the aspiration on the following vowel, the apparent action-at-a-distance effect evaporates because the aspiration overlaps phonetically with the dissimilation target. Following release of the stop closure, aspiration persists into the following vowel for 60 msecs or so, presenting the listener with an ambiguity as to whether the aspiration represents post-aspiration of the first stop or pre-aspiration of the second. Segments that are outside of each other’s co-articulatory range are not expected to dissimilate according to the CHT.

Patrik Bye

11

The second consequence of the CHT is that dissimilation cannot take the form of a quantitative change within the same category. Dissimilation should always be limited to phonologically contrastive features (cf. Grammont 1895; Kiparsky 2003). This follows directly from the assumption that what listeners are doing, when they hypercorrect, is reconstructing what they believe is the intended form, which must be a distinctive segment of the language. Assimilative changes, on the other hand, may give rise to novel structures or segments. The third consequence of the CHT is that it should not matter which direction the perceived distortion is resolved in. The CHT is neutral with respect to whether the dissimilation is progressive or regressive. In (17), an equally valid outcome would have been dissimilation of the consonant, e.g. to /yk/. The empirical substance of Ohala’s proposal consists of the following predictions. (18)

Ohala’s predictions a.

The likelihood that a given consonantal feature participates in dissimilation depends on whether the associated perceptual cues have a short or long domain. b. The domain of dissimilation should be linked to the temporal extension of the perceptual cues. c. Features whose cues are localized on the segment should not show dissimilatory behavior. On this basis, an up-to-date list of the features shown to have temporal extension, and therefore likely to dissimilate according to the CHT, is shown in Table 60.2, adapted from a corresponding table in a paper on the evolution of metathesis by Blevins and Garrett (2004: 123). Examples are incorporated from the surrounding discussion in their text. Features not likely to dissimilate according to the CHT are fricative, affricate, stop, and voice. The phonetic cues for each of these segment types are localized on the segment itself. For example, stops are cued by high amplitude bursts on release of the closure. These bursts are very short, of the order of 5 msecs to 10 msecs. The temporal extent of voicing and fricative noise is limited by the extent of the segment’s articulation phase. Examples of continuancy and voicing dissimilation nevertheless exist. Examples of continuancy dissimilation are discussed in §4.3, along with a possible phonetic motivation. When Ohala (1981) initially framed his CHT, the existence of liquid dissimilation appeared to present a problem, since at that time no work had been done on temporally extended cues for liquids. Far from being occasional, liquids are, after labials, the most likely to dissimilate. Moreover, they show pronounced action-at-a-distance, as the Georgian example at the beginning of this chapter shows. This is surprising if it is all down to the formant transitions onto neighboring vowels. Starting with Kelly and Local (1986) and Kelly (1989), however, much research has shown that liquids have temporally extended acoustic-perceptual cues. Tunley (1999) demonstrated experimentally that /l/ causes raising in F2 and F3 on neighboring high vowels, while /r/ results in lowering. These effects are moreover observable up to five syllables away from the lateral segment itself (Hawkins and Smith 2001; see also chapter 30: the representation of rhotics). West (1999b) found that when the liquid and its phonetic context were masked with white noise, speakers were nonetheless able

Dissimilation

12

Table 60.2 Temporally extended features. References to acoustic properties are from Ladefoged (1993; L), Ladefoged et al. (1988; LMJ) and Ladefoged and Maddieson (1996; LM) Feature

Acoustic property

Examples

rounding

lowering of all formants (LM 356–358)

French, English (Benguerel & Cowan 1974; Lubker & Gay 1982)

velarization

lowered F2 (LM 361–362)

Arabic (Ghazali 1977; Card 1983)

pharyngealization

lowered F3, raised Fl (LM 307)

Interior Salish (Bessell 1998a, 1998b)

palatalization

raised F2 (LM 364)

Catalan (Recasens 1984, 1987), English (Hawkins & Slater 1994), Japanese (Magen 1984), Marshallese (Choi 1992), Russian (Keating 1988), Bantu (Manuel 1987)

retroflection

lowered F3, F4, clustering of F2, F3, F4 (E 203, LM 28)

Gooniyandi (McGregor 1990), Gujarati (Dave 1977), Hindi (Stevens & Blumstein 1975), Malayalam (Dart 1991), Tiwi (Anderson & Maddieson 1994)

laryngealization

more energy in Fl, F2, more jitter (LMJ)

Cayuga (Dougherty 1993)

aspiration

more energy in F0, more noise (LMJ)

Cayuga (Dougherty 1993)

nasalization

spectral zero, nasal anti-resonance (LM 116)

English (Cohn 1990)

jaw lowering

raised F1

English (Amerman et al. 1970)

rhoticity

lowered F3 (LM 244, 313)

laterality

lateral formants (LM: 193–197), raised/ lowered F2/F3

English (Kelly & Local 1986; Kelly 1989; Tunley 1999; West 1999a, 1999b, 2000; Hawkins & Smith 2001)

to reconstruct the intended liquid from the resonances in vowels up to three syllable nuclei away. These recent findings on the phonetics of liquids thus square well with the prediction that the phonological domain of the dissimilating feature should mirror the temporal extension of the corresponding cues. Ohala does not consider dissimilation between vowels. Öhman (1966) showed that vowels may co-articulate across intervening consonants. Dissimilation of vowels across syllables is thus consistent with Ohala’s broader claims. Interestingly, though, all of the known examples of vowel dissimilation involve vowel height (chapter 21: vowel height). Vowel height dissimilation is certainly consistent with the experimental finding that lowering of the jaw co-articulates (Amerman et al. 1970), but the existence of co-articulation is apparently not a sufficient predictor of dissimilation. Indeed there is a striking complementarity of phonological patterning between vowel height, on the one hand, and vowel color (roundness

13

Patrik Bye

and backness), on the other. To date, no examples of dissimilation involving the labial or front–back dimensions have come to light. Conversely, systems of vowel harmony in which backness or rounding (or both) are active are richly attested in the literature, but the feature [low] is not frequent in vowel harmony (see Krause 1979 for a possible example from Chukchee). Further discussion of this problem may be found in Alderete and Frisch (2007). There is also still a residue of dissimilatory patterns for which the CHT does not seem to offer an explanation, including NC, long vowel, continuancy (see §4.3 below), and voicing. In the next section we will consider an alternative theory of the origins of dissimilation that does not seem to make any prediction about which features participate.

4.2 Similarity avoidance in the lexicon Several recent studies have examined statistical asymmetries in the lexicon, pointing to a preference for phonetic dissimilarity between neighboring consonants in roots (see also chapter 86: morpheme structure constraints). Berkley (2000) studied English monosyllabic words and found evidence of gradient similarity avoidance effects. Focusing on words of the shape C 1VC 2 , she found that there are significantly fewer such words containing homorganic consonants than would be expected if consonants combined randomly, i.e. had the same probability of occurring as two independent events. Under- or overrepresentation is the ratio of observed to expected frequency (see chapter 90: frequency effects). A pair is underrepresented if the observed-to-expected (or O/E) ratio is less than 1, overrepresented if greater than 1. Words in which C 1 and C 2 are homorganic – such as mop, lull, and king – are underrepresented in the English lexicon. For CVVC words with a long vowel intervening between C1 and C2, the homorganic similarity avoidance effect is also present but weaker. These results, adapted from Berkley (2000), are shown in Table 60.3. Shaded cells indicate underrepresented combinations. Frisch et al. (2004) studied similarity avoidance in the lexicon of Arabic triradical verb roots. They found a very strong effect for adjacent pairs of consonants Table 60.3 Similarity avoidance in English monosyllabic roots Labial Cor obs Cor son Dorsopbfvmw td.Ïsz nlrj guttural œÚŒ– kghIw CVC

Labial Cor obs Cor son Dorsoguttural

pbfvmw td.ÏszœÚŒ– nlrj kghIw

0.52

1.15 0.73

1.33 1.03 0.59

0.85 1.09 1.21 0.72

CVVC

Labial Cor obs Cor son Dorsoguttural

pbfvmw td.ÏszœÚŒ– nlrj kghIw

0.61

1.11 0.75

1.11 1.17 0.70

1.14 1.03 1.01 0.71

Dissimilation

14

(C1 and C2, C2 and C3), tending to categorical, as shown in Table 60.4. For nonadjacent C1 and C3 the effect was still strong, although somewhat weaker. Under both adjacency and non-adjacency the similarity avoidance effect is far stronger than the one observed by Berkley for English. Gray cells are homorganic in terms of major class (Labial, Coronal obstruent, Dorsal, Guttural, Coronal sonorant). They also found that the avoidance was stronger the more similar the consonant. Within the major class of coronals, for example, an adjacent pair of coronals was significantly more frequent if they had different values for [continuant]. Frisch et al. argue that when observed co-occurrence deviates from expected co-occurrence, the learner posits a gradient phonological constraint, which they dub the gradient Obligatory Contour Principle that encodes the generalization “roots with repeated homorganic consonants are unusual.” Statistical generalizations like these form the basis of metalinguistic judgments of relative acceptability of novel words (“word-likeness”), influencing which words are actually used, and the phonological forms of novel and borrowed words (cf. Frisch 2004: 346). For Frisch et al., the gradient OCP represents a statistical generalization over a static lexicon; it does not encode tacit phonetic knowledge directly. Despite this, Frisch et al. do propose a functional explanation for the distributional asymmetries in the lexicon. Repetition of similar consonants is difficult to process (Frisch 2004). This finds the beginnings of an explanation in neural network models that encode linearization of segments. Nodes in the network must be excited and inhibited so as to fire in the right sequence. If there is a sequence of similar segments, the periods of excitation and inhibition may overlap, whether or not there is a corresponding overlap in the acoustic signal. Given two segments, C 1 and C 2, in linear sequence that activate the same distinctive feature node, if the node encoding C 1 is still firing when C 2 is perceived, this may result in simultaneous Table 60.4 Similarity avoidance in Arabic verb roots

Adjacent

Labial Cor obs Dorsal Guttural Cor son

Non-adjacent

Labial Cor obs Dorsal Guttural Cor son

Labial

Cor obs

Dorsal

bfm

td .Ïsz t# d # s# z# œ

kgq

Guttural Cor son ø ú p # lrn h?

bfm t d t# d# . Ï s z s # z# œ kgq øú p#h? lrn

0.00

1.37 0.14

1.31 0.52 0.04

1.15 0.80 1.16 0.02

1.35 1.43 1.41 0.07 0.00

1.17 1.25 1.26 1.04 0.07 0.06

1.18 1.23 1.21 1.48 1.39 1.26 0.06

bfm t d t# d# . Ï s z s # z# œ kgq øú p#h? lrn

0.30

1.08 0.38

1.02 1.06 0.24

1.26 1.24 1.16 0.07

1.25 1.05 1.35 0.68 0.25

1.28 1.02 1.14 1.19 0.12 0.34

1.11 0.97 1.23 1.03 1.10 1.13 0.67

15

Patrik Bye

perception of C1 and C2. The resulting blend of the two percepts may result in the same kind of ambiguity that results from co-articulation in the CHT, and is presumably consistent with the same re-analytic strategies. An alternative source of dissimilation in processing may be a refractory period during which the node must be reset in order to detect a second stimulus of the same type. Unlike the CHT and the blending scenario we have just sketched, the refractory period would seem to predict asymmetries in the direction of the resulting pattern. If C 2 occurs within this refractory period of a node that has just fired for C 1, the relevant feature will be perceived on C 1 but may not be perceived on C 2. The effective result is progressive dissimilation. This processing bias may further help explain the pronounced difference in the strength of the effect in English and Arabic. Arabic has non-concatenative morphology and psychologically real abstract consonant roots like /ktb/ ‘write’. In phonological analysis, this has motivated analyses in which consonants and vowels occupy separate tiers in the representation. Roots are vulnerable to speech errors involving misordering of radical consonants. These functional mechanisms and their implications for linguistic patterning provide much fertile ground for further research. First, a greater range of languages must be studied to determine to what extent root co-occurrence constraints have a gradient character or not. Suzuki (1998) presents 16 examples of root cooccurrence restriction, which are summarized in Table 60.5, along with a couple of additional cases. A second challenge concerns the difference between similarity and identity. Arabic shows avoidance of identical radicals. In English, however, segmental identity provides an escape hatch to the OCP. CVC roots where both C 1 and C 2 are labial are generally dispreferred, but not if C 1 and C 2 are identical. For example, selecting /p/ for C 1 and C 2 and permuting the possible nuclei, almost every cell of the paradigm corresponds to an actually occurring word of English: pip, pep, pap, pop, pup, peep, poop, parp, pipe, Pape, pope. See Idsardi and Raimy (2008) for a relevant proposal that segmental identity may be represented in a data structure they call a “linked list.” Finally, the similarity avoidance approach raises anew the question of which features are expected to participate in dissimilation. Are certain features associated with longer periods of excitation in perception than others? And if that turns out to be so, is there a systematic correlation between the length of a feature’s temporal domain in the speech signal and the duration of excitation? To date, place and laryngeal features (MacEachern 1997) have been studied. These studies must therefore be extended to short-domain features, such as stops, fricatives, and voicing.

4.3 Dissimilation and cue robustness Manner dissimilation is predicted not to occur by the CHT. Despite this, a small number of languages display dissimilation of pairs of adjacent stops or fricatives. For example, Osage (Quintero 2004), a Siouan language spoken in Oklahoma, has a rule dissimilating /Ï/ to [t] following /s/, e.g. /œk6œÏa/ → [œk6œta] ‘you want’. Tsou (Szakos 1994), an Austronesian language of Taiwan, has a rule that hardens /h/ to [k] following /s/, giving alternations such as [s-in-uhnu] ‘send someone to do something (actor voice)’ ~ [skuna] (patient voice), [s-m-ohpici] ‘pinch

Dissimilation

16

Table 60.5 Root co-occurrence restrictions Feature

Case

place

Amharic (Bender & Fulass 1978), Arabic (Greenberg 1950; McCarthy 1979, 1981; Mester 1986; Yip 1989; Padgett 1991; Pierrehumbert 1993; Frisch et al. 2004), Cambodian (Yip 1989), French (Plénat 1996), Hawai’ian (McKay 1970), Hebrew (Koskinen 1964), Javanese (Uhlenbeck 1949; Mester 1986; Padgett 1991), Russian (Padgett 1991), Serbian (McKay 1970), Yucatec Mayan (Yip 1989; Lombardi 1991)

labial

Cantonese (Yip 1982, 1988), Ponapean (Rehg & Sohl 1981; Goodman 1995), Yao (Purnell 1965; Ohala 1981), Zulu (Doke 1926)

coronal

Akan (Welmers 1946; McCarthy & Prince 1995)

pharyngeal

Moses-Columbia Salish (Czaykowska-Higgins 1993)

liquid

Javanese (Uhlenbeck 1949; Mester 1986)

rhotic

American English (Hall 2009)

voice

Japanese (Itô & Mester 1986, 2003; Steriade 1987, 1995; Ishihara 1991; Archangeli & Pulleyblank 1994; Alderete 1997; Pater 1999), Bakairi (Gussenhoven & Jacobs 2005)

spread glottis

Sanskrit (Grassmann 1863; Langendoen 1966; Anderson 1970; Phelps & Brame 1973; Sag 1974, 1976; Phelps 1975; Schindler 1976; Borowsky & Mester 1983; Kaye & Lowenstamm 1985; Lombardi 1991)

high

Ngbaka (Thomas 1963; Chomsky & Halle 1968; Mester 1986)

back

Ainu (Itô 1984; Mester 1986; Archangeli & Pulleyblank 1994), Tzeltzal (Slocum 1948; Itô 1984)

length

Japanese

(actor voice)’ ~ [skopica] (patient voice). Non-sibilant fricatives such as [. x] have diffuse spectra. Harris (1958) shows that the F2 transition is required for reliable identification of the fricative. In a cluster of fricatives, one of the transitions, C–V or V–C, is missing. Dissimilating one of the fricatives to the corresponding stop has the effect of sharpening the F2 transition and adding a stop release burst, rendering the place of articulation more easily identifiable.3 In Chontal (Waterhouse 1949, 1962; Kenstowicz and Kisseberth 1979), a Hokan language of Mexico, the imperative suffix is {-la?} after voiceless segments, and {-xa?} after voiced ones, e.g. [fuœ-l j a?] ‘blow it!’, [panx-la?] ‘sit down!’, [ko-xa?] ‘say it!’, [kan-xa?] ‘leave it!’. The pattern seems to involve deleting the second [spread glottis] feature in a 3

One of the apparent cases of continuant assimilation mentioned by Ohala as a potential counterexample to the CHT may in fact turn out to be best understood as an instance of it. Dyen (1972) shows that Proto-Austronesian */s . . . s . . ./ was dissimilated across an intervening vowel in Ngaju-Dayak to /t . . . s . . ./. The evidence, however, only consists of the two words PA *sisik > ND [tisik] ‘fish-scale’ and PA *susu > ND [tuso] ‘breast’. It is perhaps relevant that the vowel immediately following the initial *s is a high vowel. High vowels are known to increase the degree of post-aspiration of a preceding voiceless stop and affrication of a preceding /t/. The initial sibilant may thus have been interpreted as the co-articulatory affrication of an intended /t/.

17

Patrik Bye

cluster of voiceless fricatives – assuming the claim of Vaux (1998) that voiceless fricatives are universally [spread glottis] – allowing the lateral to be more clearly identified as such. Dissimilation between two stops is far more rare, but González (2008) supplies an example from North-Central Spanish, which shows dissimilation of coda /k/ to [.] (and other realizations) before another stop (generally /t/, e.g. [do..’tor] ‘doctor’. González also proposes an explanation in terms of cue robustness, noting that the cues for the first stop are not as salient before another stop as before other segments, due to weaker, or absent, stop release burst and formant transitions. Similar considerations have been argued to condition metathesis in other languages, e.g. Faroese, where final /skt/ metathesizes to [kst] (Hume and Seo 2004; see chapter 59: metathesis).

4.4 The dissimilation game In a different vein, people also apply tacit knowledge of similarity to a variety of ludic and poetic ends. Indeed, the term “dissimilation” entered the field in the 19th century from rhetoric, where it had been in use to describe the variation in style required for good public speaking (cf. Brugmann 1909). The criterion of a perfect rhyme in English, such as pet – bet, is not only that the material following the onset of each stressed syllable is identical, but that the onset of each stressed syllable is different. In considering rhyme in English, we do not appear to count features; we are merely interested in contrastive segments. The pair pet – bet is thus as good a rhyme as the pair pet – set. The same requirement of non-identity turns up in echo reduplication (Alderete et al. 1999; Nevins 2005; chapter 100: reduplication), where the base is reduplicated with an onset determined by convention (fixed segmentism). In Hindi, this kind of reduplication gives a meaning ‘X and the like’ (Singh 1969; Nevins 2005). The fixed segment is /v/ unless the base also begins with a /v/, in which case the echo reduplicant begins with /œ/. Examples from Nevins (2005: 280) are shown in (19).4 (19)

Hindi echo reduplication with fixed segmentism paanii-vaani aam-vaam tras-vras yaar-vaar vakil-œakil, *vakil-vakil

‘water and the like’ ‘mangoes and the like’ ‘grief and the like’ ‘friends and the like’ ‘lawyers and the like’

Similar facts are observed in English shm- reduplication, e.g. potato–shmotato, but shmaltz–shpaltz (Nevins and Vaux 2003), Kannada (Lidz 2001), and Javanese (Yip 1995). Yip (1995, 1998) proposes that these are due to a constraint against the repetition of identical elements, *Repeat, ultimately due to Menn and MacWhinney (1984). Similar facts also turn up in secret languages. In the Kunshan secret language Mo-pa (Yip 1982: 652ff.), a base of the shape C 1V1(C 2) is mapped to a template C1[o]-GV1C 2, where G is a consonant whose value for the feature 4

The forms in (19) represent Nevins’s own fieldwork. The original source on Hindi echo words is Singh (1969).

Dissimilation

18

[continuant] is the opposite of that of C1. Examples, not glossed in the source, are given in (20). (20)

Kunshan secret language Mo-pa tHw k’e d’oI tsa vã sja nHw JHn

to lHw k’o Âe d’o loI tso za vo pã so tsja no tHw JH tuHn

Oral stops are replaced by voiced continuants, while nasals and continuants are replaced by the corresponding voiceless unaspirated stop. This continuant dissimilation is covered neither by the CHT nor cue robustness; the reason for the alternation seems to be purely ludic.

5

Conclusions

There are a number of theories of the origin of dissimilation, and dissimilation may apparently have one of several motivations. According to the Co-articulationHypercorrection Theory (Ohala 1981, 1993, 2003), dissimilation results when the listener reverses a perceived co-articulation. The central prediction of the CHT is that dissimilation should only occur with features that have elongated cues. Other theories assume a functional motivation. Frisch et al. (2004) argue that similarity avoidance effects are due to the difficulties associated with processing the sequencing of similar segments. This bias is reflected in the statistical structure of the lexicon in many languages. However, the predictions of processing-based accounts with respect to the observed featural asymmetries are not yet clear. It was suggested here that manner dissimilation in pairs of adjacent fricatives or stops is best understood as maximizing cues for place of articulation, while dissimilatory phenomena in language games fulfill an aesthetic role. Future research will hopefully extend the empirical base for the study of dissimilation phenomena, and determine more precisely what the division of labor and synergies between the factors discussed here should be.

ACKNOWLEDGMENTS I would like to thank Beth Hume, Marc van Oostendorp, and two anonymous reviewers for helpful feedback on this chapter.

REFERENCES Alderete, John. 1995. Faithfulness to prosodic heads. Unpublished ms., University of Massachusetts, Amherst (ROA-94).

19

Patrik Bye

Alderete, John. 1997. Dissimilation as local conjunction. Papers from the Annual Meeting of the North East Linguistic Society 27. 17–32. Alderete, John & Stefan A. Frisch. 2007. Dissimilation in grammar and the lexicon. In de Lacy (2007), 379–398. Alderete, John, Jill Beckman, Laura Benua, Amalia Gnanadesikan, John J. McCarthy & Suzanne Urbanczyk. 1999. Reduplication with fixed segmentism. Linguistic Inquiry 30. 327–364. Amerman, James, Raymond Daniloff & Kenneth Moll. 1970. Lip and jaw coarticulation for the phoneme. Journal of Speech and Hearing 13. 147–161. Anderson, Stephen R. 1970. On Grassmann’s Law in Sanskrit. Linguistic Inquiry 1. 387–396. Anderson, Victoria B. & Ian Maddieson. 1994. Acoustic characteristics of Tiwi coronal stops. UCLA Working Papers in Phonetics 87. 131–162. Archangeli, Diana & Douglas Pulleyblank. 1994. Grounded phonology. Cambridge, MA: MIT Press. Aronoff, Mark & Richard T. Oehrle (eds.) 1984. Language sound structure. Cambridge, MA: MIT Press. Beckman, Jill N., Laura Walsh Dickey & Suzanne Urbanczyk (eds.) 1995. Papers in Optimality Theory. Amherst: GLSA. Bender, Byron W. 1968. Marshallese phonology. Oceanic Linguistics 7. 16 –35. Bender, Byron W. 1969. Vowel dissimilation in Marshallese. University of Hawai’i Working Papers in Linguistics 1. 88–95. Bender, M. Lionel & Hailu Fulass. 1978. Amharic verb morphology: A generative approach. East Lansing: Michigan State University. Benguerel, A.-P. & H. Cowan. 1974. Coarticulation of upper lip protrusion in French. Phonetica 30. 41– 55. Bennett, Patrick R. 1967. Dahl’s Law and Thagice. African Language Studies 8. 127–159. Berkley, Deborah M. 2000. Gradient Obligatory Contour Principle effects. Ph.D. dissertation, Northwestern University. Bessell, Nicola J. 1998a. Local and non-local consonant-vowel interaction in Interior Salish. Phonology 15. 1– 40. Bessell, Nicola J. 1998b. Phonetic aspects of retraction in Interior Salish. In Ewa Czaykowska-Higgins & M. Dale Kinkade (eds.) Salish languages and linguistics: Theoretical and descriptive perspectives, 125–152. Berlin & New York: Mouton de Gruyter. Blevins, Juliette. 2004. Evolutionary phonology: The emergence of sound patterns. Cambridge: Cambridge University Press. Blevins, Juliette & Andrew Garrett. 2004. The evolution of metathesis. In Hayes et al. (2004), 117–156. Borowsky, Toni & Armin Mester. 1983. Aspiration to roots: Remarks on the Sanskrit diaspirates. In Papers from the Annual Regional Meeting, Chicago Linguistic Society 19. 52–63. Boukous, A. 1987. Phonotactiques et domaines prosodiques en berbère (parler tachelhit d’Agadi, Maroc). Thèse d’état, Université de Paris VIII. Brugmann, Karl. 1909. Das Wesen der lautlichen Dissimilation. Abhandlungen der Königlich Sächsischen Gesellschaft der Wissenschaften 57. 139 –178. Bye, Patrik. Forthcoming. Derivations. In Nancy C. Kula, Bert Botma & Kuniya Nasukawa (eds.) The Continuum handbook of phonology. London: Continuum. Bye, Patrik, Elin Sagulin & Ida Toivonen. 2009. Phonetic duration, phonological quantity and prosodic structure in Inari Saami. Phonetica 66. 199–221. Card, Elizabeth. 1983. A phonetic and phonological study of Arabic emphasis. Ph.D. dissertation, Cornell University. Choi, John D. 1992. Phonetic underspecification and target interpolation: An acoustic study of Marshallese vowel allophony. Ph.D. dissertation, University of California, Los Angeles. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row.

Dissimilation

20

Clements, G. N. 1991. Place of articulation in consonants and vowels: A unified theory. Working Papers of the Cornell Phonetics Laboratory 5. 77–123. Clements, G. N. & Elizabeth Hume. 1995. The internal organization of speech sounds. In Goldsmith (1995), 245 –306. Cohn, Abigail C. 1990. Phonetic and phonological rules of nasalization. Ph.D. dissertation, University of California, Los Angeles. Cohn, Abigail C. 1992. The consequences of dissimilation in Sundanese. Phonology 9. 199–220. Crosswhite, Katherine. 1999. Vowel reduction in Optimality Theory. Ph.D. dissertation, University of California, Los Angeles. Czaykowska-Higgins, Ewa. 1993. Cyclicity and stress in Moses-Columbia Salish (Nxa’amxcin). Natural Language and Linguistic Theory 11. 197–278. Daly, John P. 1993. The role of tone sandhi in tone analysis. Notes on Linguistics 61. 5 –24. Dart, Sarah N. 1991. Articulatory and acoustic properties of apical and laminal articulation. Ph.D. dissertation, University of California, Los Angeles. Dave, Radhekant. 1977. Retroflex and dental consonants in Gujarati: A palatographic and acoustic study. Annual Report of the Institute of Phonetics, University of Copenhagen 11. 27–156. Davy, J. I. M. & Derek Nurse. 1982. The synchronic forms of Dahl’s Law, a dissimilation process, in Central and Lacustrine Bantu languages of Kenya. Journal of African Languages and Linguistics 4. 159–195. de Lacy, Paul (ed.) 2007. The Cambridge handbook of phonology. Cambridge: Cambridge University Press. Dixon, R. M. W. 1977. A grammar of YidiJ. Cambridge: Cambridge University Press. Doke, Clement M. 1926. The phonetics of the Zulu language. Johannesburg: University of the Witwatersrand Press. Dougherty, Brian. 1993. The acoustic-phonetic correlates of Cayuga word stress. Ph.D. dissertation, Harvard University. Dyen, Isidore. 1972. Non-gradual regular phonetic changes involving sibilants. In Jacques Barrau, Lucien Bernot, George Condominas, Mariel Jean Brunhes Delamare, Francis Leroy, Alexic Rygaloff & Jacqueline M. C. Thomas (eds.) Langues et techniques natures et société, vol. 1: Approche linguistique, 95 –99. Paris: Klincksieck. Ebert, Karen. 1974. Partial vowel harmony in Kera. Studies in African Linguistics. Supplement 5. 75–80. Elmedlaoui, Mohammed. 1985. Le parler berbère chleuh d’Imdlawn: Segments et syllabation. Doctorat de troisième cycle, Université de Paris VIII, Saint-Denis. Elmedlaoui, Mohammed. 1995. Aspect des representations phonologiques dans certaines langues chamitosemitiques. Rabat: Faculté des lettres et des sciences humaines. Evans, Nicholas D. 1995. Current issues in the phonology of Australian languages. In Goldsmith (1995), 723 –761. Fallon, Paul D. 1993. Liquid dissimilation in Georgian. Proceedings of the 10th Eastern States Conference on Linguistics (ESCOL). 105 –116. Finer, Daniel. 1986. Reduplication and verbal morphology in Palauan. The Linguistic Review 6. 99 –130. Foley, William A. 1991. The Yimas language of New Guinea. Stanford: Stanford University Press. Frisch, Stefan A. 2004. Language processing and segmental OCP effects. In Hayes et al. (2004), 346–371. Frisch, Stefan A., Janet B. Pierrehumbert & Michael B. Broe. 2004. Similarity avoidance and the OCP. Natural Language and Linguistic Theory 22. 179 –228. Geytenbeek, Brian B. & Helen Geytenbeek. 1971. Gidabal grammar and dictionary. Canberra: Australian Institute of Aboriginal Studies. Ghazali, Salem. 1977. Back consonants and backing coarticulation in Arabic. Ph.D. dissertation, University of Texas, Austin.

21

Patrik Bye

Goldsmith, John A. 1976. Autosegmental phonology. Ph.D. dissertation, MIT. Goldsmith, John A. 1984. Meeussen’s Rule. In Aronoff & Oehrle (1984), 245–259. Goldsmith, John A. (ed.) 1995. The handbook of phonological theory. Cambridge, MA & Oxford: Blackwell. González, Carolina. 2008. Assimilation and dissimilation of syllable-final /k/ in North-Central Spanish. In Joyce Bruhn de Garavito & Elena Valenzuela (eds.) Selected Proceedings of the 10th Hispanic Linguistics Symposium, 170 –183. Somerville, MA: Cascadilla Press. Goodman, Beverley. 1995. Features in Ponapean phonology. Ph.D. dissertation, Cornell University. Gragg, Gene. 1976. Oromo of Wellegga. In M. Lionel Bender (ed.) The non-Semitic languages of Ethiopia, 166–195. East Lansing: Michigan State University Press. Grammont, Maurice. 1895. La dissimilation consonantique dans les langues indo-européennes et dans les langues romanes. Dijon: Darantière. Grassmann, Hermann. 1863. Ueber die Aspiraten und ihr gleichzeitiges Vorhandensein im An- und Auslaute der Wurzeln. Zeitschrift für Vergleichende Sprachforschung auf dem Gebiete des Deutschen, Griechischen und Lateinischen 12. 81–138. Greenberg, Joseph. 1950. The patterning of root morphemes in Semitic. Word 6. 162–181. Gussenhoven, Carlos & Haike Jacobs. 2005. Understanding phonology. London: Hodder Arnold. Hall, Nancy. 2009. R-dissimilation in American English. Unpublished ms., California State University, Long Beach. Harris, K. S. 1958. Cues for the discrimination of American English fricatives in spoken syllables. Language and Speech 1. 1–7. Hawkins, Sarah & Andrew Slater. 1994. Spread of CV and V-to-V coarticulation in British English: Implications for the intelligibility of synthetic speech. In Proceedings of the 3rd International Conference on Spoken Language Processing, 57–60. Yokohama: Acoustical Society of Japan. Hawkins, Sarah & R. Smith. 2001. An acoustical study of long-domain /r/ and /l/ coarticulation. In Sebastian Heid & Sarah Hawkins (eds.) Proceedings of the 5th Seminar on Speech Production: Models and Data, 77–80. Bavaria: Kloster Seeon. Hayes, Bruce, Robert Kirchner & Donca Steriade (eds.) 2004. Phonetically based phonology. Cambridge: Cambridge University Press. Holton, David. 1995. Assimilation and dissimilation of Sundanese liquids. In Beckman et al. (1995), 167–180. Hume, Elizabeth. 1992. Front vowels, coronal consonants and their interaction in nonlinear phonology. Ph.D. dissertation, Cornell University. Hume, Elizabeth & David Odden. 1996. Reconsidering [consonantal]. Phonology 13. 345–376. Hume, Elizabeth & Misun Seo. 2004. Metathesis in Faroese and Lithuanian: From speech perception to Optimality Theory. Nordic Journal of Linguistics 27. 35 –60. Idsardi, William J. & Eric Raimy. 2008. Reduplicative economy. In Bert Vaux & Andrew Nevins (eds.) Rules, constraints, and phonological phenomena, 149–184. Oxford: Oxford University Press. Ishihara, Masahide. 1991. A lexical prosodic phonology of Japanese verbs. Ph.D. dissertation, University of Arizona. Itkonen, Erkki. 1986 –91. Inarilappisches Wörterbuch. 4 vols. Helsinki: Suomalais-ugrilainen Seura. Itô, Junko. 1984. Melodic dissimilation in Ainu. Linguistic Inquiry 15. 505 –513. Itô, Junko & Armin Mester. 1986. The phonology of voicing in Japanese: Theoretical consequences for morphological accessibility. Linguistic Inquiry 17. 49 –73. Itô, Junko & Armin Mester. 2003. Japanese morphophonemics: Markedness and word structure. Cambridge, MA: MIT Press. Jebbour, Abdelkrim. 1985. La labio-vélarization en berbère dialecte tachelhit (parler de tiznit). In Mémoire de phonologie. Rabat: Université Mohammed V, Faculté des letters et des sciences humaine.

Dissimilation

22

Johnson, Lawrence. 1973. Dissimilation as a natural process in phonology. Stanford Occasional Papers in Linguistics 3. 45 –56. Joseph, Brian D. & Richard Janda (eds.) 2003. The handbook of historical linguistics. Malden, MA & Oxford: Blackwell. Josephs, Lewis S. 1975. Palauan reference grammar. Honolulu: University of Hawai’i Press. Josephs, Lewis S. 1990. New Palauan–English dictionary. Honolulu: University of Hawai’i Press. Kaisse, Ellen M. 1988. Modern Greek continuant dissimilation and the OCP. Unpublished ms., University of Washington. Kaye, Jonathan & Jean Lowenstamm. 1985. A metrical treatment of Grassmann’s Law. Papers from the Annual Meeting of the North East Linguistic Society 15. 220 –233. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud. 1985. The internal structure of phonological elements: A theory of charm and government. Phonology Yearbook 2. 305–328. Keating, Patricia. 1988. Underspecification in phonetics. Phonology 5. 275 –292. Kelly, John. 1989. On the phonological relevance of some non-phonological elements. In Tamás Szende (ed.) Proceedings of the Speech Research ’89 International Conference, 56–59. Budapest: Linguistic Institute of the Hungarian Academy of Sciences. Kelly, John & John Local. 1986. Long-domain resonance patterns in English. In Proceedings of the International Conference on Speech Input/Output: Techniques and Applications, 304–309. London: Institution of Electrical Engineers. Kenstowicz, Michael & Charles W. Kisseberth. 1977. Topics in phonological theory. New York: Academic Press. Kenstowicz, Michael & Charles W. Kisseberth. 1979. Generative phonology: Description and theory. New York: Academic Press. Kent, Ronald G. 1936. Assimilation and dissimilation. Language 12. 245 –258. Kent, Ronald G. 1945. The sounds of Latin: A descriptive and historical phonology. Baltimore: Waverley Press. Keyser, Samuel J. & Paul Kiparsky. 1984. Syllable structure in Finnish phonology. In Aronoff & Oehrle (1984), 7–31. Kiparsky, Paul. 2003. The phonological basis of sound change. In Joseph & Janda (2003), 313–342. Kisseberth, Charles W. 1970. On the functional utility of phonological rules. Linguistic Inquiry 1. 291–306. Koskinen, K. 1964. Kompatibilität in den dreikonsonantigen hebräischen Wurzeln. Zeitschrift der Deutschen Morgenländischen Gesellschaft 144. 16 –58. Krause, Scott. 1979. Topics in Chukchee phonology and morphology. Ph.D. dissertation, University of Illinois, Urbana-Champaign. Ladefoged, Peter. 1993. A course in phonetics. 3rd edn. New York: Harcourt Brace Jovanovich. Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’s languages. Oxford & Malden, MA: Blackwell. Ladefoged, Peter, Ian Maddieson & Michel T. T. Jackson. 1988. Investigating phonation types in different languages. In Osamu Fujimura (ed.) Vocal physiology: Voice production, mechanisms and functions, 292–317. New York: Raven. Langendoen, D. Terence. 1966. A restriction on Grassmann’s Law in Greek. Language 42. 7– 9. Larsi, Ahmed. 1991. Aspects de la phonologie non-lindaire du parler berbère chleuh de Tidli. Thèse de doctorat, Université de la Sorbonne Nouvelle, Paris III. Leben, William R. 1973. Suprasegmental phonology. Ph.D. dissertation, MIT. Leumann, Manu. 1977. Lateinische Grammatik von Leumann-Hofmann-Szantyr, vol. 1: Lateinische Laut- und Pormenlehre. Munich: Beck. Levergood, Barbara. 1987. Topics in Arusa phonology and morphology. Ph.D. dissertation, University of Texas, Austin.

23

Patrik Bye

Levi, Susannah V. 2008. Phonemic vs. derived glides. Lingua 118. 1956–1978. Lidz, Jeffrey. 2001. Echo reduplication in Kannada and the theory of word-formation. The Linguistic Review 18. 375 –394. Lloret, Maria-Rosa. 1988. Gemination and vowel length in Oromo morphophonology. Ph.D. dissertation, Indiana University. Lombardi, Linda. 1991. Laryngeal features and laryngeal neutralization. Ph.D. dissertation, University of Massachusetts, Amherst. Lombardi, Linda. 1995. Dahl’s law and privative [voice]. Linguistic Inquiry 26. 365 –372. Lubker, J. & T. Gay. 1982. Anticipatory labial coarticulation: Experimental, biological and linguistic variables. Journal of the Acoustical Society of America 71. 437 –448. Lynch, John. 2003. Low vowel dissimilation in Vanuatu languages. Oceanic Linguistics 42. 359–406. MacEachern, Margaret R. 1997. Laryngeal co-occurrence restrictions. Ph.D. dissertation, University of California, Los Angeles. Published 1999, New York: Garland. Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press. Magen, Harriet S. 1984. Vowel-to-vowel coarticulation in English and Japanese. Journal of the Acoustical Society of America 75. Supplement S41. Manuel, Sharon Y. 1987. Acoustic and perceptual consequences of vowel-to-vowel coarticulation in three Bantu languages. Ph.D. dissertation, Yale University. Marlett, Stephen A. & Joseph Paul Stemberger. 1983. Empty consonants in Seri. Linguistic Inquiry 14. 617–639. McCarthy, John J. 1979. Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT. McCarthy, John J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry 12. 373 –418. McCarthy, John J. 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 17. 207–263. McCarthy, John J. 1988. Feature geometry and dependency: A review. Phonetica 45. 84–108. McCarthy, John J. 2007. Derivations and levels of representation. In de Lacy (2007), 99–117. McCarthy, John J. & Alan Prince. 1993. Prosodic morphology I: Constraint interaction and satisfaction. Unpublished ms., University of Massachusetts, Amherst & Rutgers University. McCarthy, John J. & Alan Prince. 1995. Faithfulness and reduplicative identity. In Beckman et al. (1995), 249–384. McConvell, Patrick. 1988. Nasal cluster dissimilation and constraints on phonological variables in Gurindji and related languages. In Nicholas D. Evans & Steve Johnson (eds.) Aboriginal Linguistics 1, 135–165. Armidale: Department of Linguistics, University of New England. McGregor, William. 1990. A functional grammar of Gooniyandi. Amsterdam: John Benjamins. McKay, D. G. 1970. Phoneme repetition in the structure of languages. Language and Speech 13. 199 –213. Menn, Lise & Brian MacWhinney. 1984. The repeated morph constraint: Toward an explanation. Language 60. 519 –541. Mester, Armin. 1986. Studies in tier structure. Ph.D. dissertation, University of Massachusetts, Amherst. Morén, Bruce. 2003. The parallel structures model of feature geometry. Working Papers of the Cornell Phonetics Laboratory 15. 194 –270. Nevins, Andrew. 2005. Overwriting does not optimize in nonconcatenative morphology. Linguistic Inquiry 36. 275 –287. Nevins, Andrew & Bert Vaux. 2003. Metalinguistic, shmetalinguistic: The phonology and morphology of shm- reduplication. Papers from the Annual Regional Meeting, Chicago Linguistic Society 39. 702–722.

Dissimilation

24

Newton, Brian. 1971. Sibilant loss in Northern Greek. Canadian Journal of Linguistics 17. 1–15. Odden, David. 1987. Dissimilation as deletion in Chuckchi. Proceedings of the 3rd Eastern States Conference on Linguistics (ESCOL). 235 –246. Odden, David. 1988. Anti antigemination and the OCP. Linguistic Inquiry 19. 451–475. Odden, David. 1994. Adjacency parameters in phonology. Language 70. 289 –330. Ohala, John J. 1981. The listener as a source of sound change. Papers from the Annual Regional Meeting, Chicago Linguistic Society 17. 178 –203. Ohala, John J. 1993. The phonetics of sound change. In Charles Jones (ed.) Historical linguistics: Problems and perspectives, 237–278. London: Longman. Ohala, John J. 2003. Phonetics and historical phonology. In Joseph & Janda (2003), 669–686. Öhman, Sven E. G. 1966. Coarticulation in VCV utterances: Spectrographic measurements. Journal of the Acoustical Society of America 39. 151–168. Padgett, Jaye. 1991. Stricture in feature geometry. Ph.D. dissertation, University of Massachusetts, Amherst. Paradis, Carole & Jean-François Prunet. 1989. On coronal transparency. Phonology 6. 317–348. Parker, Steve 1997. An OT account of laryngealization in Cuzco Quechua. Work Papers of the Summer Institute of Linguistics, University of North Dakota 41. 47–57. Pater, Joe. 1999. Austronesian nasal substitution and other NY effects. In René Kager, Harry van der Hulst & Wim Zonneveld (eds.) The prosody–morphology interface, 310–343. Cambridge: Cambridge University Press. Pearce, Mary. 2008. Vowel harmony domains and vowel undershoot. UCL Working Papers in Linguistics 20. 115 –140. Phelps, Elaine. 1975. Sanskrit diaspirates. Linguistic Inquiry 6. 447 –464. Phelps, Elaine & Michael Brame. 1973. On local ordering of rules in Sanskrit. Linguistic Inquiry 4. 387 –400. Pierrehumbert, Janet B. 1993. Dissimilarity in the Arabic verbal roots. Papers from the Annual Meeting of the North East Linguistic Society 23. 367–381. Pitkin, Harvey. 1984. Wintu grammar. Berkeley: University of California Press. Plénat, Marc. 1996. De l’interaction des contraints: Une étude de case. In Jacques Durand & Bernard Laks (eds.) Current trends in phonology: Models and methods, 585–617. Salford: ESRI. Poser, William J. 1982. Phonological representations and action-at-a-distance. In Harry van der Hulst & Norval Smith (eds.) The structure of phonological representations, part II, 121–158. Dordrecht: Foris. Posner, Rebecca R. 1961. Consonantal dissimilation in the Romance languages. Publications of the Philological Society 18. Oxford: Blackwell. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Purnell, Herbert C. 1965. Phonology of a Yai dialect spoken in the province of Chiengrai, Thailand. Hartford, CT: Hartford Seminary Foundation. Quintero, Carolyn. 2004. Osage grammar. Lincoln, NE: University of Nebraska Press. Recasens, Daniel. 1984. Vowel-to-vowel coarticulation in Catalan VCV sequences. Journal of the Acoustical Society of America 76. 1624–1635. Recasens, Daniel. 1987. An acoustic analysis of V-to-C and V-to-V coarticulatory effects in Catalan and Spanish VCV sequences. Journal of Phonetics 15. 299 –312. Rehg, Kenneth L. & Damian G. Sohl. 1981. Ponapean reference grammar. Honolulu: University of Hawai’i Press. Rubach, Jerzy. 1993. The Lexical Phonology of Slovak. Oxford: Clarendon Press. Sag, Ivan. 1974. The Grassmann’s Law ordering pseudoparadox. Linguistic Inquiry 5. 591– 607.

25

Patrik Bye

Sag, Ivan. 1976. Pseudosolutions to the pseudoparadox: Sanskrit diaspirates revisited. Linguistic Inquiry 7. 609 –622. Schindler, Jochem. 1976. Diachronic and synchronic remarks on Bartholomae’s and Grassmann’s Laws. Linguistic Inquiry 7. 622–637. Shaw, Patricia A. 1976. Theoretical issues in Dakota phonology and morphology. Ph.D. dissertation, University of Toronto. Shaw, Patricia A. 1985. Modularisation and substantive constraints in Dakota Lexical Phonology. Phonology Yearbook 2. 173 –202. Shibatani, Masayoshi. 1990. The languages of Japan. Cambridge: Cambridge University Press. Sihler, Andrew L. 1995. New comparative grammar of Greek and Latin. Oxford: Oxford University Press. Singh, Amar Bahadur. 1969. On echo words in Hindi. Indian Linguistics 30. 185 –195. Slocum, Marianna C. 1948. Tzeltzal (Mayan) noun and verb morphology. International Journal of American Linguistics 14. 77–86. Smolensky, Paul. 1995. On the structure of the constraint component Con of UG (ROA-86). Smolensky, Paul. 1997. Constraint Conjunction II. Handout from paper presented at the Hopkins Optimality Theory Workshop/University of Maryland Mayfest 1997. Smyth, Herbert Weir. 1920. Greek grammar. Cambridge, MA: Harvard University Press. Revised 1956 by Gordon M. Messing. Soden, Wolfram von. 1969. Grundriß der akkadischen Grammatik. Rome: Pontificum Institutum Biblicum. Sohn, Ho-min. 1975. Woleaian reference grammar. Honolulu: University of Hawai’i Press. Sohn, Ho-min & Anthony P. Tawerilmang. 1976. Woleaian–English dictionary. Honolulu: University of Hawai’i Press. Steriade, Donca. 1987. Redundant values. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23(2). 339–362. Steriade, Donca. 1995. Underspecification and markedness. In Goldsmith (1995), 114–174. Stevens, Kenneth N. & Sheila Blumstein. 1975. Quantal aspects of consonant production and perception: A study of retroflex consonants. Journal of Phonetics 3. 215 –233. Suzuki, Keiichiro. 1998. A typological investigation of dissimilation. Ph.D. dissertation, University of Arizona. Szakos, József. 1994. Die Sprache der Cou: Untersuchungen zur Synchronie einer Austronesischen Sprache auf Taiwan. Ph.D. dissertation, University of Bonn. Thomas, Jacqueline. 1963. Le parler ngbaka de Bokanga. The Hague: Mouton. Trefry, David. 1969. A comparative study of Kuman and Pawaian: Non-Austronesian languages of New Guinea. (Pacific Linguistics B13.) Canberra: Australian National University. Tunley, Alison. 1999. Coarticulatory influences of liquids on vowels in English. Ph.D. dissertation, University of Cambridge. Uhlenbeck, Eugenius M. 1949. De structuur van het javaanse morpheem. Bandoeng: Nix. Vaux, Bert. 1998. The laryngeal specifications of fricatives. Linguistic Inquiry 29. 497–511. Walsh Dickey, Laura. 1997. The phonology of liquids. Ph.D. dissertation, University of Massachusetts, Amherst. Waterhouse, Viola. 1949. Learning a second language first. International Journal of American Linguistics 15. 106 –109. Waterhouse, Viola. 1962. The grammatical structure of Oaxaca Chontal. Bloomington: Indiana University Research Center in Anthropology. Welmers, William E. 1946. A descriptive grammar of Fanti. Baltimore: Linguistic Society of America. West, Paula. 1999a. The extent of coarticulation in English liquids: An acoustic and articulatory study. In John J. Ohala, Yoko Hasegawa, Manjari Ohala, Daniel Granville & Ashlee Bailey (eds.) Proceedings of the 14th International Congress of the Phonetic Sciences, 1901–1904. Berkeley: Department of Linguistics, University of California.

Dissimilation

26

West, Paula. 1999b. Perception of distributed coarticulatory properties of English /l/ and /P/. Journal of Phonetics 27. 405 –426. West, Paula. 2000. Long-distance coarticulatory effects of English /l/ and /r/. Ph.D. dissertation, University of Oxford. Wolf, Matthew. 2007. For an autosegmental theory of mutation. In Leah Bateman, Michael O’Keefe, Ehren Reilly & Adam Werle (eds.) Papers in Optimality Theory III, 315–404. Amherst: GLSA. Woodhouse, Robert. 1998. Verner’s and Thurneysen’s Laws in Gothic as evidence for obstruent development in Early Germanic. Beiträge zur Geschichte der deutschen Sprache und Literatur 120. 194–222. Wordick, Frank. 1982. The Yindjibarndi language. (Pacific Linguistics C71.) Canberra: Australian National University. Yip, Moira. 1982. Reduplication and CV skeleta in Chinese secret languages. Linguistic Inquiry 13. 637–661. Yip, Moira. 1988. The Obligatory Contour Principle and phonological rules: A loss of identity. Linguistic Inquiry 19. 65 –100. Yip, Moira. 1989. Feature geometry and cooccurrence restrictions. Phonology 6. 349 –374. Yip, Moira. 1995. Repetition and its avoidance: The case of Javanese. In Keiichiro Suzuki & Dirk Elzinga (eds.) Proceedings of the 1995 Southwestern Workshop on Optimality Theory (SWOT). Coyote Papers: Working Papers in Linguistics. Conference Proceedings vol. 5, 238–262. Tucson: University of Arizona Linguistics Circle. Yip, Moira. 1998. Identity avoidance in phonology and morphology. In Steven Lapointe, Diane Brentari & Patrick Farrell (eds.) Morphology and its relation to phonology and syntax, 216 –246. Stanford: CSLI.

61

Hiatus Resolution Roderic F. Casali

1

Overview of hiatus resolution

The term vowel hiatus is commonly used to refer to a sequence of adjacent vowels belonging to separate syllables, as in the following Hawaiian examples from Senturia (1998: 26). (Periods indicate syllable boundaries.) (1)

[ko.a.na] [li.le.a] [ku.a] [hu.e.lo] [hu.i.na] [ko.e.na]

‘space’ (name of a shell) ‘back’ ‘tail’ ‘sum’ ‘remainder’

In some languages, vowel hiatus is permitted quite freely. Other languages place much stricter limits on the contexts in which heterosyllabic vowel sequences can occur, while some disallow them entirely. Languages that do not permit vowel hiatus may employ any of several processes that eliminate it in cases where it would otherwise arise (e.g. where an underlying vowel-final morpheme directly precedes a vowel-initial morpheme). One of the most common forms of hiatus resolution involves the elision of one of the two vowels. (See chapter 68: deletion.) Vowel elision is illustrated below with examples from Yoruba, adapted from Pulleyblank (1988). (2)

/bu ata/ /gé olú/ /ta epo/ /kF èkF/ /7a DwG/

→ → → → →

[ba.ta] [gó.lú] [te.po] [ké.kE] [7D.wG]

‘pour ground pepper’ ‘cut mushrooms’ ‘sell palm oil’ ‘learn’ ‘buy a broom’

In all of these examples, it is the first of the two adjacent vowels (V1) that deletes. Though this is the more common pattern cross-linguistically, cases in which the second vowel (V2) deletes are also attested (and indeed, some instances of V2 deletion are found in Yoruba itself). The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0061

Hiatus Resolution

2

In another very common hiatus resolution process, glide formation, V1 is converted to a semivowel (see also chapter 15: glides). One well-known case, illustrated in (3), is Ganda (Tucker 1962; Katamba 1985; Clements 1986).1 (3)

/mu-iko/ /li-ato/ /mu-ezi/ /mu-ogezi/ /mi-ezi/ /mu-ana/

→ → → → → →

[mwi(.ko] [lja(.to] [mwe(.zi] [mwo(.ge.zi] [mje(.zi] [mwa(.na]

‘trowel’ ‘boats’ ‘moon’ ‘talker’ ‘moons’ ‘child’

cf. [mu-le(nzi] ‘boy’ cf. [li-ggwa] ‘thorn’

In general (we will look at an exception in §2.4 below), glide formation in Ganda applies only where V1 is high. Non-high V1’s are elided before another vowel, with compensatory lengthening of V2 (e.g. /ka-oto/ ‘small fireplace’ > [ko(to]). A third common pattern, coalescence, involves the merger of V1 and V2 to form a third vowel that combines features of both. This is illustrated in the Attic Greek examples below (de Haas 1988: 126). In these examples, various underlying sequences that combine a non-high [−ATR] vowel /a e D/ with a mid [+ATR] vowel /e o/ are realized phonetically as a long mid [−ATR] vowel that retains the backness and roundness of the original [+ATR] vowel. (4)

/gene-a/ /ti(ma-omen/ /ajdo-a/ /de(lo-e(te/ /zde(-omen/

→ → → → →

[gé.ne(] [ti(.m∑(.men] [aj.d∑(] [de.l∑(.te] [zd∑(.men]

(/ea/ > [e(]) (/ao/ > [D(]) (/oa/ > [D(]) (/oe(/ > [D(]) (/e(o/ > [D(])

‘race (nom acc pl)’ ‘honor (1pl pres ind)’ ‘shame (acc sg)’ ‘manifest (2pl pres subj)’ ‘live (1pl pres subj)’

Note that for the pairs /a o/ and /e( o/, coalescence in Attic Greek is symmetric; the phonetic result is the same for both orders of input vowels.2 Other languages with symmetric coalescence include Quebec French, Korean, Rotuman, Old Portuguese, and Classical Sanskrit (all discussed in de Haas 1988), and Afar (Bliese 1981). Symmetric coalescence is relatively uncommon, however. Much more frequently, coalescence applies only when the vowels occur in one of the two possible orders (see §2.3 below). Other languages avoid hiatus by retaining both vowels but syllabifying them into the nucleus of a single syllable, a process generally known as diphthong formation or diphthongization. This occurs in Ngiti, as illustrated in the following examples, adapted from Kutsch Lojenga (1994: 90–91). (5)

1

/abvo àji/ /tÃtG akpà/ /opi àji/ /ÃnÕÈÃ akpà/ /fà FJÈ/ /fÈkË o+i/

→ → → → → →

[a.bvoà.ji] [tÃ.tGa.kpà] [o.pià.ji] [Ãn.ÕÈÃa.kpà] [fàF.JÈ] [fÈ.kËó.+i]

‘widow’ ‘liar’ ‘Lendu woman’ ‘male goat’ ‘our food’ ‘your (pl) knives’

It is also common to find cases in which the second of two vowels becomes non-syllabic, e.g. /gene-i/ ‘race (dat sg)’ > [.ge.nej.] in Attic Greek (de Haas 1988: 126). Generally, such cases are potentially analyzable as diphthong formation. See Senturia (1998: 12–15) for examples and related discussion. 2 This is not the case in Attic for the pair /a e/: /a+e/ yields [e(], while /e+a/ yields [ea].

Roderic F. Casali

3

Kutsch Lojenga states that “both vowels must be realised as a short complex vowel nucleus on one V timing slot.” She further notes (personal communication) that the first vowel in each sequence is shorter in duration than the second vowel, though not to the point where any auditory distinctions among vowels in V1 position are neutralized. This argues against an analysis (i.e. glide formation) in which V1 is syllabified as a consonantal onset. Other languages that exhibit diphthong formation include Haitian Creole (Picard 2003), Indonesian (Rosenthall 1997), Attic Greek (Senturia 1998 and references therein), Obolo (Faraclas 1982), Bakossi (Hedinger and Hedinger 1977), Eastern Ojibwa (Howard 1973), Margi (Tranel 1992), and Larike (Rosenthall 1997). Finally, an obvious means of eliminating vowel hiatus is to epenthesize a consonant between the two vowels. One language in which this occurs is Washo, as illustrated in the examples below, adapted from Midtlyng (2005); in each example a semivowel [j] is inserted between the initial vowel of a suffix and the final vowel of a preceding morpheme. (6)

a.

’la(du-a → [’la(.du.ja] ‘in my hand’ ‘my hand-loc’ b. le’gu?u-i? → [le.’gu.?u.ji] ‘my daughter’s child’ ‘(1sg obj) mother’s mother-attrib-ag’ c. ’lemts’iha-i → [’lém.ts’i.ha.ji] ‘I am waking him up’ ‘I cause to awake-imp’ d. ’lemlu-’e(s-i → [’lém.lu.’je(.si] ‘I am not eating’ ‘I eat-neg-imp’

Though these hiatus resolution strategies have been presented independently using data from different languages, it is common to find two or more different strategies at work in the same language (see §2.5). It is also common to find that languages tolerate hiatus in some contexts but not others. A number of factors are capable of blocking or influencing hiatus resolution, including the nature of the prosodic or morphosyntactic boundary at which hiatus arises (Kaisse 1977; Baltazani 2006), prominence factors such as stress (Senturia 1998), vowel length and tone (Casali 1998: 73), minimal word length or weight conditions, the lexical or functional status of particular morphemes, rate of speech, and sensitivity to particular lexical items. Hiatus resolution also sometimes shows derived environment effects (see chapter 88: derived environment effects), in which hiatus is tolerated in vowel sequences internal to a morpheme, but is eliminated in cases where two vowels come together across a morpheme boundary. Finally, morphemes consisting of just a single vowel are sometimes resistant to loss through elision, presumably due to the loss of semantic content that could result (Casali 1997). Hiatus resolution can also arise in cases where three (or more) underlying vowels occur in sequence. Such cases are considerably less common, and it is difficult to make many strong generalizations about the resolution of /V1V2V3/ sequences. Attested outcomes include gliding of V2 (e.g. Eastern Ojibwa [Howard 1973]; Ganda [Clements 1986: 75]), and elision of both V1 and V2 (Baka [Parker 1985]). The remainder of this paper is organized as follows. In §2, I describe some major respects in which hiatus resolution processes vary across languages. §3 discusses the treatment of hiatus resolution within various theoretical models and

Hiatus Resolution

4

some associated challenges and issues. The paper concludes with a brief summary in §4.

2

Typological variation

Hiatus resolution patterns show considerable variation across languages, and any survey of this variation in a work of the present paper’s scope will necessarily be selective.3 Here we will look at certain aspects of variation involving consonant epenthesis (§2.1), vowel elision (§2.2), coalescence (§2.3), and glide formation (§2.4), as well as the co-occurrence of multiple processes within a single language (§2.5).

2.1 Consonant epenthesis A question that naturally arises in looking at hiatus resolution by consonant epenthesis is which consonants can function epenthetically as hiatus interrupters. Three possibilities seem reasonably well attested: (i) (ii) (iii)

A semivowel, usually one that is homorganic with (i.e. shares the same frontness or roundness as) V1 or V2. A glottal stop ([?]) or fricative ([h]). A coronal consonant, generally [t] or a rhotic.

By far the most common pattern (Picard 2003; Uffmann 2007) is the first one. This is sometimes explained (see for example Uffmann 2007) by assuming that homorganic glide epenthesis is in some sense different from (and less costly than) epenthesis of an entirely new segment, since the glide might be interpreted as a prolongation of phonological content that is already present. However, there are also languages – e.g. Ait Seghrouchen Berber (Senturia 1998), Galacian (Picard 2003), and Washo (Midtlyng 2005) – that consistently epenthesize [j], regardless of the featural content of adjacent vowels, and at least one language, Chamicuro (Parker 1989; de Lacy 2006), with consistent [w]-epenthesis. An example of a language with glottal stop epenthesis is Malay (Ahmad 2001). The examples below show insertion of a glottal stop between a CV prefix and vowel-initial root: (7) /di-ubah/ /sH-indah/ /sH-elok/ /di-olah/ /di-aIkat/

→ → → → →

[di?ubah] [sH?indah] [sH?elo?] [di?olah] [di?aIkat]

‘to ‘to ‘to ‘to ‘to

change (pass)’ be as beautiful as’ be as pretty as’ beguile (pass)’ lift (pass)’

Other languages that epenthesize [?] in at least some hiatus contexts include Ilokano, Selayarese, Tunica, and Indonesian (see Lombardi 2002 and references therein).

3

One topic that is not treated, for reasons of space, is the typology of diphthong formation. See Schane (1987), Sohn (1987), Rosenthall (1994), and Senturia (1998) for some discussion.

5

Roderic F. Casali

A well-known case of epenthesis of a coronal consonant in hiatus contexts is Axininca Campa (Payne 1981; Lombardi 2002; Bakovio 2003), illustrated in the examples below (Payne 1981): (8)

/i-N-koma-i/ /i-N-koma-aa-i/

→ →

[iIkomati] [iIkomataati]

‘he will paddle’ ‘he will paddle again’

These examples show an epenthetic [t] interrupting vowel hiatus in suffixal contexts; hiatus in prefixal contexts is resolved in Axininca Campa by eliding one of the vowels instead. The problem of predicting the range of possible epenthetic consonants has received significant attention in recent theoretical work. This is discussed further in §3.3.3 below.

2.2 Vowel elision A natural question that arises in connection with vowel elision is which of two adjacent vowels elides. Cross-linguistically, elision of V1 is far more common than elision of V2 (Bergman 1968; Lamontagne and Rosenthall 1996; Casali 1997, 1998). Interestingly, it turns out that the contexts in which V2 elision is well attested are not random. Clear cases of V2 elision are largely confined to two contexts: (i) the boundary between a lexical (content) word and a following function word, and (ii) stem–suffix boundaries.4 Examples of the former type, from Etsako (Elimelech 1976), are shown in (9). Note that the latter also display V1 elision of the final vowel of a preceding function word, suggesting rather strongly that it is lexical or non-lexical status, and not simple linear order, that is relevant in this case (see also chapter 104: root–affix asymmetries). (9)

/Dna the /Dna the

aru louse e:i tortoise

Dli/ that Dna/ this

→

[Dnaruli]

‘that louse’

→

[Dne:ina]

‘this tortoise’

Examples of the latter type, adapted from Okpe (Pulleyblank 1986), are shown in (10). (10)

/è-sé-ó/ → inf-fall-inf /è-dé-F/ → inf-buy-inf

[èsé]

‘to fall’

[èdé]

‘to buy’

Compare these forms with the additional Okpe words in (11), where the final V suffix is retained following an underlying high vowel, which undergoes glide formation. 4

In addition to the more common cases in which the elided vowel is one that occupies a particular position, there are also cases (see Casali 1996, 1998; Causley 1999b) in which the vowel targeted depends on the featural makeup of the two vowels.

Hiatus Resolution (11)

/è-tí-ó/ inf-pull-inf /è-sË-F/ inf-sing-inf

→

[ètjó]

‘to pull’

→

[èswF]

‘to sing’

6

At other kinds of morphosyntactic boundaries, such as that between a prefix and following root or between two content words, elision regularly targets V1. The cross-linguistically well-attested possibilities are summarized below. (See Casali 1997 for more discussion.) (12)

Vowel elision and morphosyntactic position Context Between two content words Content word before function word Prefix + root Root + suffix

Robustly attested possibilities V1 elision V1 elision or V2 elision V1 elision V1 elision or V2 elision

2.3 Coalescence As noted previously, symmetric coalescence, as in the Attic Greek data in (4), is relatively rare. By far the most common form of coalescence is a directionally asymmetric pattern, termed height coalescence in Casali (1998) (see also Lamontagne and Rosenthall 1996; Parkinson 1996), in which a non-high V1 and a high V2 coalesce to form a non-high vowel otherwise identical to V2 , e.g. /a+i/ > [e], /a+u/ > [o], as in the Xhosa examples below (Aoki 1974). (13)

/wa-inkosi/ /wa-umfazi/

→ →

[wenkosi] ‘of the chiefs’ [womfazi] ‘of the woman’

The reverse sequences /i+a/ and /u+a/ are not subject to coalescence in Xhosa, but are resolved instead by vowel elision and glide formation, respectively (see §2.5 below). Languages in which the feature [ATR] is contrastive sometimes show a slightly more elaborate form of asymmetric height coalescence, in which the [ATR] value of a non-high V1 is preserved in some cases as well. Such languages divide into two types: those in which [−ATR] is systematically preserved (e.g. /a+i/ > [e], /e+o/ > [D]), and those in which [+ATR] is preserved (e.g. /a+i/ > [e], /o+// > [e]). Languages of the former type include Owon Afa (Awobuluyi 1972) and Anufo (Adjekum et al. 1993). Languages of the latter type include several North Guang languages and Southern Sotho (Casali 1998, 2003 and references therein).5 Though asymmetric height coalescence most commonly applies to sequences in which V1 is lower than V2 , cases of “reverse height coalescence” also exist in which a higher V1 followed by a lower V2 yields a lowered version of V1 (e.g. /i+a/ > [e], /u+a/ > [o]), while the opposite sequences do not trigger coalescence. This occurs in Foodo (Kwa; Ghana; Plunkett 1991: 68), as shown below. (The initial 5

The particular [ATR] value that is preserved under height coalescence shows a strong correlation with a language’s vowel inventory structure; see Casali (1998, 2003) and Causley (1999a) for discussion.

Roderic F. Casali

7

and final /a/’s are noun class affixes. The tonal changes are due to independent processes discussed in Plunkett.) (14)

a. /i+a/ > [e(] /á-bì-á/ → [ábê(] ‘seeds’ b. /Á+a/ > [D(] /á-sË-á/ → [àsF(] ‘ears’ c. /u+a/ > [o(] /á-jù-á/ → [ájô(] ‘millet’

cf. [dí-bí-lì] ‘seed’ cf. [kÈ-sË] ‘ear’ cf. [dú-jú-lì] ‘millet’

Sequences in which V2 is high and V1 is non-high do not undergo coalescence; compare the /Á+a/ sequence in (14b) with the /a+Á/ sequence in /kÈ-tá-Ë/ ‘bow’, which is retained in the surface form, [kÈtáË]. Other languages with reverse height coalescence patterns include Tem (Tchagbale 1976; de Craene 1986), Chagga (Nurse and Philippson 1977; Saloné 1980), Ewe (Westermann 1930), Bakossi (Hedinger and Hedinger 1977), and Nkengo (Hulstaert 1970). Interestingly, such patterns seem to occur predominantly at root–suffix boundaries, a restriction that partly parallels some limitations on the distribution of V2 elision (§2.2). A further coalescence pattern that should presumably be expected to occur is one in which front unrounded and back rounded vowels coalesce to form a front rounded vowel, e.g. /i+u/ > [y], /e+o/ > [ø], etc. Patterns of this type appear to be considerably less common than height coalescence. Two possible cases, Rotuman and Korean, are discussed in de Haas (1988) (see also Sohn 1987; Rice 1995; Causley 1999a). Coalescence of /e+o/ to [ø] is also described in Obolo (Faraclas 1982).

2.4 Glide formation In Ganda (cf. (3) above) and quite a few other languages, both front and back V1’s are subject to glide formation. It is also quite common, however, to find that only back round vowels glide and that front V1’s trigger a different resolution strategy, most commonly elision. This is the case for example in Xhosa (see §2.5 below) and Chumburung (Snider 1985). Though they are seemingly less common, there are also languages (e.g. Polish; Rubach 2000) in which only front vowels glide. A second point of variation involves the height of V1. Generally, if a language has glide formation at all, high V1’s will undergo the process (Rosenthall 1994, 1997; Casali 1995). In some languages (e.g. Ebira; Adive 1989), only high V1’s glide. In quite a large number of other languages, however, mid V1’s also glide.6 One such case, Chicano Spanish, is illustrated in the examples below (from Bakovio 2007, with phonemic forms substituted for orthographic ones): (15)

a.

b.

6

/mi ultima/ /mi ob7a/ /tu epoka/ /tu alma/ /me u7xe/ /po7ke a beses/ /komo eba/ /lo abla/

→ → → → → → → →

[mjultima] [mjoß7a] [twepoka] [twalma] [mju7xe] [po7kjaßeses] [komweßa] [lwaßla]

‘my last one (fem)’ ‘my deed’ ‘your time’ ‘your soul’ ‘it is urgent to me’ ‘because sometimes’ ‘like Eva’ ‘speaks it’

In rare cases, e.g. Aghem (Hyman 1979), languages may glide the low vowel /a/ as well.

Hiatus Resolution

8

Further variation exists as well. In some languages, glide formation does not apply to sequences in which V1 and V2 share the same frontness and roundness. In Gichode (Casali 1998: 168–169), for example, glide formation of a round vowel occurs only before non-round vowels, e.g. /u+i/ > [wi] but /Á+o/ > [o] (*[wo]). (Contrast this with realization of /u+o/ as [wo] in Ganda, as in (3) above.) Glide formation is also blocked in some languages (e.g. Ganda; Clements 1986) following certain consonants.7 Typically, both sorts of restrictions can be attributed to constraints that are effective quite generally in the language (e.g. languages that fail to glide /u/ or /o/ before a round vowel typically lack [Cw] before round vowels in general). Finally, some languages impose less stringent restrictions on glide formation when V1 occurs in absolute word-initial position. In Ganda, for example, only high V1’s generally glide in word-internal /CV1-V2/sequences. Word-initially, however, mid and even low V1’s undergo glide formation (in this case without compensatory lengthening), as in the examples below (Clements 1986: 75, n. 1):8 (16)

/o-a-gula/ /a-a-gula/ /e-a-laba/

→ → →

[wagula] [jagula] [jalaba]

‘you (sg) bought’ ‘he/she bought’ ‘it (cl. 9) saw’

Rather similar patterns are reported in Nyarwanda (Kimenyi 1979). Notwithstanding the considerable variation that exists in its patterning, there is one very significant respect in which the behavior of glide formation is surprisingly regular across languages. Quite consistently, (non-word-initial) sequences in which V1 and V2 are identical regularly fail to undergo glide formation. We can illustrate this restriction with additional examples from Ganda (Clements 1986): (17) /mi-iko/ /lu-uji/

→ →

[mi(ko] [lu(ji]

*[mji(ko] *[lwu(ji]

‘trowels’ ‘side’

Moreover, sequences such as /o+u/ and /e+i/, in which V1 and V2 are both front or both round and V1 is lower than V2 , rarely if ever trigger glide formation, but are resolved instead by vowel elision or coalescence (Casali 1995, 1998: 172, n. 5). Exceptions to these generalizations clearly arise in absolute word-initial position in some languages, as in the Ganda example in (16) above. I am not aware of any languages that consistently violate these restrictions word-internally, however.

2.5 Multiple hiatus resolution strategies in the same language It is quite common to find two or more different hiatus resolution processes at work in the same language. In some such cases, different processes are operative 7

In some languages, glide formation following certain consonants triggers further changes, e.g. /siV/ and /ziV/ are realized as [œV] and [ÚV] respectively in Ebira (Adive 1989). 8 Intervocalic gliding of non-high vowels also occurs in some three-vowel sequences discussed by Clements, as in /te-a-a-gula/ ‘he/she didn’t buy’, realized as [tejagula].

Roderic F. Casali

9

in different morphosyntactic contexts. In Axininca Campa (Bakovio 2003) and Washo (Midtlyng 2005), for example, hiatus is resolved by vowel elision at a prefix–stem boundary but by epenthesis at a stem–suffix boundary. In Lugisu (Brown 1970), a sequence /a+i/ is resolved by coalescence (to [e]) across a word boundary, but by eliding /a/ word-internally. There are also many cases, however, in which multiple strategies apply in exactly the same morphosyntactic context, targeting different vowel sequences. Especially common are cases (see Casali 1998: 83 –84) in which vowel elision occurs along with glide formation, coalescence, or both. Languages with both vowel elision and glide formation (but not coalescence) include Ganda, Etsako (Elimelech 1976), Igede (Bergman 1968), and Chicano Spanish (Bakovio 2007). Languages with coalescence and vowel elision (but not glide formation) include Afar (Bliese 1981) and Owon Afa (Awobuluyi 1972). Particularly intricate patterns are found in a considerable number of languages (32 cases are listed in Casali 1998: 83 –84) that manifest all three processes. One such language is Xhosa (McLaren 1955; Aoki 1974), whose hiatus resolution alternations conform to the following generalizations: (18)

Hiatus resolution in Xhosa a. b. c.

Where V1 is non-high and V2 is high, the outcome is a [−high] version of V2. A round V1 undergoes glide formation before a following non-round vowel.9 Elsewhere, V1 elision applies.

The overall pattern corresponding to these generalizations is shown below in Table 61.1, where coalescent realizations are underlined and those involving glide formation are italicized. Note that in the case of the input /o+i/, both coalescence and glide formation apply.

Table 61.1 Glide formation, coalescence, and vowel elision (in Xhosa) V2

V1

9

i e a o u

i i e e we wi

e e e e we we

a a a a wa wa

o o o o o o

u u o o o u

Aoki’s description implies that glide formation should apply before round vowels as well, e.g. /u+o/ > [wo], but he gives no examples of such realizations. In contrast, McLaren’s data and explicit statements (1955: 10) strongly suggest that gliding of /u/, /o/ occurs only before non-round vowels. I follow McLaren’s account here.

Hiatus Resolution

10

Examples illustrating some of these realizations in Xhosa (Aoki 1974) are shown below: (19)

/esisu-ini/10 /ni-o–a/ /ndi-akha/ /ni-enza/ /wa-ejele/ /a+a-oni/ /a+a-akhi/ /wa-inkosi/ /wa-umfazi/ /esilo-ini/

→ → → → → → → → → →

[esiswini] [no–a] [ndakha] [nenza] [wejele] [a+oni] [a+akhi] [wenkosi] [womfazi] [esilweni]

‘stomach (loc)’ ‘you roast’ ‘I build’ ‘you make’ ‘he fell in’ ‘wrong doers’ ‘builders’ ‘of the chiefs’ ‘of the woman’ ‘animal’

The descriptive summary of the Xhosa patterns in (18) illustrates something that is quite typical of languages that combine vowel elision with glide formation and/or coalescence, which is that it is generally possible to regard vowel elision as a kind of default process. That is, the simplest way of describing the relevant generalizations is often to specify the conditions under which glide formation and/or coalescence apply, with a statement that vowel elision applies elsewhere. All three processes – vowel elision, glide formation, and coalescence – can occur either with or without compensatory lengthening, depending on the language (see chapter 64: compensatory lengthening). Typically, if compensatory lengthening applies with one process it will apply with the others as well. Thus, Ganda shows compensatory lengthening with both vowel elision and glide formation, while Xhosa does not show compensatory lengthening with either of these, nor with coalescence. It also appears generally true that languages (e.g. Ganda) with contrastive vowel length manifest compensatory lengthening while those with no phonemic length do not, but it remains to be seen how universal this correlation is.

3

Theoretical treatments and issues

3.1 Early generative phonology Many analyses of hiatus resolution patterns in particular languages (e.g. Brown 1970; Aoki 1974; Phelps 1975, 1979; Elimelech 1976; Halle 1978; Shaw 1980; Snider 1985) were carried out within early generative phonological frameworks conforming roughly to the model proposed in Chomsky and Halle (1968) or its offshoots. In such models, hiatus resolution processes are due to the operation of languagespecific phonological rules. To account for the Xhosa hiatus resolution patterns in (19), for example, Aoki (1974: 239) posits three ordered rules of Vowel Lowering, Glide Formation, and Vowel Deletion, which (with minor notational adjustments) are essentially those in (20): 10

Aoki (1974: 238) displays the underlying forms of the first and last forms in (19) as /isisu-ini/ and /isilo-ini/, respectively, but describes the lowering of the word-initial vowel form /i/ to [e] as a morphosyntactic replacement, suggesting that initial /e/ is present underlyingly.

Roderic F. Casali

11 (20)

a.

Vowel Lowering V → [−high] /

G V J __ I−highL

b.

Glide Formation G V J → [−vocalic] / __ V I+roundL

c.

Vowel Deletion V → Ø / __ V

Derivations illustrating the operation of these rules are shown below (Aoki 1974: 40):11 (21)

Underlying Form Vowel Lowering Glide Formation Vowel Deletion Output

/wa-umfazi/ /esisu-ini/ /esilo-ini/ wa-omfazi — esilo-eni — esisw-ini esilw-eni w-omfazi — — [womfazi] [esiswini] [esilweni]

The formal apparatus of early generative phonology frequently offered multiple possibilities for analyzing a given pattern. For example, in contrast to Aoki’s analysis of Xhosa coalescence using separate vowel lowering and elision rules, other researchers (e.g. Phelps 1975, 1979; Halle 1978) treated very similar patterns in other languages using a type of rule, known as a transformational rule, which is capable of simultaneously affecting (and, in the case of coalescence, merging the features of) two different segments. Perhaps not surprisingly, much of the literature on hiatus resolution patterns of this period focused on issues of rule formulation and the related question of when the rules for two potentially related processes might appropriately be collapsed into a single rule. Aoki’s paper, which provides extensive arguments against a transformational rule analysis of coalescence (on the grounds that it is arbitrary and unrevealing and that it leads to an unnecessary increase in the complexity and power of the theory), is itself an interesting case in point. Other relevant work includes Brown (1970), Harms (1973), Hyman (1973), Shaw (1980), Snider (1985), and an extended debate (Chomsky and Halle 1968; Phelps 1975, 1979; Halle 1978) over some particularly intricate patterns in Kasem.

3.2 Autosegmental and non-linear generative phonology The late 1970s and 1980s saw the development of alternative and greatly elaborated autosegmental or non-linear conceptions of phonological structure in which some or all phonological features are assumed to occur on separate structural tiers (see chapter 14: autosegments). A number of studies of hiatus resolution phenomena (e.g. Katamba 1985; Clements 1986; Pulleyblank 1986, 1988; Sohn 1987; de Haas 1988; Snider 1989) were carried out using such models. We will look at one representative (and influential) case in some detail, Clements’ (1986) treatment 11

The derivations in (21) differ slightly from those shown in Aoki due to an apparent typo in his derivation of [esiswini] and a minor (and irrelevant) difference in choice of underlying forms (see note 10).

Hiatus Resolution

12

of glide formation and elision in Ganda (see (3) above). Clements’ analysis employs the rules in (22).12 (22)

a.

Glide Formation C

V

V

[+high] [−cons] b. Non-high Vowel Deletion V

V

[−high] [−cons] An appealing feature of Clements’ analysis is that it provides a very straightforward account of the compensatory lengthening that accompanies both elision and glide formation in Ganda. Both rules in (22) have the effect of delinking a V element from its associated vowel features. This is illustrated below for the case of vowel elision. (23a) shows the underlying form corresponding to /ka-oto/ ([ko(to]) ‘small fireplace’ within Clements’ model, while (23b) shows the representation that results when this form is subjected to rule (22b) (Non-high Vowel Deletion), which delinks /a/ from its associated V element, in conjunction with a further (universal) convention that is assumed to delete unassociated segments (in this case the delinked /a/). (23)

C

V

V

C

V

k

a

o

t

o

b. C

V

V

C

V

o

t

o

a.

k

The parallel forms in (24) illustrate the application of the Glide Formation rule (22a). (24a) shows the underlying form of /mu-ana/ ([mwa(na]) ‘child’, and (24b) shows the result of applying Glide Formation to this form. (24)

a.

b.

12

C

V

V

C

V

m

u

a

n

a

C

V

V

C

V

m

u

a

n

a

Glide Formation as formulated in (22a) does not account for the cases where non-high vowels glide word-initially in (16). Clements proposes an additional rule to account for these, which we will not treat here.

Roderic F. Casali

13

Following the application of these rules, the forms in (23b) and (24b) both contain an unassociated V element. Clements assumes that there is a universal Linking Convention that has the effect of automatically reassociating such an unassociated V element to an accessible vowel segment (subject to a general prohibition on crossing of association lines). Applied to the representations in (23b) and (24b), this convention yields the representations in (25a) and (b), respectively. (25)

a.

C

V

k b.

V

C

V

o

t

o

C

V

V

C

V

m

u

a

n

a

In these surface representations, V2 emerges as a long vowel, since it is linked to two V elements (see chapter 54: the skeleton). This account encodes quite directly the intuition that compensatory lengthening involves the transfer of duration from one segment to another. There is a further wrinkle to the analysis. As noted in §2.4 above, glide formation does not apply in Ganda to the sequences /i+i/ and /u+u/, in which V1 and V2 are identical. To prevent the Glide Formation rule (22a) from applying to these sequences, Clements posits an additional rule of Twin Vowel Deletion that is ordered before Glide Formation and functions to remove sequences of identical high vowels as possible inputs to the latter: (26)

Twin Vowel Deletion V

V

[aF] [aF] This rule is applicable to words like /mi-iko/ [mi(ko] ‘trowels’, whose underlying form is shown in (27): (27)

C

V

V

C

V

m

i

i

k

o

Application of Twin Vowel Deletion, along with the universal convention requiring deletion of unassociated segments and the Linking Convention that accomplishes reassociation of a free V element, will convert this to (28). (28)

C m

V

V

C

V

i

k

o

Hiatus Resolution

14

Since the form in (28) does not meet the structural description for Glide Formation to apply, the analysis correctly predicts [mi(ko] and not *[mji(ko] as the surface form. While the analysis derives the correct forms, however, the need to posit the language-specific rule (26) implies that immunity of sequences of identical vowels from glide formation is an idiosyncratic characteristic of the language. As noted in §2.4, such sequences appear to be regularly exempt from glide formation in other languages as well, suggesting that something more universal than a language-specific rule (26) is at work. (Potentially, this presents an interesting challenge not only for autosegmental models like Clements’ but for other approaches as well.) A strong interest of many autosegmental theories is the specification of phonological features. Much research has been done in particular on the possibility of accounting for certain phonological patterns based on the assumption that only one value of a feature is phonologically specified (see chapter 7: feature specification and underspecification). Underspecification models of this type have potential implications for the analysis of vowel coalescence. One of the analytical questions that arises in connection with coalescence is what determines which features of the two merged vowels are preserved in the output. An interesting general answer to this question, pursued in a study by de Haas (1988) (see also Sohn 1987 and Snider 1989), is that the underlyingly specified features from both vowels are preserved in the output. Preservation of all specified features of both vowels under coalescence would presumably be impossible in cases where the two vowels have opposite values of some feature, since this would lead to a surface vowel simultaneously specified as both [+F] and [−F] for some feature [F]. Following previous work in radical underspecification theory, de Haas assumes that only one value of each feature is underlyingly specified. Consider in this regard the symmetric coalescence of /o/ and /a/ to [D] in Attic Greek, as in the relevant forms in (4) above, repeated here as (29). (29)

/ti(ma-omen/ /ajdo-a/

→ →

[ti(.m∑(.men] [aj.d∑(]

(/ao/ > [D(]) ‘honor (1pl pres ind)’ (/oa/ > [D(]) ‘shame (acc sg)’

In de Haas’s underspecification analysis, /a/ is specified only as [+low] and [+back] at the point where coalescence applies, while /o/ is specified only as [+round]. Combining all three feature values yields a [+low], [+back], [+round] vowel, which in de Haas’s analysis is equivalent to [D]. Many autosegmental treatments of hiatus resolution processes were also concerned with the relationship between hiatus resolution and syllable structure and attempted to establish a formal connection between the two. In the model of de Haas (1988), for example, (symmetric) coalescence is contingent on prior resyllabification of two adjacent vowels into a single syllable. Other autosegmental analyses that attempted to connect hiatus resolution to syllabification include Katamba (1985), Pulleyblank (1986), Walli-Sagey (1986), Schane (1987), and Sohn (1987).

3.3 Optimality Theory Analyses of hiatus resolution patterns within Optimality Theory (OT) date from the early years of the paradigm and include, among other studies, Rosenthall (1994,

15

Roderic F. Casali

1997), Casali (1995, 1997, 1998), Orie and Pulleyblank (1998), Senturia (1998), Causley (1999a, 1999b), and Bakovio (2003, 2007). Though they differ somewhat in detail, most such analyses share the following general components: (i)

(ii)

Some constraint (which must be highly ranked) that militates against heterosyllabic adjacent vowel sequences. There is some controversy (discussed below) over the exact identity of this constraint. For now, we will simply label it “NoHiatus.” Constraints that are violated by various hiatus resolution possibilities. Generally, vowel elision is assumed to violate a constraint Max, which requires underlying segments to be represented in surface forms. Epenthesis is assumed to violate a constraint Dep against insertion of material (as well as relevant markedness constraints against the features of the inserted consonant – see below). Diphthong formation violates a constraint NoDiph against diphthongs. Glide formation violates, minimally, a markedness constraint, here labeled *CG, against consonant + glide sequences.13 Coalescence violates a constraint Uniformity, which prohibits merger of two underlyingly distinct segments into a single segment in the output.

Given these assumptions, hiatus resolution is forced whenever NoHiatus is ranked sufficiently high. At a rough first approximation, the particular form of hiatus resolution that occurs is determined by the constraint that is ranked lowest. For example, epenthesis is predicted to occur if the constraint Dep is outranked by the remaining constraints, as illustrated in (30), using a hypothetical input /ku abo/. (30)

/ku abo/ a. .ku.a.bo.

NoHiatus Max NoDiph *CG Uniformity Dep *! *!

b. .kua.bo. c. .ka.bo. d. .kwa.bo. e. .ko.bo. ☞ f. .ku.?a.bo.

*! *! *! *

The simplified analysis sketched above would need to be significantly elaborated to account for the intricate patterns and interactions found in many languages.14 It does, however, illustrate one important general feature of OT analyses, which is that all phonological processes occur in response to some markedness constraint(s). In this case, the primary markedness constraint is the constraint labeled ‘NoHiatus’ in (30). One of the issues that has been debated is the exact nature of this constraint. In what follows, we will look briefly at this question and several other important issues that arise within OT approaches to hiatus resolution. 13

Under some analyses (e.g. Bakovio 2007), gliding of [−high] vowels will also incur violations of a constraint Ident[high], which prohibits changes to the feature [high], since the resulting semivowel [w]/[j] is assumed to be [+high]. 14 For some proposed constraints relevant to compensatory lengthening which are not considered in this simplified analysis, see Rosenthall (1997).

Hiatus Resolution

16

3.3.1 What drives hiatus resolution? Many descriptions and analyses of vowel hiatus resolution processes (e.g. Brown 1970; Mtenje 1980; Shaw 1980; Katamba 1985; Pulleyblank 1986; Walli-Sagey 1986; Sohn 1987; de Haas 1988; Wiltshire 1992; Balogné Bérces 2006) have suggested that such processes are motivated by factors related to canonical syllable structure, and in particular the need to avoid onsetless syllables (see chapter 33: syllable-internal structure and chapter 55: onsets). In OT, this notion has often been formalized by high ranking of a constraint Onset that requires syllables to have onsets, thus disallowing heterosyllabic V.V sequences which would arise in contexts where hiatus is maintained. An alternative view is that hiatus resolution derives from an avoidance of vowel sequences, and not a requirement that all syllables have onsets. Such a view is made plausible by the observation that vowel hiatus seemingly involves unique phonetic difficulties not found with word-initial onsetless syllables. At least two kinds of difficulty have been cited. First, mutual co-articulatory interaction in a sequence of adjacent vowels tends to perturb the quality of each vowel, potentially making accurate identification of vowel qualities more difficult (Borroff 2003). A different explanation is proposed by de Haas (1988), who sees the problem as a kind of “sonority clash” or “bad syllable contact” (see chapter 49: sonority and chapter 21: vowel height). The adjacent heterosyllabic vowels have (roughly) equal sonority, whereas the preferred transition between syllables should involve a sonority trough. Under the widespread assumption that constraints exist in response to particular phonetic challenges, these considerations lend support to the view that there should be some phonological constraint that specifically excludes hiatus. Several studies have raised novel arguments that hiatus resolution cannot always be attributed to Onset. Orie and Pulleyblank (1998) argue that attributing hiatus resolution to Onset in Yoruba misses important generalizations about the conditions that govern the distribution of different hiatus resolution strategies across different contexts. They adopt instead a constraint NoHiatus, which is violated by vowels in hiatus but not by onsetless syllables in general. Borroff (2003, 2007) presents data from a number of languages in which the same hiatus resolution patterns found with clear /VV/ sequences apply to /V?V/ sequences as well. In Chickasaw (Borroff 2007: 57, citing Ulrich 1993), for example, /VV/ hiatus is resolved by glide epenthesis, as shown in (31a). Interestingly, the same process applies to /V?V/, as in (31b). (31)

a. b.

/tof-to-a/ /bo?-a/

→ →

[toftowa] [bo?w-a]

‘to spit more than once’ ‘to be beaten’

On the assumption that an intervocalic [?] should suffice to satisfy Onset, the fact that the same glide epenthesis process applies even when an intervocalic [?] is present argues that something other than Onset is responsible for hiatus resolution in this case. Borroff (2003) argues for a constraint Vcv-Coord, motivated with reference to phonetic facts involving the sequencing of vowel gestures, which in essence requires that a consonantal target appear between two different vowels.15 15

More precisely, the label Vcv-Coord is a shorthand for a conjoined alignment constraint (see chapter 62: constraint conjunction) Align(V1, release, C1, target) & Align(C1, release, V2 , target), which is described in prose as a requirement to “align the release of the first vowel in a sequence of vowels with the achievement of the target of a consonant, and align the release of that same consonant with the achievement of the target of the second vowel of a sequence” (Borroff 2003: 11).

17

Roderic F. Casali

Though the constraint is equivalent for most purposes to NoHiatus, it is (in contrast to Onset) crucially not satisfied by an intervocalic glottal stop, which lacks an (oral) gestural target. An alternative analysis (Borroff 2007) is to assume that a prevocalic glottal stop does not in fact satisfy Onset. In either case, patterns such as these raise interesting challenges for familiar assumptions about hiatus resolution and its motivations.

3.3.2 Directionality in vowel elision Any analysis of vowel elision in hiatus contexts must account for the choice of vowel, V1 or V2 , that is elided. In rule-based models, the deleted vowel is typically specified directly in the form of the elision rule. For example, both the linear deletion rule (20c) and the autosegmental deletion rule (22b) given above stipulate deletion of the first of two adjacent vowels. In contrast, an account within Optimality Theory must assume that elision of V1 or V2 in a given context will violate different constraints, whose relative ranking determines which outcome occurs. The problem then becomes to identify the relevant constraints. The possible rankings of these constraints should also suffice to generate the V1 or V2 elision cases that are attested cross-linguistically, without predicting patterns that are unattested. Arguably, the relevant generalizations to be accounted for are at least approximately as summarized in (12) above. A possible account of these generalizations is outlined in Casali (1997). The explanation assumes that at a prefix–root juncture or a boundary between two content words, V2 is protected by a constraint MaxMI or MaxWI, demanding, respectively, preservation of morpheme- and word-initial vowels. In addition, the analysis continues to assume a generic Max constraint that is violated by deletion of a segment in any context. The analysis also assumes a constraint MaxLex requiring preservation of segments in roots and in content words. Crucially, there are no analogous Max constraints that specifically target word- or morpheme-final position, or affixes or function words. A consequence of these assumptions is that in some contexts the constraint violations incurred by elision of V1 will be a subset of those incurred by V2 elision. At a prefix–root boundary, for example, elision of V2 violates MaxMI (since V2 is the root-initial segment), MaxLex, and (ordinary) Max, while elision of V1 violates only the latter (assuming we are dealing with a minimally CV prefix, so that V1 is not morpheme-initial).16 Since the constraint violations incurred by V1 elision in this context are a subset of those arising with V2 elision, eliding V2 in this context should, all else being equal, be more costly than eliding V1. Thus, only V1 elision is ordinarily expected in this context. This is illustrated below, using a hypothetical CV prefix and VCV root. Note that there is no ranking of the constraints under which the second candidate, with V2 elision, is optimal. (32)

/CV1-V2CV/

MaxMI MaxLex Max

☞ a. CV2CV b. CV1CV 16

*!

*

* *

The full analysis in Casali (1997) actually predicts that V2 elision should be possible at prefix–root boundaries in the special case of a V prefix, since V1 is protected by MaxWI and an additional constraint MaxMS requiring preservation of monosegmental morphemes. We will ignore these complications here.

Hiatus Resolution

18

Similarly, only V1 elision is predicted when underlying vowels abut at the boundary between two content words. In this case, both V1 elision and V2 elision violate MaxLex and general Max; the two possibilities thus tie on these constraints. However, since V2 elision violates MaxWI while V1 elision does not, the former outcome is less optimal. This is illustrated below for a sequence of two hypothetical VCV content words. (33)

/VCV1 V2CV/ MaxWI MaxLex Max ☞ a. VC V2CV b. VCV1 CV

*!

* *

* *

In other contexts, elision of either vowel is predicted to be possible. For example, at a root–suffix boundary V2 elision violates MaxMI but not MaxLex. Thus, V2 elision is possible if MaxLex outranks MaxMI, as shown below using a hypothetical VCV root and VC suffix: (34)

/VCV1-V2C/ a. VCV2C

MaxLex MaxMI Max *!

☞ b. VCV1C

*

* *

V1 elision, which violates MaxLex but not MaxMI, is predicted under the opposite ranking: (35)

/VCV1-V2C/

MaxMI MaxLex Max

☞ a. VCV2C b. VCV1C

* *!

* *

For roughly analogous reasons, both V1 elision and V2 elision are predicted possibilities at the boundary between a content word and a following function word; in this context V1 elision violates MaxLex but not MaxWI, while V2 elision violates only the latter. Note that this model encodes no general context-independent preference for elision of V1; the overall statistical predominance of V1 elision noted above arises indirectly from the fact that V1 elision is predicted in a wider range of contexts. An alternative interpretation of the observed typology might suppose that there is a general context-independent preference for preservation of V2, expressible as some constraint(s), and that this can be overridden in cases where V1 occurs in a prominent position (and hence falls under the protection of some positional faithfulness constraint). The view that hiatus patterns reveal a general context-independent preference for preservation of V2 is expressed by Lamontagne and Rosenthall (1996) (see also Alderete 2003), who refer to this effect as the persistence of V2. Finding evidence to distinguish the two accounts is not easy. There is one context, however, in which the two views potentially make different predictions: where underlying vowels come together morpheme-internally, for example due to the optional deletion of an intervening consonant. While the Casali (1997) model offers no clear predictions in such cases, a model assuming general persistence of V2 should predict, all else being equal, that V1 must elide. A number of languages

19

Roderic F. Casali

do show vowel elision in such cases, and in at least some of them, this prediction is not borne out. Yoruba (Orie and Pulleyblank 1998; Pulleyblank 1998) and Igbo (Emenanjo 1972) both elide V2 , not V1 , in such cases. Though it might be premature to rule out the possible influence of other factors in these cases, these patterns at least appear to challenge the Persistence of V2 view, especially since both languages normally show V1 elision in other contexts (which could be attributed to constraints favoring preservation of initial segments).

3.3.3 Epenthetic consonants and markedness As noted in §2.1 above, only certain consonants are widely observed to function epenthetically as hiatus interrupters. An adequate phonological theory should explain why this is so. Within OT, the problem of explaining the range of possible epenthetic consonants is closely tied to the question of markedness (see chapter 4: markedness; chapter 12: coronals; chapter 22: consonantal place of articulation). Since epenthetic consonants, by definition, are not present underlyingly, their featural content is not affected by faithfulness constraints requiring preservation of phonological material. Consequently, the epenthetic consonant used to resolve hiatus in a given language should be the consonant that is optimal with respect to relevant markedness constraints alone, as these are ranked in the language. The predicted typological range of possible epenthetic consonants should thus follow from the set of universal markedness constraints posited, together with any restrictions (assumed in some models) on their possible rankings. We can illustrate the basic principles at issue with reference to markedness constraints on place of articulation (POA), which have received much attention in the recent literature. OT models have generally assumed markedness constraints targeting each major POA feature, e.g. the constraints *Lab, *Cor, *Dors, and *Glot, which ban, respectively, labial, coronal, dorsal (e.g. velar), and glottal consonants. All else being equal, the particular epenthetic consonant employed in a language is predicted to have the POA of whichever POA constraint is ranked lowest, e.g. a glottal consonant is expected if *Glot is lowest-ranked. In a theory in which the possible ranking of these POA constraints varies freely across languages, we should expect that any POA could function epenthetically in some language. However, some phonologists have assumed that certain places of articulation are universally more marked than others. For example, de Lacy (2006) assumes the fixed scale in (36), where “>” means “is more marked than.” (36)

dorsal > labial > coronal > glottal

It would be straightforward enough to translate this scale into a universally fixed ranking (i.e. one which is stipulated to hold in all languages as part of Universal Grammar) of POA constraints, as in (37). (37) *Dors >> *Lab >> *Cor >> *Glot Fixed rankings of this sort, with some disagreement over details, have played a role in a number of OT analyses (see for example Lombardi 2002). In place of such a fixed hierarchy, however, de Lacy (2006: 2) adopts a different technical implementation of the same general idea, specifically the set of freely rankable POA markedness constraints in (38):

Hiatus Resolution (38)

a. b.

*{Dors} *{Dors,Lab}

c.

*{Dors,Lab,Cor}

d. *{Dors,Lab,Cor,Glot}

20

Assign a violation for each [dorsal] feature. Assign a violation for each [dorsal] and each [labial] feature. Assign a violation for each [dorsal], each [labial], and each [coronal] feature. Assign a violation for each [dorsal], each [labial], each [coronal], and each [glottal] feature.

In this system, a consonant at a POA further to the left on the scale in (36) will always incur worse violations of these POA constraints than one further to the right. This is because the violations incurred by a POA further to the left are necessarily a superset of those incurred by a POA further to the right, regardless of how these constraints are ranked, as shown below (de Lacy 2006: 50): *{Dors} *{Dors,Lab} *{Dors,Lab,Cor} *{Dors,Lab,Cor,Glot}

(39) k p

*

* *

t ?

* *

* *

*

* *

(Crucially, there are no further POA constraints, e.g. *{Cor} or *{Cor,Glot}, targeting other individual places or place combinations.) If a fixed place markedness hierarchy of this sort were the whole story, we would predict that epenthetic consonants would always be glottals, since an epenthetic glottal consonant is always least costly according to this constraint system. This prediction is too restrictive, as it does not account for various other possibilities (e.g. coronals or a homorganic semivowel) that are reported to exist (see §2.1 above). De Lacy’s solution assumes that there are additional markedness scales that refer to dimensions other than place, and that these interact with the place markedness hierarchy to produce the observed range of typological possibilities. For example, the possibility of epenthesizing a coronal stop [t], as in Axininca Campa, follows from the assumption of an additional set of markedness constraints (this time related not to place but to manner of articulation) against high-sonority consonants in onsets, along with the further (and controversial – see Lombardi 2002; Uffman 2007) assumption that glottal consonants [?] and [h] are higher in sonority than all non-glottal consonants (see chapter 49: sonority). These assumptions motivate a constraint *__ D q/Glot prohibiting glottals in syllable margins (onsets or codas). In languages in which *__ D q/Glot is ranked above the relevant POA markedness constraints, glottals will be excluded as epenthetic hiatus interrupters, despite their (universal) optimality with respect to POA alone. With glottals ruled out, the predicted outcome (all else being equal) should be the POA that fares second best according to the constraint system (see (39)). This is coronal.17 The predicted outcome is illustrated in (40), using a hypothetical input /ai/.

17

See Lombardi (2002) for a similar proposal.

21 (40)

Roderic F. Casali /ai/

*— Dq/ *{Dors} *{Dors,Lab} *{Dors,Lab, *{Dors,Lab, Cor} Glottal Cor,Glot}

a. aki

*

b. api ☞ c. ati d. a?i

*

* *

* *

* *

*

* *

Epenthetic homorganic semivowels (e.g. [w] following /u/, [j] following /i/) are predicted in de Lacy’s theory in cases where further markedness constraints requiring consonants (including epenthetic ones) to agree in their place and manner features with adjacent vowels are highly ranked. Additional markedness constraints generate a few further predicted possibilities in languages in which they are highly ranked. In all, the model predicts the following restricted range of epenthetic consonants in hiatus contexts: [? t h P r w j]. De Lacy claims that this set corresponds to the attested range of possibilities. While de Lacy’s theory provides a detailed, plausible, and comprehensive OT account of consonant epenthesis, it is unlikely to be the last word on the subject. The topic of markedness (both with respect to epenthesis and other areas) has been an extremely complex and controversial one. Among other things, some phonologists (see Hume 2003; Rice 2007; chapter 12: coronals; chapter 22: consonantal place of articulation) have questioned the claim that glottal (or any other) place of articulation is universally unmarked, arguing that either dorsal or labial (as well as coronal) can also function as the unmarked place in some languages. If this is correct, it would suggest the possibility of epenthetic consonants such as [p] or [k] as well. It remains, perhaps, to be seen whether such cases exist. De Lacy discusses several reported cases, but argues that they are better analyzed in other terms (for example because putative epenthetic consonants in some such cases can be treated as present underlyingly). At present, a clear understanding of the typology of consonant epenthesis is arguably somewhat clouded by lack of clear consensus on relevant empirical generalizations. Considerable disagreement exists over the interpretation of patterns in some individual languages, a famous example being the question of whether the “intrusive r” phenomenon found in some English dialects (e.g. the pronunciation of saw it as [sDP/t] in some Eastern Massachusetts dialects, including my own) constitutes epenthesis (see de Lacy 2006, Lombardi 2002, and Uffman 2007 for discussion of this and other cases). Certain cross-linguistic generalizations have also been disputed. For example, while glottal stop is widely regarded as a frequent choice of hiatus interrupter, Uffmann (2007) proposes that glottal stops are not typically inserted primarily to avoid hiatus, but are generally used (German is cited as one example) to provide an onset in prosodically strong positions, e.g. word-initially or before a stressed vowel, where they function to create a maximized sonority contrast with the following vowel. (This account crucially assumes that glottal stops are the lowest sonority consonants, which is exactly the opposite of what de Lacy assumes.) Undoubtedly, there will be further debate over some of the relevant empirical generalizations, as well as their appropriate theoretical treatment.

Hiatus Resolution

22

3.4 The problem of gradience An important distinction in most phonological theories is the distinction between categorical and gradient processes (see chapter 89: gradience and categoricality in phonological theory). A categorical change involves a clear “either-or” shift in the presence of one or more segments or their features, as in a case where an underlying segment is removed completely (elision) or undergoes changes in the binary values of one or more features. Frequently, however, languages manifest gradient processes that involve changes in the degree of some feature, e.g. a phonemically oral vowel is slightly nasalized next to a nasal consonant but remains less nasal than phonemic nasal vowels in the same language. In hiatus contexts, a possible gradient change might involve the “near elision” of one of the adjacent vowels, e.g. a case where an underlying /V1 V2/ sequence is realized phonetically as V2 (perhaps with lengthening) preceded by a short and variable remnant of V1. Hiatus resolution processes have most often been described and analyzed in terms that suggest categorical changes. However, two recent instrumental studies, Baltazani (2006) and Zsiga (1993, 1997), have shown that hiatus resolution patterns (glide formation and/or vowel elision) that had previously been described as categorical in two languages, Modern Greek and Igbo, respectively, actually involve gradient and highly variable timing adjustments. For reasons of space, we will consider only the Igbo case here. Sequences of adjacent vowels arise very commonly in Igbo in cases where a word ending in a vowel precedes a word beginning in a vowel, as in the phrases shown below (from Zsiga 1997; the diacritics mark [−ATR] vowels). (41)

/ïsïtñ ïtñ/ /ñt¢ ñzñ/ /ezi ïtñ/ /ede ïtñ/

‘three sevens’ ‘another grub’ ‘three loans’ ‘three coco-yams’

Three Igbo subjects in Zsiga’s (1993) study each produced six repetitions of each of these and various similar phrases in which one of the eight Igbo vowels occurs word-finally before one of the words [ïtñ] ‘three’ or [ñzñ] ‘another’. (In all, each of the eight vowels was used in two utterances.) Vowel formant measurements of the digitized recordings showed extreme variation, even for the same utterance produced by the same speaker, in the realization of the underlying vowel sequences. These ranged from tokens showing essentially no deletion or assimilation (i.e. in which both vowels clearly surface) to those showing complete loss of V1 (i.e. with the output consisting entirely of a lengthened version of V2). If all the observed outcomes were of one of these two types, this might suggest a categorical but optional rule eliding V1 with compensatory lengthening of V2 (or a rule of total assimilation of V1 to V2). Importantly, however, the results show a range of intermediate realizations as well, in which formant values near the beginning of the vocalic span show a quality intermediate between V1 and V2. This intermediate quality varies across repetitions of the same utterance from one that is more similar to V1 to one that is more similar to V2. Zsiga argues that such findings are not easily reconcilable

23

Roderic F. Casali

with an analysis that treats hiatus resolution as optional but categorical, and that the process is better understood as an adjustment in the relative timing of V1 and V2. More specifically, achievement of the target articulatory gestures for V2 varies from relatively late (allowing for a more or less normal manifestation of a preceding V1) to relatively early (resulting in partially assimilated tokens) to virtually at the release of the preceding consonant (in which case V1 is essentially gone). Seen from this perspective, superficial instances of categorical deletion in some of the tokens are better regarded as simply the extreme endpoint of a process that applies along a continuum. Though specific proposals vary, it has been widely assumed that the familiar kinds of phonological rules and/or constraints standardly used in the analysis of categorical processes are not appropriate to the treatment of gradient sound changes. Zsiga analyzes gradient hiatus resolution in Igbo using the framework of articulatory phonology (Browman and Goldstein 1990), a model that is well suited to handling variable adjustments in the relative timing of gestures. In addition to highlighting the importance of (and need for additional) explicit theoretical treatments of gradient changes in hiatus contexts, these studies raise an important empirical issue as well. Hiatus resolution in both Igbo and Modern Greek had been described in some previous studies as categorical. This raises the possibility (see Zsiga 1997: 265) that other hiatus resolution processes that have been described as categorical in the literature might turn out to be gradient upon closer examination. Studies such as Baltazani’s and Zsiga’s underscore the need for careful attention to the possibility of gradience in the context of descriptive phonological fieldwork.

4

Summary

Hiatus resolution patterns are extremely varied. This chapter has provided a brief and necessarily selective look at some of the variation that occurs in the behavior of particular hiatus resolution processes and in their co-occurrence and interaction. The range of explanatory models that have arisen in connection with hiatus resolution phenomena is also very broad. We have looked at a sample of theoretical proposals from several time periods, including early generative treatments, autosegmental analyses, and several OT models. The central research questions have varied somewhat from model to model. Whereas rule formalism and related issues were a central concern in early generative analyses, autosegmental analyses used more elaborated phonological representations to suggest new solutions to problems such as compensatory lengthening, the featural output of coalescence, and the role of syllable structure in hiatus resolution. Issues that have arisen within OT include the primary markedness constraint that triggers hiatus resolution, the constraint rankings that determine which of two adjacent vowels elides, and the problem of accounting for the range of consonants that can function epenthetically as hiatus interrupters. Finally, we have looked briefly at an issue, gradient hiatus-related processes, which poses potentially important theoretical and empirical challenges for any approach.

Hiatus Resolution

24

ACKNOWLEDGMENTS I am grateful to Beth Hume, Marc van Oostendorp, Keith Snider, and two reviewers for their valuable comments and suggestions on an earlier version of this paper, and to Sudharsan Seshadri Nagarajan for assistance with relevant library research.

REFERENCES Adive, John Raji. 1989. The verbal piece in Ebira. Arlington: Summer Institute of Linguistics & University of Texas at Arlington. Adjekum, Grace, Mary E. Holman & Thomas W. Holman. 1993. Phonological processes in AnufD. Legon: Institute of African Studies, University of Ghana. Ahmad, Zaharani. 2001. Onset satisfaction and violation in Malay: An optimality account. In Graham W. Thurgood (ed.) Papers from the 9th Annual Meeting of the Southeast Asian Linguistics Society, 1999, 135–159. Tempe: Arizona State University Program for Southeast Asian Studies. Alderete, John. 2003. Structural disparities in Navajo word domains: A case for LexCatFaithfulness. The Linguistic Review 20. 111–157. Aoki, Paul K. 1974. An observation of vowel contraction in Xhosa. Studies in African Linguistics 5. 223 –241. Awobuluyi, Oladele. 1972. The morphophonemics of Owon Afa. Research Notes, Department of Linguistics and Nigerian Languages, University of Ibadan 5. 25 –44. Bakovio, Eric. 2003. Conspiracies and morphophonological (a)symmetry. Paper presented at the 1st Old World Conference in Phonology, Leiden. http://works.bepress.com/ cgi/viewcontent.cgi?article=1041&context=ebakovic (April 2010). Bakovio, Eric. 2007. Hiatus resolution and incomplete identity. In Fernando Martínez-Gil & Sonia Colina (eds.) Optimality-theoretic studies in Spanish phonology, 62–73. Amsterdam & Philadelphia: John Benjamins. Balogné Bérces, Katalin. 2006. What’s wrong with vowel-initial syllables? SOAS Working Papers in Linguistics 14. 15 –21. Baltazani, Mary. 2006. Focusing, prosodic phrasing, and hiatus resolution in Greek. In Louis M. Goldstein, Douglas Whalen & Catherine T. Best (eds.) Laboratory phonology 8, 473–494. Berlin & New York: Mouton de Gruyter. Bergman, Richard. 1968. Vowel sandhi in Igede and other African languages. M.A. thesis, Hartford Seminary Foundation. Bliese, Loren F. 1981. A generative grammar of Afar. Arlington: Summer Institute of Linguistics. Borroff, Marianne L. 2003. Against an Onset approach to hiatus resolution. Paper presented at the 77th Annual Meeting of the Linguistic Society of America, Atlanta (ROA-586). Borroff, Marianne L. 2007. A landmark underspecification account of the patterning of glottal stop. Ph.D. dissertation, Stony Brook University. Browman, Catherine P. & Louis Goldstein. 1990. Tiers in articulatory phonology, with some implications for casual speech. In John Kingston & Mary E. Beckman (eds.) Papers in laboratory phonology I: Between the grammar and physics of speech, 341–376. Cambridge: Cambridge University Press. Brown, Gillian. 1970. Syllables and redundancy rules in generative phonology. Journal of Linguistics 6. 1–17. Casali, Roderic F. 1995. Patterns of glide formation in Niger-Congo: An optimality account. Paper presented at the 69th Annual Meeting of the Linguistic Society of America, New Orleans.

25

Roderic F. Casali

Casali, Roderic F. 1996. A typology of vowel coalescence. UC Irvine Working Papers in Linguistics 2. 29 –42. Casali, Roderic F. 1997. Vowel elision in hiatus contexts: Which vowel goes? Language 73. 493–533. Casali, Roderic F. 1998. Resolving hiatus. New York & London: Garland. Casali, Roderic F. 2003. [ATR] value asymmetries and underlying vowel inventory structure in Niger-Congo and Nilo-Saharan. Linguistic Typology 7. 307–382. Causley, Trisha. 1999a. Complexity and markedness in Optimality Theory. Ph.D. dissertation, University of Toronto. Causley, Trisha. 1999b. Faithfulness and contrast: The problem of coalescence. Proceedings of the West Coast Conference on Formal Linguistics 17. 117–131. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clements, G. N. 1986. Compensatory lengthening and consonant gemination in LuGanda. In Wetzels & Sezer (1986), 37–77. Craene, Robert de. 1986. Le verbe conjugé en Tem. Studies in African Linguistics 17. 1–37. de Lacy, Paul. 2006. Markedness: Reduction and preservation in phonology. Cambridge: Cambridge University Press. Elimelech, Baruch. 1976. A tonal grammar of Etsako. UCLA Working Papers in Phonetics 35. Available at http://escholarship.org/uc/item/7qd5v492. Emenanjo, E. N. 1972. Vowel assimilation in Igbo. Research Notes, Department of Linguistics and Nigerian Languages, University of Ibadan 5. 7–18. Faraclas, Nicholas. 1982. Elision and other morpheme boundary phenomena in the western dialects of Obolo. Journal of West African Languages 12. 69 –82. Haas, Wim G. de. 1988. A formal theory of vowel coalescence: A case study of ancient Greek. Ph.D. dissertation, University of Nijmegen. Halle, Morris. 1978. Further thoughts on Kasem nominals. Linguistic Analysis 4. 167–185. Harms, Robert T. 1973. How abstract is Nupe? Language 49. 439 –446. Hedinger, Robert & Sylvia Hedinger. 1977. Phonology of AkDDse (Bakossi). Yaoundé: Summer Institute of Linguistics. Howard, Irwin. 1973. A directional theory of rule application in phonology. Bloomington: Indiana University Linguistics Club. Hulstaert, Gustaaf. 1970. Esquisse du parler des Nkengo. Tervuren: Musée Royal de l’Afrique Centrale. Hume, Elizabeth. 2003. Language specific markedness: The case of place of articulation. Studies in Phonetics, Phonology and Morphology 9. 295 –310. Hyman, Larry M. 1973. Nupe three years later. Language 49. 447–452. Hyman, Larry M. (ed.) 1979. Aghem grammatical structure. Southern California Occasional Papers in Linguistics 7. Kaisse, Ellen M. 1977. Hiatus in Modern Greek. Ph.D. dissertation, Harvard University. Katamba, Francis. 1985. A non-linear account of the syllable in Luganda. In Didier L. Goyvaerts (ed.) African linguistics: Essays in memory of M. W. K. Semikenke, 267–283. Amsterdam: John Benjamins. Kimenyi, Alexandre. 1979. Studies in Kinyarwanda and Bantu phonology. Carbondale, IL: Linguistic Research Inc. Kutsch Lojenga, Constance. 1994. Ngiti: A Central-Sudanic language of Zaire. Cologne: Rüdiger Köppe Verlag. Lamontagne, Greg & Sam Rosenthall. 1996. Contiguity constraints and persistent vowel parsing. Unpublished ms., University of British Columbia & Ohio State University. Lombardi, Linda. 2002. Coronal epenthesis and markedness. Phonology 19. 219 –251. McLaren, James. 1955. A Xhosa grammar. 2nd edn. Cape Town: Longmans, Green & Co. Midtlyng, Patrick J. 2005. Washo morphophonology: Hiatus resolution at the edges or let them be vowels. Santa Barbara Papers in Linguistics 16. 50 –63.

Hiatus Resolution

26

Mtenje, Al. 1980. Aspects of Chichewa derivational phonology and syllable structure. M.A. thesis, Southern Illinois University, Carbondale. Nurse, Derek & Gérard Philippson. 1977. Tones in Old Moshi (Chaga). Studies in African Linguistics 8. 49 –80. Orie, Olanike Ola & Douglas Pulleyblank. 1998. Vowel elision is not always onset-driven. Unpublished ms., Tulane University & University of British Columbia. Parker, Kirk. 1985. Baka phonology. Occasional Papers in the Study of Sudanese Languages 4. 63–85. Parker, Steve. 1989. The sonority grid in Chamicuro phonology. Linguistic Analysis 19. 3–58. Parkinson, Frederick B. 1996. The representation of vowel height in phonology. Ph.D. dissertation, Ohio State University. Payne, David L. 1981. The phonology and morphology of Axininca Campa. Austin, TX: Summer Institute of Linguistics. Phelps, Elaine. 1975. Simplicity criteria in generative phonology: Kasem nominals. Linguistic Analysis 4. 297–332. Phelps, Elaine. 1979. Abstractness and rule ordering in Kasem: A refutation of Halle’s maximizing principle. Linguistic Analysis 5. 29 –69. Picard, Marc. 2003. On the emergence and resolution of hiatus. Folia Linguistica Historica 24. 44 –57. Plunkett, Gray C. 1991. The tone system of Foodo nouns. M.A. thesis, University of North Dakota. Pulleyblank, Douglas. 1986. Underspecification and low vowel harmony in Okpe. Studies in African Linguistics 17. 119 –153. Pulleyblank, Douglas. 1988. Vowel deletion in Yoruba. Journal of African Languages and Linguistics 10. 117–136. Pulleyblank, Douglas. 1998. Yoruba vowel patterns: Deriving asymmetries by the tension between opposing constraints. Unpublished ms., University of British Columbia. Rice, Keren. 1995. On vowel place features. Toronto Working Papers in Linguistics 14. 73–116. Rice, Keren. 2007. Markedness in phonology. In Paul de Lacy (ed.) The Cambridge handbook of phonology, 79 –97. Cambridge: Cambridge University Press. Rosenthall, Sam. 1994. Vowel/glide alternation in a theory of constraint interaction. Ph.D. dissertation, University of Massachusetts, Amherst. Rosenthall, Sam. 1997. The distribution of prevocalic vowels. Natural Language and Linguistic Theory 15. 139 –180. Rubach, Jerzy. 2000. Glide and glottal stop insertion in Slavic languages: A DOT analysis. Linguistic Inquiry 31. 271–317. Saloné, Sukari. 1980. Vowel coalescence and tonal merger in Chagga (Old Moshi): A natural generative approach. Studies in African Linguistics 11. 75 –100. Schane, Sanford A. 1987. The resolution of hiatus. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23(2). 279–290. Senturia, Martha B. 1998. A prosodic theory of hiatus resolution. Ph.D. dissertation, University of California, San Diego. Shaw, Patricia A. 1980. Theoretical issues in Dakota phonology and morphology. New York: Garland. Snider, Keith L. 1985. Vowel coalescence across word boundaries in Chumburung. Journal of West African Languages 15. 3 –l3. Snider, Keith L. 1989. Vowel coalescence in Chumburung: An autosegmental analysis. Lingua 78. 217–232. Sohn, Hyang-Sook. 1987. On the representation of vowels and diphthongs and their merger in Korean. Papers from the Annual Regional Meeting, Chicago Linguistic Society 23. 307–323. Tchagbale, Zakari. 1976. Phonologie et tonologie du tem. Ph.D. dissertation, Université de la Sorbonne Nouvelle, Paris.

27

Roderic F. Casali

Tranel, Bernard. 1992–1994. Tone sandhi and vowel deletion in Margi. Studies in African Linguistics 23. 111–183. Tucker, Archibald N. 1962. The syllable in Luganda: A prosodic approach. Journal of African Languages 1. 122 –166. Uffmann, Christian. 2007. Intrusive [r] and optimal epenthetic consonants. Language Sciences 29. 451–476. Ulrich, Charles H. 1993. The glottal stop in Western Muskogean. International Journal of American Linguistics 59. 430 –441. Walli-Sagey, Elisabeth. 1986. On the representation of complex segments and their formation in Kinyarwanda. In Wetzels & Sezer (1986), 251–295. Westermann, Diedrich. 1930. A study of the Ewe language. London: Oxford University Press. Wetzels, W. Leo & Engin Sezer (eds.) 1986. Studies in compensatory lengthening. Dordrecht: Foris. Wiltshire, Caroline. 1992. Syllabification and rule application in harmonic phonology. Ph.D. dissertation, University of Chicago. Zsiga, Elizabeth C. 1993. Features, gestures, and the temporal aspects of phonological organization. Ph.D. dissertation, Yale University. Zsiga, Elizabeth C. 1997. Features, gestures, and Igbo vowels: An approach to the phonology–phonetics interface. Language 73. 227–274.

62 Constraint Conjunction Megan J. Crowhurst

. . . conjunctive interaction is from a formal point of view entirely natural in OT; indeed, in an important sense, its absence would be unnatural. By this I mean simply that without conjunction, basic OT typologies are not strongly harmonically complete, but with conjunction, they are. (Smolensky 2006: 139)

Interactions among constraints in early Optimality Theory (Prince and Smolensky 1993) were limited to a mode of strict domination, represented by the connective “>>.” However, researchers soon noted that a theory of constraint interaction relying on strict domination did not adequately account for certain well-known sound behaviors, prompting arguments in favor of additional modes of interaction. Three main proposals for combining simple constraints into more complex ones using connectives other than “>>” were advanced: local conjunction, an analog of Boolean conjunction, and material implication.

1

Classical Optimality Theory

Optimality Theory (OT) made its debut in linguistics through the work of Prince and Smolensky (1993), who adapted a constraint-based model with a long history in other fields (including evolutionary biology, the information sciences, and economics) for the formal analysis of sound phenomena in language. In Prince and Smolensky’s model, input states are mapped to output states through a procedure in which a set of potential analyses of the input (output candidates) is passed through a filter consisting of an evaluator, Eval, and a hierarchy of constraints restricting properties of the output. The constraints are members of Universal Grammar, and the priorities assigned to them vary by language. OT constraints are soft constraints – any constraint can be violated, but only under pressure from a constraint with higher priority. The predicted output is the candidate that survives the filter. This candidate is optimal in that it represents the input–output mapping that best satisfies the hierarchy by minimizing violations of higher-ranking constraints at the expense of lower-ranking ones. According to the doctrine of strict domination, if two constraints A and B interact, then constraint A outranks B (or vice versa) in the hierarchy, notated as The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0062

2

Megan J. Crowhurst

A >> B (or B >> A). An interaction A >> B is exposed when the requirements of A and B are incompatible, and there is an input whose properties are such that no viable candidate satisfies both constraints. In such cases, the higher-ranking constraint A is easy to identify because its effect is evident, while B’s effect is obscured. (Constraint B may, however, take effect when A is not at stake.) The following tableau illustrates the evaluation of a mini-set of two candidates, each violating either A or B and not the other. Being higher-ranked, A acts on the candidate set first: the candidate which best satisfies A is kept, and the less successful candidate (in regard to constraint A) is rejected. In this simple example, the candidate that fares best on A is the optimal candidate. Were the rankings of A and B reversed, candidate 2 would be the winner instead. (1)

Constraint A >> Constraint B Input ☞ a. output candidate 1 b. output candidate 2

Constraint A Constraint B * *!

Given that OT is a theory of constraint interaction, an important issue naturally concerns the ways in which constraints might interact. As noted, for Prince and Smolensky and much subsequent work, only strict domination was sanctioned as a mode of constraint interaction.1 Over time, various researchers have challenged this doctrine in considering what other relationships might hold among constraints. Some researchers have proposed a relation of non-dominance, in which neither constraint A nor B dominates the other (Crowhurst 2001). Under non-dominance, violations of A and B together are evaluated cumulatively. Others have argued that constraints can exist in a free ranking relationship to account for optional sound phenomena (e.g. Reynolds 1994; Prince 2001; Itô and Mester 2003a; Jacobs 2004). This chapter reviews proposals that have addressed the question “Can simple constraints combine with one other, and if so, how?” Or, “What connectives other than ‘>>,’ if any, define relationships that can hold among constraints?”

2

Local conjunction: Rejecting the “worst of the worst”

The first proposal for combining constraints beyond the standard mode of strict domination, introduced in Smolensky (1993, 1995, 1997) and subsequently worked out in a series of presentations culminating in Smolensky (2006), was that elemental OT constraints can be locally conjoined to form a more complex constraint that is violated only if both of its members are violated in a specified domain. Other influential discussions of the details and applications of local conjunction appeared in a series of papers by Itô and Mester (e.g. 1996, 1998, 2003a), culminating in Itô and Mester (2003b). Smolensky’s definition of local conjunction appears in (2) (Smolensky 2006: 43).

1

Discussion here is limited to proposals advanced in the core OT phonological literature. Combinatorial devices have been used uncontroversially in probabilistic models of OT.

Constraint Conjunction (2)

3

Local conjunction within a domain D *A &D *B is violated if and only if a violation of *A and a (distinct) violation of *B both occur within a single domain of type D.

In evaluating local conjunctions, Eval returns a mark “*” only when both conjuncts are violated, (3a). Some authors have observed (e.g. Hewitt and Crowhurst 1996) that local conjunction is in fact analogous to logical disjunction: “*” is equivalent to False (F), and the absence of a mark is equivalent to True (T). This can be seen by comparing (3a) with the truth table for logical disjunction. (3)

a.

Evaluation of a local conjunction C1

C1&DC2

C2

Cand1 Cand2

*

Cand3 Cand4 b.

* *

*!

*

C1

C1 C2

C2

Cand1

T

T

T

Cand2

F

T

T

Cand3

T

T

F

Cand4

F

F

F

Logical disjunction

Itô and Mester (2003b: 24) and Smolensky (2006) assume that “&” is a combinatorial operation made available by UG to individual languages, which may activate “&” to derive complex constraints on a language-specific basis. On this view, all of the constraints specified in ConUG, the universal set, plus any languagespecific local conjunctions, are mapped onto a larger, language-specific constraint set, ConG. So, “>>” determines strictly hierarchical rankings, while “&” combines constraints into “superconstraints,” which can then be inserted into hierarchies defined by “>>.”2 The role of the “&” operator in grammars is formally defined by Itô and Mester (2003b: 25) as in (4). (4)

Role of local conjunction in grammars A grammar G can expand the basic constraint set Con inherited from Universal Grammar to a superset ConG = Con ∪ {C1&C2}, for C1, C2 § Con. Expansion is potentially recursive, so that ConG can in turn be extended to a superset ConG′ by adding C3&C4 to ConG, for C3, C4 § ConG, and so on.

2

But see Itô and Mester (1998) and Bakovio (2000) for a slightly different view.

Megan J. Crowhurst

4

The main goal of local conjunction is to derive empirical generalizations about markedness from irreducible principles. Constraint conjunction and other proposals for combining constraints have been criticized on the grounds that they greatly increase the expressive power of the formal architecture of OT by exponentially expanding the constraint set. However, the insights, the improvements in precision, and to some extent the economies achieved with local conjunction have often been impressive.

2.1

Coda conditions

The earliest works to employ local conjunction (e.g. Itô and Mester 1996; Spaelti 1997; Smolensky 2006) noted that the mechanism could be used to advantage in analyzing coda conditions, constraints which impose strict conditions on syllable codas rather than penalizing them outright (chapter 33: syllable-internal structure; chapter 53: syllable contact). Coda conditions have often been captured by the schema CodaCond[a], where a denotes restricted features. A common requirement is for obstruents in coda positions to be voiceless, other factors being equal (e.g. German, Turkish, Russian; see chapter 69: final devoicing and final laryngeal neutralization). The coda condition for voicing appears in (5). (5)

CodaCond[voi] J K K G −son J K I+voiceL L q *C

(One * for any voiced obstruent syllabified exclusively as a syllable coda.) Early work promoting local conjunction noted that coda conditions can be treated as local conjunctions of two well-established constraints, NoCoda in (6a) on the one hand, and a markedness constraint such as (6b) on the other, to yield (6c), with the segment as the local domain of evaluation (Smolensky 1993, 2006; Itô and Mester 2003b; see also Morris 2002). (6)

a. b. c.

NoCoda Syllables do not have codas. (One * per syllable with a coda.) *VoiObst *[−sonorant, +voice]. (One * for any voiced obstruent.) NoCoda &SEG *VoiObst (One * for any segment which is in a syllable coda and which is a voiced obstruent.)

Local conjunctions such as (6c) capture the intuition that constraints expressing coda conditions are really more restrictive versions of NoCoda. A standard assumption has been that a local conjunction is ranked above its conjuncts, and this is consistent with the understanding that the effects of special constraints are visible only when some constraint intervenes between the special constraint and a related, less highly ranked general constraint. In the case under discussion,

Constraint Conjunction

5

restrictions on codas are visible in a grammar only when the following are true: (i) MaxIO(Seg) outranks both instantiations of NoCoda – the unconjoined version and the local conjunction in (6c) – so that the effects of the conjunction are not obscured by deletion; and (ii) IdentIO[voi] ranks above the unconjoined constraint *VoiObst, allowing a surface obstruent voicing contrast, but below (6c), so that only voiceless obstruents will occur in coda position, unless higher-ranking constraints demand otherwise (as in cases of regressive voice assimilation). A tableau making this point is given in (7) (adapted from Itô and Mester 2003b: 27–28). (7)

li(b

MaxIO(Seg) NoCoda &SEG IdentIO[voi] *VoiObst NoCoda *VoiObst

☞ a. li(p b. li(b c. li(

* *!

b

* *

*!

li(be MaxIO(Seg) NoCoda &SEG IdentIO[voi] *VoiObst NoCoda *VoiObst ☞ a. li(bH b. li(pH

b *!

Three advantages to the reformulation of (5) as (6c) are immediately apparent. First, locally conjoining *VoiObst with NoCoda avoids the redundancy that occurs when *VoiObst is stated twice, once as the general feature co-occurrence constraint in (6b), and once again, embedded in the coda condition, (5). Second, restating a given CodaCond constraint as the local conjunction of NoCoda with a standard markedness constraint explains why we have restrictions on coda consonants, but no constraints that express similar restrictions on onsets (Itô and Mester 2003b: 29). In a grammar that allows conjunction, nothing prevents the local conjunction of Onset and a markedness constraint like *VoiObst, but as tableau (8) shows, a conjunction like Onset &SEG *VoiObst could never be violated: if Onset is violated, then there is no onset consonant to check against *VoiObst. Conversely, if *VoiObst is violated by a consonant in onset position, then Onset is clearly not violated. (8)

bit a. bit ☞ b. pit c. it

Onset &SEG *VoiObst *VoiObst Onset *!

b *!

This latter result of local conjunction weighs strongly in its favor: given that *VoiObst encodes a standard generalization about markedness, and given that coda conditions stated in traditional terms following the model of (5) are common, it is notable that no evidence has been found for mirror image constraints imposing comparable restrictions on segments in onset position. The traditional approach

Megan J. Crowhurst

6

has been to assume that a formal asymmetry matches the empirical one – that is, that there are no Onset conditions, or constraints imposing CodaCond-like requirements on segments in onset position. As we have just seen, the local conjunction approach requires no formal asymmetry: local conjunctions like Onset &SEG *VoiObst might exist, but as they can never be active in selecting surface candidates, they will never be rankable, and their existence is moot.

2.2

Universal markedness hierarchies

Local conjunction has also been used successfully in explaining the fact that universal markedness hierarchies are preserved in multiple domains. Much work establishes the place harmony scale in (9a) (chapter 22: consonantal place of articulation). In OT grammars, this scale is expressed by the hierarchy in (9b). The subscript “UG” indicates that this ranking is “fixed,” or specified in universal grammar, and does not vary across languages. (9)

a. b.

Coronal > Labial, Dorsal *Lab, *Dors >>UG *Cor

The place markedness hierarchy in (9) is known to be preserved in different domains, and Smolensky (1993, 2006) advances a detailed argument as to why this might be so: given that (9b) is fixed in UG, any hierarchy in which the constraints in (9b) are locally conjoined with other constraints will be similarly fixed. To continue with the example of coda conditions, various authors, including Smolensky (1993, 2006), Zoll (1998), and Itô and Mester (2003b), have observed that locally conjoining the markedness constraints in (9b) with NoCoda produces the hierarchy in (10), which favors coronals over labials and dorsals in syllable codas. (10)

NoCoda &SEG *Lab, NoCoda &SEG *Dors >>UG NoCoda &SEG *Cor

Smolensky (2006) uses the same reasoning to explain the commonly observed proliferation of segmental contrasts among coronal consonants in many segmental inventories, relative to the labial and dorsal classes (chapter 12: coronals). As an example, consider the consonant inventory of Tohono O’odham (Uto-Aztecan) in (11). (11)

Tohono O’odham consonants labial p b m w

dental } { Ú |

coronal retroflex Õ Ï Î

palato-alveolar ΠРJ j

velar k g

glottal ? h

I

The coronal class dominates the inventory: 11 of the 19 consonants are coronals. Note also that the coronals are represented by three places and three manners of articulation (stops, fricative, and affricates) in the obstruent class, whereas the labials and velars are represented only by stops. To abbreviate Smolensky’s point, if a single-feature markedness constraint like *[+cont] (specified for fricatives

Constraint Conjunction

7

and affricates; see chapter 28: the representation of fricatives; chapter 16: affricates) can be locally conjoined with the fixed hierarchy in (9b), yielding (12), then a grammar that interposes a constraint IdentIO[Place] above *Cor &SEG *[+cont] produces an inventory that has [+continuant] coronals, but no continuant obstruents at other places of articulation. (12)

*Lab &SEG *[+cont], *Dors &SEG *[+cont] >>UG *Cor &SEG *[+cont]

2.3

Feature harmonies

Local conjunction has also been used to account for patterns of vowel and consonant harmony by locally conjoining markedness constraints to form feature domains (see also chapter 91: vowel harmony: opaque and transparent vowels; chapter 72: consonant harmony in child language; chapter 77: longdistance assimilation of consonants; chapter 118: turkish vowel harmony; chapter 123: hungarian vowel harmony). In Smolensky (2006), any contiguous sequence of segments which share one or more phonological features forms a feature domain, and any feature domain is a possible instantiation of the domain of a conjunction. An abbreviated version of Smolensky’s (2006: 64) definition of a feature domain is given in (13). (13)

Definition of feature domain (“z” stands for any feature) A maximal contiguous span of z-bearers with a common value [±z] is a [±z] feature domain D[±z]. (Thus, by definition, contiguous domains of the same z value are impossible.)

Smolensky’s use of the feature domain in OT builds on earlier work (e.g. Kirchner 1993; Smolensky 1993; Cole and Kisseberth 1994; Cassimjee and Kisseberth 1998), but the insight that feature domains can be analyzed in terms of local conjunctions of constraints is due to Smolensky. Examples of Smolensky’s (2006) use of feature domains as restrictors on local conjunction are seen in his treatments of vowel harmony and restrictions on consonant clusters. Smolensky (2006) provides an extended discussion of vowel harmony, with special attention given to a pattern of source-conditioned [ATR] harmony found in Lango (Nilotic: Okello 1975; Bavin Woock and Noonan 1979; Noonan 1992; Archangeli and Pulleyblank 1994). The Lango facts are complex; for our purposes, a fragment of Smolensky’s account will serve to make the point. Lango has the [+ATR] vowels [i e H o u] and the [−ATR] vowels [> e a D Á] (Noonan 1992). The examples in (14) show a root with a [−ATR] mid vowel /e/ combining with a suffix containing a [+ATR] high vowel, /u/ or /i/. (Tone is omitted in these examples.) In (14a), we see that /e/ assimilates [+ATR] when the suffix vowel is /i/, but not when it is /u/, as in (14b). (14)

Lango [ATR] harmony a. b.

Regressive [+ATR] harmony /dek + Ci/ dek.ki ‘your (sg) stew’ No harmony /dek + wu/ dek.wu ‘your (pl) stew’

Megan J. Crowhurst

8

In autosegmental terms, [+ATR] spreads regressively from /i/ to /e/ in [dek.ki], forming a [+ATR] feature domain whose head is the source vowel [i], as shown in (15a). (Smolensky marks heads of feature domains with a superscript “o”; we will adopt the convention of underlining heads of feature domains.) The form [dek.wu], where no assimilation has applied, has both a [−ATR] and a [+ATR] feature domain. (15)

a.

[+ATR]

b. [−ATR] [+ATR]

dek ki

dek

wu

According to Archangeli and Pulleyblank’s (1994) autosegmental analysis of Lango, [+ATR] spreads regressively from a [+high] source vowel to a target vowel if one of the following is true: (i) the source vowel is [+front] and the target is any vowel in either an open or closed syllable; (ii) the source vowel is not [+front], source and target vowels are both [+high], and the target is in either an open or closed syllable; or (iii) the source vowel is not [+front], and the target is not [+high], and the target is in an open syllable. [+ATR] spread in (14a) meets condition (i), but (14b) meets none of the conditions for harmony. Following Archangeli and Pulleyblank (1994), Smolensky’s analysis (2006) of Lango draws on the insight that ATR harmony is conditioned by the markedness of segments that combine [−ATR] and [+ATR] with other features. Well-established markedness constraints are shown in (16).3 (16)

a.

b.

c.

[ATR] and backness *[+ATR, −front] *[−ATR, +front] [ATR] and height *[+ATR, −high] *[−ATR, +high] *V[+A]C]q

(Avoid [+ATR] back vowels.) (Avoid [−ATR] front vowels.) (Avoid [+ATR] mid and low vowels.) (Avoid [−ATR] high vowels.) (No [+ATR] vowels in closed syllables.)

Returning to the examples in (14), note that in [dek.ki], in which underlying /e/ harmonizes with /i/, the outcome optimizes the constraint *[−ATR, +front], which prefers [e], but at the cost of violating *[+ATR, −high], which prefers [e]. In [dek.wu], however, we see the opposite pattern of constraint satisfaction: [e] in the output optimizes *[+ATR, −high] but violates *[−ATR, +front]. The critical difference is in the source vowel. Archangeli and Pulleyblank’s intuition, which Smolensky seeks to capture, is that less marked segments make better domain heads, and that when possible, better domain heads propagate their features through harmony. To the constraints in (16), then, Smolensky’s account adds *Hd[+ATR] and *Hd[−ATR], which penalize a segment for being the head of a [+ATR] or [−ATR] domain, respectively.

3

The constraints in (16) and similar constraints, were proposed as elements of the “grounded phonology” framework developed in Archangeli and Pulleyblank (1994).

Constraint Conjunction (17)

9

a.

Hd-L[−ATR] A [−ATR] domain must be left-headed. (No regressive [−ATR] spread.) b. (*[−ATR, +front] & *Hd[ATR]) &D[ATR] F[ATR] No [+front] head of an unfaithful [−ATR] domain. (No [−ATR] spread from a [+front] vowel.) c. *[−ATR, +high] Vowels do not combine the features [−ATR] and [+high]. (One * for either of [> Á].) d. (*[+ATR, −front] & Hd-L[ATR]) &D[ATR] (*[+ATR, −high] & *V[+ATR]C]q & F[ATR]) A [+ATR] domain with a [−front] head that is not leftmost must be faithful at a [−high] vowel at a closed syllable. (No regressive [+ATR] spread from a [−front] source onto a [−high] vowel in a closed syllable.)

The constraints in (17b) and (17d) are complex and require further explanation. The embedded local conjunction (*[−ATR, +front] & *Hd[ATR]) penalizes a vowel which is both the head of an ATR domain and has the features *[−ATR, +front]. According to this restriction, the lax vowels [> e] cannot head an ATR domain. The syllable [dek] in [dek.wu] violates this requirement. Smolensky uses the expression F[ATR] in the “macro” local conjunction more or less as the more standard Ident[ATR] would be used, to require the segments in an ATR domain to be faithful to the value for [ATR] they came with. Conjoining (*[−ATR, +front] & *Hd[ATR]) and F[ATR], taking an ATR domain as the locus of violation, has the effect of minimizing ATR domains headed by [−ATR, +front] vowels. That is, the macro-conjunction prevents [−ATR] spread from a front vowel, and this is why /dek + wu/ surfaces as [dek.wu] and not [dek.wÁ]. By itself, the first embedded local conjunction in (17d), (*[+ATR, −front] & Hd-L[ATR]), would penalize a [+ATR] domain whose head, a [−front] vowel [u], [H] or [o], is not leftmost within the domain.4 The second embedded local conjunction, (*[+ATR, −high] & *V[+ATR]C]q & F[ATR]), penalizes an unfaithful [+ATR] domain which contains a closed syllable whose vowel is [−high]. The macro-conjunction formed by locally conjoining the two smaller local conjunctions with an ATR domain as the locus of violation means just this: for a [+ATR] domain whose head is on the right and is one of the vowels [u H o] to pass the macro-conjunction, there cannot be further to the left a closed syllable with a [−high] vowel, which is unfaithful through assimilation to the head. This is why regressive [+ATR] spread does not apply in /dek + wu/, so that on the surface we find [dek.wu] and not [dek.wu]. Tableau (18) shows why /dek + wu/ surfaces as [dek.wu] and not [dek.wÁ] or [dek.wu] (i.e. why we have neither regressive [+ATR] harmony nor progressive [−ATR] harmony in this case, under Smolensky’s analysis).

4

For example, a [+ATR] domains such as e . . . u would be penalized. Three types of ATR domain are permitted by the local conjunction (*[+ATR, −front] & Hd-L[ATR]). If the head is one of the set [i e > e a D Á] (i.e. anything other than [u H o]), then the head is rightmost or leftmost in the ATR domain. The head can be one of the set [u H o] if it is domain-initial.

Megan J. Crowhurst

10 (18)

Lango: No regressive [+ATR] harmony /dek+wu/ a. dek.wu ☞ b. dek.wu

(17d)

Hd-L[−atr]

(17b)

*[–atr,hi] Agree[atr]

*! *

c. dek.wÁ

*!

*

Tableau (19) shows how the analysis works for [dek.ki] (from /dek + Ci/). In this case, progressive [−ATR] is blocked just as for [dek.wu], and for the same reason. However, in [dek.ki], regressive [+ATR] harmony is not blocked, because in this case, neither the head (a front vowel this time) nor the target of assimilation (the vowel /e/, in an open, not closed syllable) violates the macro-conjunction in (17d). (19)

Lango: Regressive [+ATR] harmony /dek+Ci/

(17d)

Hd-L[−atr]

(17b)

*[–atr,hi] Agree[atr]

☞ a. dek.ki b. dek.ki

*!

c. dek.k>

*!

*

The brief discussion offered here falls short of representing the very complicated pattern of [ATR] harmony found in Lango, or of Smolensky’s (2006) minutely detailed account of the same. The goal here has been to illustrate a proposal for combining constraints that are already complex into a macro-constraint. More detailed discussions of the Lango pattern can be found in Noonan (1992), Archangeli and Pulleyblank (1994), and Smolensky (2006). Another very common type of restriction can be seen in conditions on consonant clusters: in languages that permit consonant clusters at all, coronals tend to combine with other consonants much more permissively than do labials and dorsals (chapter 46: positional effects in consonant clusters). For example, English allows coronals to cluster with other coronals, with labials, and with dorsal consonants, as shown in (20a). However, with the exception of a few loans (e.g. Akbar, Afghan) and clusters formed across compound boundaries (e.g. [[cup][cake]], [[black][bird]]), labials and dorsals do not cluster with either labials or dorsals, so that (with the regular exception of homorganic nasal + obstruent sequences), clusters such as those in (20b) are generally disallowed. (20)

a.

Cor with Cor state holder chortle adze parlour Atlantic bolster trap

Cor with Lab apt Abner abduct help spry almond arm atmosphere

Cor with Dors acne task silk argue disclose alcohol agree Atkins

Constraint Conjunction b.

Lab with Lab *pb, *bp *pm, *bm *pf, *bv, *fp, etc.

Dors with Dors *kg, *gk

11

Dors with Lab *kp, *pk, *gb, *bg *km, *gm *kf, *fk, *gv, *vg, etc.

The asymmetric distribution of place in consonant clusters is another effect of the place hierarchy in (9). Smolensky (2006) shows that these effects can be accounted for by locally conjoining the constraints in (9b) with one another to produce a cluster place markedness hierarchy like that in (21) ((21) expands Smolensky’s hierarchy to include dorsals) (21)

Place markedness hierarchy for two-consonant clusters 1*[Lab]&CL*[Lab] 5 *[Lab]&CL*[Cor] # 2*[Dors]&CL*[Lab] 6 >>UG ! >> *[Cor]&CL*[Cor] @*[Dors]&CL*[Cor] $ UG 3*[Dors]&CL*[Dor]7

In his analysis of phonotactic conditions on consonant sequences, Smolensky (2006) identifies the domain of local conjunction as a consonant cluster, CL. However, a consonant cluster per se is not a unit of phonological structure. Although Smolensky doesn’t discuss consonant clusters in this light, any sequence of segments whose members share features can be thought of as instantiating feature domains. An obstruent cluster, for example, would instantiate the D[±z] formed by the feature [−sonorant], and a nasal–stop cluster would correspond to the feature domain [−continuant].

2.4

Counterfeeding effects

One of the more striking advantages claimed for local conjunction is its usefulness in accounting for counterfeeding effects. Moreton and Smolensky (2002) show, refining a proposal in Kirchner (1996), that local conjunction can account for synchronic chain shifts of the type A → B, B → C.5 In Western Basque, for example, when two identical vowels are juxtaposed, the low vowel /a/ raises to [e] before [a], and /e/ raises to [i] before [e], as shown in (22) (Kirchner 1996; Moreton and Smolensky 2002; Kawahara 2003). (22)

a.

b.

/a/ → [e] / __ V (violates Ident[low]) alaba bat ‘daughter (indef)’ alabea neska bat ‘girl (indef)’ neskea /e/ → [i] / __ V (violates Ident[high]) seme bat ‘son (indef)’ semie ate bat ‘door (indef)’ atie

‘daughter (def)’ ‘girl (def)’ ‘son (def)’ ‘door (def)’

The Kirchner and the Moreton and Smolensky account goes as follows: if the unmarked vowel is [i], then we’d expect both /a/ and /e/ to raise to [i] in the shifting environment, other factors being equal. However, each vowel moves only one step up, so that /a/ changes its value for the feature [low], and /e/ changes 5

See also Beckman (2003) and Smolensky (2006: §5).

12

Megan J. Crowhurst

its value for [high]. Neither adjustment results in a change of both features. This effect can be accounted for if a conjunction of the constraints Ident[low] and Ident[high] is ranked above a set of markedness constraints (presumably including at least one OCP constraint) that promote raising in a hiatus; call this set HR. Thus, either of the Ident constraints can be violated on its own, when they are ranked below HR, but no shift violating both constraints within a segment (the relevant domain) can occur. The tableaux in (23), adapted from Moreton and Smolensky (2002), illustrate the analysis for two identical sequences of vowels, /aa/ in /alaba-a/ and /ee/ in /seme-e/. (23)

/alaba-a/ Ident[low] &SEG HR Ident[low] Ident[high] Ident[high] ☞ a. alabea b. alabaa c. alabia

* * *!

*

/seme-e/ Ident[low] &SEG HR Ident[low] Ident[high] Ident[high] ☞ a. semie b. semee

* *!

In the tableaux for [alabea] and [semie], we see that the sequences /aa/ and /ee/ are ruled out by the markedness constraints (the HR set). For /alabaa/, an output [alabia], with raising to [i], is rejected by the local conjunction – the output is the candidate that violates only one of the Ident constraints (in this case Ident[high]). In the case of /semee/, [semie] with a high vowel is possible because raising in this case violates only Ident[high] and not Ident[low]. Violating only one of the Ident constraints, [semie], does not violate the local conjunction.

2.5

The conjunction of markedness and faithfulness constraints

So far, we have discussed the local conjunction of markedness constraints, and the conjunction of faithfulness constraints (chapter 63: markedness and faithfulness constraints). Bakovio (2000) and Ìubowicz (2002, 2005) argue for a special role for the conjunction of markedness with faithfulness constraints (M&F conjunction). Bakovio (2000) shows that M&F conjunction can be used to resolve what he calls the “majority rules” problem in the analysis of voicing assimilation in obstruent clusters. In cross-syllabic obstruent clusters, voicing assimilation is generally controlled by the obstruent in onset position. In Dutch, for example, the medial clusters /kd/ zakdoek ‘handkerchief’ and /dk/ in bloedkoraal ‘red coral’ surface as [gd] and [tk], respectively (Kager 1999).6 However, patterns of assimilation 6

The pattern reported here for Dutch is very common. A seemingly odd characteristic of the dialect Kager describes that voiced fricatives in onset position are always devoiced (e.g. /zv/ in kaasvorm ‘cheese mould’ surfaces as [sf]), and this requirement compels the surface voicelessness of any preceding obstruent.

Constraint Conjunction

13

in word-final obstruent clusters are not influenced by requirements on syllable onsets. Consider a language in which an obstruent voicing contrast is preserved in word-final position, which has word-final clusters of two obstruents such as /bt#/ and /pd#/, and in which the markedness constraint Agree[voice] (obstruent clusters must agree in voicing) dominates the faithfulness constraint Ident[voice]. In such a case, where all candidates passing Agree[voice] fail the faithfulness constraint, the candidate containing the voiceless cluster is selected by *VoiObst. Where the input provides a two-consonant cluster which already satisfies the agreement constraint, for example /pt/ or /bd/, input voicing is generally preserved, showing that Ident[voice] dominates *VoiObst (as long as no “spoiler” constraint intervenes). (24)

/pd/ a. pd b. bd

Agree[voice] Ident[voice] *VoiObst *! *

☞ c. pt /bt/ a. bt b. bd

* *!*

* Agree[voice] Ident[voice] *VoiObst *! *

☞ c. pt

* *!*

*

This analysis predicts a different outcome for clusters of more than two obstruents. Tableau (25) shows that in such cases the ranking Ident[voice] >> *VoiObst predicts that cluster voicing will be determined by the majority. This yields the typologically supported outcome for clusters like /skd/, in which two of three obstruents are voiceless at input (e.g. English basked). Unfortunately, it also predicts rare (if at all attested) outcomes like /zgt/ → [zgd] when two of three obstruents are voiced at input. (25)

/zgt/ Agree[voice] Ident[voice] *VoiObst a. zgt ☞ b. zgd c. skt

*! *

** ***

**!

/skd/ Agree[voice] Ident[voice] *VoiObst a. skd b. zgd ☞ c. skt

*! **! *

* ***

Megan J. Crowhurst

14

Bakovio shows that the “majority rules” problem is avoided if the constraints Ident[voice] and *VoiObst can be locally conjoined, assuming the segment as the domain of the conjunction. In (26) we see that on this analysis, of the two clusters that survive Agree[voice], [zgd] and [skt], the fully voiced cluster [zkd] is rejected even though it is more faithful, because it violates the conjunction Ident[voice] &SEG *VoiObst (which is ranked above both of its conjuncts). (26)

/zgt/ Agree[voice] a. zgd b. zgt

*!

c. zkt

*!

☞ d. skt

Ident[voice] &SEG*VoiObst

Ident[voice] *VoiObst

*!

*

*** **

*

*

*

**

Bakovio’s larger point is that the presence of marked segments (in this case voiced obstruents) on the surface is one thing, but the presence of unfaithful marked segments is another. In cases of assimilation in mixed clusters, the presence of unfaithful marked segments can be compelled by high-ranking faithfulness constraints (e.g. assimilation to a voiced onset in cross-syllabic clusters). However, when such pressures are absent, assimilation to the unmarked is the pattern attested across languages. In Bakovio’s example, an analysis that admits the local conjunction Ident[voice] &SEG *VoiObst guarantees the typologically supported outcome, whereas an analysis that relies exclusively on the unconjoined constraints admits the (unsupported) assimiliation to the marked alternative. Ìubowicz (2002, 2005) argues that M&F conjunctions are necessary to account for derived environment effects (chapter 88: derived environment effects), using palatalization in Polish (chapter 121: slavic palatalization) as an example. Other works using local conjunction to account for derived environment effects include Downing (2001) and Itô and Mester (2003a).

2.6

Local self-conjunction

As a special case of local conjunction, some researchers have argued that markedness constraints can be “self-conjoined,” and that self-conjunction offers new insights into cases of dissimilation that have often been attributed to the OCP (e.g. Fukazawa 1999, 2001; Itô and Mester 1998, 2003b; Alderete 2004; Smolensky 2006). As Itô and Mester (2003b: 29) put it: the “Obligatory Contour Principle” is by itself neither a constraint nor a formal universal in phonological theory. The culprit in OCP-type dissimilations is not the adjacency of identical feature specifications on a tier, but the multiple presence of a marked type of structure within some domain.

Constraint Conjunction

15

Self-conjunction is intended to capture the intuition that in dissimilation, one violation of a markedness constraint is tolerated, but a double violation crosses what Itô and Mester aptly call a markedness threshold, and is categorically rejected. As an example, let us consider the role of obstruent voicing in Japanese. In the lexical classes comprised by native Japanese and Sino-Japanese forms, voiced obstruents cannot co-occur within a morpheme (chapter 86: morpheme structure constraints), but neither voiceless obstruents nor voiced sonorants are subject to voicing restrictions (Itô and Mester 1998, 2003b). Thus, morphemes such as those in (27a) and (27b) are typical of Japanese (Itô and Mester 2003b: 34–35). However, native Japanese and Sino-Japanese morphemes containing more than one voiced obstruent (e.g. *[kabazi]) are not. (27)

a.

b.

kusa tako sato kaze geta hada

‘grass’ ‘octopus’ ‘village’ ‘wind’ ‘clogs’ ‘skin’

kataki hotoke tatami kasegi kagami -bakari

‘enemy’ ‘Buddha’ ‘straw mat’ ‘earning’ ‘mirror’ ‘only’

The obstruent voicing contrast in Japanese requires the constraint ranking IdentIO >> *VoiObst. The fact that multiple occurrences of voiced obstruents within a morpheme is prohibited is accounted for by self-conjoining the constraint *VoiObst, with the morpheme as the domain. The conjunction *VoiObst2MORPH (=*VoiObst &MORPH *VoiObst) is ranked above IdentIO. Tableau (28) shows how a set of candidates based on the form kaze and the hypothetical input /gaze/ are evaluated under these rankings. (28)

/kaze/ *VoiObst2MORPH IdentIO(Seg) *VoiObst ☞ a. kaze b. kase

*!

c. gase

*!*

g

*

gz

d. gaze

z

*!

/gaze/ *VoiObst2MORPH IdentIO(Seg) *VoiObst ☞ a. kaze b. kase

* **!

z

☞ c. gase

*

g

d. gaze

*!

gz

Self-conjunction can account not only for passive restrictions on the occurrence of similar structures within a domain, but also cases in which the existence of alternations makes the dissimilation more obvious. A well-known example is Lyman’s Law, which blocks the effect of Rendaku, a phenomenon seen in Japanese

16

Megan J. Crowhurst

lexical compounds. Rendaku assigns the feature [+voice] to a voiceless obstruent appearing as the initial consonant of the second member of a compound. The feature [+voice] is treated as a special morpheme in this case, and as such, its presence in the output is required by the constraint RealizeMorph, defined in (29) (Walker 2000; chapter 103: phonological sensitivity to morphological structure). (29)

RealizeMorph Every morpheme has a phonological exponent in the output. (One * for any null morpheme.)

The Rendaku effect requires RealizeMorph to be ranked more highly than IdentIO(Seg). Tableau (30) is adapted from Itô and Mester (2003b: 37) (constraint names have been changed to match the names used here). The examples used in tableaux (31) and (30) are /ori + kami/ → [origami] ‘paper folding’ and /kami + kaze/ → [kamikaze] ‘divine wind’. (30)

/ori+R+kami/ RealizeMorph IdentIO(Seg) *VoiObst ☞ a. origami b. orikami

*

g

*!

The Lyman’s Law effect occurs because the self-conjunction *VoiObst2MORPH dominates RealizeMorph, neutralizing its effect. (31)

/kami+R+kaze/ *VoiObst2MORPH Realize IdentIO(Seg) *VoiObst Morph ☞ a. kamikaze b. kamigaze

* *(gaze)!

*

z gz

Beyond the examples discussed in §2 local conjunction and its special variant, self-conjunction have been used in the analysis of a diverse set of phenomena. Local conjunction has been used in analyses of sonority distance restrictions on syllable structures (Baertsch 1998; Baertsch and Davis 2003; Smolensky 2006), the sonority hierarchy (Smolensky 2006), tone sandhi in Mandarin Chinese (Lin 2000), glide formation in German (Hall 2004, 2007), and accentual phenomena (Alderete 1999). Finally, Levelt et al. (1999) and Levelt and van de Vijver (1998) propose that local conjunction plays a role in children’s acquisition of the constraint hierarchy of the language they are learning.

3

Other modes of constraint combination

Other proposals for deriving complex constraints from simpler ones using connectives other than “&” have included constraint analogs of Boolean conjunction and two versions of material implication.

Constraint Conjunction

3.1

17

The “best of the best”

Hewitt and Crowhurst (1996) and Crowhurst and Hewitt (1997) argue that in addition to local conjunction and self-conjunction, a proper account of some phonological patterns is best achieved using complex constraints derived using a connective, “û,” with the semantics of Boolean conjunction. Crowhurst and Hewitt call this mode of interaction simply “constraint conjunction.” Here, we will call it b-conjunction, to distinguish it from local conjunction. In contrast to local conjunction, which bans the “worst of the worst,” b-conjunction insists on the “best of the best.” See the definitions and tables in (32) and (33). (32)

a.

b.

(33)

Boolean conjunction The conjunction AûB is true iff proposition A is true and proposition B is true. B-conjunction A candidate Cand passes a b-conjunction AûB iff Cand passes constraint A and Cand passes constraint B.

a. Boolean conjunction A

AûB

B

T

T

T

F

T

T

T

T

F

F

F

F

b. Evaluation of a B-conjunction C1

C 1 C2

*

*!

C2

Cand1 Cand2 Cand3 Cand4

*

*!

*

*!

*

Crowhurst and Hewitt restrict b-conjunction to constraints that share what they call a focus (or fulcrum in Hewitt and Crowhurst 1996) which identifies the locus of violation for each of the conjuncts. (34)

Focus (of a constraint)def a. b.

Every constraint has a unique focus. A constraint’s focus is identified by the universally quantified argument.

Crowhurst and Hewitt’s use of b-conjunction is exemplified by their analysis of alignment effects in the stress system of Diyari (Pama-Nyungan; Austin 1981). In Diyari, any morpheme of two or more syllables has initial stress, and secondary stress falls on the penult in morphemes longer than three syllables, as shown in

Megan J. Crowhurst

18

(35a). The cases of special interest are polymorphemic words containing monosyllabic suffixes, exemplified in (35b). Monosyllabic suffixes are never stressed. Moreover, no syllable preceding any morpheme is ever stressed (cf. [’puljudu-Ki‘maÍa], [’maÕa-Îa-Ki]). Thus, although they have the same number of syllables, the forms [’maÕa-Îa-Ki] and [’kaJa-‘waÈa] are assigned stress quite differently: the bisyllabic suffix [-’waÈa-] has stress, but the sequence of suffixes [-Îa-Ki-] does not.7 (35)

a.

b.

’kaJa ’pinadu ’Ianda‘walka ’wintara‘naja ’kaJa-‘waÈa ’taji-‘jati‘maji ’maÕa-Îa-Ki ’puljudu-Ki-‘maÍa ’pinadu-‘waÈa ’Janda-na-‘maÍa

(’ka.Ja) (’pi.na)du (’Ian.da)(‘wal.ka) (’winta)ra(‘naja) (’ka.Ja)-(‘wa.Èa) (’ta.ji)-(‘ja.ti)(‘ma.ji) (’ma.Õa)-Îa-Ki (’pu.lju)du-Ki-(‘ma.Ía) (’pi.na)du-(‘wa.Èa) (’Jan.da)-na-(‘ma.Ía)

‘man’ ‘old man’ ‘to close’ ‘how long’ ‘man-pl’ ‘to eat-opt’ ‘hill-char-loc’ ‘mud-loc-ident’ ‘old man-pl’ ‘hit-part-ident’

The lynchpin of Crowhurst and Hewitt’s account of Diyari is the conjunction in (36c) of the two alignment constraints in (36a) and (36b). The individual constraints require an edge (left or right) of a morpheme to be aligned with the same edge of a foot. In conjunction, they require each edge of a morpheme to be aligned with the same edge of a foot. (36)

a.

b.

c.

MorphemeFt-Left (MFL) Align-L(Morpheme, Ft) (One * for any morpheme whose left edge does not coincide with the left edge of some foot.) MorphemeFt-Right (MFR) Align-R(Morpheme, Ft) (One * for any morpheme whose right edge does not coincide with the right edge of some foot.) MFL ûMORPH MFR (One * per morpheme which fails MFL, MFR, or both.)

Crowhurst and Hewitt treat Diyari as a language that avoids foot structure when possible. They rank the standard constraint Parse-q below *Struc(Ft), which assigns a penalty for each foot in the output. The conjunction in (36c) outranks *Struc(Ft), resulting in a foot at each edge of morphemes whose syllable count makes this possible. FtMin(q) prevents the assignment of monosyllabic feet. Tableau (37) shows how the correct outputs (’winta)ra(‘naja) and (’ta.ji)-(‘ja.ti)(‘ma.ji) are selected under the rankings FtMin >> MFL ûMORPH MFR >> *Struc(Ft). (In the following tableaux, violations of the individual conjuncts that are promoted to 7

Note that the underived pentasyllabic form [’wintara‘naja] has initial and penultimate (not antepenultimate) stress. Hewitt and Crowhurst (1996) use this example (from Austin 1981) to show that stress in Diyari is not assigned from left to right, as many theoretical accounts of Diyari have assumed (e.g. Poser 1989; Crowhurst 1994; Kager 1994). Rather, Diyari assigns a trochaic foot at both edges of a morpheme when it can.

Constraint Conjunction

19

violations of the conjunction are shown in parentheses. Violations of the alignment constraints are shown with a superscript corresponding to the initial phoneme of the morpheme that incurred the penalty.) (37)

/taji-jatimaji/

FtMin MFL ûMORPH MFR *Struc(Ft)

☞ a. (’ta.ji)-(‘ja.ti)(‘ma.ji) b. (’ta.ji)-ja.ti(‘ma.ji) c. (’ta.ji)-(‘ja.ti).ma.ji d. (’ta.ji)-ja.ti.ma.ji /wintaranaja/

(*j)

*j!

(*j)

*j! *j!

*** ** (*j) (*j)

** *

FtMin MFL ûMORPH MFR *Struc(Ft)

☞ a. (’winta)ra(‘naja) b. (’win.ta)ra.na.ja

(*w)

*w! *w!

c. win.ta.ra(‘na.ja)

** ** (*w)

*

The crucial cases are those that contain monosyllabic affixes. In [’Jan.da-na-‘ma.Ía], the monosyllabic suffix [-na-] is not footed, but the two bisyllabic morphemes are aligned at each edge with a foot. The especially interesting case is [’ma.Õa-Îa-Ki], which has adjacent monosyllabic suffixes. The candidate (’ma.Õa)-(‘Îa)-(‘Ki), which satisfies the alignment constraints, is ruled out by FtMin. If the constraints MFL and MFR were not conjoined but were evaluated independently, then of the remaining two, the optimal candidate should be (’ma.Õa)-(‘Îa-Ki), which minimizes violations of MFL and MFR. However, the optimal candidate is in fact (’ma.Õa)Îa-Ki, in which neither of the suffixes is aligned with foot structure. When MFL and MFR are b-conjoined, however, a morpheme aligned at only one edge is no better than a morpheme aligned at both edges, so that (’ma.Õa)-(‘Îa-Ki) and (’ma.Õa)-Îa-Ki both fail the conjunction MFL ûMORPH MFR. In this case, *Struc(Ft) decides for the candidate (’ma.Õa)-Îa-Ki, in which the monosyllabic suffixes are unfooted. The analysis of [’ma.Õa-Îa-Ki] under b-conjunction is shown in (38). (38)

/maÕa-Îa-Ki/

FtMin MFL

MFR *Struc(Ft)

(*Î *K) *Î *K (*Î *K) (*K) *Î *K (*Î)

☞ a. (’ma.Õa)-Îa-Ki b. (’ma.Õa)-(‘Îa-Ki) c. (’ma.Õa)-(‘Îa)-(‘Ki)

MORPH

*!*

* * ***

The analysis just presented for [’ma.Õa-Îa-Ki] predicts that trisyllabic roots should not be aligned with foot structure, yet, like all other root morphemes, they have initial stress. Crowhurst and Hewitt’s analysis attributes the presence of initial stress to the constraint MainStress-L in (39), which insists that the head of the prosodic word stand exactly at the left edge of the prosodic word. Ranking MainStress-L above the conjunction MFL ûMORPH MFR compels the presence of an initial foot, even though that foot is poorly aligned with the root morpheme in the output.

Megan J. Crowhurst

20 (39)

MainStress-L (MSL) Align(PrWd, L; Head(PrWd), L) (Every PrWd is left aligned with the stressed syllable in its head foot; one * per misaligned PrWd.)

B-conjunction is also useful in stating templatic constraints on the size of reduplicants (chapter 100: reduplication). A constraint like Red=Ft, for example, can be stated as in (40a). However, note that this statement informally assumes the interpretation provided by b-conjunction: any reduplicant must be both left aligned and right aligned with the same foot. Under b-conjunction, the templatic effect is achieved by b-conjoining the constraints in (40b) and (40c), with the foot as the focus of the conjunction. Under this analysis, any foot will be evaluated against the conjunction. Any foot not co-extensive with Red will fail the conjunction, although it might be required by higher-ranking constraints on metrical structure. However, the constraint will be satisfied by reduplicated forms in which Red and a foot are co-extensive. (40)

a.

Red=Ft The reduplicant string is co-extensive with a foot (e.g. Downing 2000: 3). b. FootRed-L Align-L(Ft, AffixRED) One * for any reduplicant that is not left aligned with a foot. c. FootRed-R Align-R (Foot, AffixRED) One * for any reduplicant that is not right aligned with a foot. d. FootRed-L ûFOOT FootRed-R

The connective “û” is also used by Downing (1998, 2000) to account for exceptional alignment behavior found with onsetless syllables (e.g. as stress and tone bearers) and prosodic restrictions on reduplication in Kinande.

3.2

Implication

A different form of constraint combination, most closely related to material implication in logic, was proposed by Crowhurst and Hewitt (1997) as a way of analyzing effects of conflicting directionality (Kiparsky 1973; Zoll 1997) in the assignment of stress. As an example, Dongolese Nubian (Armbruster 1960) assigns a single stress, which surfaces on the rightmost of heavy syllable, if there are any (where only vowel length is relevant for syllable weight). Examples are shown in (41a). However, when only light syllables are present, as in (41b), stress falls on the initial syllable. (41)

a.

Heavy syllables ’bee.kat.t> do.’goo.g>r te.le.’graaf.k> t>n.t>.’neeI.ke.g>d maa.’suu.ra maa.’leeœ se.ree.g>r.œug.lee.re.’daag

‘to be killed’ ‘raise it’ ‘a telegram’ ‘their maternal aunt’ ‘tube, pipe’ ‘it doesn’t matter’ ‘be in the situation of having worked well’

Constraint Conjunction b.

Light syllables ’bu.run ’ta.ra.ga ’mu.go.san ’–>.J>.ran

21

‘it is a girl’ ‘page, leaf’ ‘tell to leave’ ‘tell him (her) to go and wait’

Crowhurst and Hewitt’s analysis of Dongolese uses the constraints in (42). The pattern found in words containing heavy syllables seems to require that the constraint HeavyHead in (42a) be ranked above Heads-R in (42b), which must in turn be ranked above Heads-L, (42c). (42)

a.

b.

c.

HeavyHead The head syllable of a foot is bimoraic. (One * for any stressed light syllable.) Heads-R Align-R(Head(Ft), PrWd) (One * per syllable coming between any stressed syllable and the right edge of the dominating PrWd.) Heads-L Align-L(Head(Ft), PrWd) (One * per syllable coming between any stressed syllable and the left edge of the dominating PrWd.)

The problem is that the ranking required for the heavy syllable cases doesn’t account for stress in forms containing only light syllables: these seem to require that the alignment constraints be ranked in the reverse order, Heads-L >> Heads-R. The insight Crowhurst and Hewitt (1997) seek to capture is that in a language like Dongolese, heavy syllable stress and any conditions imposed on heavy syllables under stress (in this case, Heads-R) take priority. Heads-L becomes relevant only when no heavy syllables are present. Crowhurst and Hewitt propose that HeavyHead and Heads-R combine in a complex constraint that takes the form of an implication, in which the satisfaction of one requirement is unilaterally dependent on the satisfaction of another. Under their interpretation of material implication, whether a candidate passes A > B depends primarily on the candidate’s success on constraint A, and secondarily on its success on constraint B. Crowhurst and Hewitt’s definition of implication appears in (43).8 (43)

Implication (Crowhurst and Hewitt 1997) If Cand passes A, then Cand is evaluated with respect to B: If Cand passes B, then Cand passes A > B; If Cand fails B, then Cand fails A > B. If Cand fails A, then Cand fails A > B and Cand’s success on B is irrelevant.

As in the case of conjunction, Crowhurst and Hewitt propose that only constraints which share an argument may combine to form implications. Crowhurst 8

Note that Crowhurst and Hewitt’s use of implication does not have the semantics of classical material implication in Boolean logic.

Megan J. Crowhurst

22

and Hewitt’s account of Dongolese ranks the implication HeavyHead >HEAD(FT) Heads-R above Heads-L. Tableau (44) shows how their analysis works in forms like [’mu.go.san], which contain only light syllables. All viable candidates violate HeavyHead (constraint A), and this translates to one violation each against the implication HeavyHead >HEAD(FT) Heads-R. (Heads-R cells have been shaded to indicate that, per Crowhurst and Hewitt’s definition, this constraint has no effect on the outcome.) So it is that in forms containing only light syllables, Heads-L selects the initially stressed candidate. /mugosan/

(44)

HeavyHead >HEAD(FT) Heads-R Heads-L

☞ a. (’mu.go)san b. mu(’go.san)

(*) (*)

* *

c. mu.go(’san)

(*)

*

(**) (*)

*! *!*

Tableau (45) shows how the analysis selects the correct result in forms containing long vowels, [maa.’suu.ra]. /maasuura/

(45)

HeavyHead >HEAD(FT) Heads-R Heads-L

☞ a. maa(’suu)ra b. (’maa)suu.ra c. (‘maa)(’suu)ra d. maa.suu(’ra)

(*)

*

*!

(*) (**)

*!

(***)

*

*!

In response to Crowhurst and Hewitt, Balari et al. (2000) argue essentially that Crowhurst and Hewitt’s interpretation of implication should not be allowed. These authors maintain that if complex constraints are to be derived using logical connectives, they should be connectives with classical Boolean semantics. Like Crowhurst and Hewitt, Balari et al.’s analysis of Dongolese uses Heads-L and Heads-R, but they replace HeavyHead with a constraint LightHead in (46a), which requires stress to fall on a monomoraic syllable, and they propose the (classically evaluated) implication in (46b).9 The tableaux for [’mu.go.san] and [maa.’suu.ra] under BVM’s analysis appear in (47).10 (46)

a. b.

LightHead The head q of a foot is monomoraic. LightHead → Heads-L The implication is violated only if LightHead is satisfied and Heads-L is not.

9

Balari et al. use the symbol “→” in the same way as “>” in Boolean logic. Balari et al.’s take on conflicting directionality is very close to that of Zoll (1997). Zoll proposes the constraint below to account for stress in Selkup. Note that Zoll’s constraint is an unacknowledged complex constraint, which, when unpacked, has the semantics of Balari et al.’s implicational constraint. (i) Align-L(’q[, PWd): A light stressed syllable should be word-initial. 10

Balari et al. differ from CH in assuming that Heads-L and Heads-R restricts the foot, not just the stressed syllable. (47) is taken from their work and therefore reflects this difference.

Constraint Conjunction /mugosan/

(47)

LightHead → Heads-L

☞ a. (’mu.go)san b. mu(’go.san)

*!

(*)

c. mu.go(’san)

*!

(**)

/maasuura/ ☞ a. maa(’suu)ra b. (’maa)suu.ra c. (‘maa)(’suu)ra d. maa.suu(’ra)

23

Heads-R *

LightHead → Heads-L (*)

(*)

* **!

(*)

**!*

(*) (*) *!

Heads-R

(**)

In the end, it turns out that neither version of implication is strictly necessary to account for standard effects of conflicting directionality in stress. Conflicting directionality in stress, as in Dongolese, can be analyzed using the local conjunction of the constraints *PkFT/q[ and InitialStress in (48a) and (48b) (taking the domain of the conjunction to be the prosodic word). (48)

a. b.

c.

*PkFT/q[ Avoid stress on monomoraic syllables. InitialStress Align-L(Head(PrWd), PrWd) The head of the PrWd, syllable with main stress, occurs initially in its prosodic word. *PkFT/q[ &PRWD InitialStress

Tableau (49) shows that the hierarchy *PkFT/q[ &PRWD InitialStress >> *PkFT/q[ >> Heads-R correctly accounts for Dongolese. (49)

/mugosan/

*PkFT/q[&PRWDInitialStress *PkFT/q[ Heads-R

☞ a. (’mu.go)san b. mu(’go.san)

*!

* *

c. mu.go(’san)

*!

*

/maasuura/

*PkFT/q[&PRWDInitialStress *PkFT/q[ Heads-R

☞ a. maa(’suu)ra b. (’maa)suu.ra c. maa.suu(’ra)

** *

* **! *!

*

In addition to Balari et al., other arguments in favor of constraints derived using a connective with the semantics of Boolean material implication are presented in Archangeli et al. (1998) and Ìubowicz (2005).

Megan J. Crowhurst

24

4

Different perspectives on the domain of conjunction

One distinguishing feature of work using various forms of conjunction has been researchers’ different assumptions concerning what can constitute D, the domain of a conjunction. As the definition in (3) makes clear, Itô and Mester’s (2003b) view is that D must be instantiated by a member of the set of phonological or morphological constituents. Smolensky’s use of D is similar, but perhaps more inclusive in his use of the feature domain – a “maximal contiguous span of zbearers with a common value [±z]” (see again (13)). We can say that for both Itô and Mester as well as Smolensky, the domain of a local conjunction is identified with a designated node in the phonological or morphological structure, and all material associated with this node (the association possibly being mediated by other intervening nodes). One difference between these authors seems to be that for Smolensky, but perhaps not Itô and Mester (2003b), any phonological feature can be the node that determines a domain. A different perspective on D is found in Hewitt and Crowhurst (1996), Crowhurst and Hewitt (1997), Downing (1998, 2000), Bakovio (2000), and Ìubowicz (2002, 2005), all of whom argue in one way or another that constraints can be conjoined only when they have the same locus of violation. As noted earlier, Hewitt and Crowhurst (1996), Crowhurst and Hewitt (1997), and Downing (1998, 2000) argue that b-conjunction should be limited to constraints that share an argument, which serves as the locus of violation of the conjoined constraint. In Crowhurst and Hewitt’s analysis of Diyari reviewed earlier, for example, the locus of violation of the constraints MorphemeFt-L (MFL) and MorphemeFt-R (MFR) is the morpheme, and the conjunction of these constraints produces a complex constraint, MFL ûMORPH MFR, which has the same locus of violation. An extended formal discussion of the locus of violation as a principled restriction on the domain of local conjunctions of markedness and faithfulness constraints is developed in Ìubowicz (2005). Her definitions of restricted local conjunction and of the locus for local conjunction are shown in (50). (50)

a. b.

Restricted local conjunction C = C1 & C2 is violated iff LOCC1 ∩ LOCC2 ≠ Ø. Locus for local conjunction The locus for local conjunction is the intersection of the sets LOCC1 and LOCC2.

Formally, (50a) states that the loci of violation for the conjuncts of a local conjunction must intersect and that this intersection may not be null; (50b) states that the locus of violation for the local conjunction is the intersection of sets defined in (50a). In a similar vein, Bakovio (2000) proposes that two constraints can be conjoined only if they are co-relevant, meaning that the definition of each conjunct specifies a particular feature also mentioned by the other conjunct. The co-relevance restriction plays an important role in Bakovio’s use of M&F (markedness and faithfulness) conjunctions to account for “assimilation to the unmarked” phenomena, as discussed earlier. In his words, “the net effect of a co-relevant local conjunction of

Constraint Conjunction

25

markedness and faithfulness is to specifically prohibit the unfaithful introduction of a marked segment” (Bakovio 2000: 7). Bakovio notes that the co-relevance restriction (or Ìubowicz’s more precisely defined restriction based on locus of violation) provides an answer to Itô and Mester (1998), who take the position that M&F conjunctions should not be allowed because they lead to undesirable results. Positioning the conjunction NoCoda & Ident[voice] above the fragment *[+voice] >> Ident[voice], for example, would falsely predict a language in which obstruents are voiced only in syllable codas. Bakovio notes that conjuncts used to illustrate Itô and Mester’s point are not co-relevant, and in fact uses the example to reinforce his claim that only co-relevant constraints are conjoinable.

5

Taming the beast: Curbing the expressive power of constraint conjunction

Proposals for combining simpler constraints into more complex ones using connectives other than “>>” have not been universally welcomed (see for example Orgun and Sprouse 1999; Parker 2001; Idsardi 2006; Iverson and Salmons 2007; Zhang 2007). The issue of restricting the power of a formal theory that permits the derivation of complex constraints from simpler ones has generated concern, and inspired various proposals aimed at restrictiveness. The position of some scholars that restrictions on conjunction should be encoded in restrictions on the domain of conjunction was discussed in the last section. Other ways in which the issue of restrictiveness has played into arguments for and against constraint conjunction are reviewed below.

5.1

Chain shifts: An argument against material implication

Wolf (2007) calls for restricting the set of connectives that can be used to combine constraints. He asserts that local conjunction and strict domination exhaust the ways in which constraints may interact. In particular, he argues that complex constraints with the semantics of material implication must not be allowed, because admitting such constraints would have the consequence of allowing OT to model synchronic circular chain shifts. A circular chain shift takes the form A → B → A (where B can be expanded to accommodate additional links in the chain), in which a phoneme /A/ shifts to sound [B], and phoneme /B/, also present in the language, shifts to [A] (see also chapter 73: chain shifts). The existence of a circular chain shift in which all links occur synchronically would present a problem for the OT doctrine of harmonic ascent (Prince and Smolensky 1993; McCarthy 2002). This is because a change A → B could only occur if B is less marked than A (and this would require a particular set of markedness constraints to outrank a particular set of faithfulness ones). But then, if B is less marked than A, there would be no reason for the change B → A (and the ranking that accounted for A → B could not account for the second change). Moreton (1999) provides a formal proof showing that an OT grammar that admits only faithfulness and markedness constraints is incapable of modeling circular chain shifts. As it turns out, there are no convincing synchronic examples of a circular chain

26

Megan J. Crowhurst

shift conditioned by purely phonological factors, although there do appear to be examples triggered by morphological conditions (Anderson and Brown 1973). Diachronic examples of phonologically conditioned circular chain shifts are quite common (e.g. the flip-flop /q/ → /i/ → /q/ in Siriono: Crowhurst 2000; the Germanic Kreislauf: Iverson and Salmons 2008), but these might be best handled by an analysis that assumes changes in constraint rankings at various stages of a language’s development, along the lines proposed in Holt (2003).

5.2

Other proposals for promoting restrictiveness

Several other proposals aimed at promoting restrictiveness should be mentioned. Some researchers have proposed restrictions (beyond the locus of violation) on the kinds of constraints that can combine. As noted earlier, Itô and Mester (1998) argued against the conjunction of markedness with faithfulness constraints. Fukazawa and Miglio (1998) proposed that conjunction should be limited to constraints within the same family, which would have much the same effect. These proposals have been countered by persuasive demonstrations of the benefits of conjoining faithfulness and markedness constraints (see §2.5). A commonly accepted restriction has been to allow only complex constraints whose parts are independently necessary constraints. To do otherwise would permit unnatural results. Smolensky (2006) provides the example of a rule spreading [+ATR] from a [−high] vowel (e.g. [o . . . e] → [D . . . e]). This pattern is exactly what we don’t find in ATR harmony, because [−ATR] non-high vowels are marked, and only unmarked feature bearers propagate their features (Archangeli and Pulleyblank 1994). Smolensky (2006) notes that ranking the conjunction *[+ATR, +hi] & HD-L[+ATR] above F[ATR] (the constraint requiring a feature domain to be faithful) could produce this result. However, the conjunction *[+ATR, +hi] & HD-L[+ATR] could simply never exist; one of its conjuncts *[+ATR, +hi] is not a member of Con, since from a markedness perspective, [+ATR] favors high vowels. However pernicious, conjunctive relations of this type have in fact been proposed in the literature, sometimes to account for very real phenomena. An example would be the implication in (46b), used by Balari et al. (2000) to account for directional stress effects. Their constraint LightHead, which demands that stressed syllables be light, has no independent motivation – it is in fact an anti-harmonic constraint (since stress on light syllables is less harmonic than stress on heavy syllables). If LightHead is not a plausible candidate for membership in Con, then it would seem reasonable to conclude that Balari et al.’s conjunction LightHead → Heads-L is not an admissible complex constraint. Another example would be the constraint *Lapse in (51), employed by Elenbaas and Kager (1999: 282), who are proponents of the view that only independently motivated constraints may be combined. (51)

*Lapse Every weak beat must be adjacent to a strong beat or the word edge.

Elenbaas and Kager note the disjunction in the requirement imposed by (51), but they point out that one of the disjuncts has no justification as a constraint in its own right. The first requirement, “every weak beat must be adjacent to a strong beat,” is a cross-linguistically common restriction. However, Elenbaas and Kager note that there is no clear and independent motivation for the second requirement, that

Constraint Conjunction

27

“every weak beat must be adjacent to the word edge.” For this reason, they conclude that *Lapse should not be treated as a complex constraint. However, whether acknowledged or not, *Lapse is a form of disjunction (local conjunction), and a proper definition of the constraint would have to take account (more on this below). *Lapse is more problematic than the previous example, because the need for *Lapse is well established: as we showed in the last section, the Balari et al. implication is unnecessary; alternatives are readily available.

6

Concluding remarks

Proponents of allowing devices for combining constraints have argued that they further the goal of deriving empirical generalizations about markedness from irreducible principles (Itô and Mester 2003b; Smolensky 2006) and (a related point) that they promote greater precision and formal explicitness. Resistance to accepting complex constraints has been due largely to the observation that despite the advantages claimed, devices for combining constraints greatly increase the expressive power of the formal architecture, especially in regard to the constraint inventory Con. On the other hand, Itô and Mester (2003b: 22) note that in fact, the use of connectives keeps in check the proliferation of inexplicit constraints that duplicate aspects of others without a formal account of their components and the relations among them. (That is, since the need for the constraints in (6a) and (6b) is uncontroversial, and if we can combine them as in (4c), then we don’t need the coda condition in (5).) In closing, I will put forward a consideration that may seem provocative, but which ought to be taken seriously. Smolensky (2006) points out that the statement normally given for a constraint such as (51) (of which there are many examples in the literature) may pass as a translation of the definition of a constraint into English (or pseudo-formal language blended with English), but it is not a proper definition of that constraint (i.e. stated in the precise language of formal logic). A precise definition of (51) in formal language must necessarily take account of the disjunction it contains; it cannot do otherwise. Whether complex constraints should be allowed is a non-issue. In fact, they have been used without controversy from the inception of OT, camouflaged as informal descriptions that have been accepted as definitions of constraints. The issue that deserves consideration is whether the position that complex constraints should be excluded can be maintained in light of its implications – most seriously, the implication that an enormous number of generally accepted constraints ought to be expelled from Con, or at least rethought. On the issue of overgeneration, we may perhaps leave the last word, for now, to Itô and Mester: The broadly defined outline of local conjunction theory . . . admits a huge number of conjoined constraints, only a small subset of which will turn out to play a role in grammar, and many of which are unwanted. In our view, the task of distinguishing between “reasonable,” “plausible,” “expected” conjunctions and “unreasonable,” “implausible,” “unexpected” conjunctions cannot be relegated to the syntax of conjunction, which simply provides a system for expressing derived constraints. The distinction is an issue of phonological substance and phonetic groundedness, not one of formalization (Itô and Mester 2003b: 24).

28

Megan J. Crowhurst

ACKNOWLEDGMENTS The author appreciates useful comments received from the editors, Marc van Oostendorp and Keren Rice, and from Scott Myers, Joe Salmons, and two anonymous reviewers. Their input has helped to make this a better chapter.

REFERENCES Alderete, John. 1999. Faithfulness to prosodic heads. In Ben Hermans & Marc van Oostendorp (eds.) The derivational residue in phonology, 29–50. Amsterdam & Philadelphia: John Benjamins. Alderete, John. 2004. Dissimilation as local conjunction. In John J. McCarthy (ed.) Optimality Theory in phonology: A reader, 394–406. Malden, MA & Oxford: Blackwell. Anderson, Stephen R. & Wayles Browne. 1973. On keeping exchange rules in Czech. Papers in Linguistics 6. 445–482. Archangeli, Diana & Douglas Pulleyblank. 1994. Grounded phonology. Cambridge, MA: MIT Press. Archangeli, Diana, Laura Moll & Kazutoshi Ohno. 1998. Why not *NY? Papers from the Annual Regional Meeting, Chicago Linguistic Society 34. 1–26. Armbruster, Charles Hubert. 1960. Dongolese Nubian: A grammar. Cambridge: Cambridge University Press. Austin, Peter. 1981. A grammar of Diyari, South Australia. Cambridge: Cambridge University Press. Baertsch, Karen. 1998. Onset sonority distance constraints through local conjunction. Papers from the Annual Regional Meeting, Chicago Linguistic Society 34. 1–16. Baertsch, Karen & Stuart Davis. 2003. The split margin approach to syllable structure. ZAS Papers in Linguistics 32. 1–14. Bakovio, Eric. 2000. Harmony, dominance, and control. Ph.D. dissertation, Rutgers University. Balari, Sergio, Rafael Marín & Teresa Vallverdu. 2000. Implicational constraints, defaults, and markedness. Unpublished ms., Universitat Autònoma de Barcelona (ROA-396). Bavin Woock, Edith & Michael Noonan. 1979. Vowel harmony in Lango. Papers from the Annual Regional Meeting, Chicago Linguistic Society 15. 20–29. Beckman, Jill N. 2003. The case for local conjunction: Evidence from Fyem. Proceedings of the West Coast Conference on Formal Linguistics 22. 56–69. Cassimjee, Farida & Charles W. Kisseberth. 1998. Optimal Domains Theory and Bantu tonology: A case study from Isixhosa and Shingazidja. In Larry M. Hyman & Charles W. Kisseberth (eds.) Theoretical aspects of Bantu tone, 33–132. Stanford: CSLI. Cole, Jennifer & Charles W. Kisseberth. 1994. Nasal harmony in Optimal Domains Theory. Unpublished ms., University of Illinois. Crowhurst, Megan J. 1994. Prosodic alignment and misalignment in Diyari, Dyirbal, and Gooniyandi: An optimizing approach. Proceedings of the West Coast Conference on Formal Linguistics 13. 16–31. Crowhurst, Megan J. 2000. A flip-flop in Sirionó (Tupian): The mutual exchange of /i q/. International Journal of American Linguistics 66. 57–75. Crowhurst, Megan J. 2001. Coda conditions and um infixation in Toba Batak. Lingua 111. 561–590. Crowhurst, Megan J. & Mark Hewitt. 1997. Boolean operations and constraint interactions in Optimality Theory. Unpublished ms., University of North Carolina & Brandeis University (ROA-229). Downing, Laura J. 1998. On the prosodic misalignment of onsetless syllables. Natural Language and Linguistic Theory 16. 1–52.

Constraint Conjunction

29

Downing, Laura J. 2000. Morphological and prosodic constraints on Kinande verbal reduplication. Phonology 17. 1–38. Downing, Laura J. 2001. Liquid spirantisation in Jita. Malilime: Malawian Journal of Linguistics 2. 1–27. Elenbaas, Nine & René Kager. 1999. Ternary rhythm and the lapse constraint. Phonology 16. 273–329. Fukazawa, Haruka. 1999. Theoretical implications of OCP effects on features in Optimality Theory. Ph.D. dissertation, University of Maryland at College Park (ROA-307). Fukazawa, Haruka. 2001. Local conjunction and extending sympathy theory: OCP effects in Yucatec Maya. In Linda Lombardi (ed.) Segmental phonology in Optimality Theory: Constraints and representations, 231–260. Cambridge: Cambridge University Press. Fukazawa, Haruka & Viola Miglio. 1998. Restricting conjunction to constraint families. Proceedings of the Western Conference on Linguistics 9. 102–117. Hall, T. A. 2004. German glide formation and constraint conjunction. Unpublished ms., University of Indiana. Hall, T. A. 2007. German glide formation and its theoretical consequences. The Linguistic Review 24. 1–31. Hewitt, Mark & Megan J. Crowhurst. 1996. Conjunctive constraints and templates in Optimality Theory. Papers from the Annual Meeting of the North East Linguistic Society 26. 101–116. Holt, D. Eric (ed.) 2003. Optimality Theory and language change. Dordrecht: Kluwer. Idsardi, William J. 2006. A simple proof that Optimality Theory is computationally intractable. Linguistic Inquiry 37. 271–275. Itô, Junko & Armin Mester. 1996. Rendaku 1: Constraint conjunction and the OCP. Paper presented at the Kobe Phonology Forum. Itô, Junko & Armin Mester. 1998. Markedness and word structure: OCP effects in Japanese. Unpublished ms., University of California, Santa Cruz (ROA-255). Itô, Junko & Armin Mester. 2003a. On the sources of opacity in OT: Coda processes in German. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 271–303. Cambridge: Cambridge University Press. Itô, Junko & Armin Mester. 2003b. Japanese morphophonemics: Markedness and word structure. Cambridge, MA: MIT Press. Iverson, Gregory K. & Joseph C. Salmons. 2007. Domains and directionality in the evolution of German final fortition. Phonology 24. 121–145. Iverson, Gregory K. & Joseph Salmons. 2008. Germanic aspiration: Phonetic enhancement and language contact. Sprachwissenschaft 33. 257–278. Jacobs, Haike. 2004. Rhythmic vowel deletion in OT: Syncope in Latin. Probus 16. 63–89. Kager, René. 1994. Generalized alignment and morphological parsing. Unpublished ms., University of Utrecht. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kawahara, Shigeto. 2003. On a certain type of hiatus resolution in Japanese. Phonological Studies 6. 11–20. Kiparsky, Paul. 1973. “Elsewhere” in phonology. In Stephen R. Anderson & Paul Kiparsky (eds.) A Festschrift for Morris Halle, 93–106. New York: Holt, Rinehart & Winston. Kirchner, Robert. 1993. Turkish vowel harmony and disharmony: An optimality theoretic analysis. Unpublished ms., University of California, Los Angeles (ROA-4). Kirchner, Robert. 1996. Synchronic chain shifts in Optimality Theory. Linguistic Inquiry 27. 341–350. Levelt, Clara C. & Ruben van de Vijver. 1998. Syllable types in cross-linguistic and developmental grammars. Paper presented at the 3rd Biannual Utrecht Phonology Workshop (ROA-265). Levelt, Clara C., Niels O. Schiller & Willem J. Levelt. 1999. A developmental grammar for syllable structure in the production of child language. Brain and Language 68. 291–299.

30

Megan J. Crowhurst

Lin, Hui-Shan. 2000. Layered OCP, unparsed preposition, and local constraint conjunction in Mandarin tone sandhi. Paper presented at the 33rd International Conference on Sino-Tibetan Languages and Linguistics, Ramkhamhaeng University, Thailand. Ìubowicz, Anna. 2002. Derived environment effects in Optimality Theory. Lingua 112. 243–280. Ìubowicz, Anna. 2005. Locality of conjunction. Proceedings of the West Coast Conference on Formal Linguistics 24. 254–262. McCarthy, John J. 2002. A thematic guide to Optimality Theory. Cambridge: Cambridge University Press. Moreton, Elliott. 1999. Non-computable functions in Optimality Theory. Unpublished ms., University of Massachusetts, Amherst (ROA-364). Moreton, Elliott & Paul Smolensky. 2002. Typological consequences of local constraint conjunction. Proceedings of the West Coast Conference on Formal Linguistics 21. 306–319. Morris, Richard. 2002. Coda obstruents and local constraint conjunction in north-central Peninsular Spanish. In Teresa Satterfield, Christina Tortora & Diana Cresti (eds.) Current issues in Romance languages: Selected papers from the 29th Linguistic Symposium on Romance Languages, 207–224. Amsterdam & Philadelphia: John Benjamins. Noonan, Michael. 1992. A grammar of Lango. Berlin & New York: Mouton de Gruyter. Okello, Jenny. 1975. Some phonological and morphological processes in Lango. Ph.D. dissertation, Indiana University. Orgun, Cemil Orhan & Ronald L. Sprouse. 1999. From MParse to Control: Deriving ungrammaticality. Phonology 16. 191–224. Parker, Steve. 2001. Non-optimal onsets in Chamicuro: An inventory maximized in coda position. Phonology 18. 361–386. Poser, William J. 1989. The metrical foot in Diyari. Phonology 6. 117–148. Prince, Alan. 2001. Invariance under re-ranking. Paper presented at the 20th West Coast Conference on Formal Linguistics, University of Southern California. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Reynolds, William. 1994. Variation and phonological theory. Ph.D. dissertation, University of Pennsylvania. Smolensky, Paul. 1993. Harmony, markedness, and phonological activity. Paper presented at the Rutgers Optimality Workshop 1, Rutgers University (ROA-87). Smolensky, Paul. 1995. On the structure of the constraint component Con of UG (ROA-86). Smolensky, Paul. 1997. Constraint interaction in generative grammar II: Local conjunction, or random rules in universal grammar. Paper presented at the Hopkins Optimality Theory Workshop/University of Maryland Mayfest 1997. Smolensky, Paul. 2006. Optimality in phonology II: Harmonic completeness, local constraint conjunction, and feature domain markedness. In Paul Smolensky & Géraldine Legendre (eds.) The harmonic mind: From neural computation to optimality-theoretic grammar, vol. 2: Linguistic and philosophical implications, 27–160. Cambridge, MA: MIT Press. Spaelti, Philip. 1997. Dimensions of variation in multi-pattern reduplication. Ph.D. dissertation, University of California, Santa Cruz. Walker, Rachel. 2000. Nasal reduplication in Mbe affixation. Phonology 17. 65–115. Wolf, Matthew. 2007. What constraint connectives should be permitted in OT? University of Massachusetts Occasional Papers in Linguistics 36. 151–179. Zhang, Jie. 2007. Constraint weighting and constraint domination: A formal comparison. Phonology 24. 433–459. Zoll, Cheryl. 1997. Conflicting directionality. Phonology 14. 263–286. Zoll, Cheryl. 1998. Positional asymmetries and licensing. Unpublished ms., MIT (ROA-282).

63

Markedness and Faithfulness Constraints Paul de Lacy

1

Introduction

Objects and mechanisms called “constraints” have featured in many theories of the phonological and syntactic modules. However, the explicit bifurcation into “markedness and faithfulness” constraints is specifically found in Optimality Theory (OT; Prince and Smolensky 1993) and its developments (especially McCarthy and Prince 1999), as well as in theories based on OT (Stochastic OT: Boersma and Hayes 2001; Targeted Constraint Theory: Wilson 2001; OT with Candidate Chains: McCarthy 2007; Stratal OT: Bermúdez-Otero, forthcoming; Kiparsky, forthcoming). So this chapter focuses on the things called “constraints” in OT (specifically the “classical OT” of Prince and Smolensky 1993 and McCarthy and Prince 1999). In particular, it focuses on OT constraints in the phonological module; there are also OT theories of the syntactic module and OT theories of morphology – they will not be discussed here. This chapter’s aim is to examine the basic syntax and semantics of constraints. On the syntax side: What is the form of constraints? What is the “constraint construction language”? On the semantics side: How are constraints “interpreted” – i.e. how are constraints used to assess a candidate’s violation marks? This chapter focuses on the basics of constraints, so it does not aspire to identify every constraint theory or list every constraint and constraint generator (§4.2) that has been proposed; for that, see the ongoing ConCat project.1 An OT constraint is commonly treated as a function that takes a candidate and returns “violation marks” (see Prince and Smolensky 1993). Violation marks are discrete elements; they are usually written as a string of asterisks with one asterisk per unique element (but violation marks in their formal implementation are not necessarily a string). For example, a constraint *Dorsal returns one violation mark for each instance of the representational element [dorsal] in an output representation (constraint names are usually written in small capitals). So for a candidate that includes an output representation [pakak], *Dorsal returns **, because there are two [dorsal] features in the form (one for each [k]). “Candidates” are sets of 1

Available at http://concat.wiki.xs4all.nl.

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0063

2

Paul de Lacy

forms including at least an output and input representation, but probably many other related representations as well as relations between them (§3.3; Prince and Smolensky 1993). Constraints fit into the overall phonological system thus (following Prince and Smolensky 2004). An input is drawn from the lexicon; the input consists of phonological material with morphological and syntactic annotation/structure. A generation mechanism (GEN) produces many (perhaps an infinite number of) candidates. One or more candidates is selected from the array; the selection process (EVAL) involves constraints generating violation marks for candidates and an algorithm that uses the violation marks and other factors to determine the winning candidate(s). EVAL refers to “ranking” – a total order on constraints. Ranking does not influence how violation marks are calculated; however, ranking is crucial in discovering the winning candidate. The phonetic module then takes (one of) the winner(s) and realizes it (i.e. converts it into articulatory movements that produce speech sound). In short, constraints are just one part of many mechanisms that work together to determine the winning output representation. Constraints do not determine winners on their own. The term “markedness and faithfulness” as applied to constraints was coined in Prince and Smolensky (1993: §1.4). “Markedness constraints” return violation marks based solely on the form of the output representation. *Dorsal above is a markedness constraint. Unfortunately, the term “markedness” can cause confusion because it seems to imply an inherent connection to theories of markedness (see chapter 4: markedness). However, theories of markedness are expressed in OT via both markedness and faithfulness constraints (e.g. de Lacy 2006). The term “output constraint” is therefore less confusing than “markedness constraint,” and I will use it here. However, the phrase “markedness constraint” is in such widespread use that I fear “output constraint” will never catch on, in spite of my efforts in this chapter (to add to the confusion there are constraints called “output–output constraints,” which are actually faithfulness constraints; see §3.3). As originally used, “faithfulness constraints” are those that return violation marks based on comparison of the output representation with the input (Prince and Smolensky 1993: §1.2; though strictly speaking the output includes the input, as discussed in §3.1). Later work, especially McCarthy and Prince (1999), broadens the term to include any constraint that assigned violations by comparing any pair of inter- or intra-representational forms (e.g. the base of reduplication and the reduplicant (McCarthy and Prince 1999); the derivational base and the output (Benua 1997); a designated form and the output (McCarthy 1999)). The majority of work in OT now uses McCarthy and Prince’s Correspondence Theory, so in these cases it is accurate to refer to “correspondence constraints” – i.e. those that use correspondence relations in their calculation of violation marks. However, non-correspondence faithfulness constraints exist in some versions of OT (e.g. containment theories – §3.1), so “faithfulness constraints” is still a usefully broad term. This chapter focuses on a few important issues about constraints. §2 discusses constraint form in output constraints: What are constraints made of, and how do they return violation marks? §3 deals with faithfulness and correspondence constraints. §4 discusses the source of constraints – whether they are innate and how/whether they relate to external sources.

Markedness and Faithfulness Constraints

1.1

3

“Constraint” in other theories

The term “constraint” is used in many different ways in many different theories. In some rule-based theories there are objects called “constraints” or “filters” that – if their conditions are met – doom the derivation or block rules from applying. If an input I undergoes a series of rules to create a representation z and there is a filter *z, the derivation is doomed (i.e. input I has no corresponding output). See Chomsky and Lasnik (1977) for early examples in syntax, and Itô (1986) for conditions on syllable structure. The filter/condition concept does not have a direct analogue in OT – OT constraints assign violations; EVAL is the source of (relative) doom for candidates. Occasionally, “constraint” is used to refer to the side-effects of conditions on representational primitives and to restrictions on the algorithm that generates output candidates (GEN). Obviously, output representations can only be constructed out of objects and relations that are available (i.e. prosodic nodes, features, planes, tiers; precedence, dominance). For example, there is no candidate in which a node can both precede and follow another node, because the phonological precedence relation is asymmetric (i.e. if a and b are on the same tier and a < b (i.e. a precedes b) and b < a then a = b) (see chapter 34: precedence relations in phonology for a discussion of precedence). One could informally call the asymmetry of phonological precedence a “constraint,” but it is not an OT constraint.

2

Output constraints

A constraint takes a candidate and returns violation marks.2 For example, the constraint *Dorsal returns one violation mark for [ka], two for [kax], and so on. A constraint’s violation assignment can be described in informal terms: e.g. “*k returns a violation mark for each [dorsal] segment.” This informal description is useful, but far from being a formal definition. A formal definition of a constraint must be couched in a Constraint Definition Language (CDL). A comprehensive CDL specifies representational primitives and relations and restrictions on their combination in constraints. The same distinction can be made for rule-based theories like Chomsky and Halle (1968; SPE). Suppose we observe a rule R that takes an input representation /ak/ and converts it to the representation [a?]. R could be described as “change /k/ into a [?] word-finally.” However, R must be defined in terms of a Rule Definition Language (RDL); an RDL is the elements and relations that can be used to construct a rule and limits on their combination. Most of Chomsky and Halle (1968) is devoted to developing such an RDL; R is defined as /k/ → [?] / __ [−seg, −FB, +WB] (the rightmost cluster of features is a word boundary).

2

It is common to see comments like “constraints impose a partial order on the candidate set” (Samek-Lodovici and Prince 1999: 9), “this constraint dooms the candidate,” and so on. These comments are meant as a quick way of describing the complex process of determining order among candidates; the process involves constraints, VR (§2.2), ranking, and Eval’s mechanisms; constraints are merely one part of the process of establishing the winning candidate.

4

Paul de Lacy

Some work in OT uses informal descriptions to talk about constraints. Formal objects (representational elements like prosodic nodes, features, etc.) are often mentioned in the informal descriptions, but the constraints are nevertheless not defined in terms of an overarching CDL. There have been attempts to develop a comprehensive CDL (Eisner 1997; Potts and Pullum 2002), but most work has either focused on particular groups of constraints, or treated constraints as “black boxes.” To explain, it is possible to fruitfully investigate some (perhaps many) aspects of OT theories without knowing the precise definition of constraints, but only knowing which violation marks constraints assign in which situations. After all, the winning candidate is not directly determined by constraints, but by their violation marks. So, if the violation marks are known then the winner can be determined – the exact means by which the violation marks came to be assigned is often not crucial. For this reason I believe it is fair to say that there has been less focus on developing a CDL in OT theories than on developing RDLs in rule-based theories like Chomsky and Halle (1968). Even so, there have been detailed proposals of CDLs for groups of OT constraints (see §4 below), and some proposals about aspects of the general CDL. In my own (joint) work, for example, Bye and de Lacy (2000) propose restrictions on how constraints can refer to constituent edges; de Lacy (2006) proposes that constraints cannot include both prosodic nodes and segmental features in their definitions. An explicit CDL is both useful and ultimately essential to a complete optimality theory. A CDL can tell us which constraint formulations are valid, and thus set a bound on which constraints can and cannot exist. The following subsections will discuss a CDL. There is a strong uniformity in constraint descriptions and definitions that suggests broad agreement about certain aspects of the CDL. For expository reasons, I will start with the CDL for output constraints. As a word of warning, due to space limitations I will discuss only a few CDLs, and focus on the basic components of just one. The CDL discussed below deals with a broad set of output constraints that I believe every phonologist would accept as possible constraints. I will not attempt to comprehensively discuss all extant CDLs or aspects of CDLs, but instead focus on basic CDL properties.

2.1

Output constraints: Representation

There is not a lot of explicit discussion about how constraints work in OT. It seems to me that the majority of work in OT treats constraints as functions from candidates to violation marks. Output constraints inspect the output representation in a candidate, and return a string of violation marks. So *Dorsal returns one violation for candidates with an output representation [ka], two for [kag], three for [gaxikan], and so on, leading to the description: “Assign one violation for each dorsal segment.” We are seeking a CDL in which *Dorsal can be formulated. One issue to address is the CDL’s representational primitives. For example, *Dorsal might be cast in terms of a representational theory in which velar consonants are [+back, +high] (Chomsky and Halle 1968: 303), or one in which a Place node dominates a dorsal node which dominates [+back] and [+high] terminal nodes, or one in which an oral cavity node dominates a C-place node which dominates a dorsal terminal node (see Hall 2007 for an overview of feature theories). There are many extant representational theories, but for the purposes of this chapter I will adopt

Markedness and Faithfulness Constraints

5

Clements and Hume’s model (there is no widespread consensus on which representational theory is correct, though; even less now than in 1995, I suspect; see also chapter 27: the organization of features). The CDL also must specify how representational elements can be combined in constraints. For example, a relatively lax CDL could allow several different versions of *Dorsal, as in (1). I use • for “root node” – the lowest node that dominates all segmental features. The symbol ↓ stands for the immediate dominance relation: a↓b means “a immediately dominates b” (i.e. a dominates b and there is no c such that a dominates c and c dominates b); a↓b↓c means “a immediately dominates b and b immediately dominates c.” Dominance is an asymmetric transitive relation that holds between nodes on different autosegmental tiers. The descriptions are given in (1); the violation marks the constraint assigns are shown for [k], [k(] (assuming a one-root geminate theory), and [Ik] (assuming obligatory feature sharing for adjacent elements; Schein and Steriade 1986). (1)

*Dorsal versions Return a violation for . . .

[k]

[k(]

[Ik]

a. each distinct root node • s.t. •↓CPlace↓[dorsal]

*

*

**

b. each distinct [dorsal] feature

*

*

*

c. each distinct prosodic node Ò s.t. p↓•↓CPlace↓[dorsal]

*

**

**3

I do not know which of the constraints in (1) exist. Suppose it turns out that we need only (1a). The non-existence of (1b) and (1c) could be achieved by placing restrictions on the CDL such that all output constraints must refer to a root node in their definition and no constraint may mention both prosodic nodes and segmental features. Every extant theory of representation provides a CDL with a great deal of potentially expressive power. So, it is highly likely that any CDL theory will have to incorporate extensive limitations on permissible representations in constraints; it is probably too hopeful that all limits on constraints will be a side-effect of inherent limitations on representations (see §4). The CDL must also specify how the representation is used to assess violation marks. For example, suppose a constraint mentions the structure [•↓CPlace↓ [dorsal]]. How is this structure used to assess violation marks relative to some candidate? In the constraint description (1a), I assumed that the constraint searches the candidate’s output representation and one violation mark is returned for each distinct structure that has the form [•↓CPlace↓[dorsal]]. However, could there be a constraint which returns one violation regardless of how many [•↓CPlace↓ [dorsal]] structures there were? Such a constraint would return * for [ka], [kax], and [kaxga]. Let’s turn to this issue now.

2.2

Function or representation?

I asserted without comment above that a constraint is a function: i.e. it takes a candidate as an input and returns violation marks. Conceiving of constraints as 3

(1c) could return one violation if [I] and [k] were both dominated by the same [ or q node (e.g. as in [oiIk]q).

6

Paul de Lacy

independent functions opens up the possibility that different constraints could assign violation marks in very different ways. For example, the Align schema from McCarthy and Prince (1993) takes four arguments and assesses violation marks with respect to designated prosodic constituents. The violation marks from Align(Ft, R; PrWd, R) are the sum of the number of syllables between the right edge of each foot and the right edge of the PrWd. So, the constraint returns nine violations for [(qq)(qq)(qq)q] (see McCarthy and Prince 1993: 15–16 for details; to understand this constraint see especially 1993: 10 and definitions 14–16). It is clear that the way in which this Align constraint assesses violation marks is quite different from the way in which violations of *Dorsal are assessed (cf. McCarthy 2003). If constraints are self-contained algorithms that return violation marks and the CDL is sufficiently powerful, we might see pairs of constraints that refer to the same representational structure but differ in how they calculate violation marks. For example, there could be a pair of constraints *Dorsal, where *Dorsal returned only one violation mark regardless of how many [dorsal] features there are in a form, as long as there is at least one (see Wolf 2007a for relevant discussion). The two constraints refer to the same structure – [•↓CPlace↓[dorsal]] – and differ only in terms of how that structure is used to assess violation marks from a candidate. A pair of constraints like this – i.e. that refer to the same representational structure but differ only in their quantification – would be strong evidence that each constraint is an independent algorithm that assigns violations (or at least that there are several groups of constraints that differ in how they assign violations). However, my impression is that the constraints-as-functions approach is too powerful. The output constraints that have been proposed in phonological literature are often very similar: they essentially have the form *R, where R is a representation; one violation mark is assigned for each distinct occurrence of R in a candidate’s output representation. *Dorsal is an example of such a constraint. The apparent uniformity in how constraints assess violations suggests that it is worthwhile considering an alternative theory of constraints in which constraints are not functions but solely representations. In such an approach, there would be a single algorithm, Violation Assigner (VR). VR takes as its input an output constraint and a candidate and returns violation marks. VR works the same way for all constraints, thus imposing uniformity in how violation marks are assigned. So the constraint *Dorsal is really a representation [•↓CPlace↓[dorsal]]; *Dorsal itself does not assess violation marks. There are many ways to formulate a VR algorithm that does the job described above. For example, one could take the set of all sub-representations of a candidate’s output representation and compare each member of the set to a constraint representation; the number of violation marks returned for a particular constraint would be the number of subsets that were equivalent to the constraint’s representation. I will instead discuss a somewhat more efficient algorithm that does a similar job.4

4

See www.pauldelacy.net/VR for software which allows the user to see the VR below in action and try out various constraints and representations.

Markedness and Faithfulness Constraints (2)

7

Violation Assigner (VR): Outline Inputs: – a constraint C; – a candidate that includes an output representation R; Output: – a set of violation marks (a set of unique identifiers) indexed to C and the candidate. a. b.

c.

Take a node c in C. For each node r in R: If r is the same type and value as c, then check whether r is connected to a structure equivalent to C. – if it is, return a violation mark, and continue to the next r. Add no other violation marks to the result.

For example, take a constraint *[+voice], which consists of one [voice] node with a value “+.” Each node in the candidate’s output representation is checked. If a node is a [voice] node and has the value “+,” a violation mark is added to the result. The VR might seem straightforward, but it has interesting complexities, particularly in the procedure that checks whether a node is “connected to a structure equivalent to [the constraint] C.” Take a more complex constraint – one that involves two nodes: e.g. *q[[, “Don’t have bimoraic syllables.” *q[[ has three nodes (q, [1, [2) and three relations (q↓[1, q↓[2, [1 < [2). A node is selected from the constraint (it doesn’t matter which one) – let’s say q in this case. The output representation is searched for q nodes. When one is found, the next step is to check whether q is connected to a structure equivalent to the constraint. The implementation of this checking procedure is that q is checked to see if it is in any of the relations mentioned by the constraint: i.e. does the particular q in the representation dominate two different [ nodes? If it does, then the [ nodes that are dominated by q are checked to see whether their relations have equivalents in the constraint. After nodes and relations are found in the output representation that are equivalent to those in the constraint, a violation mark is returned. The procedure that checks whether n is connected to a structure equivalent to C means that constraints cannot be unconnected. For example, suppose that there is a constraint *[1, [2 which is violated if a word contains two (not necessarily adjacent) moras; these moras are unconnected in this constraint – there is no precedence relation between them, nor is there a node that dominates them both. VR can evaluate such a constraint, but the constraint will never return any violation marks. VR checks relations between nodes: i.e. VR will search for a [1 node, and then check its relations. Since [1 has no connection to [2 via either precedence or dominance, VR will never find any structure in any R that is equivalent to the structure described by the constraint. So the VR algorithm itself, through how it compares the constraint’s structure to structures in the output representation, imposes a weak connectedness requirement on constraints. For a constraint C ever to return a violation mark, every node in C must be connected to every other node. Nodes x and y are “connected” here if it is possible to trace a direct route through precedence and dominance relations from x to y.5

8

Paul de Lacy

The connectedness requirement that results from VR is weak, because a much stronger requirement could be imagined and implemented: i.e. in a constraint, every pair of nodes on the same tier must be in a precedence relation, and every pair of nodes on different tiers must be in a dominance relation. The difference between weak and strong connectedness can be seen in a less connected version of *q[[: one where q↓[1 and q↓[2 , but there is no precedence relation between [1 and [2. Such a constraint would not be evaluated by a VR that imposes strong connectedness because it has two nodes on the same tier ([1, [2) that are not in a precedence relation. However, it is perfectly acceptable in the weak-connectedness VR described above, because every node is connected to every other node. Weak connectedness allows constraints of the form *[x . . . x]D, where there are two nodes of type x within a particular domain D (e.g. Suzuki 1998; see also the discussion of local conjunction in §4.2). The connectedness side-effect of VR is desirable – as far as I am aware, no one has proposed constraints that have completely unconnected elements. The larger point here is that the nature of the algorithm that assigns violation marks is crucial in any theory of constraints. The algorithm not only determines how violation marks are assigned, but whether particular constraints will ever assign violation marks (i.e. it effectively puts restrictions on constraint form just as VR means that constraints must contain connected representations). If constraints are representations and there is a single VR, there should be broad uniformity in the way that violation marks are assigned. For example, the constraint [•↓CPlace↓[dorsal]] will assign violations for each occurrence of its representation in the candidate’s output representation. In contrast, there is no way to formulate a constraint like *>dorsal: if the constraint consists of the representation [•↓CPlace↓[dorsal]], then the VR will return a violation for each occurrence of [•↓CPlace↓[dorsal]]; it cannot be limited to assigning one violation regardless of the number of occurrences of [•↓CPlace↓[dorsal]]. So the VR theory means that output constraints should all assign violations in fundamentally the same way, while the constraints-as-functions theory allows for significant differences. It is even possible that there is a middle ground: there could be several violation assignment algorithms, with VR being just one of them. With several VRs, we would expect to see uniformity in how violations are assigned, but only within particular groups of constraints. Which view is correct? How much uniformity in violation mark assignment is there?

2.3

Regularities and irregularities in violation assignment

There is a great deal of regularity in the way that violation marks are assessed in output constraints. In fact, there is so much regularity that I am sure there would be no confusion among phonologists about how a constraint like *œ7 works (even though I concocted it just now): it would return a violation mark for each Let us define a transitive symmetric relation ⊕; x⊕y if xIdent[nasal] does not return any violations, because for every input segment (i.e. /ã/) there is some pair that preserves the [nasal] value (i.e. ). Alderete (2001) argues that at least some faithfulness constraints have “antifaithfulness” counterparts. For example, OO-Ident[voice] returns a violation for each corresponding segment that has different values of [voice] (for the OO-part, see §3.3). However, ¬OO-Ident[voice] returns a violation if there is no pair of correspondents that disagrees on [voice] values. The constraints differ in terms of how they assess violations rather than the representations they refer to. If Struijke’s (2000) and Alderete’s (2001) proposals are correct, they pose a serious challenge to an approach that seeks to find a single F-VR algorithm. Their proposals mean that there are sets of constraints that differ only in terms of the procedure of violation mark assignment, not in the representation and relations they refer to.

3.3

Developments in correspondence constraints

The theory of correspondence constraints has been reduced, altered, and extended since McCarthy and Prince (1993). For example, Keer (1999) argues that Uniformity does not exist, with the effect that coalescence is obligatory in certain situations. Similarly, a number of authors have argued that Dep does not exist (e.g. Bernhardt and Stemberger 1998). They observe that output constraints do a similar job to IO-Dep; output constraints

Markedness and Faithfulness Constraints

15

(usually) prefer less representational structure over more, and so does IO-Dep. However, Gouskova (2007) argues that the effects of IO-Dep can be distinguished from structural constraints. Most work has focused on extending correspondence to new nodes (§3.3.1) and dimensions (§3.3.2).

3.3.1 Loci of correspondence McCarthy and Prince (1993: 14) proposed that correspondence holds between segments. I have adopted a particular theory of representation (autosegmental theory) in this chapter that does not provide an easy way to define “segment”; in this theory, the most natural understanding of McCarthy and Prince’s proposal is to say that correspondence holds between “root nodes.” McCarthy and Prince also suggest that correspondence could hold between other nodes: tonal nodes, prosodic nodes, and terminal and non-terminal feature nodes. Myers (1997) develops this idea for tonal nodes. For example, IO-Max-T requires that every input tone node correspond to some output tone node. The most significant effect of the proposal is that tones can survive even when their segmental sponsors are deleted. See Yip (2007) for an introduction to tone constraints; also chapter 45: the representation of tone. McCarthy (2000) argues for a variety of constraints that require (at least some) prosodic nodes to be in correspondence. Since (most) prosodic structure is apparently absent in inputs, evidence for correspondence between syllables and feet comes from identity across other forms (e.g. base–reduplicant, base–derivative; see §3.3.2). A widely discussed extension has been feature correspondence. Lombardi (2001) argues that feature-to-feature correspondence is essential in explaining differences between how place features and voice features differ in their behavior. However, coalescence can be achieved without feature correspondence (e.g. Pater 1999; de Lacy 2002: chapter 8), and a general concern with Max-feature approaches is the lack of observed feature autonomy. In several theories, features do not seem to have the same kind of independence as tones: while tones can survive if their sponsors are deleted, there may not be similar effects for features (featural morphemes are special cases, however; see Wolf 2007b: §2.2 and chapter 82: featural affixes for discussion).

3.3.2 Dimensions of correspondence The discussion above has focused on correspondence between inputs and outputs. However, there have been many proposals that extend the reach of correspondence. The proposals fall into two categories: intra-representational correspondence and inter-representational correspondence. McCarthy and Prince (1999) propose that intra-representational correspondence is found in reduplication (see also chapter 100: reduplication). A reduplicant morpheme has no input content, but its output segments can correspond to certain other output segments (the reduplicant’s “base”). For example, one of the reduplicated forms of Maori [parau] ‘baffled’ is [pa7apa7au]. The reduplicated segments correspond to other output segments thus: [p1a273a4p1a273a4u5]. McCarthy and Prince (1993) argue that constraints on Base–Reduplicant (BR) correspondence have the same form as constraints on IO correspondence. BR-Max requires every base element to have some correspondent in the reduplicant

16

Paul de Lacy

(violated once in [p1a273a4p1a273a4u5]), BR-Dep requires every reduplicant segment to have a correspondent in the base, and BR-Ident[F] regulates featural identity between base and reduplicant. What is surprising about the extension of correspondence to the Base– Reduplicant dimension is that there is essentially one formal mechanism that accounts for both the input–output relation and reduplication. Other theories of reduplication conceive of the phenomenon as involving templates or a type of long-distance assimilation, perhaps through autosegmental spreading (see e.g. McCarthy and Prince 1986). Urbanczyk (2007) gives an overview of BR correspondence and reduplication. Other intra-representational correspondence relations have been proposed. Kitto and de Lacy (1999) argue that epenthetic segments can correspond to other output segments, resulting in “copy epenthesis”: e.g. Winnebago [bo(pjnjs] ‘hit at random’ (Miner 1992). The reason for proposing correspondence here is “overapplication”: nasal vowels only occur after nasal consonants in Winnebago, except when epenthetic vowels copy a post-nasal vowel, as above. Such “overapplication” is expected with correspondence, since featural identity of corresponding elements can trump phonotactic restrictions; and is also found in reduplication and other types of correspondence (see Urbanczyk 2007 for discussion of under- and overapplication in reduplication; and Benua 1997 for output–output correspondence). Hansson (2001) and Rose and Walker (2004) go further in arguing that any output segment can correspond to another output segment. The effect is seen in long-distance agreement. For example, in Chumash sibilants agree in anteriority within a word: /s-ilakœ/ → [œilakœ] ‘it is soft’; cf. [s-ixut] ‘it burns’. There have been many proposals for inter-representational correspondence, too. Benua (1997) proposes that segments in the output representation can correspond to segments in the “trans-derivational base” of that output. The trans-derivational base of a word is basically the word minus its structurally outermost affix. So, the base of original [[origin]al] is origin. Original itself is the base of originality, and origin is also the base of originate. OO-correspondence can be used to explain why some morphologically complex words do not follow expected phonological patterns, but instead remain similar to their base. For example, in my idiolect (and in many other English-based idiolects) the head foot avoids final syllables in nouns, but otherwise is drawn to the right edge of a PrWd: [Hd(’mHsH)bÁ] admissible, [HdmHsH(’bHlH)7i] admissibility ([H] can be stressed in my dialect, and /l/ → [Á] outside onsets). However, with some affixes the foot does not get drawn rightward as expected: [Hd(’mHsH)bÁnHs] admissibleness, *[HdmH(’sHbÁ)nHs]. When ness appears in a word, it subjects the candidate to an OO-faithfulness requirement that has the effect of forcing the correspondent of the base’s head syllable to also be a head. So *[HdmH(’sHbÁ)nHs] loses to [Hd(’mHsH)bÁnHs] because the corresponding head syllable in the base [Hd(’mHsH)bÁ] is [mH], not [sH]. Further work on interword relationships has argued that candidates should consist of entire output paradigms of related word forms. See McCarthy (2005) for references and discussion. Yet other work has proposed correspondence relations between the output representation and another output representation that is identified by a special selection mechanism, with the aim being to account for phonological opacity.

Markedness and Faithfulness Constraints

4

17

Possible and impossible constraints

Even after defining the CDL’s representational elements, relations, and algorithm(s) assigning violation marks, there remains the question of which constraints actually exist. I wish I could list all the phonological constraints that exist in the human brain here. Unfortunately, there is no agreed-upon list. Many constraints have been proposed, and many algorithms too. Given the rapid changes in phonological theories and variety of constraint proposals, it is more useful to discuss general intrinsic and extrinsic restrictions on theories of constraints. Every CDL imposes intrinsic limits on possible constraints. The nature of the elements and relations by which constraints are defined means that some imaginable constraints could not occur. For example, suppose the CDL has no disjunction operator. A constraint that assigns a violation to a segment if it is either [+voice] or [labial] is then not possible – it is impossible to formulate using the CDL’s syntax. Similarly, the VR itself may impose “restrictions” on constraints in the sense that certain constraints might be well formed in the CDL, but not assign violation marks. In the VR discussed above, constraints that had unconnected representational elements would not assign violation marks; so, while such unconnected constraints could exist, they are effectively inert and will never be observed to have an effect on selecting a winning candidate. Other cases were discussed in §1.1 and §2.2. It is also possible (even likely) that there are extrinsic limits on constraints. An extrinsic limit is a restriction on particular types of constraint even though the constraint would have a well-formed syntactic structure in the CDL. For example, suppose there was a CDL that made it possible to define a constraint that banned syllable onsets, yet such a constraint did not exist: an extrinsic limit would have to be responsible. The alternative is to suppose that there are almost no significant extrinsic limits: the set of constraints includes every constraint definable using the CDL (up to a certain level of complexity). The issue of extrinsic limits is a very difficult one. The first issue addressed below is methodological: how can we tell whether there are extrinsic limits on constraints (§4.1)? §4.2 discusses where those extrinsic limits come from.

4.1

Evidence for extrinsic restrictions

The majority of OT theories and subtheories do propose many extrinsic restrictions on constraints, specified by the CDL (cf. discussion in Blevins 2004). The evidence comes from restrictions and requirements that cannot be attributed to non-cognitive mechanisms. To explain, suppose we never observe a particular phonological property in any human language, like an epenthetic [k] (e.g. /iti/ never surfaces as [kiti] in any grammar). It is possible that the lack of [k]-epenthesis is due to constraints. For [k] to be epenthetic, there has to be a (set of) output constraint(s) that returns violation marks for every segment except [k]. Without such a constraint, epenthetic [k] won’t occur (see e.g. de Lacy 2006: ch. 3). However, there are potentially non-CDL reasons why epenthetic [k] is never observed. Some other part of the phonological component could be responsible

18

Paul de Lacy

(see §4.2). There is also luck – e.g. war, pestilence, or plague – which may have accidentally wiped out all speakers of languages with epenthetic [k]. After all, every theory of phonology predicts many tens of thousands of distinct phonological systems, and only a few thousand have existed and will ever exist. For epenthetic [k] to be observed, it must also be learnable. Actuation of a phonological change comes about through learner misperception or misarticulation. So if epenthetic [k] cannot come about through such a situation, it won’t be observed. Even if a sound change can be actuated easily, it will quickly disappear if it cannot be transmitted effectively. In this particular case, though, there is evidence that learners could misperceive vowel hiatus as involving a [k] (Kingston and de Lacy 2006). So, suppose a phonological situation P never occurs. Suppose further that the lack of P cannot be ascribed to non-constraint grammatical processes or extraphonological mechanisms. In such a case, phonological extrinsic restrictions on constraints are responsible for the lack of P. For epenthetic [k], it is easy to come up with a set of constraints that penalize everything except [k] (e.g. *Labial, *Coronal, *Glottal), so the CDL must not permit this set of constraints (or at least, this set of constraints with free ranking). There are several other methods of determining that a particular phonological situation is due to constraints. See Kingston and de Lacy (2006: §3.3) and references cited for discussion.

4.2

Origins and universality of constraints

If there are restrictions on possible constraints, where do those restrictions come from? There are fundamentally two different proposals: (a) innateness and (b) constraint-construction mechanisms that refer to phonology-external structures. The innateness view is that constraints are hard-wired into the brain (i.e. part of our genetic make-up). The “hard-wired” view comes in two versions. One is that each constraint is specified independently. In this version, only those constraints that are hard-wired into the brain exist, so extrinsic limits on constraints boil down to genetics. The other version is that there are hard-wired algorithms that automatically generate constraints – “constraint generators” (sometimes called “schemas”). For example, there would be an “Ident[F]“ constraint generator that produces constraints with the form D-Ident[F] where D is a pair of dimensions (input–output, base–reduplicant, etc.), and F is a subsegmental node. The generator is “complete” in that it would generate constraints for every D and every F (Green 1993). The constraint-construction algorithms determine which constraints exist. The alternative is to propose mechanisms that are derived from phonologyexternal mechanisms, or at least can take phonology-external factors into account. A growing body of work argues that there are many algorithms that take phonetic factors like ease of articulation and perceptual distinctiveness into account in evaluating which phonological constraints to generate. In this view, limits on constraints are a combination of the inherent limits of the constraint-construction algorithm and the restrictions imposed by the phonology-external factors that those algorithms refer to. For example, Hayes (1995) discusses phonological constraints on voiced stops. Phonetic voicing in stops is hard to maintain; the further back the stop is, the more

Markedness and Faithfulness Constraints

19

difficult it is to maintain voicing during the closure phase: it is harder to maintain voicing during the closure phase of [g] than for [d], and it is harder for [d] than for [b]. Suppose there is a constraint generator that produces constraints on voicing in stops. It could imaginably generate many constraints, e.g. *g, *d, *b, *g/d, *g/b, *d/b, *g/d/b (where *x/y means “Return a violation for any segment that is x or y”). However, if the mechanism referred to articulation in a way that reflected voicing difficulty, the constraints would be winnowed down to *g, *g/d, *g/d/b. Hayes (1995) further observes that the CDL’s intrinsic representational restrictions could impose further limits on the possible constraints: *g/d, for example, is not definable in some feature theories, as there is no feature that [g] and [d] share to the exclusion of [b]; with such representational theories, the only constraints generated by the mechanism would be *g and *g/d/b (i.e. *[+voice, −continuant, −nasal]). To summarize, the majority of work in OT adopts the idea that there are constraint generators. However, there is ongoing disagreement over whether constraint generators can refer to phonology-external factors like ease of articulation and perceptual difficulty. Gordon (2007) provides discussion and references; see also chapter 98: speech perception and phonology. A related issue is constraint universality. A constraint is “universal” if it exists in every grammar. A constraint can exist in every grammar because it is hardwired into CON (the set of constraints), or because it is produced by a constraint generator (see §4.2) that produces the same constraints in the same way for every grammar. A “language-specific” constraint is one that exists in only some languages; it must be learned. For specific discussions of constraint universality, see Green (1993), Prince and Smolensky (1993), and McCarthy (2002: §1.2.1, §3.1.5.2). There is an important nuance to constraint universality/language specificity. It is possible that constraints are not universal, but rather constraint generators are. For example, Align is a constraint generator that exists in every grammar. However, if Align is allowed to take individual morphemes (or morphs) as arguments, it could produce language-specific constraints like Align([um]Af, L; Stem, L), “The affix um occurs stem-initially, is a prefix” for Tagalog, and Align([ka]Af, L; Ft’, R), “The affix ka follows, is a suffix to, the head foot” for Ulwa (McCarthy and Prince 1993). So, while Align([um]Af, L; Stem, L) does not exist in every language, the constraint generator that created it does. The point could be extended to other constraint generators, and even those that refer to phonology-external factors. If a constraint generator refers to an articulatory or acoustic factor that varies among speakers, it could be that the same constraint generator will produce speakerspecific constraints.

5

Summary

This chapter has left a vast number of issues about constraints untouched and only barely skimmed over a few others. However, a few points about constraints emerge. A formal theory of constraint form – a “Constraint Definition Language” – provides valuable insight into which constraints can and cannot exist. There is fairly widespread (if tacit) agreement on many aspects of such a CDL, but also many disagreements about both fundamental issues and the details. Constraints are only one part of a complex system that determines phonological winners;

20

Paul de Lacy

constraints themselves do not determine winners, and no constraint or set of constraints has any predictive power on its own. Only when the entire collection of constraints, GEN, EVAL, and the phonetic module interface are examined together can anything be asserted about the predictive power or restrictive nature of the theory.

ACKNOWLEDGMENT My thanks to Marc van Oostendorp, Keren Rice, Matt Wolf, and an anonymous reviewer for their extensive comments.

REFERENCES Alderete, John. 2001. Dominance effects as transderivational anti-faithfulness. Phonology 18. 201–253. Bach, Emmon. 1968. Two proposals concerning the simplicity metric in phonology. Glossa 2. 128–149. Bakovio, Eric. 2007. Local assimilation and constraint interaction. In de Lacy (2007), 335–351. Benua, Laura. 1997. Transderivational identity: Phonological relations between words. Ph.D. dissertation, University of Massachusetts, Amherst. Bermúdez-Otero, Ricardo. Forthcoming. Stratal Optimality Theory. Oxford: Oxford University Press. Bernhardt, Barbara H. & Joseph P. Stemberger. 1998. Handbook of phonological development from the perspective of constraint-based nonlinear phonology. San Diego: Academic Press. Blevins, Juliette. 2004. Evolutionary Phonology: The emergence of sound patterns. Cambridge: Cambridge University Press. Boersma, Paul & Bruce Hayes. 2001. Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry 32. 45–86. Bye, Patrik & Paul de Lacy. 2000. Edge asymmetries in phonology and morphology. Papers from the Annual Meeting of the North East Linguistic Society 30. 121–135. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Chomsky, Noam & Howard Lasnik. 1977. Filters and control. Linguistic Inquiry 8. 425–504. Clements, G. N. & Elizabeth Hume. 1995. The internal organization of speech sounds. In John A. Goldsmith (ed.) The handbook of phonological theory, 245–306. Cambridge, MA & Oxford: Blackwell. de Lacy, Paul. 2002. The formal expression of markedness. Ph.D. dissertation, University of Massachusetts, Amherst. de Lacy, Paul. 2006. Markedness: Reduction and preservation in phonology. Cambridge: Cambridge University Press. de Lacy, Paul (ed.) 2007. The Cambridge handbook of phonology. Cambridge: Cambridge University Press. Eisner, Jason. 1997. What constraints should OT allow? Handout from paper presented at the 71st Annual Meeting of the Linguistic Society of America, Chicago (ROA-204). Elías Ulloa, José. 2006. Theoretical aspects of Panoan metrical phonology: Disyllabic footing and contextual syllable weight. Ph.D. dissertation, Rutgers University (ROA-804). Gordon, Matthew. 2007. Functionalism in phonology. In de Lacy (2007), 61–78. Gouskova, Maria. 2007. Dep: Beyond epenthesis. Linguistic Inquiry 38. 759–770. Green, Thomas. 1993. The conspiracy of completeness. Unpublished ms., MIT (ROA-8).

Markedness and Faithfulness Constraints

21

Hall, T. A. 2007. Segmental features. In de Lacy (2007), 311–333. Hansson, Gunnar Ólafur. 2001. Theoretical and typological issues in consonant harmony. Ph.D. dissertation, University of California, Berkeley. Hayes, Bruce. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. Itô, Junko. 1986. Syllable theory in prosodic phonology. Ph.D. dissertation, University of Massachusetts, Amherst. Kager, René, 1999. Optimality Theory. Cambridge: Cambridge University Press. Kager René, Harry van der Hulst & Wim Zonneveld (eds.) 1999. The prosody–morphology interface. Cambridge: Cambridge University Press. Keer, Edward. 1999. Geminates, the OCP and the nature of Con. Ph.D. dissertation, Rutgers University. Kingston, John & Paul de Lacy. 2006. Synchronic explanation. Unpublished ms., University of Massachusetts, Amherst & Rutgers University. Available (July 2010) at www. pauldelacy.net/webpage/publications.html. Kiparsky, Paul. Forthcoming. Paradigms and opacity. Stanford: CSLI. Kitto, Catherine & Paul de Lacy. 1999. Correspondence and epenthetic quality. In Catherine Kitto & Carolyn Smallwood (eds.) Proceedings of AFLA VI: The Sixth Meeting of the Austronesian Formal Linguistics Association, 181–200. Toronto: Department of Linguistics, University of Toronto. Lombardi, Linda. 2001. Why Place and Voice are different: Constraint-specific alternations in Optimality Theory. In Linda Lombardi (ed.) Segmental phonology in Optimality Theory: Constraints and representations, 13–45. Cambridge: Cambridge University Press. Ìubowicz, Anna. 2005. Locality of conjunction. Proceedings of the West Coast Conference on Formal Linguistics 24. 254–262. McCarthy, John J. 1999. Sympathy and phonological opacity. Phonology 16. 331–399. McCarthy, John J. 2000. The prosody of phase in Rotuman. Natural Language and Linguistic Theory 18. 147–197. McCarthy, John J. 2002. A thematic guide to Optimality Theory. Cambridge: Cambridge University Press. McCarthy, John J. 2003. OT constraints are categorical. Phonology 20. 75–138. McCarthy, John J. 2005. Optimal paradigms. In Laura J. Downing, T. A. Hall & Renate Raffelsiefen (eds.) Paradigms in phonological theory, 170–210. Oxford: Oxford University Press. McCarthy, John J. 2007. Hidden generalizations: Phonological opacity in Optimality Theory. London: Equinox. McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. McCarthy, John J. & Alan Prince. 1993. Generalized alignment. Yearbook of Morphology 1993. 79–153. McCarthy, John J. & Alan Prince. 1999. Faithfulness and identity in prosodic morphology. In Kager et al. (1999), 218–309. McCarthy, John J. & Matthew Wolf. 2005. Less than zero: Correspondence and the null output. Unpublished ms., University of Massachusetts, Amherst (ROA-722). Miner, Kenneth L. 1992. Winnebago accent: The rest of the data. Indiana University Linguistics Club Twenty-Fifth Anniversary Volume, 28–53. Bloomington: Indiana University Linguistics Club. Myers, Scott. 1997. OCP effects in Optimality Theory. Natural Language and Linguistic Theory 15. 847–892. Pater, Joe. 1999. Austronesian nasal substitution and other NY effects. In Kager et al. (1999), 310–343. Potts, Christopher & Geoffrey K. Pullum. 2002. Model theory and the content of OT constraints. Phonology 19. 361–393.

22

Paul de Lacy

Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA & Oxford: Blackwell. Rose, Sharon & Rachel Walker. 2004. A typology of consonant agreement as correspondence. Language 80. 475–531. Samek-Lodovici, Vieri & Alan Prince. 1999. Optima. Unpublished ms., University College London & Rutgers University (ROA-363). Schein, Barry & Donca Steriade. 1986. On geminates. Linguistic Inquiry 17. 691–744. Struijke, Caro. 2000. Existential faithfulness: A study of reduplicative TETU, feature movement and dissimilation. Revision of Ph.D. dissertation, University of Maryland, College Park. Suzuki, Keiichiro. 1998. A typological investigation of dissimilation. Ph.D. dissertation, University of Arizona. Urbanczyk, Suzanne. 2007. Reduplication. In de Lacy (2007), 473–493. Wilson, Colin. 2001. Consonant cluster neutralisation and targeted constraints. Phonology 18. 147–197. Wolf, Matthew. 2007a. What constraint connectives should be permitted in OT? University of Massachusetts Occasional Papers in Linguistics 36. 151–179. Wolf, Matthew. 2007b. For an autosegmental theory of mutation. In Leah Bateman, Michael O’Keefe, Ehren Reilly & Adam Werle (eds.) Papers in Optimality Theory III, 315–404. Amherst: GLSA. Yip, Moira. 2007. Tone. In de Lacy (2007), 195–227.

64

Compensatory Lengthening Randall Gess

1

Types of compensatory lengthening

Compensatory lengthening is the lengthening of one segment, referred to here as the “target,” in compensation for the loss or reduction of another, referred to here as the “trigger” (see also chapter 37: geminates and chapter 20: the representation of vowel length). The segments are usually in close proximity to one another – either adjacent or in adjacent syllables. Theoretically, a consonant can be lengthened in compensation for the loss or reduction of another consonant or vowel, and a vowel can be lengthened in compensation for the loss or reduction of another vowel or consonant. In fact, an argument can be made that all types of compensatory lengthening exist, although as Table 64.1 indicates, some types are far more common than others. There is also a problem in the classification of some types as compensatory lengthening proper rather than as instantiations of other processes, such as total assimilation (2A, B) or rhythmic lengthening (1D). In Table 64.1, Row 1 lists cases in which the target for lengthening is a vowel, and Row 2 lists cases in which the target for lengthening is a consonant. Column A lists cases in which the trigger for lengthening (the reduced or deleted segment) is a reduced consonant following the target, Column B those in which the trigger is a reduced consonant preceding the target, Column C those in which the trigger is

Table 64.1 Types of compensatory lengthening Trigger A / __ C Target

B / C __

C / __ (X )V

D / V(X) __

1: V

numerous

limited

numerous

limited

2: C

limited

numerous

isolated

limited

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0064

Randall Gess

2

a reduced vowel following the target, and Column D those in which the trigger is a reduced vowel preceding the target. Examples representative of each cell in Table 64.1 are provided in the following brief sections.

1.1 Type 1A (Target V; Trigger / __ C) (1)

Old French (Gess 1998, 1999) a. b. c.

blasmer angle large

[blazmer] > [ãnglH] > [laú–H] >

[bla(mer] [ã(glH] [la(–H]

‘to blame’ ‘angel’ ‘wide’

The type of compensatory lengthening in (1) is common. Kavitskaya (2002: App. 1) lists 58 languages in which it is manifested. Other cases not mentioned by Kavitskaya can be found in Gordon (1999) and Beltzung (2008). More exotic types of compensatory lengthening triggered by a following consonant are ones in which the trigger is an intervocalic consonant, or in which the target and trigger are separated by an intervening consonant, i.e. in which the triggering consonant is the second of a sequence of two intervocalic consonants (e.g. Ancient Greek *odwos > East Ionic /o(dos/ (Wetzels 1986: 310). Hayes (1989) refers to this type of compensatory lengthening as a “double flop,” a term which has gained currency in the literature (see Beltzung 2008 for an extensive discussion of “exotic” types of compensatory vowel lengthening triggered by consonant loss).

1.2 Type 1B (Target V; Trigger / C __) (2)

Samothraki Greek (Katsanis 1996, as reported in Topintzi 2006; see also chapter 55: onsets) a. b. c. d. e. f. g.

’rafts ’ruxa ’rema ’protos ’vrisi me’trun ’extra

> > > > > > >

’a(fts ’u(xa ’e(ma ’po(tus ’vi(s’ mi’tu(n ’exta(

‘tailor (masc)’ ‘clothes’ ‘stream’ ‘first’ ‘tap’ ‘they count’ ‘hostility’

The type of lengthening shown in (2) is rare and somewhat controversial (see Beltzung 2008 for an overview), as it is predicted not to occur by the framework of Moraic Phonology developed in Hayes (1989). In Samothraki Greek this process is limited to rhotics in word-initial or post-consonantal position – it does not occur when the segment is in intervocalic position or in a coda position (there is no deletion of syllable-final /r/ in Samothraki Greek). According to Beltzung (2008), the segments implicated in this type of compensatory lengthening are rhotics, pharyngeals, and laryngeals.

Compensatory Lengthening

3

1.3 Type 1C (Target V; Trigger / __ CV) (3)

Hungarian (Kálmán 1972, as reported in Kavitskaya 2002) a. b. c. d. e. f. g.

*wizi *tyzy *utu *ludu *neÚi *modoru *teheni

> > > > > > >

vi(z ty(z u(t lu(d ne(Ú modo(r tehe(n

‘water’ ‘fire’ ‘road’ ‘goose’ ‘four’ ‘bird’ ‘cow’

Cases like the one illustrated here are relatively common, but appear to be more phonologically restricted and less widespread than the type shown in §1.1. Kavitskaya (2002: App. 2) lists 21 languages (neglecting to include Yapese (Jensen 1977), where it appears to be a synchronic process) in which this process occurs or has occurred, whereas she lists 58 languages (App. 1) in which the CVC type of compensatory lengthening shown in §1.1 is manifested. As indicated in Table 64.1, this type of compensatory lengthening does not always involve an intervening consonant (e.g. Old French [fi(] < [fiH]; Pope 1952: 205).

1.4 Type 1D (Target V, Trigger / VC __) According to Hayes (1989: 284), this type of compensatory lengthening “appears not to exist.” However, a process in Macuxi may prove problematic for this claim. In a section entitled “Compensatory length,” Carson (1981: 50) states that “[w]hen a short vowel is suppressed, the vowel that immediately precedes a stop consonant in its vicinity is lengthened.” Representative data are shown in (4). (4)

Macuxi (Carson 1981) kasa’pan kusupa’ra ’wakqrq’pe ’mqkq-’rq

→ → → →

ksaa’pan ’ksuu’pra ’wakrqq’pe ’mqq’krq

‘sand’ ‘machete’ ‘agreeable’ ‘he’

It appears in all the examples but the last one that the lengthened vowel is the one that follows the deleted one. However, the absence of otherwise expected *[’mq’krqq] in the last example may be attributed to what Pessoa (2006: 78, 2009: 117), citing Abbott (1991), refers to as the absence of “sílaba alongada fonológica na última posiçao,” since final syllables are already rhythmically lengthened. According to Kavitskaya (2002: 149), lengthening in Macuxi (as described in Kager 1997) is not a case of compensatory lengthening, and “should be attributed to the properties of iambic systems” since it “happens regardless of syncope” (see also chapter 44: the iambic–trochaic law). She cites the following examples from Kager (1997: 466 –467). (5)

More examples from Macuxi /piripi/ /waimujamq/

(‘pri().(’pi() (‘wai).(mja().(’mi()

‘spindle’ ‘rats’

4

Randall Gess

The first example is taken by Kager (1997) from Hawkins (1950: 89), and the second from Abbott (1991: 147). In both examples, the final syllable of the form in question is apparently lengthened, and there is no vowel syncope to point to as a trigger. However, with respect to the first form, no lengthening in any position is indicated by Hawkins (1950), who says nothing more about final vowels than that “[t]he last vowel in each stress contour in the basic form of any utterance is always retained,” and he transcribes the form as ‘pripí’, with the acute accent marking “the end point of contours” (Hawkins 1950: 89). It may be that length is present and not noted by Hawkins, but if so, it is just as easily described as phrase-final lengthening that is independent of iambic (foot-level) lengthening. According to Abbott (1991: 145), “[t]he final CV in a phonological phrase (i.e. a phrase bounded by a pause) is always long and stressed.” Abbott describes rhythmically derived length on even-numbered V or CV syllables, counting from the left, as well as on final syllables. Again, though, Hawkins (1950) says nothing about lengthening in any context, and does not indicate it in any transcriptions. Nevertheless, Kager (1997) introduces foot-level lengthening systematically in forms taken from both authors, as well as syncope in forms where Abbott indicates none. Hawkins (1950) does discuss “stress contours” in which the contour consists of “a stretch of speech marked by loud stress on the last vowel,” and he notes that “[w]hen more than one word occurs in a stress contour, the last vowel in each non-final word in the contour is retained.” However, retention of a vowel is a far cry from lengthening. Carson (1981) does not indicate final lengthening either. Kager (1997) chooses to ignore the data from Carson (1981), cited above, noting that she posits “lexical tone rather than stress,” and that her data must be “based on a different dialect than those studied by Hawkins and Abbott” (1997: 466). In fact Carson describes lexical pitch accent (1981: 42–45), which may be “disturbed” at the phrasal level (1981: 46). If Carson is correct, that the variety she documented manifests pitch accent, then the lengthening she describes cannot appropriately be attributed to iambic lengthening. Two points arise from the preceding discussion. First, it is not clear that rhythmic lengthening, independent of vowel reduction or deletion, does occur in the Macuxi variety described by Abbott (1991) (and Hawkins 1950, if lengthening other than in final syllables even occurs in the variety he describes). Second, the lengthening described by Carson (1981) cannot be dismissed on the grounds suggested by Kavitskaya (2002). An important question arises with respect to the first point in the preceding paragraph: If rhythmic lengthening is always tied to vowel reduction or deletion, can it properly be considered compensatory lengthening? Kavitskaya’s point is that it is not – that it is better in this case to consider it a property of iambic systems, together with rhythmic vowel deletion. It is unclear, however, why a foot-based process that can be described as CVCV > CvCV( (where “v” represents a reduced vowel) should be treated any differently than one that can be described as CVCV > CV(Cv, i.e. the fairly common type of compensatory lengthening described in (3) (1C in Table 64.1), that is uncontroversially labeled as such. Two other types of compensatory lengthening can result from a triggering vowel preceding the target vowel, in these cases with no intervening consonant. The first of these is compensatory lengthening through glide formation (typically from high vowels), a relatively common synchronic process in Bantu languages (e.g. Ganda

Compensatory Lengthening

5

/li+ato/ ‘boat’ → [ljaato]; Clements 1986: 47). The other type of process is also attested in Ganda, involving the deletion of non-high vowels in prevocalic position (e.g. /ka+oto/ ‘fireplace (dim)’ → [kooto]; Clements 1986: 49).

1.5 Type 2A (Target C; Trigger / __ C) (6)

Semitic (LipiUski 2001: 195) *us#tabbit *at#tarad *it#talaba *?HtzHkkar *jilkHdenhu( *gHma(lathu( *wesfi *nisf

> > > > > > > >

us#s#abbit at#t#arad it#t#alaba ?HzzHkkar jilkHdennu( gHma(lattu( wessi nHs#s#

‘he imprisoned’ ‘I sent’ ‘he sought’ ‘I remember’ ‘he shall capture him’ ‘she weaned him’ ‘awl’ ‘half’

(Assyro-Babylonian) (Assyro-Babylonian) (Arabic) (Ge’ez) (Hebrew) (Hebrew) (Gurage) (Colloquial Arabic)

This type of compensatory lengthening appears to be relatively uncommon, and like the type described in the following section, it is not formally distinguishable from total assimilation.

1.6 Type 2B (Target C; Trigger / C __) (7) Bengali (Hayes and Lahiri 1991: 81) bDrœa bDrdi bhorti korŒ he kor-lo

~ ~ ~ ~ ~

bDœœa bDddi bhotti ko-Œ he kol-lo

‘rainy season’ ‘elder sister’ ‘full’ ‘do-3pres’ ‘do-3fut’

Although this type of compensatory lengthening is quite common, unlike the type illustrated in (6), it shares with that process the lack of any formal distinctiveness from total assimilation.

1.7 Type 2C (Target C; Trigger / __ V) (8)

Bulgarian (Shishkov 2002) ’balite er’genite ku’œarite ’belezite ’babinata ven’Œiloto ameri’kancite done’sa

> > > > > > > >

’bal’(te er’gen’(te ku’œar’(te ’beles(te ’babin(ta ven’Œil(tu amer(’kancite don(’se

‘the bales’ ‘the bachelors’ ‘the (sheep) pens’ ‘the scars’ ‘the grandmother’s (things)’ ‘the wedding’ ‘the Americans’ ‘bring (3sg)’

Cases such as the one illustrated in (8), in which a consonant is lengthened before a following reduced vowel, appear to be isolated. It is noteworthy that the

6

Randall Gess

consonants involved in the Bulgarian process are of relatively high sonority – only sonorants and /z/, although there is at least one attested case of synchronic compensatory lengthening in this category in which sonority does not appear to be relevant (compensatory lengthening resulting from glide formation, as in Ilokano /’luto+en/ → /lutt’w-en/ ‘cook-goal focus’; Hayes 1989: 269).

1.8 Type 2D (Target C; Trigger / V __) (9)

Ganda (adapted from Clements 1986: 62–63) /li + kubo/ /li + tabi/ /li + daala/

> > >

[kkubo] [ttabi] [ddaala]

According to Clements (1986: 6), the synchronic rule deriving geminates from a CV prefix is “a restructuring of the historical situation, in which a phonetically motivated rule is replaced by a morphologically conditioned one.” Clements assumes the geminates to have arisen historically from earlier *VÔ sequences (Ô represents an upper high front vowel), “with the process giving rise to consonant gemination [being] one in which the articulation of a consonant is anticipated on a preceding postvocalic *Ô” (1986: 65). They are now associated with a certain class of nominal prefixes.

2

Approaches to compensatory lengthening

Most documented cases of compensatory lengthening, at least those formally distinguishable from total assimilation of adjacent consonants, involve the compensatory lengthening of vowels. Furthermore, the most common types of compensatory lengthening of vowels involve those in which the trigger follows, rather than precedes, the target – i.e. the types of cases illustrated in §1.1 and §1.3. These two types of compensatory lengthening are commonly referred to as CVC and CVCV compensatory lengthening, respectively. In this section, I focus on these most common types and, since synchronic cases of compensatory lengthening are derived from historical ones, I focus on the diachronic instantiation of the processes. I first summarize three general approaches to compensatory lengthening, all of which have in common an implicit assumption that the phenomenon is speaker-controlled. A fourth section outlines an alternative approach put forth in Kavitskaya’s (2002) quite comprehensive treatment of compensatory lengthening, which may be considered somewhat radical in proposing a strictly listener-oriented account of the process. The relevance of the various approaches to synchronic cases of compensatory lengthening, as well as to the other types illustrated in §1, is discussed in §4. The first three approaches to be examined in this section fall into two categories, as described by Kavitskaya (2002): one that treats compensatory lengthening as a type of conservation and one that does not. The first category is the most common, and assumes that compensatory lengthening is fundamentally teleological in that its goal is to preserve length present in the input in the output string. Being

Compensatory Lengthening

7

the most common category, it comprises two of the three approaches: a phonetic conservation approach and a phonological conservation approach. The third approach, in a category of its own, is the non-conservation approach, which denies the existence of any intrinsic connection between the loss of a segment and the lengthening of another.

2.1 Phonetic conservation approach In a phonetic conservation approach, compensatory lengthening is viewed as a goal-oriented process functioning to preserve some or all of the physical duration of lost segmental material. Timberlake (1983) discusses a case in Slavic in which a number of modern dialects have long or tense vowels in syllables that preceded a weak jer in Late Common Slavic (see also chapter 122: slavic yers). This is a case of CVCV compensatory lengthening (Timberlake does not discuss CVC cases). According to Timberlake, a long reflex of a vowel in a syllable before a Late Common Slavic weak jer “is in some way a result of the phonetic weakening and the eventual phonemic loss of the following jer vowel” (1983: 293). He suggests that: Late Common Slavic was subject to a constraint on the preservation of word timing, such that phonetic reduction in one syllable (containing the “weak” jer) was compensated for by increased phonetic duration in the preceding, “strong” syllable.

In Timberlake’s model, compensatory lengthening takes place phonologically through re-analysis. Re-analysis depends upon both the phonemic elimination of jers and the surpassing of a “critical duration” on the part of the phonetically lengthened preceding vowel. If, when jers are “eliminated phonemically, either by identification with another vowel or by identification with null” (1983: 299), phonetically lengthened vowels are sufficiently lengthened, the latter are reanalyzed as phonemically long (or tense). Timberlake sets the critical duration for re-analysis arbitrarily at anything beyond 1.5 times “full duration (nearly or exactly 1.0 morae. [. . .] numerical values for duration [. . .] are intended to be highly approximate)” (1983: 298). Timberlake’s model is an additive one, “in which the duration of vowels is adjusted by adding or subtracting increments of duration depending on various factors.” The various factors at play in Late Common Slavic were the consonant intervening between the jer and the lengthened preceding vowel; the position of the CVCH sequence in the word (final or internal); and the accent of the lengthened vowel. The phonetic process of compensatory lengthening is described by Timberlake (1983: 298) as in (10). (10)

Compensatory lengthening as a phonetic process /CVCH/ > [CV1.0+aCH−a]

The formula in (10) states that for any reduction of value a in the phonetic length of a jer, a preceding vowel is realized at full duration (1.0) plus a.

Randall Gess

8

In order to model the gradual nature of phonetic lengthening, Timberlake breaks the process down into discrete stages, arbitrarily shown in 0.2 increments, as illustrated in (11), from Timberlake (1983: 298). (11)

Jer reduction and compensatory lengthening a. b. c. d.

/CVCH/ /CVCH/ /CVCH/ /CVCH/

> > > >

[CV 1.2CH−0.2] [CV 1.4CH−0.4] [CV 1.6CH−0.6] [CV 1.8CH−0.8]

{a {a {a {a

= = = =

0.2} 0.4} 0.6} 0.8}

Finally, as indicated above, Timberlake assumes re-analysis at anything beyond 1.5 times full duration. Regarding the cut-off, Timberlake (1983: 299) explains: When reduced jers were eliminated phonemically, the phonetic phase of CL was necessarily interrupted, and the lengthened variant of a vowel in strong position had to be identified as phonemically long (tense) or short (lax).

This view of re-analysis is illustrated schematically in (12). (12)

Phonemic analysis a. b. c. d.

[CV 1.2CH−0.2] [CV 1.4CH−0.4] [CV 1.6CH−0.6] [CV 1.8CH−0.8]

⇒ ⇒ ⇒ ⇒

/CVC/ /CVC/ /CV(C/ /CV(C/

{a {a {a {a

= = = =

0.2} 0.4} 0.6} 0.8}

Timberlake’s account of CVCV compensatory lengthening can be straightforwardly extended to CVC compensatory lengthening. A demonstration of this is provided in Gess (forthcoming). Explicit criticisms of phonetic conservation approaches to compensatory lengthening are minimal. The most obvious thing to point out here is the limited relevance of this approach to synchronic cases of compensatory lengthening. That is, while the approach may be useful in showing how compensatory lengthening may arise historically, it is not well suited for modeling the processes in synchronic grammars unless the process at hand is, or at least could be, a change in progress (i.e. it is a gradient, post-lexical process). Another problem with the phonetic conservation approach, pointed out by Gess (forthcoming), is that its extension to CVC cases of compensatory lengthening, while straightforward in a mechanical sense, seems to entail at least implicitly the problematic assumption that moras associated with consonants are equivalent in duration to those associated with vowels. This problem could be overcome by making assumptions regarding the formalisms used more explicit.

2.2 Phonological conservation approach In a phonological conservation approach, compensatory lengthening is viewed as a goal-oriented process functioning to preserve some aspect of the phonological representation (a suprasegmental unit) associated with the loss of segmental

Compensatory Lengthening

9

material. In fact, compensatory lengthening phenomena crucially informed the debate as to the best way to characterize the prosodic tier (or timing tier) assumed in autosegmental phonology, i.e. in terms of C- and V-slots, X-slots, or moras (McCarthy 1979; Clements and Keyser 1983; Hyman 1984, 1985; Levin 1985; Lowenstamm and Kaye 1986; McCarthy and Prince 1986; Hayes 1989; Beltzung 2008 (especially Chapter 3); see also chapter 54: the skeleton). In order to explain constraints on compensatory lengthening (e.g. that triggers in CVC cases are only coda consonants and not onset consonants [an assumption later proven problematic; Topintzi 2006; Beltzung 2008], and even more specifically that coda consonants are triggers only when they contribute to syllable weight in the language in question), Hayes (1989), in probably the most influential single article on compensatory lengthening, suggests that lengthening only occurs when deletion results in an empty prosodic position and that only a prosodic frame defined in terms of moras yields the correct typological results. (Hayes 1989: 260–261 also provides a simple and straightforward demonstration of the inability of a linear approach to account for compensatory lengthening.) Hayes (1989) accounts for CVC compensatory lengthening as illustrated in (13), with Latin [kasnus] > [ka(nus] ‘dog’. (13)

Compensatory lengthening in CVC sequences (Hayes 1989: 262) a.

/s/-deletion s → Ø / __

b.

G+sonJ (segmental tier only) I+ant L

Compensatory lengthening [ [′

where [′ is a segmentally unaffiliated mora

a c.

q

q [

k a

q

[

[

[ →

[

s n

u

s

a

k

q [ n

q

[

[ →

[

u

s

a

k

q [ n

[

[ = [ka(nus]

u

s

A theory assuming a prosodic frame defined in terms of X-slots can account for the example above, but not for the fact that in the same language (Latin), /s/deletion does not trigger lengthening when it is word-initial, as in snurus > nurus (the same problem holds for the type of compensatory lengthening illustrated in §1.2). In a segmental theory based on X-slots, any deleted segment should trigger lengthening, whereas in a moraic theory only those segments that are mora bearing will do so. The moraic theory of compensatory lengthening accounts for CVCV cases as illustrated in (14) with Middle English [talH] > [ta(l] ‘tale’ (see Minkova 1982 for an in-depth discussion of this case).

Randall Gess

10 (14)

Compensatory lengthening in CVCV sequences (Hayes 1989: 268–269) a.

t b.

t c.

q

q

[

[

a

l

H

q

q

[

[

a

Parasitic delinking [

[

d. .

a

l Compensatory lengthening

q [ t

e.

[ l

a

Resyllabification

q [ t

a

Schwa drop

l

q

t

Input

[ l

= [ta(l]

Parasitic delinking, illustrated in (14c), is a principle that eliminates ill-formed syllable structure, caused in this case by the loss of the vowel segment via schwa drop. A very positive aspect of Hayes (1989) is that a wide range of cases of compensatory lengthening are discussed (although a potentially problematic empirical gap is discussed below). Besides the classic CVC and CVCV cases, Hayes (1989) treats the so-called “double flop” cases in which the deletion of a glide triggers compensatory lengthening of a vowel in a preceding syllable, as in Ancient Greek *odwos > /o(dos/ (1989: 265–266), and compensatory lengthening from glide formation, as in the Ilokano case mentioned earlier /’luto+en/ → /lutt’w-en/. He also mentions (without providing a formal treatment) “straightforward” cases like compensatory lengthening through progressive and regressive total assimilation of consonants, compensatory lengthening through prenasalization, and so-called “inverse compensatory lengthening,” which involves the lengthening of a consonant triggered by the shortening or loss of a vowel (1989: 279 –281).

Compensatory Lengthening

11

Fox (2000) points out a number of what he sees as problems for Hayes’s (1989) approach. The first is that the principle of parasitic delinking is “a radical measure which is not required in most other processes of Compensatory Lengthening.” The second relates to the required linking of the vowel of the first syllable to the mora stranded by deletion of the second syllable. According to Fox (2000: 100 –102), this is: unmotivated by the normal principles of the model, since, according to one view at least, the syllable would be perfectly well-formed without this linking; the final mora would be linked to the final consonant and is thus not left stranded.

Finally, Fox (2000) suggests that Hayes’s (1989) principle of mora conservation is inappropriate as a motivation for CVCV compensatory lengthening. This is because Hayes defines the mora as “the basic unit for syllable weight” (1989: 285) and syllable weight is not maintained in these cases. Rather, what is maintained is the length of the foot (2000: 101). One might also argue that a problem for phonological conservation approaches generally is that they are ill equipped to deal with the gradual nature of diachronic compensatory lengthening. For example, the Middle English case discussed above did not take place as a discrete change in one fell swoop. According to Minkova (1982: 50): Before becoming identified with existing long vowels or developing into new ones, i.e. prior to the establishment of a phonological length contrast, the short vowels in the environment / __ C11e# undergo phonetic lengthening. [. . .] In a situation where forms with and without the second syllabic element, the -e, are both available to the speaker, there will be a negative correlation between it and the first syllabic element. Phonetically, “the word as a whole has a certain duration that tends to remain relatively constant.” (Lehiste 1970: 40)

One way to show intermediate length in moraic phonology is to have segments share a mora. In this case, a standard interpretation of the formalism prevents this because it would involve the crossing of association lines, as illustrated in (15). (15)

Potential inadequacy of Hayes’s mora conservation approach with respect to modeling gradual change: CVCV cases *

t

q

q

[

[

a

l

H

If one adopts a strict and unnuanced view of the ban on crossed association lines, since parasitic delinking (illustrated in (14c)) is triggered only when ill-formed syllable structure is present, the moraic account of compensatory lengthening is only able to succeed if the final vowel is entirely deleted and parasitic delinking applies. That is, the account is unable to account for the allophonic lengthening that must be assumed to precede phonemic lengthening. However, a more relaxed view

Randall Gess

12

might interpret the ban on crossed association lines as applying separately to distinct C and V tiers. Gradual change might also be seen as a problem for the mora conservation approach in diachronic cases involving CVC compensatory lengthening. Hock (1986) mentions a case reported in Brockelman (1908) from Tunisian Arabic (with similar instances in Ge’ez and Tigrinya), in which a preconsonantal glottal stop is reduced (not deleted), with compensatory lengthening on the preceding vowel. The examples provided are shown in (16). (16)

Compensatory lengthening triggered by segmental reduction in Tunisian Arabic (Hock 1986: 444) œeffa?ni sma?tkum

> >

œeffa(?ni sma(?tkum

In autosegmental representation, the output of this process would be as illustrated in (17). (17) Potential inadequacy of Hayes’s mora conservation approach with respect to modeling gradual change: CVC cases *

…f

q

q [

[

a

?

[ n

i

While the shared mora representation in (17) is adequate for representing an intermediate stage between a fully moraic glottal segment following the short vowel and a long vowel with no following glottal, one might argue that it cannot represent any more than a single such stage, whereas more stages might well be warranted. However, this potential criticism disregards the possibility of a single phonemic representation having different phonetic interpretations at different periods (or indeed across speakers at a single period). We must also note (as have others) one apparent empirical weakness with the phonological conservation approach, as put forth in Hayes (1989). This involves cases in which compensatory lengthening is triggered by a prevocalic consonant that in normal circumstances would not be associated with a mora. Some such cases are discussed in Hock (1986), a paper cited by Hayes (1989), but without mention of these specific examples. Strangely, the cases are also problematic for Hock, although he does not treat them as such. Hock’s interest in the cases is that they involve compensatory lengthening triggered not by deletion of a segment, but by its weakening only (as in the case illustrated in (17), but in intervocalic position). The first case is from Tyrone Irish (as discussed in Stockman and Wagner 1965), where: vowels are dialectally distinctively lengthened before the highly reduced glottalfricative outcome of earlier voiceless fricatives (as well as before sonorant + consonant etc. and in “ordinary” CL environments, but not in open syllables). (Hock 1986: 443–444, emphasis added – RG)

Compensatory Lengthening

13

The fact that Hock takes care to note that compensatory lengthening does not take place in open syllables suggests that the reduced segments in question might be ambisyllabic, but this is not made explicit anywhere, including in the original source. If it is the case that the segments in question are ambisyllabic, then an argument could be made that they are doubly linked – to a mora in the first syllable and to the onset of the second. The relevant data are shown in (18). (18)

Compensatory lengthening triggered by segmental reduction in Tyrone Irish (Hock 1986: 444) srathar tachas

[stra(hHr] [tD(hHs]

The second problematic case mentioned by Hock (1986) is from the Westphalian dialect of Soest, as reported in Holthausen (1886: 28 –29). In this dialect, as illustrated in (19), compensatory lengthening occurs before highly reduced, voiced labial and velar fricatives, and before a deleted “secondary” (analogically reinserted) voiced alveolar stop. (19)

Compensatory lengthening triggered by segmental reduction in Westphalian (Hock 1986: 444) *hege *seven *snede

‘hedge’ ‘seven’ ‘slice’

> > >

(*)hiHgH (*)siHv(e)n (*)sniHde

> > >

hi(H:H si(Hvn sni(H

(with HH > H)

Again, there is no suggestion that the consonants triggering compensatory lengthening in these cases were ambisyllabic, nor is there any reason to believe that they were. This case therefore represents an apparent problem for Hayes’s (1989) mora conservation approach, aside from the issue of the gradient nature of the triggering segmental reduction. (The case also poses a problem for Hock 1986, which represents a mora conservation approach as well, although one not couched in an autosegmental framework.) As mentioned earlier, other cases of compensatory lengthening triggered by the loss of intervocalic consonants are discussed by Beltzung (2008: especially ch. 2). Finally, the phonological conservation approach à la Hayes (1989), particularly in the case of CVC compensatory lengthening, has proven difficult to model in Optimality Theory (OT). The basic problem is that in order for lengthening to occur, consonants must be assigned weight before deletion happens, thus suggesting a serial analysis. Getting around this problem has necessitated the abandonment of some of the basic tenets of OT. For example, one could simply assume that consonants are moraic in the input (Sprouse 1997), but this requires a sidestepping of OT’s principle of Richness of the Base, whereby output well-formedness is determined solely by constraints and their ranking, and not by restrictions on input. Other ways of handling the problem involve treatments designed to handle opacity more generally, such as stratal OT (Kiparsky 2000), which rejects strict serialism, Turbidity Theory (Goldrick 2001), Sympathy Theory (McCarthy 2003), or OT with candidate chains (McCarthy 2005; Shaw 2009), which require reference to what amounts to one or more intermediate representations.

Randall Gess

14

2.3 Non-conservation approach In an influential article, de Chene and Anderson (1979) take a novel approach to compensatory lengthening by rejecting the notion that such a process exists as “an independent mechanism of phonetic change” (1979: 505) (they discuss only cases of CVC compensatory lengthening). For them, putative cases of compensatory lengthening can be decomposed into two independent processes: weakening of the consonant in question to a glide, and subsequent monophthongization of the resulting vowel + glide sequence. De Chene and Anderson further contend that monophthongization will result in a long vowel only if the language in question has a pre-existing vowel length distinction. The latter claim is about a structurepreserving condition on compensatory lengthening and not about the process itself. Discussion of it is explored further in §3. For now, let us look in more detail at the first claim. The proposal that cases labeled compensatory lengthening are in fact the result of two unrelated processes has generated much discussion. Gess (1998) points out that it has been challenged by numerous scholars, including Hock (1986), Poser (1986), Sezer (1986), and Gildea (1995). According to Gess (1998: 353), “[E]ach of these scholars provides a strong case against the view that compensatory lengthening is always decomposable into two distinct stages. The ensemble of their arguments renders this claim simply untenable.” Without laboring the point, then, we simply illustrate de Chene and Anderson’s hypothesis with one straightforward example. According to de Chene and Anderson (1979: 512): In Latin, compensatory lengthening involving loss of a dental spirant is limited in source to *Vz[C, +dent] sequences, where *z is the reconstructed allophone of *s before a voiced segment. Thus we have *ni-sd-o > ncdus ‘nest’ and *si-sd-d > scdd ‘I sit down’, both involving the zero grade of *sed ‘to sit’ (cf. seded ‘I sit’).

De Chene and Anderson continue: “Our posited intermediate development involves the loss of occlusion in (preconsonantal) *[z], leading to the voiced glottal spirant [Â]” (1979: 512). In this type of analysis, de Chene and Anderson were not in fact alone. Jeffers and Lehiste (1979) propose the analysis in (20) for the remarkably similar change from Proto-Indo-European (PIE) *nisdo to Sanskrit /ni(Õa-/. (20)

Jeffers and Lehiste’s analysis of PIE *nisdo > Sanskrit /ni(Õa-/ (as presented in Hock 1986: 435) nisdnizdnisdnisÕnijÕni(Õ-

voicing assimilation retroflexion retroflex assimilation gliding contraction

In noting the similarity in analyses, Hock (1986: 435) points out that what distinguishes Jeffers and Lehiste’s analysis from de Chene and Anderson’s is the fact that the former is “not proposed as explanations for all cases traditionally labeled loss with compensatory lengthening, but only for a certain subset, however poorly that subset may be defined.” (Note that Kavitskaya 2002: 38 incorrectly

Compensatory Lengthening

15

interprets this sentence as referring to de Chene and Anderson 1979, rather than to Jeffers and Lehiste 1979.) Hock (1986: 435) continues by saying that he is: ready to concede that many instances of what has traditionally been called loss with compensatory lengthening may well be ambiguous, and can be analyzed either as weakening-cum-assimilation or as cases of loss-with-mora-retention. However, in light of the fact that there are [. . .] cases of loss with CL which cannot be explained in terms of weakening-cum-assimilation, any theory which recognizes only the latter process must be considered insufficient.

One of the examples provided by Hock that is not amenable to explanation under de Chene and Anderson’s non-conservation approach is from Icelandic, and is provided in (21). (21)

Compensatory lengthening in Icelandic (Hock 1986: 442) *liugan *keosan (*)priar (*)se(an

‘lie’ ‘choose’ ‘three-fem’ ‘see’

> > > >

lju(ga kjo(sa prja(r sja(

In this case, which is similar to the Ganda case at the end of §1.4, although there is weakening to a glide, it involves a preceding vowel rather than a following consonant, and there is no monophthongization involved. Hock’s (1986: 435) reproach of de Chene and Anderson (1979) for proposing an alternative explanation of all cases traditionally called compensatory lengthening while neglecting to treat all types extends to other cases as well. For example, while de Chene and Anderson are aware of CVCV compensatory lengthening, they choose not to discuss it (1979: 506, n. 1).

2.4 Kavitskaya (2002) Kavitskaya (2002) puts forth a model of compensatory lengthening that can be considered a radical departure from previous treatments, in that it assumes the process to be entirely listener-oriented (see chapter 98: speech perception and phonology). In this respect, her model is representative of the overall approach to phonological change espoused in Blevins (2004, 2006), Evolutionary Phonology. This is a model that rejects any explanations for historical phenomena that involve the synchronic phonologies of speakers (e.g. by assuming a role for phonological rules or markedness constraints) when there is an alternative, diachronic explanation available. This essentially removes the speaker from the story of phonological change, except as a source of variation from which potential changes may or may not take root through “innocent misperception” on the part of the listener. This variation is constrained by speaker-specific anatomical differences, and within the speech of a given speaker, due to phonetic transforms of speech dependent (at least) on: rate of speech; degree of physical effort involved; and the humanly physical impossibility of making exactly the same sound twice. (Blevins 2006: 125 –126)

16

Randall Gess

According to Kavitskaya’s (2002) listener-oriented account of compensatory lengthening: diachronic CL through consonant loss [CVC > CV(] ultimately has its origin in the phonetic lengthening of vowels in the environment of neighboring consonants; the subsequent loss of a consonant conditioning such length causes the length to be re-analyzed as phonological. (Kavitskaya 2002: 8)

Further, according to Kavitskaya, with respect to diachronic compensatory lengthening through vowel loss [CVCV > CV(C]: Prior to the deletion of the final vowel, the longer vowel duration characteristic of open syllables is correctly parsed by listeners as a phonetic consequence of syllable structure in the first syllable of a CVCV sequence, and is discounted [. . .] Upon deletion of the final vowel, however, the duration of the vowel in the newly-closed syllable becomes inexplicable, since it is longer than is expected in the closed syllable. (Kavitskaya 2002: 9)

If Kavitskaya’s arguments are right, then compensatory lengthening is not really compensatory in nature. For the process to be compensatory, the compensatory aspect would rely on a role for the speaker, as is assumed at least implicitly in all other models of compensatory lengthening. In so far as Kavitskaya (2002) is representative of the Evolutionary Phonology framework proposed in Blevins (2004, 2006), it is susceptible to the general criticisms that have been leveled against that framework. Lindblom (2006) criticizes the Evolutionary Phonology framework for its reliance on so-called “extraphonological” explanations over phonological accounts. According to Blevins (2006: 20), “principled extra-phonological explanations for sound patterns have priority over competing phonological explanations unless independent evidence demonstrates that a purely phonological account is warranted.” Lindblom takes exception to this stance on the grounds that it highlights a “phonetics/phonology split” and traps the framework in “the conceptual prison of the form/substance distinction” (2006: 242). As the title of his response to Blevins (2006) declares very loudly, Lindblom (2006) rejects the phonetics/phonology split. Lindblom admonishes us to: Deduce sound structure from language use. Anchor theory construction in the universal conditions under which all speech communication must take place. Start from “first principles” and not circularly from the data to be explained (cf. “markedness”). At the level of the individual user, model phonological structure, not as autonomous form, but as an emergent organization of phonetic substance acquired by each native speaker in the context of socially shared, ambient knowledge. At the population level, model this knowledge as a use- & user-dependent process that undergoes change along the historical time scale. Get rid of the distinction between “phonological” and “extraphonological.” Here is a key step: Make the “intrinsic content” an integral part of the theory from scratch. Treat “intrinsic content” as the source that helps generate discrete structure and that constrains both synchronic and diachronic phonological patterning. (2006: 243)

In §4, we will explore how a rejection of the phonetics/phonology split might be helpful in accounting for the many types of compensatory lengthening as a

Compensatory Lengthening

17

unified phenomenon. We conclude this section by looking at an empirical challenge to Kavitskaya’s listener-oriented approach – a synchronic case of compensatory lengthening that suggests an explanation in terms of speaker-controlled behavior. McRobbie-Utasi (1999) provides evidence for a synchronic case of compensatory lengthening that is apparently speaker-controlled and that suggests the relevance of a principle of isochrony in a synchronic production grammar. In an acoustic analysis of quantity in Skolt Saami, McRobbie-Utasi shows a clear connection between the distribution of duration in V1, intervening C, and V2 sequences in disyllabic groups, and a phonological process realized as “an optional rule that either reduces word-final short vowels or deletes them” (1999: 111). Deletion of the final vowel is a feature of casual speech. The relevant optional rule is shown in (22). (22)

Word-final vowel deletion in Skolt Saami (McRobbie-Utasi 1999: 111) V → Ø / __ # Vowel deletion rule A word-final vowel is optionally deleted in Type 1–5 disyllabics.

It is important to note that in the V1, intervening C, and V2 sequences, the intervening C can be long and ambisyllabic (in four of the five types mentioned), or short and affiliated as the onset of the second syllable. The principal relevant passage from McRobbie-Utasi (1999) is shown below, where the “stress-group locations” referred to are the V1 and following C in the relevant sequences. V2 constitutes a third “stress-group location.” According to McRobbie-Utasi (1999: 114 –115): From the [. . .] measurements an important tendency can be deduced: namely, that the presence or absence (or reduced duration) of the vowel in the second syllable has clear consequences for the distribution of duration in the first syllabic vowel and the consonant(s) following it. Thus, an increase in duration takes place as a result of compensatory lengthening. It will be recalled that second syllabic vowel durations were constant in all the structural types once they were realized as full vowels, with an average of 87 msec [. . .]; also, that durations signaling differences between the structural types and/or gradation types are manifested in the first syllabic vowel and the consonant(s) following it. The fact that the presence or absence of the second syllabic has a considerable effect on these durational distributions in the segments preceding has important implications. The durational changes noticeable in these two stress-group locations (i.e. first syllabic vowel and the consonant(s) following) must be recognized as exemplifying the phenomenon of compensatory lengthening. The absence or reduced status of the second syllabic vowel results in an increase of duration in both of the stress-group locations referred to above.

Compensatory lengthening in Skolt Saami, triggered by the reduction or deletion of a final vowel, affects both the preceding vowel and consonant in four of five types (those in which the consonant is long and ambisyllabic), and the lengthening that occurs does so in a way that precisely preserves the overall V/C ratio. In the remaining type, in which the consonant is short and syllabified as the onset of the second syllable, “reduced duration of the second syllabic vowel results in compensatory lengthening in the first syllabic vowel only. There is practically

18

Randall Gess

no durational increase in the consonant following this vowel” (McRobbie-Utasi 1999: 118). It is difficult, although perhaps not impossible, to reconcile McRobbie-Utasi’s (1999) findings with a listener-based approach. Although McRobbie-Utasi’s study involved only two speakers of Skolt Saami, their behavior with respect to the 550 test words used (recorded three times by both speakers, for a total of 3079 usable tokens) was remarkably consistent. Nor do the types of sequences involved lend themselves readily to Kavitskaya’s line of explanation for CVCV compensatory lengthening, since they do not involve (except for Type 3) phonetically lengthened vowels in open syllables. (As expected, V1 in Type 3 sequences is longer than in other types, both when V2 is fully realized and when it is not.) Nor has any reanalysis occurred (whatever that might look like given the sequences involved and their variety [five types]) since the trigger for compensatory lengthening is still synchronically recoverable. Rather, it appears that the speakers are guided directly or indirectly by a principle of isochrony with respect to the disyllabic group. Other empirical problems for Kavitskaya’s approach, from historical French (manifesting types 1A and 1C), are discussed in Gess (forthcoming).

3

A putative constraint on compensatory lengthening

This section briefly explores the second claim made by de Chene and Anderson (1979): that compensatory lengthening can only occur in a language with a pre-existing vowel length contrast – i.e. that it is strictly structure preserving (chapter 76: structure preservation: the resilience of distinctive information). This issue is discussed in detail in Gess (1998), which treats the very data from Old French on which de Chene and Anderson base their claim, thus adding a particularly severe blow to a claim already questioned in other work (for example, in Hock 1986, Hayes 1989, Morin 1994, and Lin 1997, as well as two further cases discussed more recently in Beltzung 2008: 20–21). According to de Chene and Anderson (1979: 517), “a necessary condition for the development of contrastively long vowels through monophthongization is the independent existence of a length contrast in the language.” With respect to historical French, de Chene and Anderson compare two distinct processes (in the ninth and sixteenth centuries) of monophthongization of the diphthong [aw]. At the earlier stage, the resulting monophthong [o] was short. However, at the later stage, the outcome was the long vowel [o(]. (Strangely, de Chene and Anderson (1979: 519) also suggest a sixteenth-century date for the loss of preconsonantal [l] – the same century in which they contend that monophthongization of the vowel + glide sequence resulting from its loss had occurred. However, Gess (1999) provides strong evidence for a much earlier date for the loss of syllable-final [l], after the latter part of the eleventh century – and many scholars assume a much earlier date still.) The difference in outcomes in the monophthongization of derived [aw] was due, according to de Chene and Anderson, to the introduction of vowel length into the language via the loss of intervocalic consonants, in the late ninth and early tenth centuries (1979: 521). This introduction of vowel length also allowed for compensatory lengthening, according to de Chene and Anderson, following the loss (through an intermediate stage as a glide) of syllable-final [z], [s], and nasals.

Compensatory Lengthening

19

Loss of the latter is incorrectly dated by de Chene and Anderson in the sixteenth century, while loss of the former, [z] and [s], is dated in the twelfth and thirteenth centuries. According to Gess (1999), loss of nasal consonants dates from the thirteenth century, and loss of [z] and [s] dates from the eleventh to the thirteenth centuries. De Chene and Anderson (1979: 522) make the following claim with respect to the establishment of long vowels in Old French: There is a solid body of long vowels, however, that were established by 1100 through deletion of the consonant in original ViCVi sequences. In these cases, no leveling or assimilation being necessary, a long vowel is the automatic result of loss of the consonant.

They go on to provide a list of several words illustrating the relevant consonant loss and the resulting putative long vowels. However, Gess (1998) found each of the forms listed by de Chene and Anderson in twelfth- and thirteenth-century Old French poetry and, in each case, the forms are clearly treated as consisting of two syllables. Gess (1998: 358) “found many other examples of orthographic geminate vowels in 12th and 13th-century Old French poetry, all of which are treated as bisyllabic.” The fact that sequences of two vowels were still counted as bisyllabic in the thirteenth century, when the loss of [z] at the very least had occurred, with compensatory lengthening, shows that a pre-existing vowel length contrast in the language was not a prerequisite for compensatory lengthening to take place. Rather, Hayes’s (1989) assumption is likely the right one, that a syllable weight distinction in the language in question is necessary and, crucially, sufficient for compensatory lengthening to take place. Gess (1998: 364) points out that from an optimality-theoretic perspective this would be a rather unsurprising consequence of the general principle of minimal violation, in this case of faithfulness to the input. While a given constraint ranking may allow for the erosion of segmental features, it may still protect prosodic structure.

4

Assessment and recent directions

§2 outlined various approaches to compensatory lengthening: a phonetic conservation approach, a phonological conservation approach, a non-conservation approach, and a listener-based approach. We saw that the non-conservation approach, proposed only in the context of CVC compensatory lengthening, is basically untenable, both because it fails to account for any other type of compensatory lengthening and because there are instances of CVC compensatory lengthening that appear not to be decomposable into the stages suggested by de Chene and Anderson (1979). This leaves us with two conservation approaches, both suggestive of a speaker-based process, and a listener-based approach. We have noted problems with each of these approaches, which I will summarize briefly here. We have observed that the phonetic conservation approach proposed in Timberlake (1983) is most relevant to those instances of compensatory lengthening that are gradient in nature and that may be characterized as changes in progress. This approach seems ill suited for dealing with synchronic cases of

20

Randall Gess

compensatory lengthening that involve complete loss of the trigger. Without some refinement, the approach also has difficulty with CVC cases of compensatory lengthening given the assumption implicit in its formalism that moras associated with consonants are of equal duration to those associated with vowels. With respect to synchrony/diachrony dichotomy, the phonological conservation approach suffers the opposite problem from the phonetic conservation approach. That is, while it can account for most, if not all, cases of synchronic compensatory lengthening, it is not ideally suited to account for compensatory lengthening as a gradual diachronic process or as a process that involves a trigger that is only reduced and not entirely lost. Other criticisms of the phonological conservation approach relate to formalisms (e.g. motivation for parasitic delinking in a rule-based approach and required mechanisms for dealing with opacity in an OT approach) and empirical weakness (its inability to account for compensatory lengthening involving non-moraic segments). The listener-based approach is explicitly an historical approach. That is, it aims to account only for the diachronic development of compensatory lengthening and does not attempt to model synchronic instantiations of the process. For synchronic cases, it is compatible with (but also limited by) the formalisms required by the phonological conservation approach so long as the process involves a trigger that is synchronically recoverable (see Kavitskaya 2002: ch. 5). The principal problem with the listener-based approach is its inability to account for cases that do not lend themselves to re-analysis via misperception. One such case is the synchronic process in Skolt Saami, which involves a prevocalic VC complex as the target, with lengthening affecting the complex as a whole, and with VC ratios precisely maintained. Also problematic for the listener-based approach is left-to-right CVCV compensatory lengthening (1D), which is why it is so important for Kavitskaya to dismiss such cases as instances of non-compensatory, rhythmic lengthening. Further research is necessary to shed light on this particular issue. The listener-based approach will also have difficulty accounting for left-to-right compensatory lengthening processes in which both target and trigger are consonants (2A). Note that this difficulty will obtain for an Evolutionary Phonology inspired approach whether one labels the process as compensatory lengthening or as total assimilation since the second consonant in such sequences will normally be perceptually stronger. Further challenges for the listener-based approach come from the compensatory lengthening of consonants triggered by either a following vowel (2C) or a preceding one (2D). Whether the approach can be developed sufficiently to meet these challenges will be interesting to see. It may be worthwhile for future work to explore the compatibility of the different approaches to compensatory lengthening. Note, for example, that a phonetically based speaker-oriented analysis neither denies a role for the listener, nor necessarily discounts phonological (e.g. moraic) structure. Timberlake (1983), for example, makes clear reference to the mora as the “full duration” of a vowel. Hock (1986: 432), who does not cite Timberlake (1983), also points out that “at least some traditional historical linguists have offered a phonetic explanation of CL in terms of the concept ‘mora’.” Note that mora-based phonological approaches do not necessarily deny a gradual, phonetic aspect either, at least those prior to autosegmentalist accounts. Minkova (1982) provides a clear phonological account of compensatory lengthening in Middle English, based on syllables and rhythmic units (metrical feet) “described [. . .] with reference to their phonological/moric

Compensatory Lengthening

21

composition” (1982: 48), but is careful to “complete the picture by adding some considerations of purely phonetic nature” (1982: 50). Minkova also touts as an advantage of her revised environment for Middle English Open Syllable Lengthening the fact that “it is the only way in which the shift from allophonic to phonemic length of the stressed vowel can be accounted for” (1982: 51). Hock (1986: 434) cites the “striking extent” to which historical evidence coincides with “fine-phonetic” experimental data. Indeed, Hock goes even further (1986: 445), citing the apparent fact “that CL may set in before the complete loss of a segment, simply as the result of TC [temporary compensation] for the reduction of the segment” (emphasis in the original). An important consequence of this is that: the situation just described requires an important modification of the notion “mora”: Rather than referring to a temporal unit measureable in terms of segment length, it must – at least for CL – be permitted to refer to time spans which are fractions of ordinary segment length. (Hock 1986: 445)

This view of the “mora” is entirely in keeping with an analysis along the lines of the one proposed in Timberlake (1983). It is also in keeping with the spirit of Lindblom’s (2006) view of phonological structure as non-autonomous and emergent from phonetic substance, at least if we assume both that the mora is an abstract temporal unit and that reference is permitted to time spans that may be fractions thereof. A rejection of the phonetics/phonology split with respect to the mora may be the only way to achieve real explanatory adequacy with respect to compensatory lengthening. It allows us to explain the gradual nature of compensatory lengthening – a clearly phonetic aspect of the process. On the phonological side, it also accounts for the fact that CVC compensatory lengthening tends to occur mostly in languages with moraic consonants and for the fact that the process in general functions to preserve moraic structure. In conceptual terms, it is the phonological status of the mora, as an abstract unit of weight functioning in the grammar, that provides the motivation for preserving it when an associated segment is subject to gradual reduction and eventual elimination. On the other hand, it is the physical timing associated with moraic elements that guides the actual articulatory implementation of reduction with concomitant compensatory lengthening, a process that is gradual (and variable) in nature. Since all segments have physical timing associated with them, the only (unsurprising) assumption we have to make is that preservation of timing associated with weight-bearing units is generally privileged over the preservation of timing associated with units that do not bear weight. Recent work by Topintzi (2006) and Beltzung (2008) demonstrates the continuing relevance of compensatory lengthening for phonological theorizing. It has directly tackled the problems that compensatory lengthening poses for OT and manages to maintain the basic tenets of the framework. Both pieces of work must be categorized as phonological conservation approaches, both seek to expand the empirical coverage of previous approaches (notably to account for non-moraic consonant triggers of compensatory vowel lengthening), and, interestingly, both demonstrate the need for formal appeal to the preservation of segment positions in addition to moras. Both appear compatible, therefore, with a rejection of the phonetics/phonology split as described above (whether or not the authors

22

Randall Gess

themselves agree with such a rejection). Both also appear compatible with a potential complementary, phonetically based OT approach that might focus on the functional motivation for and phonetic implementation of non-categorical reduction with concomitant lengthening, as it has for assimilation (Jun 1995) and lenition (Kirchner 1998, 2004; Gess 2003, 2004, 2009) (again, whether or not the authors themselves agree with such a move, which is not part of the OT orthodoxy). It seems safe to conclude that compensatory lengthening will continue to be a topic of some interest in the phonological community.

REFERENCES Abbott, Miriam. 1991. Macushi. In Desmond C. Derbyshire & Geoffrey K. Pullum (eds.) Handbook of Amazonian languages, vol. 3, 23–160. Berlin & New York: Mouton de Gruyter. Beltzung, Jean-Marc. 2008. L’allongement compensatoire dans les representations phonologiques: Nature, contraintes et typologie. Ph.D. dissertation, Université Paris III, Sorbonne-Nouvelle. Blevins, Juliette. 2004. Evolutionary Phonology: The emergence of sound patterns. Cambridge: Cambridge University Press. Blevins, Juliette. 2006. A theoretical synopsis of Evolutionary Phonology. Theoretical Linguistics 32. 117–166. Brockelmann, Carl. 1908. Grundriss der vergleichenden Grammatik der semitischen Sprachen, vol. 1. Hildesheim: Georg Olm. Carson, Neusa M. 1981. Phonology and morphosyntax of Macuxi (Carib). Ph.D. dissertation, University of Kansas. Chene, Brent de & Stephen R. Anderson. 1979. Compensatory lengthening. Language 55. 505–535. Clements, G. N. 1986. Compensatory lengthening and consonant gemination in LuGanda. In Wetzels & Sezer (1986), 37–77. Clements, G. N. & Samuel J. Keyser. 1983. CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Fox, Anthony. 2000. Prosodic features and prosodic structure: The phonology of suprasegmentals. Oxford: Oxford University Press. Gess, Randall. 1998. Compensatory lengthening and structure preservation revisited. Phonology 15. 353–366. Gess, Randall. 1999. Rethinking the dating of Old French syllable-final consonant loss. Diachronica 16. 261–296. Gess, Randall. 2003. On re-ranking and explanatory adequacy in a constraint-based theory of phonological change. In D. Eric Holt (ed.) Optimality Theory and language change, 67–90. Dordrecht: Kluwer. Gess, Randall. 2004. Phonetics, phonology and phonological change in Optimality Theory: Another look at the reduction of three-consonant sequences in Late Latin. Probus 16. 21–41. Gess, Randall. 2009. Reductive sound change and the perception/production interface. Canadian Journal of Linguistics 54. 229 –253. Gess, Randall. Forthcoming. Compensatory lengthening in Old French: The role of the speaker. In Deborah Arteaga (ed.) Old French: The state of the research. Dordrecht: Kluwer. Gildea, Spike. 1995. A comparative description of syllable reduction in the Cariban language family. International Journal of American Linguistics 61. 62–102. Goldrick, Matthew. 2001. Turbid output representations and the unity of opacity. Papers from the Annual Meeting of the North East Linguistic Society 30(1). 231–245.

Compensatory Lengthening

23

Gordon, Matthew. 1999. Syllable weight: phonetics, phonology, and typology. Ph.D. dissertation, University of California, Los Angeles. Hawkins, W. Neil. 1950. Patterns of vowel loss in Macushi (Carib). International Journal of American Linguistics 16. 87–90. Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20. 253–306. Hayes, Bruce & Aditi Lahiri. 1991. Bengali intonational phonology. Natural Language and Linguistic Theory 9. 47–96. Hock, Hans Henrich. 1986. Compensatory lengthening: In defense of the concept “mora.” Folia Linguistica 20. 431–460. Holthausen, Ferdinand. 1886. Die Soester Mundart. Norden & Leipzig: Soltau. Hyman, Larry M. 1984. On the weightlessness of syllable onsets. Proceedings of the Annual Meeting, Berkeley Linguistics Society 10. 1–14. Hyman, Larry M. 1985. A theory of phonological weight. Dordrecht: Foris. Jeffers, Robert J. & Ilse Lehiste. 1979. Principles and methods for historical linguistics. Cambridge, MA: MIT Press. Jensen, John T. 1977. Yapese reference grammar. Honolulu: University Press of Hawaii. Jun, Jongho. 1995. Perceptual and articulatory factors in place assimilation: An Optimality Theoretic approach. Ph.D. dissertation, University of California, Los Angeles. Kager, René. 1997. Rhythmic vowel deletion in Optimality Theory. In Iggy Roca (ed.) Derivations and constraints in phonology, 463–499. Oxford: Clarendon. Kálmán, Bela. 1972. Hungarian historical phonology. In Loránd Benk4 & Samu Imre (eds.) The Hungarian language, 49 –83. The Hague: Mouton. Katsanis, Nikolaos. 1996. Sn ckwqqijó idíw[a rgp Ra[nhoájgp [The dialect of Samothraki Greek]. Dg[np Ra[nhÙájgp [Municipality of Samothraki]. Kavitskaya, Darya. 2002. Compensatory lengthening: Phonetics, phonology, diachrony. London & New York: Routledge. Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17. 351–365. Kirchner, Robert. 1998. An effort-based approach to consonant lenition. Ph.D. dissertation, University of California, Los Angeles. Published 2001. New York & London: Routledge. Kirchner, Robert. 2004. Consonant lenition. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.) Phonetically based phonology, 313–345. Cambridge: Cambridge University Press. Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, MA: MIT Press. Lehiste, Ilse. 1977. Isochrony reconsidered. Journal of Phonetics 5. 253–263. Levin, Juliette. 1985. A metrical theory of syllabicity. Ph.D. dissertation, MIT. Lin, Yen-Hwei. 1997. Syllabic and moraic structures in Piro. Phonology 14. 403–436. Lindblom, Björn. 2006. Rejecting the phonetics/phonology split. Theoretical Linguistics 32. 237–243. LipiUski, Edward. 2001. Semitic languages: Outline of a comparative grammar. 2nd edn. Leuven: Peeters. Lowenstamm, Jean & Jonathan Kaye. 1986. Compensatory lengthening in Tiberian Hebrew. In Wetzels & Sezer (1986), 97–132. McCarthy, John J. 1979. Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT. McCarthy, John J. 2003. Sympathy, cumulativity, and the Duke-of-York gambit. In Caroline Féry & Ruben van de Vijver (eds.) The syllable in Optimality Theory, 23–76. Cambridge: Cambridge University Press. McCarthy, John J. 2005. Candidate chains. Paper presented at the 2nd Old World Conference on Phonology, University of Tromsø. McCarthy, John J. & Alan Prince. 1986. Prosodic morphology. Unpublished ms., University of Massachusetts, Amherst & Brandeis University. McRobbie-Utasi, Zita. 1999. Quantity in the Skolt (Lappish) Saami language: An acoustic analysis. Bloomington: Indiana University.

24

Randall Gess

Minkova, Donka. 1982. The environment for open syllable lengthening in Middle English. Folia Linguistica Historica 3. 29 –58. Morin, Yves-Charles. 1994. Phonological interpretations of historical lengthening. In Wolfgang U. Dressler, Martin Prinzhorn & John R. Rennison (eds.) Phonologica 1992, 135–155. Turin: Rosenberg & Selier. Pessoa, Katia N. 2006. Fonologia Taurepang e comparação preliminar da fonologia de línguas do grupo Pemóng (família Caribe). M.A. thesis, Universidade Federal de Pernambuco. Pessoa, Katia N. 2009. O accento ritmico na língua Taurepang (família Karíb). Guavira 9. 114–126. Pope, Mildred K. 1952. From Latin to Modern French with especial consideration of Anglo-Norman: Phonology and morphology. Manchester: Manchester University Press. Poser, William J. 1986. Japanese evidence bearing on the compensatory lengthening controversy. In Wetzels & Sezer (1986), 167–186. Sezer, Engin. 1986. An autosegmental analysis of compensatory lengthening in Turkish. In Wetzels & Sezer (1986), 227–250. Shaw, Jason. 2009. Compensatory lengthening via mora preservation in OT-CC: Theory and predictions. Papers from the Annual Meeting of the North East Linguistic Society 38. 323–336. Shishkov, Peter. 2002. Elision of unstressed vowels in the ErkeL dialect. Retrieved from http://escholarship.org/uc/item/7tm865v7. Sprouse, Ronald. 1997. A case for enriched inputs. Ms. (ROA-193.) Stockman, Gerard & Heinrich Wagner. 1965. Contributions to a study of Tyrone Irish. Lochlann 3. 43–236. Timberlake, Alan. 1983. Compensatory lengthening in Slavic 2: Phonetic reconstruction. In M. S. Flier (ed.) American Contributions to the 9th International Congress of Slavists, vol. 1: Linguistics. 293–319. Topintzi, Nina. 2006. A (not so) paradoxical instance of compensatory lengthening. Journal of Greek Linguistics 7. 71–119. Wetzels, W. Leo. 1986. Phonological timing in Ancient Greek. In Wetzels & Sezer (1986), 279–344. Wetzels, W. Leo & Engin Sezer (eds.) 1986. Studies in compensatory lengthening. Dordrecht: Foris.

65

Consonant Mutation Janet Grijzenhout

1

Introduction

The phenomenon of “consonant mutation” occurs in a wide array of unrelated languages and comprises changes that are also known as “consonant weakening” (or lenition), “consonant strengthening” (or fortition), and “nasalization.” In this chapter, “consonant mutation” will be defined as a change in one phonetic property of a consonant that affects its degree of sonority and that does not depend on the position of the consonant within a prosodic domain (i.e. neutralization and enhancement phenomena are excluded), nor on the position immediately adjacent to a segment with which it forms a natural class (i.e. progressive and regressive voicing and place assimilations are not regarded as instances of “consonant mutations”). More specifically, the term “consonant mutation” refers to a class of processes by which a consonant turns into a segment with a different degree of voicing, continuancy, or nasality that is not due to neutralization or assimilation to a neighboring segment of the same natural class.1 Some types of consonant mutation can be described as alternations that take place in a particular phonological environment; for instance, an oral stop may turn into a fricative between a sonorant consonant and a vowel. Other types of consonant mutation take place in a certain morphological or lexical context; for example, stem-initial oral stops are realized as continuants under certain morphosyntactic conditions, while continuants are deleted or realized as a laryngeal sound under the same conditions in Modern Irish (e.g. Ní Chiosáin 1991; chapter 117: celtic mutations). The interesting aspect of consonant mutation in general is a diachronic one: what starts out as a purely phonological alternation induced by neighboring segments may gradually turn into a morphological alternation for which the phonological context is no longer transparent (chapter 93: sound change). In the course of this chapter, we will encounter various examples of such developments. 1

Note that changes in voicing, continuancy, or nasality make a consonant either more or less sonorant. Consonant mutation processes thus have in common that they alter a consonant’s degree of sonority.

The Blackwell Companion to Phonology. Edited by Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume, and Keren Rice. © 2011 John Wiley & Sons, Ltd. Published 2011 by John Wiley & Sons, Ltd. DOI: 10.1002/9781444335262.wbctp0065

2

Janet Grijzenhout

Typically, mutations are “scalar.” In the languages of the world, we find consonant mutations where a consonant’s degree of stricture decreases (e.g. in Archaic Irish and Finnish an underlying geminate stop is realized as a singleton stop, while an underlying singleton stop is realized as a continuant sound) or increases. Another example of a scalar mutation is one in which a consonant’s degree of laryngeal specificity and/or nasality increases (e.g. in Old Irish, an aspirated voiceless stop is realized as an unaspirated one in the same context in which an unaspirated oral stop is realized as a prenasalized stop). This chapter first discusses possible consonant alternations in more detail (§2). As examples of languages that have relatively many types of consonant mutations (i.e. spirantization, gemination, nasalization, and/or prenasalization), we discuss Southern Paiute (§3) and Fula (§4). Balto-Finnic, Sami, and some Australian languages show scalar mutations (§5). §6 points out the merits and drawbacks of some theoretical accounts of consonant mutation that exist in phonological literature. §7 summarizes the discussion.

2

Consonant alternations within prosodic and morphological domains

Consonants are highly adaptable elements that may change their properties for a variety of reasons. This section will focus on some phonological and morphological environments that may trigger a change in one phonetic property of a consonant. We start with consonant alternations that are characterized by the fact that a phonological opposition is neutralized in a certain prosodic environment, viz. final devoicing and debuccalization. We will also briefly consider consonant alternations that occur at the left edge of a prosodic domain. Next, five consonant alternations that are not triggered merely by a prosodic environment (i.e. that are independent of the position within a prosodic domain) and that fall under the rubric of “consonant mutation” are introduced: (a) stopping (Soninke), (b) obstruent voicing (Burmese), (c) spirantization or fricativization (Djapu; the first stage of Grimm’s Law), (d) devoicing (the second stage of Grimm’s Law), and (e) deaspiration (the third stage of Grimm’s Law). Other types of consonant mutation that are frequently encountered in languages – both diachronically and synchronically – are gemination, degemination, and (pre)nasalization. The latter phenomena will be discussed in later sections. Many consonant alternations are characterized by the fact that a phonological opposition is neutralized in a certain prosodic environment. In a variety of unrelated languages, e.g. Catalan, Czech, Dutch, German, Ojibwa, Polish, Russian, Turkish, and Wolof, we find that the opposition between voiced and voiceless obstruents is neutralized in one particular environment only, viz. at the end of a prosodic domain (usually the syllable; see Brockhaus 1995 and chapter 69: final devoicing and final laryngeal neutralization). In other languages, place of articulation contrasts are neutralized at the end of a prosodic domain – usually the syllable coda – and this phenomenon is known as “debuccalization” (chapter 80: mergers and neutralization). In some generative frameworks, alternations at the end of a prosodic domain are described as processes where consonants “lose” their underlying marked specification for laryngeal features or place of

Consonant Mutation

3

articulation features.2 Conversely, in other positions of the word, consonants may become reinforced phonetically, e.g. by initial aspiration (which is seen as a form of enhancement, for instance, by Keyser and Stevens 2006: 42ff.). Consonant alternations that involve neutralization of an opposition in a particular prosodic context (e.g. the contrast between voiced and voiceless obstruents in Dutch and German is neutralized at the end of a prosodic domain), or that involve adding a feature to enhance an opposition in a particular prosodic domain (e.g. the contrast between voiced and voiceless stops is enhanced by adding aspiration for the voiceless plosives at the left word edge in English) do not fall under the category of “consonant mutations” as understood here. In the examples in (1), stem-final consonants optionally change their laryngeal properties to become more similar to the neighboring consonants within a phonological phrase (e.g. Berendsen 1983 for Dutch) and in (2) the consonant /n/ changes its place of articulation to the same place of articulation as the following obstruent (chapter 81: local assimilation). (1)

Regressive laryngeal assimilation in obstruents a.

b.

(2)

Dutch zeep + doos kas + boek zak + doek Hungarian zseb + kendo

→ → →

zee[bd]oos ka[zb]oek za[gd]oek

soap + box cash + book pocket + cloth

‘soap-box’ ‘cash-book’ ‘handkerchief’

→

zse[pk]endõ

pocket + cloth

‘handkerchief’

Regressive place assimilation of n- (marker of noun classes 9 and 10) in Kisukuma (data from Batibo 2000: 169) n + buli n + dama n + guzu

→ → →

mbuli ndama Iguzu

‘goat’ ‘calf’ ‘strength, energy’

Cases where consonants change a laryngeal or place feature under the influence of an adjacent consonant within a certain prosodic domain are most commonly referred to as “assimilations” rather than “mutations,” and these processes are relatively easy to describe in theoretical frameworks, e.g. in autosegmental theory as spreading of laryngeal or place features or in Optimality Theory (OT) as phenomena that are the result of ranking Agree[feature] and *[afeature] constraints higher than the corresponding Ident[feature] constraint (e.g. Lombardi 1996). The reason for assimilation is not to increase or decrease the degree of sonority of a segment, but rather to become “more similar” with respect to laryngeal or place properties to an immediately adjacent obstruent (in the Dutch or Hungarian examples) or a stop (in Kisukuma). 2

Note that the assumption about laryngeal neutralization being a case of “weakening” in the sense that the consonant “loses” an underlying feature is highly controversial, as can be seen, for example, by the German terminology Auslautverhärtung (“final hardening”) for syllable-final devoicing. Foley (1970), for instance, claims that a change from voiced to voiceless obstruent should be considered a case of strengthening (“fortition”) rather than weakening (“lenition”). For further discussion on the issue of what exactly constitutes strengthening or weakening, I refer to chapter 66: lenition.

Janet Grijzenhout

4

The set of data below illustrates a process whereby the oral stricture of initial consonants of nouns increases, i.e. voiceless fricatives or voiced continuants are realized as oral stops (3a), /l/ (3b), or nasal stops (3c) after nasal segments:3 (3)

Consonant alternations after nasals in Soninke a.

b. c.

fare si xore raqqe wulle jaaxe

‘donkey’ ‘horse’ ‘charcoal’ ‘mouth’ ‘dog’ ‘eye’

n n n n n n

pare Œi gore laqqe Iulle Jaaxe

‘my ‘my ‘my ‘my ‘my ‘my

donkey’ horse’ charcoal’ mouth’ dog’ eye’

In the examples in (3), all continuant consonants are affected and the trigger of the change is always the same, i.e. a nasal element. Intuitively, we may formulate the consonant alternation as a process that increases a consonant’s oral stricture after a nasal stop or nasalized vowel. Nevertheless, it is not easy to describe this process as one in which a single feature of the stem-initial consonant changes under the influence of a preceding nasal vowel or stop.4 Furthermore, some nouns (especially names) “resist” change and do not participate in the consonant alternation process. Moreover, under some morphological conditions, the nasal trigger is absent, but the change takes place nonetheless, for instance in some imperative forms, e.g. /pagu/ ‘fill up!’ and /Œi/ ‘shave!’ (cf. /si/ ‘to shave’). According to Kendall and Bird (1982), the language is thus in a transitional stage in which the phonologically triggered process of consonant change has developed into a process that is no longer purely phonological; there are exceptions as well as overapplications, i.e. cases where the consonantal change takes place without an overt phonological trigger. We next consider another type of consonant alternation where the phonological context determines the shape of a consonant: in Burmese, voiceless stops are voiced in intervocalic position (4a), (4b) or following a nasal (4c). (4)

Intersonorant stop gradation in Burmese (data from Campbell 1995: 98–102) a. b. c.

3

hwa + pa hwa + tD + lu kauI + kauI

→ → →

hwàba hwàdGlu kauIgauI

‘please go’ ‘the man who is going’ ‘to be good’

Soninke is a Mende language spoken in West Africa. All Soninke data presented in this chapter are from Kendall and Bird (1982). The nasals in (3a) assimilate in place of articulation to the following segment. Kendall and Bird (1982: 1, 3) state that the same consonant alternation occurs both after nasal consonants and after nasalized vowels, e.g. /r/ in /ri/ ‘to come’ is realized as /l/ after a nasalized vowel in /nhli/ ‘I came’. 4 In SPE (Chomsky and Halle 1968), we could formulate a rewrite rule [+continuant] → [−continuant] / [+nasal] __, with some additional rules to account for the fact that voiced continuants turn into a lateral or nasal stop. In autosegmental phonology, there is no single feature that nasal consonants and vowels have in common that could spread unto the following consonant. In OT, it is possible to formulate a constraint that bans continuant consonants after a nasal consonant or vowel (e.g. *[+nas][+cont, –voc]), but this constraint would be an ad hoc one and leaves open the possibility of other ad hoc constraints such as *[+strid][−cont] or *[+nas][lab], etc.

Consonant Mutation

5

In the Australian language Djapu, stops are only realized as such in lexical words after obstruents and nasal stops. Thus the dative suffix /-ku/ appears as such in [buurutj-ku] ‘mosquito’; when preceded by a vowel or a liquid, however, the labial and velar stops become a labial-velar glide (e.g. [Iajmil-wu] ‘Ngaymil clan’) and the dental stop becomes a palatal glide (Morphy 1983). Cases where consonants change in an intersonorant environment are most often referred to as “gradations” or “mutations.” When glides become fricatives, when underlying continuants becomes non-continuant segments (as in Soninke), or when singleton stops are geminated, the mutation is often referred to as “consonant hardening” or “fortition.” If the mutated consonant increases its degree of sonority, this type of mutation is often referred to as “consonant weakening” or “lenition.” The environment for gradation in Soninke can be characterized as “after a nasal”; gradation in Burmese is “in intervocalic position or between a nasal stop and a vowel,” whereas the environment for gradation in Djapu is “in intervocalic position or between a liquid and a vowel.” chapter 66: lenition provides more examples of intervocalic voicing and spirantization. The interesting problem that consonant mutations pose for phonological theory is that they change the degree of sonority of a segment (chapter 49: sonority) and that they are not easily accounted for by means of autosegmental processes such as spreading, inserting, or deleting a phonological feature or class of features within a natural phonological context (chapter 14: autosegments). Neither is it easy to account for them by means of wellformedness constraints that correctly predict surface forms in non-mutation contexts and the corresponding alternating forms in mutation contexts. Consider in this respect that a possible constraint that penalizes intersonorant voiceless stops could be *[+son][−son, –voice][+son] (i.e. no voiceless obstruent in between two sonorants). This constraint would be violated in a potential output [hwàpa]. If the markedness constraint in question is ranked higher than the faithfulness constraint Ident[voice] – which, presumably would be the case in Burmese – the output [hwàba] would win. Now consider the fact that the winning candidate in Burmese would be a form that is disallowed in Spanish (see e.g. Harris 1969). Whereas Spanish allows intervocalic voiceless stops – suggesting that *[+son][−son, −voice][+son] is low-ranked – it does not have output forms with intervocalic voiced stops (resembling the winning candidates for Burmese). In an optimality-theoretical framework, we could again propose a constraint, e.g. *[+son][−son, −cont, +voice][+son] (i.e. no voiced stop between two sonorants), which would have to be ranked higher than a faithfulness constraint, e.g. Ident[−son, −cont] to generate the correct output for Spanish. Apart from the fact that there is an ad hoc flavor to the OT accounts suggested immediately above, a further complicating factor is the fact that some languages exhibit both consonant alternations in their grammars. In Northern Corsican, for example, voiceless stops become voiced where voiced stops spirantize. Moreover, we find exactly the same consonant alternations in contexts that cannot be described in a straightforward way as being “inter-sonorant.” Rather, as will be shown in subsequent sections, the same consonant alternations as described here appear in different morphological contexts or are lexicalized in various unrelated languages. The cases presented so far all reflect synchronic processes. The interest in consonant mutations, however, first arose with respect to diachronic changes such

6

Janet Grijzenhout

as the first consonant shift in West Germanic (also known as Grimm’s Law) and early Celtic consonant mutations (Pedersen 1897; Thurneysen 1898).5 Grimm’s Law is often formulated as follows: the Indo-European stops /p t k kw/ spirantized and became fricatives, the unaspirated stops /b d g gw/ became voiceless aspirated stops, and so-called “breathy voiced” stops /bh dh g h gwh/ were replaced by voiced unaspirated stops (chapter 73: chain shifts). Iverson and Salmons’s (1995) account of the first shift in Grimm’s Law runs as follows. The voiceless aspirated stops became voiceless fricatives when aspiration was audible (i.e. only when the stop was released, so that stops following /s/, geminates and stop–stop clusters did not undergo the shift). The unaspirated stops – in which voicing was optional – had a “stronger and longer” closure phase, which made them unlikely candidates for fricativization. Thus, the outcome of the first shift in Grimm’s Law is a system in which we find fricatives (specified for [stiff vocal folds]), unaspirated stops (unspecified for laryngeal features), and voiced aspirated stops (specified for [slack vocal folds, spread glottis]). The second stage of Grimm’s Law is the process whereby the contrast between stops unmarked for laryngeal features (/b d g gw/) and stops marked as “slack with aspiration” (/bh dh g h gwh/) is increased by introducing [stiff vocal folds] (which is available already in the fricative series) for the unmarked stops. In the system where [stiff vocal folds] stops contrast with [slack vocal folds] ones, there is no need to maintain the aspiration contrast and hence the feature [spread glottis] for breathy voiced stops gradually loses its place in the stop system. The three subsequent mutations may thus be characterized as follows: (a) fricativization of voiceless stops (/p t k kw/ → /f h x xw/), (b) fortition or devoicing of lax stops (/b d g gw/ → /p t k kw/), and (c) deaspiration of aspirated lax stops (/bh dh g h gwh/ → /b d g gw/). This section has introduced five consonant alternations that affect the degree of sonority and that fall under the rubric of “mutations”: stopping, obstruent voicing, spirantization or fricativization, devoicing, and deaspiration. Other mutations that are frequently encountered in languages – both diachronically and synchronically – are gemination, degemination, and (pre)nasalization. In the following sections we will discuss informative examples of synchronic consonant mutations. We start with Southern Paiute, a language that exhibits spirantization, gemination, and nasalization.

3

Southern Paiute consonant mutations

Sapir (1930) describes the Shoshonean dialect Kaibab Paiute, as it was spoken in southwestern Utah and northwestern Arizona during the 1910s. In this dialect, consonants can appear in one form when in suffix-initial position after a consonant or in word-initial position, and in various alternating forms when – by the process of derivation or compounding – they are immediately preceded by a vowel. The particular process that a suffix undergoes is dependent on the lexical item it is attached to and is not phonologically predictable in any obvious way. 5

For a discussion on the history and present-day exponents of Celtic consonant mutations the reader is referred to chapter 117: celtic mutations.

Consonant Mutation (5)

7

Southern Paiute consonant mutations (Sapir 1930: 62)6 underlying spirantized geminated p ß p( t P t( k : k( kw :w kw( ts/Œ ts/Œ ts(/Œ( s/œ s(/œ( m Iw m( n n(

nasalized mp nt Ik Ikw nts/nΠm( n(

Under spirantization, oral stops change into voiced continuants, affricates do not change, and the labial nasal becomes a “back palatal” labialized nasal consonant (represented as /Iw/ by Sapir). Harms (1966) attempts to describe Paiute spirantization as a regular phonological rule ([+consonantal, –vocalic, –strident] → [+continuant, +voice] / [−consonantal, +vocalic, +voice] __), but such a rule only accounts for part of the alternation, and leaves the change from labial nasal to dorsal labialized nasal unexplained. Before we turn to another analysis of Southern Paiute consonant mutations, we first introduce some examples. To illustrate the effect of mutation, Sapir (1930: 63, 67) mentions the verbalizing suffix /-ka/ and the agentive suffix /-pi/, with their respective mutated initial consonants following adjectival (6a)–(6c) and nominal stems (6d)–(6e): (6)

Consonant mutations affecting morpheme-initial /k/ and /p/ in Southern Paiute a. b. c. d. e.

aIka k‚Œa paq nD taIa

+ + + + +

-ka -ka -ka -pi -pi

→ → → → →

aIka:a k‚Œak(a paqIka nDvi taIampi

‘to be red’ ‘to be gray’ ‘to be smooth’ ‘carrier’ ‘kicker’

spirantization gemination prenasalization spirantization prenasalization

Note that the stem /aIka/ ‘red’ triggers spirantization in a following suffix (6a), but is followed by a geminated consonant in compounds (7): (7)

Consonant mutations affecting stem-initial /k/ and /p/ in Southern Paiute compounds aIka aIka

+ +

kani pajq

→ →

aIkak(anê aIkap(ajq

red + house red + fish

‘red house’ ‘trout’

As a possible explanation for the difference in the choice of consonant mutation between derivation and compounding in some nouns, Sapir (1930: 70) suggests that “the tendency to use geminated consonants in composition is probably due to the greater phonetic similarity thus brought about between a simplex and its compound.” 6

The consonants and the consonant alternations mentioned by Sapir (1930) are represented here by means of the corresponding IPA symbols. According to Sapir, spirantized consonants following voiced vowels are voiced, and those following voiceless vowels are voiceless. I present the voiced allophones only. The glides /w/ and /j/ and the nasal /I/ do not occur in initial positions and are hence not affected by mutations. The glottal stop does not undergo mutations. The three blanks in (5) indicate that [s/œ] and [n] do not have a spirantized counterpart and that [s/œ] lacks a nasalized form.

8

Janet Grijzenhout

McLaughlin (1984: 70–71) notes that prefixes are followed by one fixed mutation, i.e. all prefixes affect the initial consonant of a following stem, and Southern Paiute distinguishes “spirantizing,” “geminating,” and “nasalizing” prefixes. In compounds, noun stems may trigger one of three mutations, but in most cases nouns in compounds trigger gemination only. In verbal compounds, the verb of the second member has its initial consonant geminated in the majority of cases. To McLaughlin, these facts suggest that prefixes show the strongest reflection of an earlier stage in the language, viz. one in which a prefix-final phoneme triggered a change in the following consonant. The phoneme in question is lost, but the mutation that it triggered is still in effect. Nouns also show this effect, but are in a transient stage; nominal stems are cautiously on their way to induce one particular mutation only, i.e. gemination. Verbs have developed even further and generally geminate a following consonant. To account for Southern Paiute consonant mutations, McLaughlin (1984) proposes that all prefixes, most noun stems, some adjectival stems, and a few verb stems end either in (a) a vowel, (b) an unspecified stop C, or (c) an unspecified nasal N.7 The last two are reflexes of final stops and nasals that used to be present in earlier stages of the language. If a stem is preceded by a prefix or another stem ending in a vowel, its first consonant will undergo a rule of spirantization. A suffix or stem following an unspecified stop or nasal, however, will undergo a place assimilation rule (e.g. /taIaN + pi/ → [taIampi]). Finally, stems following an unspecified stop will also be subject to a lengthening rule, so that the result will be a geminate consonant. Another rule may subsequently degeminate the stop in question when it follows an unstressed vowel. McLaughlin’s analysis of spirantization thus involves a regular spirantization rule, which turns stops into voiced spirants in inter-sonorant position. The analysis of gemination and nasalization, on the other hand, depends on “ghost” elements, i.e. a final stop or nasal that never surfaces, but has a geminating or nasalizing effect if another consonant follows in the next morpheme (i.e. a suffix in the case of derivation and a stem in the case of compounding). These processes are thus lexical in the sense that in the lexicon, the morphemes in question (prefixes, most noun stems, and some adjectival stems) end either in a vowel or in an unspecified stop or nasal. The mutation triggered by verb stems is becoming more and more morphologized; i.e. morphological leveling is producing ever larger numbers of geminating verb stems. Since oral stops become continuants under spirantization, we would perhaps expect the labial nasal stop to turn into a labial approximant [w] (chapter 28: the representation of fricatives), as is the case in Celtic (see chapter 117: celtic mutations). The question arises why this is not the case. With respect to the quirky character of spirantized /m/, McCarthy and Prince (1995: 349ff.) make the following suggestion. They first observe that the segments [w] and [Iw] are in complementary distribution: the former is found word initially and the latter occurs post-vocalically.8 This distribution follows from the following 7

For similar proposals to account for Celtic consonant mutation, see chapter 117: celtic mutations. The only exception is the context of reduplication, where “w” surfaces between vowels due to an identity constraint that says that a base and a reduplicated form cannot have different values for the feature [nasal]. The fact that [Iw] is not allowed to occur in word-initial positions (*[Iw) rules out a candidate like hypothetical *[Iwa-Iwaxipija]. In *[wa-Iwaxipija] the constraint Ident-BR[nasal] is violated and the candidate [wa-waxipija] is the winner even though it has a “w” after a vowel (i.e. the constraints mentioned in this footnote are ranked as follows: *[Iw, Ident-BR[nasal] >> *VwV). 8

Consonant Mutation

9

ranking of constraints: *VwV >> *Iw >> *w. Even though this ranking accounts for the fact that we are more likely to find [Iw] rather than [w] in intervocalic position, it is not immediately clear why /m/ alternates with [Iw] in the first place, i.e. why /m/ surfaces as a nasal with a “back palatal” place of articulation in a spirantization context. If spirantization means that oral or nasal stops are realized as continuants, and if there is an operation or constraint (*VwV) that prevents the nasal /m/ from being realized as the corresponding continuant [w], it is still not evident that /m/ undergoes a transformation whereby it turns into a labialized dorsal nasal stop. I will leave this issue for further research.

4

Fula consonant mutation

Fula (also known as Fulani, Fulbe, Fulfulde, Pular, Pulaar, and Peul) is spoken in West Africa; the majority of the speakers live in Nigeria (Campbell 1995: 178). In all dialects of Fula, initial consonants of verbal radicals and nominal, numeral and verbo-nominal stems can have different forms. In contrast to Southern Paiute – where stem-initial consonants change their form depending on the mutating requirements of the preceding prefix or stem – one cannot argue that in Fula a segment has a different surface form depending on a preceding vowel or a morpheme specified for a mutating feature. Arnott (1970) shows that in the Gombe dialect of Fula, stem-initial consonants surface as homorganic stops, continuants or prenasals, depending on the adjective or noun class. (8)

Fula consonant mutations (based on Arnott 1970: 42–43)9 stop spirant p f b w d r 10 œ s Á j k h g j/w11

prenasal p mb nd œ Jj k Ig

According to Arnott (1970), adjectives and nouns are marked as belonging to one of 25 possible classes. Class membership is indicated by a suffix12 and a particular manner of articulation of the initial consonant of the stem: 11 classes are marked by a stem-initial voiceless or voiced stop, six classes are marked by 9

The consonant alternations mentioned by Arnott are represented here by means of the corresponding IPA symbols. Note that the glottalized consonants /+ e —/, the nasals /m n J I/, and the coronal consonants /t c l/ are “invariable” and do not alternate. 10 Note that this sound is a fricative, but in the mutation system it functions as a stop. In most analyses of Fula consonant mutations that can be found in the literature, this sound is usually transcribed as the palatal stop /c/, but this does not tally with Arnott’s (1970) description of this sound. 11 The palatal glide is found before the front vowels /i e/ and the labial-velar one before the back vowels /u o a/. 12 Suffixes belong to certain grades, indicated as grades A, B, C, or D in Arnott (1970). In the phrase /gude daneeje/ ‘white cloths’, both the noun and the adjective belong to class 24 (i.e. one of the classes that are marked by an initial stop), but the noun /gude/ ‘cloths’ has a grade A suffix /-e/ and the adjective /daneeje/ ‘white’ has a grade B suffix /-je/.

10

Janet Grijzenhout

an initial spirant and eight are marked by an initial voiceless oral stop or a prenasalized stop. The marker of noun class 1 (singular nouns referring to persons), for instance, is a suffix ending in the vowel /o/ and a stem-initial stop. The marker of noun class 2 (plural nouns referring to persons) is a spirant-initial stem consonant and the suffix /-+e/. The marker of class 7 is a suffix ending in the vowel /a/ and a stem-initial oral voiceless stop or prenasalized stop for consonants that alternate with a voiced stop in class 1: (9)

Stem-initial consonant alternations in Fula nouns (data from Arnott 1970: 98–99) class 1 pull-o beer-o dim-o œook-o Áuul-eo kor-eo gim-eo

class 2 ful-+e weer-+e rim-+e sook-+e juul-+e hor-+e jim-+e

class 7 pul-a mbeer-a ndim-a œook-a Jjuul-Iga kor-ga Igim-Iga

‘Fula’ ‘host’ ‘free man’ ‘poor man’ ‘Moslim’ ‘female slave’ ‘person’

In an autosegmental framework that assumes radical underspecification (chapter 7: feature specification and underspecification), Wiswall (1989) proposes that Fula voiceless continuants are underlyingly specified for [−voice], but not for manner of articulation. The voiceless stops are specified for [−voice] as well, and all underlying oral stops have a specification for [−continuant]. Underlying non-alternating nasal stops have a specification for [+sonorant] and [+nasal]. Wiswall furthermore suggests that noun class markers in Fula have floating features that associate to the initial consonant of the stem (chapter 82: featural affixes); the stop classes have a floating [−continuant] feature (which is problematical for the sound /œ/), the prenasal classes have a floating [+nasal] feature (which presumably cannot dock onto sounds underlyingly specified as being [−voice]), but the spirant classes do not have a floating feature. After association of the floating features, redundancy rules (e.g. [−continuant] → [−sonorant]), and default rules (Ø → [+continuant]) fill in the unspecified feature values. Grijzenhout (1991) points out some problems relating to the rule ordering (chapter 74: rule ordering) that Wiswall has to assume, especially related to the account of prenasals, which would involve a late counterintuitive redundancy rule [+nasal] → [−sonorant]. For an alternative account of the data involving the theory of “charm and government,” the reader is referred to work by Paradis (e.g. Paradis 1987a, 1987b, 1992). Elzinga (1996) provides an OT account of Fula consonant mutation, which makes use of alignment and parsing constraints on the mutating features [continuant] and [nasal]. He furthermore employs morpheme constraints, i.e. constraints that indicate the type of stem (“invariable,” “partially variable,” or “fully variable”). The morpheme constraints are placed in between the feature alignment and parsing constraints, so that invariable stems and suffixes are ranked higher in the hierarchy than the mutation constraints, partially variable stems and suffixes are ranked below some mutation constraints and above others, and fully variable stems and suffixes are ranked lower than the mutation constraints. We now turn to cases of scalar mutations, i.e. mutations that cause a change from one underlying consonant to another consonant, which also mutates into a different one.

Consonant Mutation

5

11

Balto-Finnic and Sami consonant gradation

Most languages belonging to the Balto-Finnic group (i.e. Finnish, Estonian, Votic, Ingrian, and Karelian; excluding Livonian and Veps), as well as northern and eastern dialects of the closely related Sami group, exhibit consonant gradations. It is striking that a similar phenomenon, by which long stops were shortened and short stops were turned into spirants, occurred in the history of Iwaidjan languages in Australia (e.g. Evans 1998). The phonological condition that generated consonant gradation in languages belonging to the Balto-Finnic group was originally the following: after a vowel or sonorant consonant in a stressed syllable, stops in the so-called “strong grade” that appeared in the onset of a syllable that was closed by certain inflectional or derivational endings were mutated such that they appeared in the corresponding “weak grade.” Under this condition, underlying long (or “geminate”) stops were reduced to short (or “singleton”) stops and underlying short stops were replaced by voiced and fricated consonantal variants. Thus, in the Balto-Finnic languages, long /p(/ alternated with short /p/ while short /p/ alternated with /b/ or /v/, and long /t(/ alternated with short /t/ while short /t/ alternated with /d/ or /Ï/. Similarly, in the history of Iwaidjan languages, long intervocalic stops were shortened and short intervocalic stops became approximants or liquids (e.g. /p(/ → /p/ → /w/ and /c(/ → /c/ → /j/). Different varieties of Sami and Estonian now have a three-way opposition between the strongest grade or quantity, the strong grade, and the weak grade, but the phonemes involved in the alternations are somewhat different for each language or language variety. We most often find that in contexts where gradation applies in these languages, (a) “overlong stops” (usually written as , corresponding to the sounds /p(( t(( k((/) are realized as “long stops,” (b) “long stops” (orthographic